@cliwant/mcp-sam-gov 0.2.1 → 1.0.0

package/dist/server.js

@@ -1,9 +1,9 @@
 #!/usr/bin/env node
 /**
  * @cliwant/mcp-sam-gov — Model Context Protocol server for SAM.gov
- * + USAspending + Federal Register + eCFR + Grants.gov.
+ * + USAspending + Federal Register + eCFR + Grants.gov + GAO + wage/pricing.
  *
- * 34 keyless tools wrapping every public federal-contracting data
+ * 52 keyless tools wrapping every public federal-contracting data
  * source that doesn't require an API key. Compatible with:
  *   - Claude Desktop  (claude_desktop_config.json)
  *   - Claude Code     (.mcp.json or `claude mcp add`)
@@ -20,13 +20,48 @@ import { Server } from "@modelcontextprotocol/sdk/server/index.js";
 import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
 import { CallToolRequestSchema, ListToolsRequestSchema, } from "@modelcontextprotocol/sdk/types.js";
 import { z } from "zod";
-import { SamGovClient } from "./sam-gov/index.js";
+import { SamGovClient, daysUntilResponse, applyResponseDeadlineWindow, } from "./sam-gov/index.js";
 import * as usas from "./usaspending.js";
 import * as fedreg from "./federal-register.js";
 import * as ecfr from "./ecfr.js";
+import * as far from "./far.js";
 import * as grants from "./grants.js";
+import * as pricing from "./pricing.js";
+import * as integrity from "./integrity.js";
+import * as gao from "./gao.js";
+import * as gsaCsv from "./gsa-csv.js";
+import * as sba from "./sba.js";
+import * as treasury from "./treasury.js";
+import * as edgar from "./edgar.js";
+import * as socrata from "./socrata.js";
+import * as ckan from "./ckan.js";
+import * as echo from "./echo.js";
+import * as datagov from "./datagov.js";
+import * as datagovCatalog from "./datagov-catalog.js";
+import * as govinfo from "./govinfo.js";
+import * as fpds from "./fpds.js";
+import * as nih from "./nih.js";
+import * as nsf from "./nsf.js";
+import * as clinicaltrials from "./clinicaltrials.js";
+import * as census from "./census.js";
+import * as fema from "./fema.js";
+import * as fdic from "./fdic.js";
+import * as bls from "./bls.js";
+import * as ofac from "./ofac.js";
+import * as nvd from "./nvd.js";
+import * as nppes from "./nppes.js";
+import * as cms from "./cms.js";
+import * as fac from "./fac.js";
+import * as usitc from "./usitc.js";
+import { fetchAttachmentText } from "./attachments.js";
+import { toToolError, ToolErrorCarrier, errorFromResponse } from "./errors.js";
+import { buildMeta, isMetaBundle, withMeta, } from "./meta.js";
+import { pathToFileURL, fileURLToPath } from "node:url";
+import { realpathSync } from "node:fs";
 const SERVER_NAME = "mcp-sam-gov";
-const SERVER_VERSION = "0.2.1";
+// Kept in lockstep with package.json / manifest.json / server.json.
+// Keep in sync with package.json "version" (asserted at release; see CHANGELOG).
+const SERVER_VERSION = "1.0.0";
 // ─── Tool input schemas (Zod) ────────────────────────────────────
 const SamSearchInput = z.object({
     query: z.string().optional().describe("Free-text title query"),
@@ -42,6 +77,39 @@ const SamSearchInput = z.object({
     setAside: z.array(z.string()).optional().describe("Set-aside codes: SBA, 8A, HZS, SDVOSBC, WOSB, EDWOSB, VSA, VSS"),
     limit: z.number().min(1).max(50).optional(),
 });
+// Pre-solicitation shaping radar (doc 06 §3.1). Surfaces Sources Sought /
+// Presolicitation / Special Notices BEFORE the RFP exists — the free, real-time
+// analogue of paid agency-forecast feeds. Keyless via the notice_type facet.
+const SamSearchShapingInput = z.object({
+    query: z.string().optional().describe("Free-text title query"),
+    ncode: z.string().optional().describe("NAICS code, e.g. '541512'"),
+    organizationName: z
+        .string()
+        .optional()
+        .describe("Issuing agency canonical name (e.g. 'Department of Veterans Affairs'). NOTE: the keyless endpoint has NO organization-name filter — it is sent best-effort and flagged in _meta.filtersDropped; filter client-side on the returned `agency`."),
+    state: z
+        .string()
+        .optional()
+        .describe("Place-of-performance state, 2-letter, e.g. 'MD'"),
+    setAside: z.array(z.string()).optional().describe("Set-aside codes: SBA, 8A, HZS, SDVOSBC, WOSB, EDWOSB, VSA, VSS"),
+    noticeType: z
+        .array(z.enum(["r", "p", "s", "k", "i", "u"]))
+        .optional()
+        .describe("Pre-solicitation notice-type codes to include. r=Sources Sought, p=Presolicitation, s=Special Notice (the DEFAULT shaping window = ['r','p','s']); k=Combined Synopsis/Solicitation, i=Intent to Bundle, u=Justification (J&A) are opt-in adjacency/incumbent tells. Ranked r/p over s via noticeTypeCode."),
+    responseDeadlineFrom: z
+        .string()
+        .optional()
+        .describe("ISO date lower bound for responseDeadline. APPLIED CLIENT-SIDE over the fetched page (the keyless feed ignores rdlfrom/rdlto) — disclosed in _meta.filtersDropped. A notice with no deadline is excluded from a windowed query."),
+    responseDeadlineTo: z
+        .string()
+        .optional()
+        .describe("ISO date upper bound for responseDeadline. APPLIED CLIENT-SIDE over the fetched page (see responseDeadlineFrom)."),
+    activeOnly: z
+        .boolean()
+        .optional()
+        .describe("Only currently-active notices (default true)."),
+    limit: z.number().min(1).max(50).optional().describe("Page size (default 25, max 50)."),
+});
 const SamGetOpportunityInput = z.object({
     noticeId: z.string().describe("32-char hex notice id"),
 });
@@ -53,6 +121,32 @@ const SamAttachmentUrlInput = z.object({
         .string()
         .describe("Resource id from sam_get_opportunity → resourceLinks (URL-tail hex)"),
 });
+const SamFetchAttachmentTextInput = z.object({
+    url: z
+        .string()
+        .url()
+        .refine((u) => {
+        try {
+            const p = new URL(u);
+            const h = p.hostname.toLowerCase();
+            return (p.protocol === "https:" &&
+                (h === "sam.gov" || h === "api.sam.gov" || h.endsWith(".sam.gov")));
+        }
+        catch {
+            return false;
+        }
+    }, {
+        message: "url must be an https:// SAM attachment download URL on sam.gov / api.sam.gov (from sam_get_opportunity's attachments[].url). Arbitrary hosts are refused (SSRF hygiene).",
+    })
+        .describe("SAM attachment download URL from sam_get_opportunity → attachments[].url / resourceLinks (https://sam.gov/api/prod/opps/v3/opportunities/resources/files/{id}/download). Must be a sam.gov / api.sam.gov host."),
+    maxChars: z
+        .number()
+        .int()
+        .min(1000)
+        .max(500_000)
+        .optional()
+        .describe("Cap on returned text characters (default 200000, max 500000). Truncation is disclosed in _meta (truncated:true)."),
+});
 // USAspending — awards & recipients
 const UsasFiltersBase = z.object({
     agency: z.string().optional().describe("Canonical agency name"),
@@ -89,16 +183,93 @@ const UsasSubawardsInput = z.object({
 const UsasExpiringInput = z.object({
     agency: z.string().optional(),
     naics: z.string().optional(),
-    fiscalYear: z.number().int().min(2007).optional(),
+    // M2 (W3-1 honesty): `fiscalYear` removed. The recompete radar windows on the
+    // current PoP end date around TODAY, not an obligation FY, so it was
+    // inapplicable — advertised here, then validated and silently discarded by
+    // searchExpiringContracts (never forwarded) with empty filtersDropped. Dropping
+    // it from the schema stops the validated-then-discarded arg at the door.
     monthsUntilExpiry: z.number().min(1).max(36).optional(),
     minAwardValue: z.number().optional(),
     limit: z.number().min(1).max(20).optional(),
 });
+const UsasRecompetesInput = z.object({
+    agency: z
+        .string()
+        .optional()
+        .describe("Canonical awarding toptier agency name (use usas_lookup_agency)"),
+    naics: z.string().optional().describe("6-digit NAICS code, e.g. '541512'"),
+    pscCodes: z
+        .array(z.string())
+        .optional()
+        .describe("Product/Service Codes to filter on, e.g. ['DA01','R425']"),
+    setAside: z
+        .enum(["SBA", "8A", "HZS", "SDVOSBC", "WOSB", "EDWOSB", "VSA", "VSS"])
+        .optional()
+        .describe("USAspending set_aside_type_code (honored server-side)"),
+    windowStartDays: z
+        .number()
+        .int()
+        .optional()
+        .describe("Lower edge of the recompete window in days from today (default -90 = include contracts that ended up to 90 days ago)."),
+    windowEndDays: z
+        .number()
+        .int()
+        .min(1)
+        .optional()
+        .describe("Upper edge of the window in days from today (default 548 ≈ 18 months)."),
+    minAwardValue: z
+        .number()
+        .min(0)
+        .optional()
+        .describe("Minimum Award Amount ($) to include (default 0)."),
+    includePotentialEnd: z
+        .boolean()
+        .optional()
+        .describe("Also return the potential (option-inclusive) PoP end date + extendableDays (default false)."),
+    actionDateLookbackYears: z
+        .number()
+        .int()
+        .min(1)
+        .max(10)
+        .optional()
+        .describe("action_date lower bound in years (default 3). Contracts with no recorded action in this span are excluded — this bound makes the End-Date sort reach the window."),
+    page: z.number().int().min(1).optional().describe("1-based page (default 1)."),
+    pageSize: z
+        .number()
+        .int()
+        .min(1)
+        .max(100)
+        .optional()
+        .describe("Rows per page (default 25, max 100)."),
+    scanBudgetPages: z
+        .number()
+        .int()
+        .min(1)
+        .max(20)
+        .optional()
+        .describe("Max 100-row pages to scan before giving up (default 8). If exhausted before the window ends, results are a lower bound and totalAvailable is null."),
+});
 const UsasAwardDetailInput = z.object({
     generatedInternalId: z
         .string()
         .describe("From spending_by_award results — e.g. CONT_AWD_*"),
 });
+const UsasAnalyzeIncumbentInput = z.object({
+    generatedInternalId: z
+        .string()
+        .describe("The ONE award to analyze — generatedInternalId from usas_search_individual_awards / usas_search_awards_by_recipient / usas_search_recompetes (e.g. CONT_AWD_*)."),
+    includeOtherAwards: z
+        .boolean()
+        .optional()
+        .describe("Also return the incumbent's other awards in the same agency×NAICS via one bounded recipient search (default true)."),
+    otherAwardsLimit: z
+        .number()
+        .int()
+        .min(1)
+        .max(50)
+        .optional()
+        .describe("Cap on incumbentOtherAwards (default 15, max 50)."),
+});
 const UsasSpendingOverTimeInput = z.object({
     group: z.enum(["fiscal_year", "quarter", "month"]).optional(),
     agency: z.string().optional(),
@@ -176,242 +347,3708 @@ const FedRegSearchInput = z.object({
     type: z
         .enum(["RULE", "PRORULE", "NOTICE", "PRESDOCU"])
         .optional()
-        .describe("Document type"),
-    publicationDateFrom: z.string().optional().describe("YYYY-MM-DD"),
-    publicationDateTo: z.string().optional().describe("YYYY-MM-DD"),
-    effectiveDateFrom: z.string().optional().describe("YYYY-MM-DD"),
-    perPage: z.number().min(1).max(100).optional(),
+        .describe("Document type"),
+    publicationDateFrom: z.string().optional().describe("YYYY-MM-DD"),
+    publicationDateTo: z.string().optional().describe("YYYY-MM-DD"),
+    effectiveDateFrom: z.string().optional().describe("YYYY-MM-DD"),
+    perPage: z.number().min(1).max(100).optional(),
+});
+const FedRegGetDocInput = z.object({
+    documentNumber: z
+        .string()
+        .describe("Federal Register document number, e.g. '2026-08333'"),
+});
+const FedRegListAgenciesInput = z.object({
+    perPage: z.number().min(1).max(500).optional(),
+});
+const FedRegPublicInspectionInput = z.object({
+    mode: z
+        .enum(["current", "date", "search"])
+        .optional()
+        .describe("Retrieval surface (default current): 'current' = all documents on public inspection now; 'date' = a specific inspection day (requires `date`); 'search' = server-side full-text over the on-inspection set (via `term`)."),
+    date: z
+        .string()
+        .optional()
+        .describe("YYYY-MM-DD; REQUIRED iff mode='date'. Rides conditions[available_on] as a query param (never a path segment). Validated (real calendar date, 1994..currentYear+1) before any fetch."),
+    term: z
+        .string()
+        .optional()
+        .describe("Full-text query; VALID only in mode='search'. Rides conditions[term] (server-side)."),
+    type: z
+        .enum(["RULE", "PRORULE", "NOTICE", "PRESDOCU"])
+        .optional()
+        .describe("Client-side document-type filter (applied in all modes)."),
+    agency: z
+        .string()
+        .regex(/^[a-z0-9-]+$/)
+        .optional()
+        .describe("Client-side agency-slug filter; matches ANY of a doc's agencies[].slug. Resolve slugs via fed_register_list_agencies."),
+    specialOnly: z
+        .boolean()
+        .optional()
+        .describe("Client-side filter keeping only filing_type='special' (off-cycle/emergency — a stronger, sooner signal)."),
+    limit: z.number().min(1).max(200).optional().describe("Page size (default 20)."),
+    offset: z.number().min(0).optional().describe("Page offset (default 0)."),
+});
+// eCFR
+const EcfrSearchInput = z.object({
+    query: z.string(),
+    titleNumber: z
+        .number()
+        .optional()
+        .describe("CFR title (1-50). e.g. 48 = FAR (Federal Acquisition Regulation), 2 = Federal financial assistance."),
+    perPage: z.number().min(1).max(20).optional(),
+});
+const EcfrListTitlesInput = z.object({});
+// FAR / DFARS clause lookup (eCFR versioner full endpoint)
+const FarClauseLookupInput = z.object({
+    clauseNumber: z
+        .string()
+        .regex(
+    // Accept an optional FAR/DFARS prefix + the NN.NNN-N / NNN.NNN-NNNN core.
+    /^\s*(?:d?far[s]?\b[\s.:#-]*)?\d{1,3}\.\d{3,4}-\d{1,4}\s*$/i, "clauseNumber must be a FAR/DFARS clause like '52.212-4', '252.204-7012', or '52.204-25' (an optional 'FAR '/'DFARS ' prefix is allowed).")
+        .describe("FAR or DFARS clause/provision number, e.g. '52.212-4', '252.204-7012', '52.204-25'. An optional 'FAR '/'DFARS ' prefix is stripped."),
+    includePrescription: z
+        .boolean()
+        .optional()
+        .describe("Also fetch the prescribing section parsed from the clause's 'As prescribed in …' opener (the rule for WHEN the clause applies). Default true."),
+    asOfDate: z
+        .string()
+        .regex(/^\d{4}-\d{2}-\d{2}$/, "asOfDate must be YYYY-MM-DD.")
+        .optional()
+        .describe("Point-in-time codification date (YYYY-MM-DD). Defaults to Title 48's current up_to_date_as_of."),
+});
+// FAR compliance matrix (composes far_clause_lookup over a cited-clause list)
+const FarComplianceMatrixInput = z.object({
+    clauses: z
+        .array(z
+        .string()
+        .regex(
+    // Same clause grammar as FarClauseLookupInput (optional FAR/DFARS prefix).
+    /^\s*(?:d?far[s]?\b[\s.:#-]*)?\d{1,3}\.\d{3,4}-\d{1,4}\s*$/i, "each clause must be a FAR/DFARS clause like '52.212-4', '252.204-7012', or '52.204-25' (an optional 'FAR '/'DFARS ' prefix is allowed)."))
+        .min(1)
+        .max(25)
+        .describe("The FAR/DFARS clause numbers a solicitation cites (e.g. from its 52.252-2 'Clauses Incorporated by Reference' list), 1–25. Deduped case-insensitively. e.g. ['52.212-4','52.204-25','252.204-7012']."),
+    asOfDate: z
+        .string()
+        .regex(/^\d{4}-\d{2}-\d{2}$/, "asOfDate must be YYYY-MM-DD.")
+        .optional()
+        .describe("Point-in-time codification date (YYYY-MM-DD) — typically the solicitation issue date. Defaults to Title 48's current up_to_date_as_of."),
+    includePrescription: z
+        .boolean()
+        .optional()
+        .describe("Also fetch each clause's prescribing section (the 'As prescribed in …' rule for WHEN it applies). Default true."),
+    flagGates: z
+        .boolean()
+        .optional()
+        .describe("Tag resolved rows that are pass/fail award-eligibility gates (Section 889, CMMC, limitations on subcontracting) with a gate label; others get gate:null. Default true. false ⇒ all gate:null."),
+});
+// FAR/DFARS-scoped search (composes ecfr_search, filtered to FAR/DFARS + deduped)
+const FarSearchInput = z.object({
+    query: z
+        .string()
+        .min(1, "query must be a non-empty search string.")
+        .describe("What to search FAR/DFARS text for, e.g. 'limitations on subcontracting', 'covered defense information', 'commercial item'."),
+    scope: z
+        .enum(["far", "dfars", "both"])
+        .optional()
+        .describe("Which corpus to search: 'far' (Title 48 chapter 1, the default), 'dfars' (chapter 2), or 'both'. Excludes GSAM/agency supplements."),
+    dedupeVersions: z
+        .boolean()
+        .optional()
+        .describe("Collapse each section's historical versions to the current (in-force) one. Default true. false ⇒ return all raw rows incl. historical."),
+    partsOnly: z
+        .array(z.number().int())
+        .optional()
+        .describe("Restrict results to these FAR/DFARS parts, e.g. [52] for clause text only, [12] for commercial-item policy."),
+    perPage: z
+        .number()
+        .int()
+        .min(1)
+        .max(20)
+        .optional()
+        .describe("Number of DISTINCT sections to return (1–20). Default 5."),
+});
+// SBA size standards
+const SbaSizeStandardInput = z.object({
+    naics: z
+        .string()
+        .regex(/^\d{6}$/, "naics must be a 6-digit NAICS code (e.g. '541512').")
+        .describe("6-digit NAICS code to look up the SBA small-business size standard for (e.g. '541512')."),
+});
+// Grants.gov
+const GrantsSearchInput = z.object({
+    keyword: z.string().optional(),
+    cfda: z.string().optional().describe("CFDA program number, e.g. '10.500'"),
+    agency: z
+        .string()
+        .optional()
+        .describe("Grants.gov agency code, e.g. 'DHS-FEMA'"),
+    oppNum: z.string().optional().describe("Specific opportunity number"),
+    oppStatuses: z
+        .array(z.enum(["forecasted", "posted", "closed", "archived"]))
+        .optional()
+        .describe("Defaults to forecasted+posted"),
+    rows: z.number().min(1).max(50).optional(),
+});
+const GrantsGetInput = z.object({
+    opportunityId: z.string().describe("Grants.gov opportunity id (numeric string)"),
+});
+// SAM.gov organization lookup (federal hierarchy)
+const SamLookupOrgInput = z.object({
+    organizationId: z.string().describe("SAM.gov federal-organization id (numeric)"),
+});
+// Pricing tier — wage determinations + GSA CALC labor-rate benchmarks
+const WageSearchInput = z.object({
+    coverage: z
+        .enum(["sca", "dba"])
+        .describe("Which wage-determination law: 'sca' (Service Contract Act — services) or 'dba' (Davis-Bacon Act — construction). 'dba' is normalized to the API's 'dbra' index."),
+    state: z
+        .string()
+        .optional()
+        .describe("2-letter USPS state code (e.g. 'VA'), applied SERVER-SIDE. A full name is applied client-side instead."),
+    county: z
+        .string()
+        .optional()
+        .describe("County name (substring match), applied CLIENT-SIDE over the fetched page only (the API has no county filter)."),
+    query: z
+        .string()
+        .optional()
+        .describe("Matches the WD NUMBER/TITLE only — NOT occupation/job title (q=guard returns 0)."),
+    activeOnly: z.boolean().optional().describe("Only currently-active WDs (default true)."),
+    standardOnly: z
+        .boolean()
+        .optional()
+        .describe("Only standard (non-non-standard) WDs (default true)."),
+    limit: z.number().min(1).max(50).optional().describe("Page size (default 20, max 50)."),
+    page: z.number().min(0).optional().describe("0-based page index (default 0)."),
+});
+const WageRatesInput = z.object({
+    reference: z
+        .string()
+        .describe("fullReferenceNumber of the wage determination (e.g. '2015-4093' for SCA, 'IA20260028' for DBA) from sam_search_wage_determinations."),
+    revision: z
+        .number()
+        .optional()
+        .describe("Revision number. Omit to resolve the latest ACTIVE revision via /history."),
+    coverage: z
+        .enum(["sca", "dba"])
+        .optional()
+        .describe("Optional hint (sca|dba) to disambiguate the parser; inferred otherwise."),
+    format: z
+        .enum(["parsed", "raw", "both"])
+        .optional()
+        .describe("'parsed' (structured rates, default), 'raw' (the full document text), or 'both'. Use 'raw'/'both' when parseConfidence is low."),
+});
+const BenchmarkLaborInput = z.object({
+    laborCategory: z
+        .string()
+        .describe("Labor category to benchmark (e.g. 'Program Manager', 'Software Engineer'). Matched exactly against CALC's labor_category."),
+    businessSize: z
+        .enum(["S", "O"])
+        .optional()
+        .describe("Business size filter: 'S' (small) or 'O' (other-than-small)."),
+    educationLevel: z
+        .string()
+        .optional()
+        .describe("Education filter — use CALC's SHORT CODES (e.g. 'HS','AA','BA','MA','PHD'); the displayed education_level field may show full words."),
+    minYearsExperience: z
+        .number()
+        .optional()
+        .describe("Minimum years of experience filter."),
+    experienceRange: z
+        .string()
+        .optional()
+        .describe("Experience range as 'min,max' (e.g. '5,10')."),
+    sin: z.string().optional().describe("Schedule SIN filter (e.g. '54151S')."),
+    priceRange: z
+        .string()
+        .optional()
+        .describe("Ceiling-price range as 'min,max' (e.g. '50,150')."),
+    maxSamplePages: z
+        .number()
+        .min(1)
+        .max(10)
+        .optional()
+        .describe("How many 20-row pages to sample for the distribution (default 3, max 10)."),
+});
+// Integrity / teaming
+const CheckExclusionsInput = z.object({
+    query: z
+        .string()
+        .optional()
+        .describe("Firm or individual name to screen (drives the server-side exclusions text search). Provide at least one of query/uei/cage."),
+    uei: z
+        .string()
+        .optional()
+        .describe("SAM UEI to match. Used as the text query when it is the sole selector; post-filtered against results when combined with a name query."),
+    cage: z
+        .string()
+        .optional()
+        .describe("CAGE code to match (post-filtered against results, or used as the text query when sole)."),
+    activeOnly: z
+        .boolean()
+        .optional()
+        .describe("Only currently-active exclusions (default true). false includes terminated exclusions."),
+    classification: z
+        .enum(["Firm", "Individual", "Special Entity Designation", "any"])
+        .optional()
+        .describe("Filter by excluded-party classification (default 'any')."),
+    page: z.number().min(0).optional().describe("0-based page index (default 0)."),
+    size: z.number().min(1).max(100).optional().describe("Page size (default 25, max 100)."),
+});
+const IntegrityLookupInput = z.object({
+    uei: z
+        .string()
+        .optional()
+        .describe("SAM UEI of the entity to screen (PREFERRED — most precise). Provide at least one of uei/cage/name."),
+    cage: z
+        .string()
+        .optional()
+        .describe("CAGE code of the entity to screen."),
+    name: z
+        .string()
+        .optional()
+        .describe("Legal entity name to screen (drives the keyless exclusions text search; normalized-name gated). Provide at least one of uei/cage/name."),
+});
+const TeamingPartnersInput = z.object({
+    // ENUM-VALIDATED: a bogus recipient_type_names value is SILENTLY accepted by
+    // USAspending (HTTP 200, 0 results), so this enum is the guardrail — only the
+    // spellings LIVE-VERIFIED to narrow a populated NAICS are accepted.
+    cert: z
+        .enum([
+        "small_business",
+        "8a_program_participant",
+        "woman_owned_business",
+        "women_owned_small_business",
+        "economically_disadvantaged_women_owned_small_business",
+        "service_disabled_veteran_owned_business",
+        "veteran_owned_business",
+        "historically_underutilized_business_firm",
+    ])
+        .describe("Socioeconomic certification (award-derived, NOT the SBA registry of record). One of: small_business, 8a_program_participant, woman_owned_business, women_owned_small_business, economically_disadvantaged_women_owned_small_business, service_disabled_veteran_owned_business, veteran_owned_business, historically_underutilized_business_firm (HUBZone)."),
+    naics: z.string().optional().describe("NAICS code to scope the search (e.g. '541512')."),
+    agency: z
+        .string()
+        .optional()
+        .describe("Awarding agency canonical toptier name (e.g. 'Department of Veterans Affairs'). Use usas_lookup_agency to resolve abbreviations."),
+    subagency: z
+        .string()
+        .optional()
+        .describe("Awarding sub-agency name. Requires `agency` to also be set (a subagency alone is dropped)."),
+    lookbackYears: z
+        .number()
+        .min(1)
+        .max(20)
+        .optional()
+        .describe("Action-date lookback window in years (default 3)."),
+    excludeDebarred: z
+        .boolean()
+        .optional()
+        .describe("Screen the top-ranked candidates via sam_check_exclusions and drop active exclusions (default true; bounded + disclosed in _meta)."),
+    minAwards: z
+        .number()
+        .min(1)
+        .optional()
+        .describe("Minimum scanned award count for a firm to be listed (default 1)."),
+    limit: z.number().min(1).max(50).optional().describe("Candidates per page (default 25, max 50)."),
+    page: z.number().min(1).optional().describe("1-based page index (default 1)."),
+    screenCap: z
+        .number()
+        .min(1)
+        .max(25)
+        .optional()
+        .describe("Max candidates to exclusion-screen per page (default 10, max 25)."),
+    scanPages: z
+        .number()
+        .min(1)
+        .max(10)
+        .optional()
+        .describe("Award-value-sorted pages (100 rows each) to scan before aggregating by recipient (default 4, max 10)."),
+});
+// OFAC denied-party sanctions screening (keyless bulk SDN + Consolidated lists)
+const OfacScreenInput = z.object({
+    name: z
+        .string()
+        .describe("REQUIRED. The entity / individual / vessel / aircraft name to screen against OFAC's published SDN + Consolidated lists. Trimmed; empty is rejected (invalid_input) — never a no-op empty screen."),
+    type: z
+        .enum(["individual", "entity", "vessel", "aircraft"])
+        .optional()
+        .describe("Optional post-filter on the matched party's OFAC type. A blank OFAC type is inferred as 'entity' (disclosed). Omit to screen all types. Only trims the returned matches — it never turns a real name hit into no_name_match."),
+    program: z
+        .string()
+        .optional()
+        .describe("Optional case-insensitive filter to one sanctions PROGRAM code (e.g. 'CUBA', 'IRAN', 'SDGT'). Applied LOCALLY to matched rows (never on the wire); only trims returned matches (a hit under another program still yields potential_matches)."),
+    list: z
+        .enum(["sdn", "consolidated", "all"])
+        .optional()
+        .describe("Which OFAC list(s) to screen: 'sdn' (SDN + its AKAs), 'consolidated' (non-SDN programs + AKAs), or 'all' (default — the correct default for a real screen). Every list required for the scope loads-or-throws (a partial set is never screened)."),
+    minMatchQuality: z
+        .enum(["exact", "strong", "weak"])
+        .optional()
+        .describe("Floor of match quality to RETURN (default 'weak'). This ONLY trims the returned matches[]; existence is computed at the lowest quality FIRST, so result is 'potential_matches' whenever ANY match exists regardless of this value (suppressed matches are disclosed)."),
+    limit: z
+        .number()
+        .min(1)
+        .max(200)
+        .optional()
+        .describe("Max matches returned (default 50, max 200). Over-limit truncation is disclosed, never silent."),
+});
+// ━━━ NVD + CISA KEV — the IT/cyber-compliance lane (2) ━━━ ADR-0035
+// cve_lookup: NVD CVE detail/search JOINED with CISA KEV status.
+const CveLookupInput = z.object({
+    cveId: z
+        .string()
+        .optional()
+        .describe("Exact CVE identifier CVE-YYYY-NNNN (^CVE-\\d{4}-\\d+$, validated client-side). Exact-lookup mode; a malformed cveId is rejected (invalid_input) — a malformed cveId 404s upstream. At least one of cveId/keyword/cpeName/cvssV3Severity/a date range is REQUIRED."),
+    keyword: z
+        .string()
+        .optional()
+        .describe("Free-text keyword search (NVD keywordSearch) over CVE descriptions (e.g. 'log4j', 'apache struts'). Control chars stripped, length-capped; rides only as a query param (SSRF-safe)."),
+    cpeName: z
+        .string()
+        .optional()
+        .describe("A CPE 2.3 formatted string to match affected products (cpe:2.3:[aho]:… — e.g. cpe:2.3:a:apache:log4j:2.14.1:*:*:*:*:*:*:*). Non-CPE input is rejected (invalid_input)."),
+    cvssV3Severity: z
+        .enum(["LOW", "MEDIUM", "HIGH", "CRITICAL"])
+        .optional()
+        .describe("Filter to a CVSS v3 base severity band (LOW|MEDIUM|HIGH|CRITICAL)."),
+    pubStartDate: z
+        .string()
+        .optional()
+        .describe("Publication-date window START (ISO YYYY-MM-DD). PAIRED with pubEndDate (both required together — NVD 404s a lone bound). A span >120 days is clamped forward to 120 days BEFORE the request and disclosed."),
+    pubEndDate: z
+        .string()
+        .optional()
+        .describe("Publication-date window END (ISO YYYY-MM-DD). Paired with pubStartDate."),
+    lastModStartDate: z
+        .string()
+        .optional()
+        .describe("Last-modified window START (ISO YYYY-MM-DD). PAIRED with lastModEndDate (both required together). A span >120 days is clamped + disclosed."),
+    lastModEndDate: z
+        .string()
+        .optional()
+        .describe("Last-modified window END (ISO YYYY-MM-DD). Paired with lastModStartDate."),
+    kevOnly: z
+        .boolean()
+        .optional()
+        .describe("When true, return ONLY rows listed in the CISA KEV catalog. ★If the KEV catalog cannot be loaded, this THROWS (a KEV-membership filter is unanswerable without a loaded catalog) — it NEVER returns a silently-empty set (which would falsely read as 'none on the mandatory-remediation list')."),
+    resultsPerPage: z
+        .number()
+        .min(1)
+        .max(2000)
+        .optional()
+        .describe("Rows per page (default 50, max 2000 — NVD's cap). Over-cap is refused, never silently clamped."),
+    startIndex: z
+        .number()
+        .min(0)
+        .optional()
+        .describe("Zero-based page offset (default 0). Pagination derives from NVD's exact totalResults, never the page length."),
+});
+// cisa_kev_lookup: filter the KEV catalog standalone.
+const CisaKevLookupInput = z.object({
+    cveId: z
+        .string()
+        .optional()
+        .describe("Exact CVE identifier CVE-YYYY-NNNN to check for KEV membership. A miss returns found:false + the not-in-KEV≠safe caveat (absence is NOT a safety clearance)."),
+    vendorProject: z
+        .string()
+        .optional()
+        .describe("Case-insensitive substring filter on the vendor/project (e.g. 'Microsoft', 'Apache')."),
+    product: z
+        .string()
+        .optional()
+        .describe("Case-insensitive substring filter on the product (e.g. 'Log4j', 'Exchange Server')."),
+    ransomwareOnly: z
+        .boolean()
+        .optional()
+        .describe("When true, keep only entries with knownRansomwareCampaignUse === 'Known'."),
+    addedSince: z
+        .string()
+        .optional()
+        .describe("Keep only entries with dateAdded >= this ISO date (YYYY-MM-DD)."),
+    dueBefore: z
+        .string()
+        .optional()
+        .describe("Keep only entries with dueDate < this ISO date (YYYY-MM-DD) — the CISA-mandated remediation deadline."),
+    limit: z
+        .number()
+        .min(1)
+        .max(1000)
+        .optional()
+        .describe("Max matches returned (default 100, max 1000)."),
+    offset: z.number().min(0).optional().describe("Zero-based page offset (default 0)."),
+});
+// ━━━ NPPES NPI Registry — the healthcare-provider identity/credentialing lane (1) ━━━ ADR-0036
+// nppes_lookup_provider: exact NPI detail OR search over CMS/HHS's keyless public
+// registry of every US healthcare provider (npiregistry.cms.hhs.gov/api, version=2.1
+// REQUIRED). Mode is inferred from `number` (no mode flag). ★M1: an exact NPI is
+// looked up by number ALONE on the wire (a co-supplied filter AND-combines and would
+// falsely zero a real active provider → found:false); co-filters are checked
+// client-side and disclosed via data.filterMatch. ★S2: the required-one gate is
+// {number, first_name, last_name, organization_name, taxonomy_description, city,
+// postal_code}; state + enumeration_type are REFINERS ONLY (rejected alone). The
+// grammars (10-digit NPI + CMS Luhn, USPS state enum, NPI-1/NPI-2 enum, trailing-`*`
+// wildcard ≥2 leading chars) are the SSRF + silent-foot-gun guards.
+const NppesLookupInput = z.object({
+    number: z
+        .string()
+        .regex(/^\d{10}$/)
+        .optional()
+        .describe("Exact NPI — 10 digits (^\\d{10}$). Triggers EXACT-NPI mode: the wire query carries number (+version) ALONE (any co-supplied filter is DROPPED from the wire and checked client-side, disclosed in data.filterMatch — NPPES AND-combines a number with filters, so a mismatched filter would falsely zero a real active provider). Also client-side CMS-Luhn-validated (Luhn over 80840+first-9): a typo'd NPI ⇒ invalid_input, NEVER a fake 'does not exist'. e.g. '1104130236'."),
+    enumeration_type: z
+        .enum(["NPI-1", "NPI-2"])
+        .optional()
+        .describe("REFINER only (NPI-1 = individual, NPI-2 = organization). Never sufficient alone (⇒ invalid_input) — must accompany a required criterion."),
+    first_name: z
+        .string()
+        .min(1)
+        .max(100)
+        .optional()
+        .describe("Individual provider first name (a required-one criterion). A trailing '*' wildcard needs ≥2 leading literal chars. e.g. 'John'."),
+    last_name: z
+        .string()
+        .min(1)
+        .max(100)
+        .optional()
+        .describe("Individual provider last name (a required-one criterion). Trailing '*' wildcard: ≥2 leading chars. e.g. 'Smith'."),
+    organization_name: z
+        .string()
+        .min(1)
+        .max(100)
+        .optional()
+        .describe("Organization (NPI-2) name (a required-one criterion). Trailing '*' wildcard: ≥2 leading chars. e.g. 'Mayo Clinic'."),
+    taxonomy_description: z
+        .string()
+        .min(1)
+        .max(100)
+        .optional()
+        .describe("Provider taxonomy/specialty description (a required-one criterion). e.g. 'Internal Medicine'."),
+    city: z
+        .string()
+        .min(1)
+        .max(100)
+        .optional()
+        .describe("Address city (a required-one criterion). e.g. 'Baltimore'."),
+    postal_code: z
+        .string()
+        .min(1)
+        .max(100)
+        .optional()
+        .describe("Address postal/ZIP code (a required-one criterion; a prefix like '212' is allowed). e.g. '21218'."),
+    state: z
+        .enum(nppes.NPPES_STATES)
+        .optional()
+        .describe("US state/territory 2-letter USPS code — a REFINER only (never sufficient alone ⇒ invalid_input; NPPES rejects 'state' as the sole criterion). e.g. 'MD'."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(200)
+        .optional()
+        .describe("Providers per page, 1..200, default 10. NPPES silently clamps >200; this tool rejects it loudly. Search mode only."),
+    skip: z
+        .number()
+        .int()
+        .min(0)
+        .max(1000)
+        .optional()
+        .describe("0-based pagination offset, 0..1000 (default 0). ★POLICY cap: this vetting tool reaches at most the first ~1,200 matches/query (a deliberate targeted-lookup boundary — NPPES itself no longer enforces a skip ceiling); skip > 1000 ⇒ invalid_input. Search mode only."),
+});
+// ━━━ CMS Open Payments — the healthcare spend/transparency lane (2) ━━━ ADR-0037
+// A NEW keyless DKAN 2.x datastore adapter (a sibling of ckan/socrata) over
+// openpaymentsdata.cms.gov. ★M2: the DCAT metastore ignores limit/offset — the
+// whole catalog is sliced CLIENT-SIDE (totalAvailable is the exact post-q size).
+// ★M1: the results-array drift guard is conditioned on the effective `results` mode
+// (results:false omits rows). ★S2: `count` is NOT a caller toggle (always count=true
+// on the wire). ★S3: offset ≤ 2000 reach cap (PII boundary, mirrors NPPES).
+// SSRF (load-bearing): datasetId (36-char lowercase UUID) + index interpolate into
+// the URL PATH — the Zod grammars are the primary path-injection guard.
+const CmsSearchDatasetsInput = z.object({
+    q: z
+        .string()
+        .max(200)
+        .optional()
+        .describe("Client-side case-insensitive substring filter over each dataset's title + description (the DKAN metastore returns the ENTIRE catalog in one response; q is applied client-side and totalAvailable is the exact post-filter catalog size). e.g. 'research payment'."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(100)
+        .optional()
+        .describe("Datasets per page, 1..100, default 20 (client-side slice of the full catalog)."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .optional()
+        .describe("0-based pagination offset (default 0), applied client-side against the known catalog length (never a server offset)."),
+});
+const CmsQueryDatasetInput = z.object({
+    datasetId: z
+        .string()
+        .length(36)
+        .regex(/^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/)
+        .describe("REQUIRED — the DKAN datasetId, a 36-char LOWERCASE UUID. ★SSRF: it interpolates into the URL PATH, so this strict grammar (no uppercase, no %2F/../, no trailing newline) is the load-bearing path-injection guard. e.g. 'f0d1de67-6852-4093-a036-c9328c256a05' (2025 Research Payment Data)."),
+    index: z
+        .number()
+        .int()
+        .min(0)
+        .max(50)
+        .optional()
+        .describe("Distribution index (default 0 = the primary CSV). Also interpolates into the URL path (int 0..50)."),
+    conditions: z
+        .array(z.object({
+        property: z
+            .string()
+            .min(1)
+            .max(128)
+            .regex(/^[a-z0-9_]+$/)
+            .describe("Column name (snake_case lowercase alnum). A bad column ⇒ HTTP 400 ⇒ invalid_input (never a silent drop). e.g. 'recipient_state'."),
+        value: z
+            .union([z.string().max(200), z.number()])
+            .describe("Filter value. e.g. 'CA'."),
+        operator: z
+            .enum(["=", "<>", "<", ">", "<=", ">=", "like", "in"])
+            .optional()
+            .describe("Comparison operator (default '='). AND-combined across conditions."),
+    }))
+        .max(10)
+        .optional()
+        .describe("Server-side filters (≤10, AND-combined) that provably narrow the EXACT count. Each either applies or the call errors — filtersDropped is always empty."),
+    properties: z
+        .array(z.string().min(1).max(128).regex(/^[a-z0-9_]+$/))
+        .optional()
+        .describe("Optional column projection (snake_case column names). Omit for all columns."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(500)
+        .optional()
+        .describe("Rows per page, 1..500, default 100. 500 is the HARD DKAN cap (the API 400s over it; this tool rejects >500 loudly)."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .max(2000)
+        .optional()
+        .describe("0-based row offset (default 0). ★POLICY reach cap ≤ 2000 (a deliberate targeted-lookup boundary — Open Payments names physicians + amounts); offset > 2000 ⇒ invalid_input."),
+    results: z
+        .boolean()
+        .optional()
+        .describe("Default true (return rows). Set false for COUNT/SCHEMA-discovery mode: no rows, pagination disabled, but the EXACT count + every column's schema are returned. (`count` is NOT a toggle — count=true is always on the wire.)"),
+});
+// ─── FAC (Federal Audit Clearinghouse) Single Audit — ADR-0038 ──────
+// ★ PII crux: the tools expose ONLY structured, validated filters — NO caller
+// `select`/`order`/free-column param (that is why v1 uses purpose-built tools,
+// not a generic fac_query). The select-allowlist is a hardcoded module constant
+// in fac.ts; no caller value can name a column (PostgREST would project a
+// personal-contact column the instant `select=` named it).
+const FacSearchAuditsInput = z.object({
+    auditeeUei: z
+        .string()
+        .regex(/^[A-Z0-9]{12}$/)
+        .optional()
+        .describe("Filter by 12-char SAM UEI (^[A-Z0-9]{12}$; → auditee_uei=eq. — the PRIMARY join key to SAM/USAspending/EDGAR). e.g. 'ZQGGHJH74DW7'."),
+    auditeeState: z
+        .string()
+        .regex(/^[A-Z]{2}$/)
+        .optional()
+        .describe("Filter by 2-letter US state code (uppercase; → auditee_state=eq.). e.g. 'CA'."),
+    auditYear: z
+        .number()
+        .int()
+        .min(2016)
+        .max(2100)
+        .optional()
+        .describe("Filter by audit year (int, → audit_year=eq.). e.g. 2024."),
+    totalExpendedMin: z
+        .number()
+        .finite()
+        .optional()
+        .describe("Minimum total federal awards expended (USD, → total_amount_expended=gte.)."),
+    totalExpendedMax: z
+        .number()
+        .finite()
+        .optional()
+        .describe("Maximum total federal awards expended (USD, → total_amount_expended=lte.)."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(100)
+        .default(25)
+        .describe("Rows per page, 1..100, default 25."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .default(0)
+        .describe("0-based row offset for pagination (default 0)."),
+});
+const FacGetFindingsInput = z
+    .object({
+    auditeeUei: z
+        .string()
+        .regex(/^[A-Z0-9]{12}$/)
+        .optional()
+        .describe("Filter by 12-char SAM UEI (^[A-Z0-9]{12}$; → auditee_uei=eq.)."),
+    reportId: z
+        .string()
+        .regex(/^[0-9A-Za-z-]+$/)
+        .max(64)
+        .optional()
+        .describe("Filter by FAC report_id (^[0-9A-Za-z-]+$; → report_id=eq. — from a fac_search_audits row)."),
+    auditYear: z
+        .number()
+        .int()
+        .min(2016)
+        .max(2100)
+        .optional()
+        .describe("Filter by audit year (int, → audit_year=eq.)."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(100)
+        .default(50)
+        .describe("Rows per page, 1..100, default 50."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .default(0)
+        .describe("0-based row offset for pagination (default 0)."),
+})
+    .refine((v) => v.auditeeUei !== undefined || v.reportId !== undefined, {
+    message: "fac_get_findings requires at least one of `auditeeUei` or `reportId` (an empty query would scan the whole 670K-row findings table).",
+    path: ["auditeeUei"],
+});
+// GAO bid-protest lookup (keyless RSS + decision-page parse)
+const GaoProtestInput = z.object({
+    agency: z
+        .string()
+        .optional()
+        .describe("Client-side substring filter on the recent-protest feed (matched against the decision title + description). NOTE: filters the RECENT feed window only — not a historical agency search."),
+    protester: z
+        .string()
+        .optional()
+        .describe("Client-side substring filter on the protester name (feed title/description)."),
+    solicitationNumber: z
+        .string()
+        .optional()
+        .describe("Client-side substring filter on the solicitation number (matched in the feed description)."),
+    outcome: z
+        .enum(["sustained", "denied", "dismissed", "withdrawn", "any"])
+        .optional()
+        .describe("Filter by protest disposition (default 'any'). Determined from each decision page, so it applies only when enrich is true."),
+    bNumber: z
+        .string()
+        .optional()
+        .describe("Fetch ONE specific decision directly by GAO B-number (e.g. 'B-424377' or 'b-424249.2'), bypassing the feed. Use to pull a decision that has aged out of the recent feed window."),
+    limit: z
+        .number()
+        .min(1)
+        .max(50)
+        .optional()
+        .describe("Max decisions to return (default 20, max 50). The feed itself carries ~25 recent legal products."),
+    enrich: z
+        .boolean()
+        .optional()
+        .describe("Fetch each decision's page to fill agency/outcome/solicitation/PDF (default true). Set false for a fast feed-only list (those fields will be null)."),
+});
+// GSA daily-CSV keyless backbone — batch page-completing enrichment
+const SamLookupNoticeFieldsInput = z.object({
+    noticeIds: z
+        .array(z.string())
+        .min(1)
+        .max(100)
+        .describe("1..100 32-char hex noticeIds (the ids returned by sam_search_opportunities) to enrich in ONE batch. Completes a whole search page's null naics/setAside/place-of-performance/deadline/type from the cached GSA daily CSV. OFF BY DEFAULT — enable by setting SAM_GOV_CSV_CACHE (a cache dir) or SAM_GOV_ENABLE_CSV=1."),
+});
+// ─── US Treasury Fiscal Data (keyless) — input schemas ───────────
+// ADR-0002. `dataset` is an ENUM of the 5 live-confirmed paths (F5: no free
+// path → no SSRF surface). pageSize max 500 (upstream page[size] ceiling).
+const TreasuryDatasetEnum = z
+    .enum([
+    "debt_to_penny",
+    "avg_interest_rates",
+    "mts_table_1",
+    "rates_of_exchange",
+    "debt_outstanding",
+])
+    .describe("Which confirmed Treasury Fiscal Data dataset to query: debt_to_penny (daily total public debt), avg_interest_rates (avg rate by security type), mts_table_1 (Monthly Treasury Statement receipts/outlays/deficit), rates_of_exchange (quarterly FX by currency), debt_outstanding (historical fiscal-year-end debt).");
+const TreasuryQueryDatasetInput = z.object({
+    dataset: TreasuryDatasetEnum,
+    fields: z
+        .string()
+        .optional()
+        .describe("Optional CSV column projection (e.g. 'record_date,exchange_rate'). An unknown column ⇒ upstream HTTP 400 ⇒ invalid_input (surfaced as an error, never silently dropped)."),
+    filter: z
+        .string()
+        .optional()
+        .describe("Optional CSV of upstream filters 'col:op:val' (ops: lt|lte|gt|gte|eq|in), AND-combined — e.g. 'record_date:gte:2024-01-01,country_currency_desc:eq:Canada-Dollar'."),
+    sort: z
+        .string()
+        .optional()
+        .describe("Optional CSV sort columns; prefix '-' for descending (e.g. '-record_date')."),
+    pageSize: z
+        .number()
+        .int()
+        .min(1)
+        .max(500)
+        .default(100)
+        .describe("Rows per page (upstream page[size]); 1..500, default 100."),
+    pageNumber: z
+        .number()
+        .int()
+        .min(1)
+        .default(1)
+        .describe("1-based page number (upstream page[number]); default 1."),
+});
+const TreasuryDebtToPennyInput = z.object({
+    latest: z
+        .boolean()
+        .default(true)
+        .describe("true (default) ⇒ only the single most-recent day (page[size]=1). false ⇒ the startDate/endDate range, newest-first."),
+    startDate: z
+        .string()
+        .optional()
+        .describe("Range mode only: ISO YYYY-MM-DD lower bound on record_date (inclusive)."),
+    endDate: z
+        .string()
+        .optional()
+        .describe("Range mode only: ISO YYYY-MM-DD upper bound on record_date (inclusive)."),
+    pageSize: z.number().int().min(1).max(500).default(100).describe("Range mode: rows per page, 1..500, default 100."),
+    pageNumber: z.number().int().min(1).default(1).describe("Range mode: 1-based page number, default 1."),
+});
+const TreasuryMonthlyStatementInput = z.object({
+    startDate: z
+        .string()
+        .optional()
+        .describe("ISO YYYY-MM-DD lower bound on record_date (inclusive). Default: trailing ~12 months."),
+    endDate: z
+        .string()
+        .optional()
+        .describe("ISO YYYY-MM-DD upper bound on record_date (inclusive)."),
+    excludeSummaryRows: z
+        .boolean()
+        .default(true)
+        .describe("true (default) excludes fiscal-year PARENT/SUMMARY rows (parent_id/amounts all null) via the server-side filter current_month_gross_outly_amt:gt:0, so only real child line-items (and totalAvailable) remain. false includes the null-amount summary rows."),
+    pageSize: z.number().int().min(1).max(500).default(100).describe("Rows per page, 1..500, default 100."),
+    pageNumber: z.number().int().min(1).default(1).describe("1-based page number, default 1."),
+});
+const TreasuryAvgInterestRatesInput = z.object({
+    securityType: z
+        .string()
+        .optional()
+        .describe("Optional exact security_type_desc filter (e.g. 'Marketable', 'Non-marketable', 'Interest-bearing Debt')."),
+    latest: z
+        .boolean()
+        .default(true)
+        .describe("true (default) ⇒ the most-recent month's full breakdown across security types (pinned to the latest record_date, memoized). false ⇒ the startDate/endDate range."),
+    startDate: z.string().optional().describe("Range mode only: ISO YYYY-MM-DD lower bound on record_date (inclusive)."),
+    endDate: z.string().optional().describe("Range mode only: ISO YYYY-MM-DD upper bound on record_date (inclusive)."),
+    pageSize: z.number().int().min(1).max(500).default(100).describe("Range mode: rows per page, 1..500, default 100."),
+    pageNumber: z.number().int().min(1).default(1).describe("Range mode: 1-based page number, default 1."),
+});
+// ─── SEC EDGAR (keyless) — input schemas ─────────────────────────
+// ADR-0003. Keyless capital-markets source over data.sec.gov / efts.sec.gov.
+// The join key is the 10-digit SEC CIK (NOT SAM UEI/DUNS — see edgar.ts caveat).
+const EdgarLookupCikInput = z.object({
+    query: z
+        .string()
+        .min(1)
+        .describe("Company ticker (exact, case-insensitive) or a company-name substring to resolve to a 10-digit SEC CIK via company_tickers.json. e.g. 'AAPL' or 'apple'. Returns up to 50 matches (found:false on none)."),
+});
+const EdgarCompanyFilingsInput = z.object({
+    cikOrTicker: z
+        .string()
+        .min(1)
+        .describe("A 10-digit (or unpadded) SEC CIK, or a ticker/company-name resolvable via company_tickers.json (e.g. '320193', 'CIK0000320193', 'AAPL')."),
+    forms: z
+        .array(z.string())
+        .optional()
+        .describe("Optional form-type filter (e.g. ['10-K','10-Q','8-K']); case-insensitive exact match on the filing's form. Omit for all forms."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(100)
+        .default(20)
+        .describe("Max filings to return, 1..100, default 20 (offset pagination over the recent window)."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .default(0)
+        .describe("0-based offset into the (form-filtered) recent filings, default 0."),
+    fullHistory: z
+        .boolean()
+        .optional()
+        .describe("When true, ALSO fetch the older filings.files[] submission shards (newest-first, up to maxShards) and assemble the COMPLETE filing history (recent ++ shard001..N, descending order preserved). Default false ⇒ recent window only (byte-identical to omitting it). A capped/failed fan-out is disclosed as PARTIAL — never a capped set claimed complete."),
+    maxShards: z
+        .number()
+        .int()
+        .min(1)
+        .max(100)
+        .optional()
+        .describe("Fan-out cap when fullHistory is true: at most this many older shards are fetched (newest-first), 1..100, default 10. Bounds wall-time (each shard is one throttle-gated GET, serialized through the SHARED edgar gate) + payload. When totalShards > maxShards the response is PARTIAL-BY-CAP (hasMore:true; older un-fetched shards reached by RAISING maxShards, not by nextOffset). Ignored when fullHistory is false."),
+});
+const EdgarCompanyFactsInput = z.object({
+    cikOrTicker: z
+        .string()
+        .min(1)
+        .describe("A 10-digit (or unpadded) SEC CIK, or a ticker/company-name resolvable via company_tickers.json (e.g. '320193', 'AAPL')."),
+    concepts: z
+        .array(z.string())
+        .optional()
+        .describe("Optional XBRL us-gaap/dei concept tags to extract (e.g. ['Assets','NetIncomeLoss']). Default: the 6 curated USD concepts (Revenues/RevenueFromContractWithCustomerExcludingAssessedTax, Assets, Liabilities, StockholdersEquity, NetIncomeLoss, CashAndCashEquivalentsAtCarryingValue). A concept absent for the filer is OMITTED (never 0)."),
+    unit: z
+        .string()
+        .default("USD")
+        .describe("XBRL unit to extract, default 'USD'. A concept present only in another unit (e.g. EarningsPerShareBasic in 'USD/shares') is reported under wrongUnit with a note — never a silent 0."),
+    latest: z
+        .boolean()
+        .default(false)
+        .describe("true ⇒ reduce each concept to its single most-recent data point (by period end). false (default) ⇒ the full reported time series."),
+});
+// ADR-0017. Keyless cross-filer XBRL cross-section over data.sec.gov/api/xbrl/
+// frames. taxonomy/tag/unit/period are RAW PATH SEGMENTS, so the enum + the three
+// regexes below ARE the SSRF guard (re-run belt-and-suspenders in the handler's
+// URL builder). `limit`/`offset` window the already-fully-fetched cross-section
+// CLIENT-SIDE (they do NOT reduce the fetch). taxonomy enum = live-confirmed only.
+const EdgarXbrlFramesInput = z.object({
+    tag: z
+        .string()
+        .regex(/^[A-Za-z0-9]+$/, "tag must be alphanumeric only — an EXACT XBRL concept tag (e.g. 'Assets', 'Revenues'). Slash/dot/space/percent/'..' are rejected (path-segment injection guard).")
+        .describe("XBRL concept tag — EXACT, alphanumeric only (e.g. 'Assets', 'Revenues', 'NetIncomeLoss', 'EarningsPerShareBasic'). A non-matching tag ⇒ upstream 404 ⇒ found:false (never a fabricated 0)."),
+    period: z
+        .string()
+        .regex(/^CY\d{4}(Q[1-4]I?)?$/, "period must be CY{yyyy} (annual flow, e.g. CY2023), CY{yyyy}Q{n} (quarterly flow, e.g. CY2023Q1), or CY{yyyy}Q{n}I (instant, trailing I, e.g. CY2023Q4I).")
+        .describe("Calendar period frame: CY2023 (annual flow) · CY2023Q1 (quarterly flow, no I) · CY2023Q4I (instant / balance-sheet, trailing I). Instant concepts (e.g. Assets) REQUIRE the trailing I; a mismatch ⇒ 404 ⇒ found:false."),
+    taxonomy: z
+        .enum(edgar.FRAMES_TAXONOMIES)
+        .default("us-gaap")
+        .describe("XBRL taxonomy namespace (a fixed enum — the SSRF guard for this segment): 'us-gaap' (financial statements, default) or 'dei' (entity/document info, e.g. EntityCommonStockSharesOutstanding, EntityPublicFloat). Live-confirmed members only."),
+    unit: z
+        .string()
+        .regex(/^[A-Za-z0-9-]+$/, "unit must match ^[A-Za-z0-9-]+$ — hyphen allowed (e.g. 'USD-per-shares'); slash/dot/percent forbidden (never the 'USD/shares' companyfacts key form).")
+        .default("USD")
+        .describe("XBRL unit of measure, as a path segment: 'USD' (default), 'shares', 'USD-per-shares' (EPS — HYPHEN, never 'USD/shares'), 'pure'. A valid-shaped but wrong unit ⇒ 404 ⇒ found:false."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("CLIENT-SIDE page size over the already-fully-fetched cross-section (1..1000, default 100). Does NOT reduce the upstream fetch — the whole frame is fetched in one call; this only windows the returned rows (page via _meta.pagination.nextOffset)."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .default(0)
+        .describe("0-based client-side offset into the fetched cross-section (default 0). Page via _meta.pagination.nextOffset to reach every filer."),
+    includeStats: z
+        .boolean()
+        .default(false)
+        .describe("When true, compute a summary distribution { count, min, max, sum, mean, median, p25, p75, nonFiniteExcluded } over the FULL cross-section (ALL rows, BEFORE the client-side slice), using linear-interpolated percentiles over the FINITE vals only. count===0 (no finite vals) ⇒ every stat is null (never 0/NaN/Infinity)."),
+});
+const EdgarFullTextSearchInput = z.object({
+    q: z
+        .string()
+        .min(1)
+        .describe("Full-text query over EDGAR filings (2001-present). Wrap a phrase in double quotes for an exact match (e.g. '\"climate risk\"')."),
+    forms: z
+        .array(z.string())
+        .optional()
+        .describe("Optional form-type filter (e.g. ['10-K','8-K'])."),
+    startdt: z
+        .string()
+        .optional()
+        .describe("Optional ISO YYYY-MM-DD filing-date lower bound (sets dateRange=custom)."),
+    enddt: z
+        .string()
+        .optional()
+        .describe("Optional ISO YYYY-MM-DD filing-date upper bound (sets dateRange=custom)."),
+    ciks: z
+        .array(z.string())
+        .max(50)
+        .optional()
+        .describe("Optional: pin filings BY one or more entities, by NUMERIC SEC CIK (each is zero-padded to 10 digits — an EXACT-entity match). Multiple CIKs are AND-of-OR (any of the listed entities). A ticker/company name / CIK-0 entry is rejected as invalid_input — use `entityName` or resolve the CIK first with edgar_lookup_cik."),
+    entityName: z
+        .string()
+        .min(1)
+        .max(200)
+        .optional()
+        .describe("Optional FUZZY filer-name narrowing (matches the filer's display name; NOT CIK-exact — can match related entities, e.g. multiple 'Apple*' filers). Combine with `ciks` for an exact-entity result."),
+    from: z
+        .number()
+        .int()
+        .min(0)
+        .default(0)
+        .describe("0-based result offset for pagination; page size is FIXED at 100 (there is no size param). Must be <= 9900 (from+100 ≤ 10000 upstream window); a larger from is rejected as invalid_input."),
+});
+// ADR-0026. Keyless BULK cross-filer capability on the EXISTING edgar source: the
+// SEC EDGAR quarterly full-index (www.sec.gov/Archives/edgar/full-index/<year>/
+// QTR<n>/master.idx). year/quarter are the ONLY path segments (bounded integers —
+// the SSRF guard, re-checked belt-and-suspenders pre-fetch in edgar's
+// buildFullIndexUrl). ALL other params are CLIENT-SIDE filters over the whole
+// downloaded body (ZERO query string reaches the wire). The year UPPER bound
+// (≤ current UTC year) is enforced at CALL time in the handler, NOT baked into Zod,
+// so the tools/list snapshot stays deterministic across a year rollover.
+const EdgarFilingIndexInput = z.object({
+    year: z
+        .number()
+        .int()
+        .min(1993)
+        .describe("Filing year (>= 1993 — EDGAR full-index begins 1993 Q1). Must be <= the current year; a future year is rejected as invalid_input with 0 fetch. Path segment."),
+    quarter: z
+        .number()
+        .int()
+        .min(1)
+        .max(4)
+        .describe("Calendar quarter 1..4 (path segment QTR<quarter>). A same-year FUTURE quarter returns a well-formed EMPTY result (genuine-empty, complete:true), NOT an error."),
+    formType: z
+        .string()
+        .min(1)
+        .max(30)
+        .optional()
+        .describe("Optional CLIENT-SIDE filter: case-insensitive EXACT match on the Form Type column (e.g. '8-K', '10-K'). '8-K' does NOT match '8-K/A' — pass each amendment variant separately."),
+    cik: z
+        .union([z.string().regex(/^\d{1,10}$/), z.number().int().nonnegative()])
+        .optional()
+        .describe("Optional CLIENT-SIDE filter: numeric SEC CIK (1-10 digits or a number), matched leading-zero-safe via padCik on both sides (so '320193' and '0000320193' match the same filer)."),
+    companyContains: z
+        .string()
+        .min(1)
+        .max(200)
+        .optional()
+        .describe("Optional CLIENT-SIDE filter: case-insensitive LITERAL substring on the Company Name column. A multi-word value matches as ONE contiguous string (NOT AND/OR-tokenized)."),
+    dateFrom: z
+        .string()
+        .regex(/^\d{4}-\d{2}-\d{2}$/)
+        .optional()
+        .describe("Optional CLIENT-SIDE filter: keep filings whose Date Filed >= this ISO YYYY-MM-DD (string compare; the column is already YYYY-MM-DD)."),
+    dateTo: z
+        .string()
+        .regex(/^\d{4}-\d{2}-\d{2}$/)
+        .optional()
+        .describe("Optional CLIENT-SIDE filter: keep filings whose Date Filed <= this ISO YYYY-MM-DD."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("Page size over the FILTERED, full-scanned matches (1..1000, default 100). Does NOT reduce the download — the whole quarter is scanned; this only windows the returned rows (page via _meta.pagination.nextOffset)."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .default(0)
+        .describe("0-based offset into the filtered matches (default 0)."),
+});
+// ─── EDGAR daily-index (ADR-0027) — input schema ─────────────────
+// The per-DAY sibling of edgar_filing_index. A single required ISO `date`
+// (YYYY-MM-DD) is the ONLY caller value that shapes the URL — the handler derives
+// year/quarter/yyyymmdd from it and re-validates them belt-and-suspenders in
+// edgar's buildDailyIndexUrl. dateFrom/dateTo are DROPPED (a single-day file has one
+// date). ALL other params are CLIENT-SIDE filters over the whole downloaded body
+// (ZERO query string reaches the wire). The date's EXACT calendar-day round-trip +
+// the future-date rejection are enforced at CALL time in the handler (M2), so the
+// tools/list snapshot stays deterministic across a day rollover.
+const EdgarDailyFilingIndexInput = z.object({
+    date: z
+        .string()
+        .regex(/^\d{4}-\d{2}-\d{2}$/)
+        .describe("Required calendar day ISO YYYY-MM-DD (>= 1994-01-01 — EDGAR daily-index begins 1994 Q1). The handler derives year/quarter/yyyymmdd. A malformed / non-real day (2024-02-30, non-leap 2023-02-29) or a FUTURE date is rejected as invalid_input with 0 fetch. TODAY is allowed (its index may not be posted until ~22:00 US-Eastern)."),
+    formType: z
+        .string()
+        .min(1)
+        .max(30)
+        .optional()
+        .describe("Optional CLIENT-SIDE filter: case-insensitive EXACT match on the Form Type column (e.g. '8-K', '10-K'). '8-K' does NOT match '8-K/A' — pass each amendment variant separately."),
+    cik: z
+        .union([z.string().regex(/^\d{1,10}$/), z.number().int().nonnegative()])
+        .optional()
+        .describe("Optional CLIENT-SIDE filter: numeric SEC CIK (1-10 digits or a number), matched leading-zero-safe via padCik on both sides (so '320193' and '0000320193' match the same filer)."),
+    companyContains: z
+        .string()
+        .min(1)
+        .max(200)
+        .optional()
+        .describe("Optional CLIENT-SIDE filter: case-insensitive LITERAL substring on the Company Name column. A multi-word value matches as ONE contiguous string (NOT AND/OR-tokenized)."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("Page size over the FILTERED, full-scanned matches (1..1000, default 100). Does NOT reduce the download — the whole day is scanned; this only windows the returned rows (page via _meta.pagination.nextOffset)."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .default(0)
+        .describe("0-based offset into the filtered matches (default 0)."),
+});
+// ADR-0041. Keyless WITHIN-SOURCE DEPTH on the existing edgar source: one filer ×
+// one XBRL concept × the COMPLETE reported time-series (data.sec.gov/api/xbrl/
+// companyconcept/CIK{cik10}/{taxonomy}/{Concept}.json). cik (via resolveCik→padCik) +
+// taxonomy (enum) + concept (alnum regex) are the THREE validated PATH SEGMENTS (the
+// SSRF guard, re-checked belt-and-suspenders pre-fetch in edgar's buildConceptUrl).
+// `unit` is a BODY key filtered CLIENT-SIDE (NOT a path segment — no unit regex);
+// unit/form/fy/canonicalOnly/limit/offset are all client-side over the fully-fetched set.
+const EdgarCompanyConceptInput = z.object({
+    cikOrTicker: z
+        .string()
+        .min(1)
+        .describe("A 10-digit (or unpadded) SEC CIK, or a ticker/company-name resolvable via company_tickers.json (e.g. '320193', 'CIK0000320193', 'AAPL')."),
+    concept: z
+        .string()
+        .regex(/^[A-Za-z0-9]+$/, "concept must be alphanumeric only — an EXACT XBRL tag (e.g. 'Assets', 'Revenues', 'NetIncomeLoss'). Slash/dot/space/percent/'..' are rejected (path-segment injection guard).")
+        .describe("XBRL concept tag — EXACT, alphanumeric CamelCase (e.g. 'Assets', 'Revenues', 'NetIncomeLoss', 'Liabilities'). A tag the filer never reported ⇒ upstream 404 ⇒ found:false (never a fabricated 0)."),
+    taxonomy: z
+        .enum(edgar.CONCEPT_TAXONOMIES)
+        .default("us-gaap")
+        .describe("XBRL taxonomy namespace (a fixed enum — the SSRF guard for this segment): 'us-gaap' (financial statements, default), 'dei' (entity/document info, e.g. EntityCommonStockSharesOutstanding), or 'ifrs-full' (IFRS filers, e.g. a foreign private issuer). Live-confirmed members only."),
+    unit: z
+        .string()
+        .min(1)
+        .max(40)
+        .optional()
+        .describe("Optional CLIENT-SIDE filter on the returned units{} keys (NOT a path segment — 'USD', 'shares', 'USD/shares', 'EUR', 'pure'). Restricts rows to that unit but STILL discloses the other units via unitsAvailable + a note. A unit not present ⇒ 0 rows + the available-units note (never a fabricated pick)."),
+    form: z
+        .string()
+        .min(1)
+        .max(30)
+        .optional()
+        .describe("Optional CLIENT-SIDE filter: case-insensitive EXACT match on a row's `form` (e.g. '10-K' for annual values only, '10-Q' for quarterly)."),
+    fy: z
+        .number()
+        .int()
+        .optional()
+        .describe("Optional CLIENT-SIDE filter: keep only rows whose fiscal year `fy` equals this integer (e.g. 2023)."),
+    canonicalOnly: z
+        .boolean()
+        .default(false)
+        .describe("When true, reduce to ONE row per distinct (unit,start,end) period — the frame-tagged canonical value, or (for a not-yet-consolidated period) the latest-filed row (marked canonical:false). SUPERSEDED/amendment rows are REMOVED (fully disclosed via a note). Default false ⇒ ALL rows incl. the amendment/restatement history. A same-`end` different-`start` pair is a DIFFERENT period (both kept), NOT a duplicate."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("CLIENT-SIDE page size over the already-fully-fetched, (unit,start,end)-keyed time-series (1..1000, default 100). Does NOT reduce the upstream fetch (SEC does not paginate companyconcept); page via _meta.pagination.nextOffset."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .default(0)
+        .describe("0-based client-side offset into the filtered time-series (default 0)."),
+});
+// ─── Socrata / SODA (keyless SLED + E-rate) — input schemas ──────
+// ADR-0004. First SLED source. `domain` is a curated allowlist ENUM (the SSRF
+// core — no free host); `datasetId` is a strict 4x4 with .length(9) (M2 — blocks
+// a trailing-newline the regex `$` would admit). SoQL params are raw upstream-
+// validated strings (a bad column ⇒ upstream 400 ⇒ invalid_input, surfaced).
+const SocrataDomainEnum = z
+    .enum(socrata.SOCRATA_DOMAINS)
+    .describe("Which allowlisted Socrata portal to query (curated .gov hosts + USAC E-rate .org; the SSRF host allowlist — no free host). e.g. data.ny.gov, data.texas.gov, data.wa.gov, opendata.usac.org.");
+const SocrataQueryInput = z.object({
+    domain: SocrataDomainEnum,
+    datasetId: z
+        // M2 — a strict 4x4. NOTE: deliberately NO .trim(): Zod applies .trim()
+        // BEFORE .length(9), so `.trim().length(9)` (the ADR's literal wording)
+        // would STRIP a trailing "\n" to a valid 9-char id and ACCEPT it —
+        // empirically confirmed — defeating the very newline rejection M2 wants.
+        // Dropping .trim() makes .length(9) see the raw string, so any trailing
+        // char (incl. "\n", which the regex `$` alone would admit) is rejected.
+        .string()
+        .length(9)
+        .regex(/^[a-z0-9]{4}-[a-z0-9]{4}$/)
+        .describe("The dataset's Socrata 4x4 id, e.g. 'kwxv-fwze' (from socrata_discover_datasets). Exactly [a-z0-9]{4}-[a-z0-9]{4} (9 chars; no surrounding whitespace)."),
+    select: z
+        .string()
+        .optional()
+        .describe("Optional SoQL $select (column projection / aggregate), e.g. 'agency,SUM(amount)'."),
+    where: z
+        .string()
+        .optional()
+        .describe("Optional SoQL $where filter, e.g. \"fiscal_year='2024' AND amount>1000\". A bad column ⇒ upstream HTTP 400 ⇒ invalid_input (surfaced, never silent)."),
+    order: z
+        .string()
+        .optional()
+        .describe("Optional SoQL $order, e.g. 'amount DESC'."),
+    q: z
+        .string()
+        .optional()
+        .describe("Optional SoQL $q full-text search across the row."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("Rows per page ($limit), 1..1000, default 100."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .default(0)
+        .describe("0-based row offset ($offset) for pagination, default 0."),
+    withTotal: z
+        .boolean()
+        .default(true)
+        .describe("true (default) ⇒ issue a count(*) companion query so totalAvailable is exact. false ⇒ skip it (one fewer request); totalAvailable is null and a note discloses results may be truncated at $limit."),
+});
+const SocrataDiscoverDatasetsInput = z.object({
+    q: z
+        .string()
+        .min(1)
+        .describe("Keyword(s) to find datasets, e.g. 'procurement', 'vendor payments', 'checkbook'."),
+    domain: SocrataDomainEnum.optional().describe("Optional: scope discovery to ONE allowlisted portal. Omit to search the whole allowlist. NOTE: the federated catalog does not index every host (e.g. USAC E-rate returns 0) — those remain queryable via socrata_query with a known 4x4."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(100)
+        .default(20)
+        .describe("Max datasets to return, 1..100, default 20."),
+});
+// ─── CKAN datastore_search (keyless SLED) — input schemas ───────
+// ADR-0006. Second SLED source; FIRST on the R2 DataSource port. `host` is a
+// curated allowlist ENUM (the SSRF core — no free host); `resourceId` is a strict
+// 36-char lowercase UUID with .length(36) (m1 — the regex `$` already blocks a
+// trailing "\n"; length is belt-and-suspenders; NO .trim(), NO `i` flag). `filters`
+// is a CONSTRAINED object we JSON.stringify (Q3 — never a raw string). A bad
+// filter field / sort ⇒ upstream 409 ⇒ invalid_input (surfaced, never silent).
+const CkanHostEnum = z
+    .enum(ckan.CKAN_HOSTS)
+    .describe("Which allowlisted CKAN portal to query (curated .gov hosts — the SSRF host allowlist, no free host): data.ca.gov (CA), data.virginia.gov (VA — eVA), data.boston.gov (City of Boston Checkbook).");
+const CkanQueryInput = z.object({
+    host: CkanHostEnum,
+    resourceId: z
+        // m1 — a strict 36-char lowercase UUID. NO .trim() (Zod trims BEFORE
+        // .length, which would strip a trailing "\n" to a valid id and defeat the
+        // guard); the regex `$` alone rejects a trailing newline in JS, and
+        // .length(36) is belt-and-suspenders. Lowercase-only ⇒ no `i` flag.
+        .string()
+        .length(36)
+        .regex(/^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/)
+        .describe("The datastore resource_id, a 36-char lowercase UUID e.g. 'bb82edc5-9c78-44e2-8947-68ece26197c5' (from ckan_discover_datasets, a datastoreActive:true resource)."),
+    q: z
+        .string()
+        .optional()
+        .describe("Optional full-text search across the record (CKAN `q`)."),
+    filters: z
+        .record(z.string(), z.union([z.string(), z.number(), z.array(z.union([z.string(), z.number()]))]))
+        .optional()
+        .describe("Optional structured field filters, e.g. {\"Fiscal Year\":\"2013-2014\"}. A constrained object (string/number/array values only) that we JSON.stringify; a bad field ⇒ upstream HTTP 409 ⇒ invalid_input (surfaced, never silent)."),
+    sort: z
+        .string()
+        .optional()
+        .describe("Optional sort, e.g. '_id asc' or 'amount desc'. A bad field ⇒ 409 ⇒ invalid_input."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("Rows per page, 1..1000, default 100."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .default(0)
+        .describe("0-based row offset for pagination, default 0."),
+});
+const CkanDiscoverDatasetsInput = z.object({
+    host: CkanHostEnum,
+    q: z
+        .string()
+        .min(1)
+        .describe("Keyword(s) to find datasets, e.g. 'procurement', 'checkbook', 'vendor'."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(100)
+        .default(20)
+        .describe("Max datasets (packages) to return, 1..100, default 20."),
+});
+// ─── FDIC BankFind Suite (keyless institution directory + financials) ──────────
+// ADR-0028. First OFF-EDGAR entity source. KEYLESS, fixed host api.fdic.gov +
+// FIXED endpoint constants (the SSRF core — no free host/path). Structured
+// inputs → a server-side filters/search builder (NAME/CITY route through FDIC's
+// full-text `search`, NOT `filters` — M1). name/city char-class is a sanity
+// boundary; the injection defense is the module's backslash-first phrase-escape
+// (M2). sortBy is a Zod enum + a Set.has recheck in the builder (an unknown sort
+// field is invalid_input before fetch).
+// name/city char-class — includes `( ) #` so real bank names like 'Mizuho Bank
+// (USA)' are NOT false-rejected (S2); these are literal inside the phrase-quoted
+// + escaped `search` value, never ES metachars there.
+const FdicNameCity = z
+    .string()
+    .regex(/^[A-Za-z0-9 .,&'\/()#\-]+$/)
+    .describe("Free-text bank name / city fragment (letters, digits, spaces and . , & ' / ( ) # -). Matched by FDIC's case-insensitive full-text `search` (token match; a multi-word value is matched per-token, may be broader than a literal substring).");
+const FdicSearchInstitutionsInput = z.object({
+    state: z
+        .string()
+        .regex(/^[A-Z]{2}$/)
+        .optional()
+        .describe("Filter by 2-letter US state code (uppercase; → STALP filter). e.g. 'VA'."),
+    activeOnly: z
+        .boolean()
+        .optional()
+        .describe("Filter to active (true → ACTIVE:1) or inactive (false → ACTIVE:0) institutions; omit for both."),
+    cert: z
+        .number()
+        .int()
+        .min(1)
+        .optional()
+        .describe("Filter by FDIC certificate number (the STABLE entity key; → CERT filter)."),
+    name: FdicNameCity.optional().describe("Filter by institution NAME via FDIC full-text `search` (case-insensitive token match; NOT case-sensitive exact-keyword — that is why we route to `search`, not `filters`)."),
+    city: FdicNameCity.optional().describe("Filter by CITY via FDIC full-text `search` (case-insensitive token match)."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("Rows per page, 1..1000, default 100."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .max(100000)
+        .default(0)
+        .describe("0-based row offset for pagination, 0..100000, default 0."),
+    sortBy: z
+        .enum(["NAME", "CERT", "ASSET", "ESTYMD", "STALP", "CITY", "ACTIVE"])
+        .optional()
+        .describe("Optional sort field (an allowlisted enum; an unknown field is rejected before fetch)."),
+    sortOrder: z
+        .enum(["ASC", "DESC"])
+        .default("ASC")
+        .describe("Sort direction when sortBy is set, ASC (default) or DESC."),
+});
+const FdicInstitutionFinancialsInput = z.object({
+    cert: z
+        .number()
+        .int()
+        .min(1)
+        .describe("REQUIRED FDIC certificate number of the institution (→ CERT filter). From fdic_search_institutions."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("Rows per page, 1..1000, default 100."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .max(100000)
+        .default(0)
+        .describe("0-based row offset for pagination, 0..100000, default 0."),
+    sortBy: z
+        .enum(["REPDTE", "ASSET", "DEP", "NETINC"])
+        .default("REPDTE")
+        .describe("Sort field (allowlisted enum; default REPDTE = report date)."),
+    sortOrder: z
+        .enum(["ASC", "DESC"])
+        .default("DESC")
+        .describe("Sort direction, default DESC (newest quarter first)."),
+});
+// ADR-0029 — the 3rd FDIC tool (source 23 unchanged; snapshot 89→90) reading
+// /banks/failures. v2 (cycle-33) live review: the state field is PSTALP (NOT
+// STALP — a false-empty landmine), and /failures IGNORES the `search` param (a
+// name/city query floods the whole dataset) → NO name/city filter here; name
+// lookup is the 2-step CERT linkage via fdic_search_institutions.
+const FdicBankFailuresInput = z.object({
+    state: z
+        .string()
+        .regex(/^[A-Z]{2}$/)
+        .optional()
+        .describe("Filter by 2-letter US state code (uppercase; → PSTALP filter — the /failures state field is PSTALP, NOT STALP). e.g. 'CA'."),
+    failYear: z
+        .number()
+        .int()
+        .min(1934)
+        .max(new Date().getUTCFullYear())
+        .optional()
+        .describe("Filter by year of failure (→ FAILYR filter). 1934..current UTC year. e.g. 2023 → the 5 real 2023 failures (Silicon Valley Bank, Signature Bank, First Republic Bank, Heartland Tri-State Bank, Citizens Bank)."),
+    cert: z
+        .number()
+        .int()
+        .min(1)
+        .optional()
+        .describe("Filter by FDIC certificate number (the STABLE entity key; → CERT filter). Resolve a bank's CERT via fdic_search_institutions."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("Rows per page, 1..1000, default 100."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .max(100000)
+        .default(0)
+        .describe("0-based row offset for pagination, 0..100000, default 0."),
+    sortBy: z
+        .enum(["FAILDATE", "COST", "QBFASSET", "QBFDEP", "NAME", "FAILYR"])
+        .default("FAILDATE")
+        .describe("Sort field (allowlisted enum; default FAILDATE = failure date). An unknown field is rejected before fetch."),
+    sortOrder: z
+        .enum(["ASC", "DESC"])
+        .default("DESC")
+        .describe("Sort direction, default DESC (most-recent failures first)."),
+});
+// ADR-0030 — the 4th FDIC tool (source 23 unchanged; snapshot 90→91) reading
+// /banks/history, the institution-level STRUCTURAL-CHANGE event log (mergers,
+// absorptions, failures, name/location/charter/regulator changes, branch open/close,
+// etc.) — completing the FDIC entity cluster. Live review: the state field is PSTALP
+// (NOT STALP — a false-empty landmine), and /history's `search` param returns 0 for
+// INSTNAME (a false-empty) → NO name/city filter; name lookup is the 2-step CERT
+// linkage via fdic_search_institutions. All inputs optional, AND-combined; `cert` is
+// the primary path.
+const FdicInstitutionHistoryInput = z.object({
+    cert: z
+        .number()
+        .int()
+        .min(1)
+        .optional()
+        .describe("Filter by FDIC certificate number (the STABLE entity key; → CERT filter — the PRIMARY lookup). Resolve a bank's CERT via fdic_search_institutions. e.g. 3510 → Bank of America's 13,794-row structural-change history."),
+    changeCode: z
+        .number()
+        .int()
+        .min(1)
+        .optional()
+        .describe("Filter by FDIC structural-change code (→ CHANGECODE filter). e.g. 223 = Merger (Without Assistance), 211 = Failure (Whole Institution), 721 = Branch Closing, 520 = Change in Physical Location, 110 = New Institution. Each row also carries FDIC's own changeDescription (CHANGECODE_DESC)."),
+    effYear: z
+        .number()
+        .int()
+        .min(1782)
+        .max(new Date().getUTCFullYear())
+        .optional()
+        .describe("Filter by the year the structural change took effect (→ EFFYEAR filter). 1782..current UTC year (1782 = the oldest observed EFFYEAR)."),
+    state: z
+        .string()
+        .regex(/^[A-Z]{2}$/)
+        .optional()
+        .describe("Filter by 2-letter US state code (uppercase; → PSTALP filter — the /history state field is PSTALP, NOT STALP). e.g. 'CA'."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("Rows per page, 1..1000, default 100."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .max(100000)
+        .default(0)
+        .describe("0-based row offset for pagination, 0..100000, default 0."),
+    sortBy: z
+        .enum(["EFFDATE", "PROCDATE", "CHANGECODE", "TRANSNUM"])
+        .default("EFFDATE")
+        .describe("Sort field (allowlisted enum; default EFFDATE = effective date). An unknown field is rejected before fetch."),
+    sortOrder: z
+        .enum(["ASC", "DESC"])
+        .default("DESC")
+        .describe("Sort direction, default DESC (newest structural change first)."),
+});
+// ADR-0031 — the 5th FDIC tool (source 23 unchanged; snapshot 91→92) reading
+// /banks/summary, the FDIC's OWN aggregate/statistical roll-ups (the FIRST
+// AGGREGATE tool on this source; the 4 existing are per-ENTITY on CERT). Live
+// review: the state field is STALP (NOT PSTALP — a per-endpoint difference), the
+// number-of-institutions field is BANKS (NOT NUMINST), /summary serves NO ratio
+// fields, and NIM is net interest INCOME in $thousands (NOT the margin ratio). Each
+// row crosses charter (CB_SI) × geography (STALP), where STALP ∈ {USA,US,OT,PI} are
+// geographic ROLL-UPS surfaced via a derived scope/isRollup so a roll-up never
+// masquerades as a state. NO name/city filter (the `search` param is a no-op). A
+// non-int year is rejected pre-fetch (a malformed year is a live HTTP-200 total:0
+// false-empty). All inputs optional, AND-combined.
+const FdicIndustrySummaryInput = z.object({
+    year: z
+        .number()
+        .int()
+        .min(1934)
+        .max(new Date().getUTCFullYear())
+        .optional()
+        .describe("Filter by aggregate YEAR (→ YEAR filter). 1934..current UTC year. e.g. 2023 → the 121 (charter × geography) aggregate rows for 2023. A non-int is rejected pre-fetch (a malformed year is a live HTTP-200 total:0 false-empty)."),
+    state: z
+        .string()
+        .regex(/^[A-Z]{2,3}$/)
+        .optional()
+        .describe("Filter by geography via the STALP code (uppercase 2-or-3 letters; → STALP filter — the /summary state field is STALP, NOT PSTALP). Accepts a jurisdiction USPS code (TX, CA, DC, GU, PR…) OR a ROLL-UP code: USA (all states+territories), US (states+DC), OT (all territories), PI (Pacific Islands). The output scope/isRollup disambiguates every returned row."),
+    charterClass: z
+        .enum(["CB", "SI"])
+        .optional()
+        .describe("Filter by charter class (→ CB_SI filter): CB = commercial banks, SI = savings institutions. Omit to return BOTH charter rows for the geography — there is NO pre-combined 'all institutions' row (a geography's total = its CB row + its SI row)."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("Rows per page, 1..1000, default 100."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .max(100000)
+        .default(0)
+        .describe("0-based row offset for pagination, 0..100000, default 0."),
+    sortBy: z
+        .enum(["YEAR", "ASSET", "DEP", "NETINC", "BANKS"])
+        .default("YEAR")
+        .describe("Sort field (allowlisted enum; default YEAR = aggregate year). An unknown field is rejected before fetch."),
+    sortOrder: z
+        .enum(["ASC", "DESC"])
+        .default("DESC")
+        .describe("Sort direction, default DESC (newest year / largest first)."),
+});
+// ADR-0040 — the 6th & 7th FDIC tools (source 30 unchanged; snapshot 103→105) —
+// WITHIN-SOURCE DEPTH on the already-wired src/fdic.ts adapter. (6) fdic_risk_ratios
+// projects the curated counterparty RISK-RATIO catalog on the ALREADY-wired
+// /banks/financials endpoint (per-field units in the output key; percent ratios
+// verbatim via num, tier-1 capital ×1000; ★M1 the CBLR RBCRWAJ=0 sentinel → null via
+// FDIC's CBLRIND flag + a per-row cblrFramework, never a false 0% capital). (7)
+// fdic_branch_deposits reads /banks/sod (Summary of Deposits) — ONE new fixed endpoint
+// constant. Both keyed on the numeric CERT; C118-quoted state; NO name/city search.
+const FdicRiskRatiosInput = z.object({
+    cert: z
+        .number()
+        .int()
+        .min(1)
+        .describe("REQUIRED FDIC certificate number of the institution (→ CERT filter). From fdic_search_institutions."),
+    reportDate: z
+        .number()
+        .int()
+        .min(19000101)
+        .max(29991231)
+        .optional()
+        .describe("Optional report date (→ REPDTE filter), a quarter-end as a YYYYMMDD integer (e.g. 20240630). Omit for the full quarterly ratio time-series."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("Rows per page, 1..1000, default 100."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .max(100000)
+        .default(0)
+        .describe("0-based row offset for pagination, 0..100000, default 0."),
+    sortBy: z
+        .enum(["REPDTE", "ROA", "ROE", "RBCRWAJ", "EEFFR"])
+        .default("REPDTE")
+        .describe("Sort field (allowlisted enum; default REPDTE = report date). An unknown field is rejected before fetch."),
+    sortOrder: z
+        .enum(["ASC", "DESC"])
+        .default("DESC")
+        .describe("Sort direction, default DESC (newest quarter first)."),
+});
+const FdicBranchDepositsInput = z.object({
+    cert: z
+        .number()
+        .int()
+        .min(1)
+        .optional()
+        .describe("Filter by FDIC certificate number (the STABLE entity key; → CERT filter). Resolve a bank's CERT via fdic_search_institutions."),
+    state: z
+        .string()
+        .regex(/^[A-Z]{2}$/)
+        .optional()
+        .describe("Filter by 2-letter branch state code (uppercase; → STALPBR filter — the SOD branch-state field). e.g. 'OR'. C118-quoted so Oregon is Lucene-operator-safe."),
+    year: z
+        .number()
+        .int()
+        .min(1934)
+        .max(new Date().getUTCFullYear())
+        .optional()
+        .describe("Filter by Summary-of-Deposits survey YEAR (→ YEAR filter), the annual June-30 snapshot year. 1934..current UTC year."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("Rows per page, 1..1000, default 100."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .max(100000)
+        .default(0)
+        .describe("0-based row offset for pagination, 0..100000, default 0."),
+    sortBy: z
+        .enum(["YEAR", "DEPSUMBR"])
+        .default("YEAR")
+        .describe("Sort field (allowlisted enum; default YEAR = snapshot year). An unknown field is rejected before fetch."),
+    sortOrder: z
+        .enum(["ASC", "DESC"])
+        .default("DESC")
+        .describe("Sort direction, default DESC (newest snapshot / largest deposits first)."),
+});
+// ─── OpenFEMA (keyless disaster declarations + emergency-assistance spend) ──────
+// ADR-0016. KEYLESS, fixed host www.fema.gov + a PINNED dataset registry
+// {entityName, version} (the SSRF core — no free host/path/version). Filters are
+// MODULE-BUILT from a per-tool, LIVE-VERIFIED field whitelist into an OData
+// `$filter` (NO raw `$filter` arg exists — no injection surface); the module
+// ALWAYS sends `$inlinecount=allpages` so metadata.count is the EXACT filtered
+// total. Per-dataset field names DIFFER (M1): the PA tool's `state` maps to
+// `stateAbbreviation`, the declarations tool's `state` maps to `state` — a shared
+// user-facing arg over two different real OData fields. limit → $top (≤1000),
+// offset → $skip.
+// A 2-letter US state/territory code (uppercase; e.g. CA, LA, PR, DC). A bad value
+// is NOT an SSRF vector (the fixed host + dataset registry are) — it just returns
+// an honest genuine-empty (metadata.count:0), never a silent unfiltered set.
+const FemaStateSchema = z
+    .string()
+    .regex(/^[A-Za-z]{2}$/)
+    .describe("2-letter US state / territory code (e.g. 'CA', 'LA', 'PR', 'DC'). Uppercase per FEMA; a bad code returns an honest empty (count 0).");
+// ISO date 'YYYY-MM-DD' or a full ISO datetime. A bare date = midnight-UTC start.
+const FemaDateSchema = z
+    .string()
+    .regex(/^\d{4}-\d{2}-\d{2}(T\d{2}:\d{2}:\d{2}(\.\d{1,3})?Z?)?$/)
+    .describe("ISO date 'YYYY-MM-DD' or full ISO datetime; a bare date is midnight-UTC start of that day.");
+const FemaSearchPublicAssistanceInput = z.object({
+    state: FemaStateSchema.optional().describe("Filter by applicant state (→ stateAbbreviation eq 'XX'). 2-letter code."),
+    disasterNumber: z
+        .number()
+        .int()
+        .positive()
+        .optional()
+        .describe("Filter by FEMA disaster number (→ disasterNumber eq N)."),
+    applicantId: z
+        .string()
+        .min(1)
+        .optional()
+        .describe("Filter by applicant id (→ applicantId eq '...'). e.g. '015-UF5E0-00'."),
+    damageCategoryCode: z
+        .string()
+        .min(1)
+        .max(2)
+        .optional()
+        .describe("Filter by PA damage category code (→ damageCategoryCode eq 'X'). e.g. 'B' = Emergency Protective Measures, 'C'–'G' = permanent work."),
+    incidentType: z
+        .string()
+        .min(1)
+        .optional()
+        .describe("Filter by incident type (→ incidentType eq '...'). e.g. 'Flood', 'Hurricane', 'Severe Storm'."),
+    minProjectAmount: z
+        .number()
+        .optional()
+        .describe("Minimum project amount (→ projectAmount ge N)."),
+    maxProjectAmount: z
+        .number()
+        .optional()
+        .describe("Maximum project amount (→ projectAmount le N)."),
+    declaredDateFrom: FemaDateSchema.optional().describe("Earliest declaration date, inclusive (→ declarationDate ge 'ISO')."),
+    declaredDateTo: FemaDateSchema.optional().describe("Latest declaration date, inclusive (→ declarationDate le 'ISO')."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("Rows per page ($top), 1..1000, default 100."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .default(0)
+        .describe("0-based row offset ($skip) for pagination, default 0."),
+});
+const FemaDisasterDeclarationsInput = z.object({
+    state: FemaStateSchema.optional().describe("Filter by state (→ state eq 'XX'). 2-letter code."),
+    incidentType: z
+        .string()
+        .min(1)
+        .optional()
+        .describe("Filter by incident type (→ incidentType eq '...'). e.g. 'Flood', 'Hurricane', 'Winter Storm'."),
+    declarationType: z
+        .enum(["DR", "EM", "FM"])
+        .optional()
+        .describe("Filter by declaration type (→ declarationType eq 'XX'): DR (major disaster), EM (emergency), FM (fire management)."),
+    fyDeclared: z
+        .number()
+        .int()
+        .optional()
+        .describe("Filter by fiscal year declared (→ fyDeclared eq N). e.g. 2024."),
+    disasterNumber: z
+        .number()
+        .int()
+        .positive()
+        .optional()
+        .describe("Filter by FEMA disaster number (→ disasterNumber eq N)."),
+    declaredDateFrom: FemaDateSchema.optional().describe("Earliest declaration date, inclusive (→ declarationDate ge 'ISO')."),
+    declaredDateTo: FemaDateSchema.optional().describe("Latest declaration date, inclusive (→ declarationDate le 'ISO')."),
+    paProgramDeclared: z
+        .boolean()
+        .optional()
+        .describe("Filter to declarations where the Public Assistance program was declared (→ paProgramDeclared eq true/false)."),
+    iaProgramDeclared: z
+        .boolean()
+        .optional()
+        .describe("Filter to declarations where the Individual Assistance program was declared (→ iaProgramDeclared eq true/false)."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("Rows per page ($top), 1..1000, default 100."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .default(0)
+        .describe("0-based row offset ($skip) for pagination, default 0."),
+});
+// ─── EPA ECHO REST (keyless facility compliance/enforcement) — input schemas ──
+// ADR-0009. KEYLESS, single fixed host (echodata.epa.gov) + three fixed service
+// paths (the SSRF core — no free host/path). `state` is a curated US state/
+// territory ENUM: it scopes the query (national is ~5.6M rows) AND is the
+// silent-zero guard (ECHO does NOT validate filter VALUES, so an unknown value
+// returns QueryRows:0 — indistinguishable from a genuine-empty). M2 (LIVE-verified
+// 2026-07-12): `sic` DOES narrow (a real filter); `naics` is DROPPED upstream —
+// exposed as BEST-EFFORT, disclosed in _meta.filtersDropped + notes, never
+// silently presented as filtered.
+const EchoStateEnum = z
+    .enum(echo.ECHO_STATES)
+    .describe("US state / territory 2-letter code to scope the search (REQUIRED — an unscoped national query is ~5.6M rows; the enum is also the SSRF value guard + the silent-zero guard). e.g. 'DC', 'TX', 'CA', 'PR'.");
+const EchoSearchFacilitiesInput = z.object({
+    state: EchoStateEnum,
+    naics: z
+        .string()
+        .regex(/^[0-9]{2,6}$/)
+        .optional()
+        .describe("BEST-EFFORT industry filter (2–6 digit NAICS). WARNING: ECHO DROPS the NAICS filter upstream (live-verified 2026-07-12) — the returned facilities are NOT guaranteed to match this code; it is reported in _meta.filtersDropped + a note. Use `sic` (which DOES narrow) to scope by industry."),
+    sic: z
+        .string()
+        .regex(/^[0-9]{2,4}$/)
+        .optional()
+        .describe("Industry filter (2–4 digit SIC code). A REAL filter — ECHO narrows by SIC (live-verified). A code with no facilities returns 0 (silent-zero — verify the code)."),
+    facilityName: z
+        .string()
+        .min(1)
+        .optional()
+        .describe("Facility-name substring filter (p_fn). NOTE: not validated by ECHO — a typo silently returns 0 results, not an error."),
+    majorOnly: z
+        .boolean()
+        .optional()
+        .describe("true ⇒ only EPA 'major' facilities (p_maj=Y)."),
+    federalOnly: z
+        .boolean()
+        .optional()
+        .describe("true ⇒ only federal facilities (p_ff=Y)."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("Facilities per page (→ responseset), 1..1000, default 100."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .default(0)
+        .describe("0-based offset for pagination, default 0. MUST be an exact multiple of `limit` (ECHO pages on fixed boundaries; a non-multiple ⇒ invalid_input)."),
+});
+const EchoFacilityReportInput = z.object({
+    registryId: z
+        .string()
+        .regex(/^[0-9]{9,12}$/)
+        .describe("The facility's FRS RegistryID (from echo_search_facilities rows' RegistryID) — an all-digit id, 9–12 digits (e.g. '110059768461'). A bad/unknown id ⇒ not_found (never a fabricated report)."),
+});
+// ─── api.data.gov keyed trio (Regulations.gov + Congress.gov) — input schemas ──
+// ADR-0007. The project's FIRST KEYED source. The key is read from env
+// (DATA_GOV_API_KEY, else the public DEMO_KEY) and travels ONLY in the X-Api-Key
+// header — NEVER in a param here. M4 (Zod-first): page/limit bounds + sort/type
+// enums + ISO date-time formats are validated LOCALLY so bad params fail as
+// invalid_input BEFORE any fetch (Regulations.gov's page[number] is HARD-capped at
+// 40 = a 10,000-record ceiling; page[size] must be 5..250).
+const RegulationsSearchInput = z.object({
+    searchTerm: z
+        .string()
+        .optional()
+        .describe("Full-text search term (filter[searchTerm]), e.g. 'artificial intelligence'."),
+    query: z
+        .string()
+        .optional()
+        .describe("Alias for `searchTerm` (either is accepted; both feed filter[searchTerm])."),
+    agencyId: z
+        .string()
+        .optional()
+        .describe("Filter by posting agency acronym (filter[agencyId]), e.g. 'EPA', 'FDA'."),
+    docketId: z
+        .string()
+        .optional()
+        .describe("Filter by docket id (filter[docketId]), e.g. 'EPA-HQ-OAR-2021-0257'."),
+    documentType: z
+        .enum(datagov.REGULATIONS_DOCUMENT_TYPES)
+        .optional()
+        .describe("Filter by document type (documents only): Rule / Proposed Rule / Notice / Supporting & Related Material / Other."),
+    withinCommentPeriod: z
+        .boolean()
+        .optional()
+        .describe("true ⇒ only documents currently open for comment (documents only; filter[withinCommentPeriod])."),
+    postedDateGe: z
+        .string()
+        .regex(/^\d{4}-\d{2}-\d{2}$/)
+        .optional()
+        .describe("Posted on/after this date, YYYY-MM-DD (filter[postedDate][ge])."),
+    postedDateLe: z
+        .string()
+        .regex(/^\d{4}-\d{2}-\d{2}$/)
+        .optional()
+        .describe("Posted on/before this date, YYYY-MM-DD (filter[postedDate][le])."),
+    sort: z
+        .enum(datagov.REGULATIONS_SORTS)
+        .optional()
+        .describe("Sort order (default '-postedDate'). Live-verified set: -postedDate/postedDate/-lastModifiedDate/lastModifiedDate/-commentEndDate (non-exhaustive)."),
+    pageNumber: z
+        .number()
+        .int()
+        .min(1)
+        .max(40)
+        .default(1)
+        .describe("1-based page number, 1..40 (HARD cap — page[number] max is 40; the reachable window is 40×pageSize ≤ 10,000 records)."),
+    pageSize: z
+        .number()
+        .int()
+        .min(5)
+        .max(250)
+        .default(25)
+        .describe("Records per page (page[size]), 5..250, default 25."),
+});
+// ADR-0044 — Regulations.gov DOCKETS (the rulemaking/nonrulemaking CONTAINER +
+// its cross-source `rin` join key). Within-source depth on the SAME api.data.gov
+// keyed adapter. The docketType enum + ISO date formats fail LOCALLY as
+// invalid_input BEFORE any fetch; `limit` exposes a friendly count while the wire
+// page[size] floor of 5 is handled by the handler (a limit<5 is client-sliced).
+const RegulationsSearchDocketsInput = z.object({
+    searchTerm: z
+        .string()
+        .optional()
+        .describe("Full-text search term (filter[searchTerm]) over docket title/abstract, e.g. 'endangered species'."),
+    query: z
+        .string()
+        .optional()
+        .describe("Alias for `searchTerm` (either is accepted; both feed filter[searchTerm])."),
+    agencyId: z
+        .string()
+        .optional()
+        .describe("Filter by owning agency acronym (filter[agencyId]), e.g. 'EPA', 'BLM', 'TREAS-FINCEN'."),
+    docketType: z
+        .enum(datagov.REGULATIONS_DOCKET_TYPES)
+        .optional()
+        .describe("Filter by docket type: Rulemaking / Nonrulemaking (filter[docketType])."),
+    lastModifiedDateGe: z
+        .string()
+        .regex(/^\d{4}-\d{2}-\d{2}$/)
+        .optional()
+        .describe("Docket last modified on/after this date, YYYY-MM-DD (filter[lastModifiedDate][ge])."),
+    lastModifiedDateLe: z
+        .string()
+        .regex(/^\d{4}-\d{2}-\d{2}$/)
+        .optional()
+        .describe("Docket last modified on/before this date, YYYY-MM-DD (filter[lastModifiedDate][le])."),
+    sort: z
+        .enum(datagov.REGULATIONS_DOCKET_SORTS)
+        .optional()
+        .describe("Sort order (default '-lastModifiedDate'). Set: -lastModifiedDate/lastModifiedDate/title/-title (first two DEMO_KEY-verified)."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(250)
+        .default(20)
+        .describe("Requested rows, 1..250, default 20. NOTE: the API's page[size] floor is 5 — a limit<5 fetches page[size]=5 upstream and returns the first `limit` rows client-side (disclosed in _meta.notes); totalAvailable stays the EXACT server total."),
+    pageNumber: z
+        .number()
+        .int()
+        .min(1)
+        .max(40)
+        .default(1)
+        .describe("1-based page number, 1..40 (HARD cap — page[number] max is 40; the reachable window is 40×page[size] ≤ 10,000 records)."),
+});
+const RegulationsGetDocketInput = z.object({
+    docketId: z
+        .string()
+        .regex(/^[A-Za-z0-9_.-]+$/, "docketId may contain only letters, digits, '_', '.', '-' (no slashes/spaces/%).")
+        .refine((v) => /[A-Za-z0-9]/.test(v), "docketId must contain an alphanumeric")
+        .describe("The docket id — the ONLY path-segment value, charclass-validated (rejects '../', '%2F', spaces, pure-dot) — e.g. 'BLM-2026-0001', 'TREAS-FINCEN-2008-0008'. A bad id ⇒ invalid_input (0 fetch); a nonexistent id ⇒ not_found (never a fabricated docket)."),
+});
+const CongressSearchBillsInput = z.object({
+    query: z
+        .string()
+        .optional()
+        .describe("Keyword — NOTE: Congress.gov /v3/bill has NO keyword search, so this is NOT applied (disclosed in _meta.filtersDropped). Use congress/billType/date filters instead."),
+    congress: z
+        .number()
+        .int()
+        .min(1)
+        .max(999)
+        .optional()
+        .describe("Congress number, e.g. 118 (scopes the path to /v3/bill/{congress})."),
+    billType: z
+        .enum(datagov.CONGRESS_BILL_TYPES)
+        .optional()
+        .describe("Bill type: hr/s/hjres/sjres/hconres/sconres/hres/sres. Requires `congress` (path /v3/bill/{congress}/{billType})."),
+    fromDateTime: z
+        .string()
+        .datetime({ offset: true })
+        .optional()
+        .describe("Filter to bills updated at/after this ISO-8601 date-time with offset, e.g. '2024-01-01T00:00:00Z'."),
+    toDateTime: z
+        .string()
+        .datetime({ offset: true })
+        .optional()
+        .describe("Filter to bills updated at/before this ISO-8601 date-time with offset, e.g. '2024-12-31T23:59:59Z'."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .default(0)
+        .describe("0-based record offset for pagination, default 0."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(250)
+        .default(20)
+        .describe("Records per page, 1..250, default 20."),
+});
+const CongressGetBillInput = z.object({
+    congress: z
+        .number()
+        .int()
+        .min(1)
+        .max(999)
+        .describe("Congress number, e.g. 117."),
+    billType: z
+        .enum(datagov.CONGRESS_BILL_TYPES)
+        .describe("Bill type: hr/s/hjres/sjres/hconres/sconres/hres/sres."),
+    billNumber: z
+        .number()
+        .int()
+        .min(1)
+        .describe("Bill number, e.g. 3076 (for H.R.3076)."),
+});
+// ─── data.gov v4 Catalog API (api.gsa.gov — CKAN-retirement replacement) ─
+// ADR-0046, resilience Phase 3. Federal dataset DISCOVERY. Same DATA_GOV_API_KEY/
+// DEMO_KEY/X-Api-Key discipline as the datagov trio (shared datagovKey.ts seam),
+// but a DIFFERENT host (api.gsa.gov). The opaque `cursor` is charclass-validated
+// here AND re-guarded in the handler (a bad token ⇒ invalid_input, 0 fetch).
+const DatagovSearchDatasetsInput = z.object({
+    query: z
+        .string()
+        .min(1)
+        .max(500)
+        .optional()
+        .describe("Free-text search over the dataset catalog (→ _q), e.g. 'wildfire'. LIVE-CONFIRMED to narrow."),
+    organization: z
+        .string()
+        .min(1)
+        .max(200)
+        .optional()
+        .describe("Publisher organization SLUG filter (→ organization), e.g. 'epa-gov', 'noaa-gov'. An org catalog lists that agency's published datasets."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(100)
+        .default(20)
+        .describe("Datasets per page (→ _size), 1..100, default 20."),
+    cursor: z
+        .string()
+        .min(1)
+        .max(4096)
+        .regex(datagovCatalog.DATAGOV_CURSOR_RE)
+        .optional()
+        .describe("Opaque continuation cursor (→ after) — pass back the _meta.nextCursor from the previous page. Pagination is a cursor, NOT a numeric offset (offset/nextOffset are null); nextCursor:null means the last page. A bad token (spaces/'../'/'%') ⇒ invalid_input pre-fetch."),
+});
+// ─── GovInfo (api.govinfo.gov — the api.data.gov keyed trio's 3rd API) ─
+// ADR-0010. Same DATA_GOV_API_KEY/DEMO_KEY/X-Api-Key discipline as the datagov
+// trio (shared datagovKey.ts seam). `collection` is grammar-checked here AND
+// validated against the live /collections catalog in the handler (the silent-empty
+// guard). The novel piece is the OPAQUE `offsetMark` cursor: continuation rides in
+// _meta.nextCursor (passed back as pageMark), never a numeric offset.
+const GovinfoListCollectionsInput = z.object({});
+const GovinfoSearchPackagesInput = z.object({
+    collection: z
+        .string()
+        .regex(govinfo.GOVINFO_COLLECTION_RE, "collection must be an uppercase alpha GovInfo code like BILLS/CFR/FR/PLAW ([A-Z]{2,10}).")
+        .describe("GovInfo collection code (uppercase alpha), e.g. BILLS, PLAW, CREC, USCODE, CFR, FR, BUDGET, GAOREPORTS. Validated against the live /collections catalog — an unknown code returns invalid_input listing valid codes (never a misleading empty). Use govinfo_list_collections to discover codes."),
+    startDate: z
+        .string()
+        .regex(govinfo.GOVINFO_DATE_RE, "startDate must be YYYY-MM-DD or YYYY-MM-DDTHH:MM:SSZ.")
+        .describe("Lower bound on lastModified (the record's last-update date, NOT dateIssued), YYYY-MM-DD (normalized to T00:00:00Z) or a full ISO datetime. e.g. '2024-01-01'."),
+    endDate: z
+        .string()
+        .regex(govinfo.GOVINFO_DATE_RE, "endDate must be YYYY-MM-DD or YYYY-MM-DDTHH:MM:SSZ.")
+        .optional()
+        .describe("Optional upper bound on lastModified (same format as startDate)."),
+    pageSize: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(100)
+        .describe("Rows per page (upstream pageSize), 1..1000, default 100."),
+    pageMark: z
+        .string()
+        .regex(govinfo.GOVINFO_PAGE_MARK_RE, 'pageMark must be "*" or the opaque _meta.nextCursor from the previous page (≤4096 base64/URL-safe chars).')
+        .default("*")
+        .describe('Opaque continuation cursor. Default "*" (first page). To page, pass back the previous response\'s _meta.nextCursor (NOT a numeric offset — GovInfo uses an opaque cursor).'),
+});
+const GovinfoGetPackageInput = z.object({
+    packageId: z
+        .string()
+        .regex(govinfo.GOVINFO_PACKAGE_ID_RE, "packageId must be a GovInfo id like BILLS-118hr1enr / CFR-2023-title1-vol1 ([A-Za-z0-9][A-Za-z0-9._-]{2,}).")
+        .describe("GovInfo packageId from govinfo_search_packages (e.g. 'BILLS-118hr1enr', 'PLAW-117publ58', 'CFR-2023-title1-vol1', 'GAOREPORTS-GAO-24-106221')."),
+});
+// ─── FPDS-NG (www.fpds.gov ezSearch ATOM — keyless XML/ATOM) ─── ADR-0012
+// The FIRST XML/ATOM source. Structured filters ONLY (NO raw-q — a typo'd field
+// name is a SILENT ZERO in FPDS, so the module builds the fielded `q`). At least
+// one filter is REQUIRED (refuse a bare unbounded scan). Dates are ISO
+// YYYY-MM-DD (reformatted to YYYY/MM/DD internally). `offset` is the 0-indexed
+// page start (page size fixed at 10); keyless deep-paging past ~200K is
+// unreliable so it is capped at 600000.
+const FPDS_ISO_DATE = z
+    .string()
+    .regex(/^\d{4}-\d{2}-\d{2}$/, "date must be ISO YYYY-MM-DD.");
+const FpdsSearchAwardsInput = z
+    .object({
+    naics: z
+        .string()
+        .optional()
+        .describe("Principal NAICS code (→ PRINCIPAL_NAICS_CODE), e.g. '541511'."),
+    vendorName: z
+        .string()
+        .optional()
+        .describe("Vendor/contractor name phrase (→ VENDOR_NAME), e.g. 'LOCKHEED MARTIN'."),
+    piid: z
+        .string()
+        .optional()
+        .describe("Contract/order PIID (→ PIID) — returns that action's full base+mod chain."),
+    departmentId: z
+        .string()
+        .optional()
+        .describe("4-digit contracting DEPARTMENT_ID, e.g. '9700' (DoD), '4700' (GSA)."),
+    contractingAgencyName: z
+        .string()
+        .optional()
+        .describe("Contracting agency name phrase (→ CONTRACTING_AGENCY_NAME), e.g. 'DEPT OF DEFENSE'."),
+    signedDateFrom: FPDS_ISO_DATE.optional().describe("Signed-date range START (ISO YYYY-MM-DD); pair with signedDateTo (→ SIGNED_DATE:[from,to])."),
+    signedDateTo: FPDS_ISO_DATE.optional().describe("Signed-date range END (ISO YYYY-MM-DD); pair with signedDateFrom."),
+    lastModifiedFrom: FPDS_ISO_DATE.optional().describe("Last-modified range START (ISO YYYY-MM-DD); pair with lastModifiedTo (→ LAST_MOD_DATE:[from,to])."),
+    lastModifiedTo: FPDS_ISO_DATE.optional().describe("Last-modified range END (ISO YYYY-MM-DD); pair with lastModifiedFrom."),
+    keyword: z
+        .string()
+        .optional()
+        .describe("Free-text keyword (bare full-text term; FPDS FIELD: operators are stripped for safety)."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .max(600_000)
+        .default(0)
+        .describe("0-indexed page start (page size fixed at 10). Keyless deep-paging past ~200K is unreliable."),
+})
+    .refine((a) => a.naics !== undefined ||
+    a.vendorName !== undefined ||
+    a.piid !== undefined ||
+    a.departmentId !== undefined ||
+    a.contractingAgencyName !== undefined ||
+    (a.signedDateFrom !== undefined && a.signedDateTo !== undefined) ||
+    (a.lastModifiedFrom !== undefined && a.lastModifiedTo !== undefined) ||
+    a.keyword !== undefined, {
+    message: "Provide at least one filter (naics, vendorName, piid, departmentId, contractingAgencyName, a signedDate range, a lastModified range, or keyword) — a bare unbounded FPDS scan is refused.",
+});
+// ─── NIH RePORTER v2 (api.reporter.nih.gov — keyless POST/JSON) ─── ADR-0014
+// The R2 getJson port's FIRST non-GET consumer. A NEW capability axis: federal
+// research-GRANT funding footprint by organization / UEI / state (recipient
+// enrichment, joinable to SAM/USAspending via primary_uei). The SSRF surface is
+// a compile-time-CONSTANT URL; all filters ride in the MODULE-BUILT POST body.
+// Only LIVE-CONFIRMED-narrowing criteria are exposed (M1): orgStates / orgNames /
+// fiscalYears. agency_ic_codes is EXCLUDED (it silently no-ops upstream). The
+// 15,000-record retrieval window is enforced by offset .max(14_999) (offset ≥
+// 15,000 ⇒ invalid_input, 0 fetch) + limit .max(500) — a cap on RETRIEVAL, not
+// on the exact meta.total count.
+const NIH_CURRENT_YEAR = new Date().getUTCFullYear();
+const NihSearchProjectsInput = z.object({
+    orgStates: z
+        .array(z.enum(nih.NIH_ORG_STATES))
+        .max(20)
+        .optional()
+        .describe("Recipient-organization US state/territory 2-letter USPS codes (UPPERCASE — the enum is the SSRF value guard + the silent-zero guard: a lowercase 'ca' or an unknown 'ZZ' silently returns zeros, so a typo is an invalid_input, never read as 'no NIH funding'). LIVE-CONFIRMED to narrow. e.g. ['CA','MA']. Max 20."),
+    orgNames: z
+        .array(z.string().min(1).max(512))
+        .max(20)
+        .optional()
+        .describe("Recipient-organization name filter values (each ≤512 chars, max 20). LIVE-CONFIRMED to narrow. e.g. ['MASSACHUSETTS INSTITUTE OF TECHNOLOGY']. A value matching no org returns a genuine total:0."),
+    fiscalYears: z
+        .array(z.number().int().min(1985).max(NIH_CURRENT_YEAR + 1))
+        .max(20)
+        .optional()
+        .describe(`NIH fiscal years to include (int array, ${1985}..${NIH_CURRENT_YEAR + 1}, max 20). LIVE-CONFIRMED to narrow. e.g. [2023,2024].`),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(500)
+        .default(50)
+        .describe("Projects per page (upstream hard cap 500), 1..500, default 50."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .max(14_999)
+        .default(0)
+        .describe("0-based offset into the result set. HARD-CAPPED at 14,999: NIH caps keyless retrieval at the first 15,000 records (offset 0..14,999), so offset ≥ 15,000 is refused (invalid_input) — narrow criteria to reach records beyond the window. The count (totalAvailable) stays EXACT past the window."),
+});
+// ─── NSF Awards API (api.nsf.gov — keyless GET) ─── ADR-0020 (source #20)
+// The grant-SIBLING of NIH RePORTER on a DIFFERENT agency: federal research-GRANT
+// award records with recipient / PI / UEI enrichment, strengthening the WEAK
+// entity/recipient layer (ueiNumber/parentUeiNumber join to SAM/USAspending). The
+// SSRF surface is a compile-time-CONSTANT host+path; all filters ride in a
+// module-built URLSearchParams from a validated whitelist. Only LIVE-CONFIRMED-
+// narrowing filters ship (M1); agency + printFields are EXCLUDED (agency = the
+// NSF-only-corpus zeros-out foot-gun; printFields = a proven no-op). The 10,000-
+// record retrieval window (offset+rpp ≤ 10,000) is enforced by offset .max(9999)
+// + a module-side outgoing-rpp clamp. Dates are STRICT mm/dd/yyyy (a wrong format
+// is silently mis-parsed by NSF); a multi-word keyword is OR-tokenized (disclosed).
+const NSF_UEI_RE = /^[A-Za-z0-9]{12}$/;
+const NSF_MMDDYYYY_RE = /^(0[1-9]|1[0-2])\/(0[1-9]|[12]\d|3[01])\/\d{4}$/;
+const NsfSearchAwardsInput = z.object({
+    keyword: z
+        .string()
+        .min(1)
+        .max(200)
+        .optional()
+        .describe("Free-text search over title/abstract. NOTE: NSF OR-tokenizes a MULTI-WORD keyword (matches ANY word, not the phrase — 'machine learning' = machine OR learning, a far broader set; disclosed in _meta.notes). Use a single distinctive word or add a scoping filter for a precise set."),
+    awardeeStateCode: z
+        .enum(nsf.NSF_STATES)
+        .optional()
+        .describe("Awardee-organization US state/territory 2-letter USPS code (UPPERCASE — the enum is the SSRF value guard + the silent-zero guard: a non-state typo silently returns 0 awards on NSF, indistinguishable from 'no NSF funding', so it is an invalid_input). LIVE-CONFIRMED to narrow. e.g. 'CA'."),
+    awardeeName: z
+        .string()
+        .min(2)
+        .max(200)
+        .optional()
+        .describe("Awardee-organization name filter (2..200 chars). LIVE-CONFIRMED to narrow (a top recipient like 'Johns Hopkins University' may still saturate at the 10,000 count cap)."),
+    ueiNumber: z
+        .string()
+        .regex(NSF_UEI_RE)
+        .optional()
+        .describe("Awardee UEI — a 12-char alphanumeric SAM/USAspending Unique Entity ID (uppercase-normalized before sending). LIVE-CONFIRMED an EXACT recipient-graph filter (the clean SAM/USAspending join). e.g. 'FTMTDMBR29C7' (Johns Hopkins)."),
+    parentUeiNumber: z
+        .string()
+        .regex(NSF_UEI_RE)
+        .optional()
+        .describe("Parent-organization UEI — a 12-char alphanumeric UEI for the awardee's parent entity (uppercase-normalized). LIVE-CONFIRMED an EXACT narrow (the parent-org roll-up join). e.g. 'GS4PNKTRNKL3'."),
+    pdPIName: z
+        .string()
+        .min(2)
+        .max(120)
+        .optional()
+        .describe("Principal-investigator name filter (2..120 chars). LIVE-CONFIRMED to narrow. e.g. 'Bell'."),
+    dateStart: z
+        .string()
+        .regex(NSF_MMDDYYYY_RE)
+        .optional()
+        .describe("Award ACTION-date lower bound (the initial award/obligation date, NOT the project startDate — live-verified). STRICT mm/dd/yyyy; a wrong format (yyyy-mm-dd) is silently mis-parsed by NSF (not an error), so it is rejected. e.g. '01/01/2024'."),
+    dateEnd: z
+        .string()
+        .regex(NSF_MMDDYYYY_RE)
+        .optional()
+        .describe("Award ACTION-date upper bound. STRICT mm/dd/yyyy (same semantics/foot-gun as dateStart). e.g. '12/31/2024'."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(100)
+        .default(25)
+        .describe("Awards per page (→ NSF rpp), 1..100, default 25. The OUTGOING page size is clamped so offset+rpp ≤ 10,000 (crossing NSF's retrieval window triggers a FATAL)."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .max(9999)
+        .default(0)
+        .describe("0-based offset. HARD-CAPPED at 9,999: NSF caps keyless retrieval at the first 10,000 records (offset+rpp ≤ 10,000), so offset ≥ 10,000 is refused (invalid_input) — narrow criteria to bring the set under 10,000."),
+});
+const NsfGetAwardInput = z.object({
+    awardId: z
+        .string()
+        .regex(/^\d{5,9}$/)
+        .describe("NSF award id — an all-digit id (5..9 digits; NSF ids are 7-digit numeric, live-verified). Returns the ONE full award record INCLUDING abstractText; a nonexistent id ⇒ found:false (never a fabricated record). e.g. '2545697'."),
+});
+// ─── ClinicalTrials.gov API v2 (ADR-0021, source #21) ────────────
+const ClinicaltrialsSearchStudiesInput = z.object({
+    "query.term": z
+        .string()
+        .min(1)
+        .max(200)
+        .optional()
+        .describe("Broad free-text search across the study record. MULTI-WORD is AND-conjunctive — ALL tokens must co-occur ('breast cancer' = breast AND cancer; disclosed in _meta.notes). LIVE-CONFIRMED to narrow. e.g. 'cancer'."),
+    sponsor: z
+        .string()
+        .min(1)
+        .max(200)
+        .optional()
+        .describe("Lead-sponsor / collaborator NAME search (→ query.spons; a fuzzy full-text name search, NOT an exact-entity join — the name is free text, not a UEI). MULTI-WORD is AND-conjunctive. LIVE-CONFIRMED to narrow. e.g. 'Pfizer'."),
+    condition: z
+        .string()
+        .min(1)
+        .max(200)
+        .optional()
+        .describe("Condition / disease filter (→ query.cond). MULTI-WORD is AND-conjunctive. LIVE-CONFIRMED to narrow. e.g. 'diabetes'."),
+    location: z
+        .string()
+        .min(1)
+        .max(200)
+        .optional()
+        .describe("Study-location filter (→ query.locn), e.g. a country or city. LIVE-CONFIRMED to narrow. e.g. 'Germany'."),
+    overallStatus: z
+        .enum(clinicaltrials.CT_STATUSES)
+        .optional()
+        .describe("Recruitment/overall status (→ filter.overallStatus). A frozen 14-value enum (COMPLETED, RECRUITING, TERMINATED, …); an unlisted value LOUD-fails at HTTP 400 upstream, so it is rejected pre-fetch. LIVE-CONFIRMED to narrow. e.g. 'RECRUITING'."),
+    funderType: z
+        .enum(clinicaltrials.CT_FUNDER_TYPES)
+        .optional()
+        .describe("Funding-source facet (→ aggFilters=funderType:<v>) — the FEDERAL-funding axis. A frozen 4-value enum: nih, fed, industry, other (the B2G-relevant nih/fed narrow to federally-sponsored trials). An UNLISTED value silently returns totalCount:0 at HTTP 200 (a fake-empty trap), so it is rejected pre-fetch (invalid_input). funderType is an OVERLAPPING facet — counts MUST NOT be summed into a total."),
+    pageSize: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(20)
+        .describe("Studies per page, 1..1000, default 20 (ClinicalTrials.gov clamps a larger request to 1000)."),
+    pageToken: z
+        .string()
+        .min(1)
+        .max(4096)
+        .regex(clinicaltrials.CT_TOKEN_RE)
+        .optional()
+        .describe("Opaque continuation cursor — pass back the _meta.nextCursor from the previous page. Pagination is a cursor, NOT a numeric offset (offset/nextOffset are null); nextCursor:null means the last page. A bad token loud-fails at HTTP 400."),
 });
-const FedRegGetDocInput = z.object({
-    documentNumber: z
+const ClinicaltrialsGetStudyInput = z.object({
+    nctId: z
         .string()
-        .describe("Federal Register document number, e.g. '2026-08333'"),
+        .regex(clinicaltrials.CT_NCT_RE)
+        .describe("NCT id — the form NCT followed by exactly 8 digits (e.g. NCT02403869). Returns the ONE full study record INCLUDING briefSummary; a nonexistent id ⇒ found:false (never a fabricated record). Injection-safe (validated before the path is built)."),
 });
-const FedRegListAgenciesInput = z.object({
-    perPage: z.number().min(1).max(500).optional(),
+// ADR-0024 — the facet-counts field enum is DERIVED from clinicaltrials.CT_FACET_FIELDS
+// (single source of truth; the handler re-validates each element against the frozen
+// Set INLINE — the [ssrf] re-guard). 1..11 whitelisted ENUM fields, comma-joined
+// module-built into fields=<v1,v2,…> (NO raw passthrough); a non-member ⇒ invalid_input
+// pre-fetch (0 fetch). No filter/scope/page param (they HTTP-400 here).
+const ClinicaltrialsFacetCountsInput = z.object({
+    fields: z
+        .array(z.enum(clinicaltrials.CT_FACET_FIELDS))
+        .min(1)
+        .max(clinicaltrials.CT_FACET_FIELDS.length)
+        .describe("1..11 ClinicalTrials.gov ENUM facet fields (deduped in-handler): OverallStatus, StudyType, Phase, LeadSponsorClass (★ the funding-SOURCE-class distribution — NIH/FED/OTHER_GOV/INDUSTRY/…, distinct from the search tool's 4-value funderType filter), Sex, DesignAllocation, DesignPrimaryPurpose, DesignInterventionModel, DesignMasking, DesignObservationalModel, DesignTimePerspective. Each returns the EXACT whole-registry per-value study-count distribution. An unlisted field ⇒ invalid_input pre-fetch (0 fetch). Phase is ARRAY-valued (counts OVERLAP — see _meta)."),
 });
-// eCFR
-const EcfrSearchInput = z.object({
-    query: z.string(),
-    titleNumber: z
+// ─── USITC Harmonized Tariff Schedule (hts.usitc.gov — keyless REST) ─── ADR-0039
+// The IMPORT-TARIFF / supply-chain PRICE lane: a good's HTS classification + its
+// Column-1 General / Special (preferential/FTA) / Column-2 duty-rate TEXT + the
+// Chapter-99 additional-duty provisions. Fixed host `hts.usitc.gov` + FIXED path
+// `/reststop/search` (the SSRF core — no free host/path); the single `query` rides
+// `keyword=` via URLSearchParams (percent-encoded). ★M2 — a MINIMUM query floor
+// (≥3 non-whitespace chars) is enforced at BOTH the Zod boundary (below) and a
+// handler belt, so a 1–2 char query is rejected before the fetch (a single char can
+// serve 10,000–16,000+ rows / several MB). The full array is fetched once and paged
+// CLIENT-SIDE (the endpoint serves no total and IGNORES offset).
+const HtsLookupInput = z.object({
+    query: z
+        .string()
+        .trim()
+        .min(3)
+        .max(100)
+        .describe("REQUIRED — a KEYWORD (e.g. 'laptop', 'cotton shirt') OR an HTS number (e.g. '8471.30' or '8471.30.01.00'); both ride the `keyword=` search. Must be ≥3 non-whitespace chars (a 1–2 char/single-char fragment can make USITC serve 10,000–16,000+ rows / several MB). Returns the matching classification rows across the HTS hierarchy with the Column-1 General / Special / Column-2 duty-rate TEXT + Chapter-99 additional-duty provisions."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(200)
+        .default(50)
+        .describe("Rows per page (CLIENT-SIDE slice over the served array), 1..200, default 50."),
+    offset: z
         .number()
+        .int()
+        .min(0)
+        .default(0)
+        .describe("0-based row offset for CLIENT-SIDE pagination over the served array (the endpoint has no server-side pagination), default 0."),
+});
+// ─── BLS Public Data API v1/v2 (api.bls.gov — keyless POST/JSON) ─── ADR-0032
+// A NEW capability axis: the PRICING / ESCALATION layer (CPI-U & ECI drive EPA-
+// clause escalation; PPI benchmarks materials; CES gives labor-rate context). The
+// SECOND POST-batch getJson-port consumer (after NIH). SSRF surface = a compile-
+// time-CONSTANT host+path; seriesids ride in the module-built POST body. `series`
+// is a FROZEN 9-key curated enum (the SSRF value guard + the units-label source);
+// `seriesId` is the raw passthrough, charclass-validated ^[A-Z0-9]{1,20}$. Years
+// are bounded ints (1900..currentYear+1); the span is clamped to the tier cap
+// (v1 ~10y) BEFORE the fetch + disclosed. An OPTIONAL free BLS_API_KEY rides ONLY
+// in the POST body (v2, ~500/day) — never a URL/header/label/_meta/log.
+const BlsTimeseriesInput = z.object({
+    series: z
+        .array(z.enum(bls.BLS_SERIES_KEYS))
+        .max(bls.BLS_SERIES_KEYS.length)
         .optional()
-        .describe("CFR title (1-50). e.g. 48 = FAR (Federal Acquisition Regulation), 2 = Federal financial assistance."),
-    perPage: z.number().min(1).max(20).optional(),
+        .describe("One or more CURATED series enum keys (typo-proof; each carries a meaning + units label): cpi_u_all (CPI-U all items NSA, index), cpi_u_core (CPI-U core NSA, index), ppi_final_demand (PPI final demand NSA, index), eci_total_comp (ECI total comp — ★12-MO % CHANGE, not an index), eci_wages (ECI wages — ★12-MO % CHANGE), unemployment_rate (SA, percent), labor_force_participation (SA, percent), employment_total_nonfarm (SA, thousands of persons), avg_hourly_earnings (SA, dollars/hour). NSA CPI-U is the escalation/EPA-clause reference. At least one of series/seriesId is required; both may be combined."),
+    seriesId: z
+        .array(z.string().regex(/^[A-Z0-9]{1,20}$/))
+        .max(bls.BLS_SERIES_KEYS.length + 50)
+        .optional()
+        .describe("One or more RAW BLS series IDs (power-user passthrough for the un-curatable space — OEWS area×occupation, local-area unemployment LAUCN…, SA/regional CPI variants). Charclass ^[A-Z0-9]{1,20}$ (uppercase alnum; punctuation/whitespace/lowercase rejected — SSRF + 'verify the ID' honesty). A raw ID has units:null (consult BLS). A nonexistent/typo'd ID returns BLS success + empty data (the ambiguity is disclosed, not asserted as 'no data'). At least one of series/seriesId is required."),
+    startYear: z
+        .number()
+        .int()
+        .min(1900)
+        .max(bls.YEAR_MAX)
+        .optional()
+        .describe(`Inclusive start year (1900..${bls.YEAR_MAX}). Default: endYear − 9 (a ~10-year window). The span is CLAMPED to the active tier's cap (v1 ~10 years/query) BEFORE the request and disclosed in _meta.notes (never a silently truncated range).`),
+    endYear: z
+        .number()
+        .int()
+        .min(1900)
+        .max(bls.YEAR_MAX)
+        .optional()
+        .describe(`Inclusive end year (1900..${bls.YEAR_MAX}). Default: the current year. Must be ≥ startYear.`),
 });
-const EcfrListTitlesInput = z.object({});
-// Grants.gov
-const GrantsSearchInput = z.object({
-    keyword: z.string().optional(),
-    cfda: z.string().optional().describe("CFDA program number, e.g. '10.500'"),
-    agency: z
+// ─── BLS OEWS — occupational wage benchmarking (2nd tool on api.bls.gov) ── ADR-0033
+// The LEVEL layer next to bls_timeseries's ESCALATION layer: mean/median annual &
+// hourly wages + employment by SOC occupation × geography. OEWS series IDs are 25
+// chars — they EXCEED the bls_timeseries raw-seriesId cap (^[A-Z0-9]{1,20}$), so
+// this tool BUILDS the 25-char ID INTERNALLY from validated structured inputs
+// (area/occupation/datatype), reusing the same POST/JSON transport + honesty
+// layer. NO year input (OEWS serves only the latest annual release). The module
+// re-validates every component (belt-and-suspenders behind these schemas).
+const BlsOewsWagesInput = z.object({
+    occupation: z
+        .array(z.enum(bls.BLS_OEWS_OCCUPATION_KEYS))
+        .max(bls.BLS_OEWS_OCCUPATION_KEYS.length)
+        .optional()
+        .describe("One or more CURATED occupation enum keys (typo-proof; each carries an SOC + official label): all_occupations, software_developer (15-1252), computer_systems_analyst, info_security_analyst, management_analyst, project_mgmt_specialist, logistician, accountant_auditor, general_ops_manager, civil_engineer, electrical_engineer, mechanical_engineer, industrial_engineer, lawyer, technical_writer, admin_assistant. The ~830-SOC long tail is reachable via `soc`. At least one of occupation/soc is required."),
+    soc: z
+        .array(z.string().regex(/^\d{6}$/))
+        .max(50)
+        .optional()
+        .describe("One or more RAW 6-digit SOC codes (the long-tail passthrough) — HYPHENLESS (use 151252, not 15-1252; the hyphen is rejected). A raw soc that matches a curated occupation is auto-labeled; otherwise key/label are null. At least one of occupation/soc is required."),
+    area: z
+        .array(z.string())
+        .max(51)
+        .optional()
+        .describe('One or more geographies (default ["national"]). Each element is "national", a 2-letter USPS state code (e.g. CA, TX, DC — the curated state enum), OR a 5-digit CBSA metropolitan code (^\\d{5}$, e.g. 19100 for Dallas-Fort Worth). Resolved internally to the OEWS areatype + zero-padded area code; an unknown token is rejected (invalid_input, never a malformed series ID on the wire).'),
+    datatype: z
+        .array(z.enum(bls.BLS_OEWS_DATATYPE_KEYS))
+        .max(bls.BLS_OEWS_DATATYPE_KEYS.length)
+        .optional()
+        .describe('One or more measures (default ["annual_mean"]): annual_mean (dollars/year), annual_median (dollars/year), hourly_mean (dollars/hour), hourly_median (dollars/hour), employment (count jobs). Each row carries measure.units from this map (H3 — never mislabel).'),
+});
+// ─── BLS QCEW — county×NAICS market-size / wages / location-quotient ── ADR-0042
+// A THIRD BLS tool but a SECOND, DIFFERENT, keyless, un-rate-limited BLS DOMAIN:
+// the QCEW Open Data Access CSV files on data.bls.gov/cew (NOT the rate-limited
+// api.bls.gov/publicAPI timeseries API the two tools above share). SSRF surface =
+// a compile-time-CONSTANT host + charclass-validated path segments (year `^\d{4}$`,
+// quarter `^[1-4]$`, mode enum {area,industry}, area `^[0-9A-Za-z]{1,6}$`, industry
+// DIGIT-ONLY `^[0-9]{1,6}$` — a hyphenated NAICS 31-33 404s). Client-side filters
+// (ownership/aggregationLevel/sizeCode/narrow) NEVER touch the URL. Honesty crux:
+// the block/code/field-scoped disclosure→null (never 0). NO BLS_API_KEY on this
+// keyless path; a NEW self-throttle gate key ("bls_qcew"), NOT "bls".
+const BlsQcewInput = z.object({
+    mode: z
+        .enum(["area", "industry"])
+        .describe("REQUIRED — the slice shape: 'area' (all industries × ownership × aggregation levels for ONE area_fips) or 'industry' (all areas for ONE NAICS). A fixed enum interpolated as a LITERAL path segment."),
+    area: z
         .string()
+        .regex(/^[0-9A-Za-z]{1,6}$/)
         .optional()
-        .describe("Grants.gov agency code, e.g. 'DHS-FEMA'"),
-    oppNum: z.string().optional().describe("Specific opportunity number"),
-    oppStatuses: z
-        .array(z.enum(["forecasted", "posted", "closed", "archived"]))
+        .describe("The area_fips (^[0-9A-Za-z]{1,6}$): county 01005, statewide 01000, national US000, MSA C1018, CSA CS122. REQUIRED when mode=area (the path segment). When mode=industry it is an OPTIONAL client-side narrow (keep only rows for this area_fips)."),
+    industry: z
+        .string()
+        .regex(/^[0-9]{1,6}$/)
         .optional()
-        .describe("Defaults to forecasted+posted"),
-    rows: z.number().min(1).max(50).optional(),
+        .describe("The NAICS code (DIGIT-ONLY ^[0-9]{1,6}$): 5415, or the aggregate 10. REQUIRED when mode=industry (the path segment). When mode=area it is an OPTIONAL client-side narrow (keep only rows for this NAICS). A hyphenated NAICS supersector (31-33, 44-45) 404s on QCEW — pass its digit aggregate code, never the hyphenated form."),
+    year: z
+        .number()
+        .int()
+        .min(1990)
+        .max(bls.YEAR_MAX)
+        .describe(`REQUIRED — the 4-digit year (1990..${bls.YEAR_MAX}). QCEW Open Data coverage begins ~1990; a pre-coverage or future year is an honest per-tuple HTTP 404 (found:false), NOT zero establishments.`),
+    quarter: z
+        .enum(["1", "2", "3", "4"])
+        .describe("REQUIRED — the quarter '1'|'2'|'3'|'4' (all four live-servable). The annual 'a' is not enabled this build."),
+    ownership: z
+        .string()
+        .regex(/^[0-9A-Za-z]{1,3}$/)
+        .optional()
+        .describe("Optional CLIENT-SIDE filter on own_code (e.g. 0=Total, 1=Federal, 2=State, 3=Local, 5=Private). Never on the URL (no SSRF surface)."),
+    aggregationLevel: z
+        .string()
+        .regex(/^[0-9A-Za-z]{1,3}$/)
+        .optional()
+        .describe("Optional CLIENT-SIDE filter on agglvl_code (e.g. 70=total-all-industries, 78=6-digit-NAICS-by-ownership). Filter to ONE agglvl_code for a coherent, non-double-counted total."),
+    sizeCode: z
+        .string()
+        .regex(/^[0-9A-Za-z]{1,3}$/)
+        .optional()
+        .describe("Optional CLIENT-SIDE filter on size_code."),
+    limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(1000)
+        .default(50)
+        .describe("Rows per page (CLIENT-SIDE window over the fetched-once slice), 1..1000, default 50."),
+    offset: z
+        .number()
+        .int()
+        .min(0)
+        .default(0)
+        .describe("0-based row offset for CLIENT-SIDE pagination over the filtered set (QCEW has no server-side pagination), default 0."),
 });
-const GrantsGetInput = z.object({
-    opportunityId: z.string().describe("Grants.gov opportunity id (numeric string)"),
+// ─── US Census Geocoder (keyless source #22) — input schemas ──────
+// ADR-0023. KEYLESS, single fixed host (geocoding.geo.census.gov) + two fixed
+// endpoint paths (the SSRF core — no free host/path; NO id in the path). benchmark /
+// vintage are frozen enums (the Zod source of truth = census.CENSUS_BENCHMARKS /
+// census.CENSUS_VINTAGES). [M2] CENSUS_VINTAGES is the UNION of live-valid vintages
+// across the four benchmarks (25 distinct) so a VALID non-default pair is not Zod-
+// rejected; an INVALID (benchmark,vintage) pair fails-closed at HTTP 400. GEOIDs are
+// strings (leading zeros survive). Coordinate finiteness is re-guarded in the handler.
+const CensusBenchmarkEnum = z
+    .enum(census.CENSUS_BENCHMARKS)
+    .optional()
+    .describe("Address-range benchmark (default Public_AR_Current — a MOVING benchmark). One of Public_AR_Current / Public_AR_ACS2025 / Public_AR_LUCA / Public_AR_Census2020. vintage MUST be compatible with this benchmark (a matrix); an incompatible pair fails-closed with an HTTP 400.");
+const CensusVintageEnum = z
+    .enum(census.CENSUS_VINTAGES)
+    .optional()
+    .describe("Geography vintage (default Current_Current — a MOVING vintage; the same address may return a different tract/CD across cycles). The valid vintage set DEPENDS on the benchmark (a matrix — this enum is the UNION across all four benchmarks); an incompatible (benchmark, vintage) pair fails-closed with an HTTP 400 (invalid_input), never a silent mis-resolution. e.g. Census2020_Census2020 (with Public_AR_Census2020), Census2010_Current.");
+const CensusGeocodeAddressInput = z.object({
+    address: z
+        .string()
+        .min(1)
+        .max(500)
+        .describe("A one-line US address, e.g. '600 Dexter Ave, Montgomery, AL 36104'. An unmatched/under-specified address is NOT an error — it returns matches:[] / matchCount:0 (a genuine empty; add city, state, ZIP). An ambiguous address may return MULTIPLE matches, each with its own matchedAddress + geographies."),
+    benchmark: CensusBenchmarkEnum,
+    vintage: CensusVintageEnum,
 });
-// SAM.gov organization lookup (federal hierarchy)
-const SamLookupOrgInput = z.object({
-    organizationId: z.string().describe("SAM.gov federal-organization id (numeric)"),
+const CensusGeographiesByCoordinatesInput = z
+    .object({
+    longitude: z
+        .number()
+        .min(-180)
+        .max(180)
+        .optional()
+        .describe("Longitude (x), a finite number in [-180, 180]. Alias of `x`. e.g. -86.301883."),
+    latitude: z
+        .number()
+        .min(-90)
+        .max(90)
+        .optional()
+        .describe("Latitude (y), a finite number in [-90, 90]. Alias of `y`. e.g. 32.377612."),
+    x: z
+        .number()
+        .min(-180)
+        .max(180)
+        .optional()
+        .describe("Longitude — the Census API's own name for longitude (alias of `longitude`)."),
+    y: z
+        .number()
+        .min(-90)
+        .max(90)
+        .optional()
+        .describe("Latitude — the Census API's own name for latitude (alias of `latitude`)."),
+    benchmark: CensusBenchmarkEnum,
+    vintage: CensusVintageEnum,
+})
+    .refine((v) => v.longitude !== undefined || v.x !== undefined, {
+    message: "longitude (or its alias x) is required.",
+    path: ["longitude"],
+})
+    .refine((v) => v.latitude !== undefined || v.y !== undefined, {
+    message: "latitude (or its alias y) is required.",
+    path: ["latitude"],
 });
-const TOOLS = [
-    // ━━━ SAM.gov (5) ━━━
-    {
+// Build a ToolDef whose `handler` is type-checked against the schema's inferred
+// input `I` at the call site (e.g. `input.searchText` is known-present). The
+// `I` binding is erased to `any` in the ToolDef[] array, so entries without a
+// handler need not use this helper.
+function defineTool(d) {
+    return d;
+}
+// Exported for the offline registry-introspection fault fixtures (W3-1): the
+// harness asserts a tool's advertised inputSchema/description directly (e.g.
+// usas_search_expiring_contracts no longer carries `fiscalYear`; usas_search_awards
+// no longer promises "+ count"). Export-only — does NOT change tools/list output
+// or any dispatch behavior (main() stays entry-point-gated).
+export const TOOLS = [
+    // ━━━ SAM.gov (8) ━━━
+    defineTool({
         name: "sam_search_opportunities",
         description: "Search SAM.gov federal contracting opportunities (keyless HAL). Returns up to 50 active notices with title, agency, NAICS, noticeId. Use for discovery — narrow with NAICS / agency / set-aside / state.",
         inputSchema: SamSearchInput,
-    },
-    {
+        handler: async (input, { sam }) => {
+            const r = await sam.searchOpportunities({
+                ...input,
+                setAside: input.setAside,
+            });
+            // OUTAGE HONESTY (C19). r.degraded is set ONLY when EVERY access tier
+            // threw (HAL down / network / 5xx-after-retry) — a total outage, NOT a
+            // confirmed zero. searchOpportunities otherwise returns the real result
+            // (incl. a genuine 0). Emit an explicitly-incomplete `_meta`: we do NOT
+            // know the count (totalAvailable:null, NEVER 0), the source is flagged
+            // degraded, the data count is null (not a fake 0 that reads as a real
+            // count), and a note tells the AI to retry rather than conclude "no
+            // matching notices". The genuine-zero + healthy paths below are UNCHANGED.
+            if (r.degraded) {
+                return withMeta({
+                    totalRecords: null,
+                    returned: 0,
+                    opportunities: [],
+                }, {
+                    source: "sam.gov/sgs/v1 (keyless HAL) (DEGRADED — search backend unavailable)",
+                    keylessMode: sam.isKeyless,
+                    complete: false,
+                    totalAvailable: null,
+                    returned: 0,
+                    filtersApplied: [],
+                    filtersDropped: [],
+                    fieldsUnavailable: [],
+                    notes: [
+                        r.degraded.reason +
+                            " This is a service outage, not a confirmed zero — retry.",
+                    ],
+                });
+            }
+            const data = {
+                totalRecords: r.totalRecords,
+                returned: r.opportunitiesData.length,
+                opportunities: r.opportunitiesData.map((o) => ({
+                    noticeId: o.noticeId,
+                    title: o.title,
+                    agency: o.fullParentPathName,
+                    solicitationNumber: o.solicitationNumber,
+                    responseDeadline: o.responseDeadLine,
+                    naics: o.naicsCode,
+                    setAside: o.typeOfSetAside,
+                    uiLink: o.uiLink,
+                })),
+            };
+            // A1 — filter honesty. Contrary to the earlier assumption, the keyless
+            // HAL list endpoint DOES honor the structured facets server-side —
+            // VERIFIED LIVE (2026-07): `naics`, `set_aside`, `pop_state` and `q` each
+            // narrow the result set AND every returned notice's detail matches the
+            // filter (the earlier "ignores facets" reading tested the WRONG param
+            // name `ncode`, which is silently dropped; the real param is `naics`).
+            // The one facet with no keyless param is organization-name (ignored).
+            // Separately, the list PAYLOAD still omits each notice's
+            // naics/set-aside/place-of-performance VALUES (null even when the filter
+            // applied), so filtering is real but reading those field values needs
+            // sam_get_opportunity. See spec §1.2 A1, §2.4.
+            if (sam.isKeyless) {
+                // NOTE: these truthiness checks must mirror EXACTLY the conditions under
+                // which client.searchPublic() actually appends each param (it uses
+                // `if (filters.x)` / `filters.setAside?.length`). Using `!== undefined`
+                // here would over-report: an empty-string facet is not sent by the
+                // client, so it must not appear in filtersApplied.
+                const filtersApplied = [];
+                if (input.query)
+                    filtersApplied.push("query");
+                if (input.ncode)
+                    filtersApplied.push("ncode");
+                if ((input.setAside?.length ?? 0) > 0)
+                    filtersApplied.push("setAside");
+                if (input.state)
+                    filtersApplied.push("state");
+                // organization-name is the only requested facet the keyless endpoint
+                // cannot honor — flag it dropped so results aren't read as org-filtered.
+                const filtersDropped = input.organizationName
+                    ? ["organizationName"]
+                    : [];
+                // ── GSA-CSV inline enrichment (opt-in, non-blocking) ──────────────
+                // The keyless HAL list payload nulls each row's naics/setAside/PoP/
+                // deadline/type. When the GSA-CSV backbone is ENABLED and its index is
+                // already warm, fill those nulls from today's snapshot in ONE lookup —
+                // instead of N sam_get_opportunity detail calls. HARD guarantees:
+                //   - DISABLED (default) → resolveCsvConfig().enabled is false, so we
+                //     never enter this branch: `data`, fieldsUnavailable and notes are
+                //     byte-for-byte the pre-enrichment behavior, and ZERO network hits
+                //     the CSV.
+                //   - NON-BLOCKING → tryGetReadyIndex returns the already-loaded index
+                //     or null immediately (kicking a background warm); a cold CSV NEVER
+                //     stalls the search on a 225 MB download.
+                //   - A CSV error can't fail the search: tryGetReadyIndex swallows and
+                //     returns null → we degrade to the un-enriched page + a note.
+                const csvCfg = gsaCsv.resolveCsvConfig();
+                // Widen the opportunities element type so the enriched rows (which may
+                // carry the added type/placeOfPerformance keys) are assignable; the
+                // original `data` (narrower) widens into this cleanly.
+                let enrichedData = data;
+                // The fields still null after enrichment (rebuilt truthfully below).
+                let fieldsUnavailable = ["naics", "setAside", "placeOfPerformance"];
+                const enrichmentNotes = [];
+                let source = "sam.gov/sgs/v1 (keyless HAL)";
+                let freshness = undefined;
+                if (csvCfg.enabled) {
+                    const ready = data.returned > 0 ? gsaCsv.tryGetReadyIndex(csvCfg) : null;
+                    if (ready) {
+                        const outcome = gsaCsv.enrichSearchOpportunities(enrichedData.opportunities, ready);
+                        enrichedData = { ...data, opportunities: outcome.opportunities };
+                        freshness = outcome.freshness;
+                        source = "sam.gov/sgs/v1 (keyless HAL) + gsa-csv (daily bulk CSV snapshot)";
+                        // Rebuild fieldsUnavailable: a field is only "unavailable" if it was
+                        // NOT filled on the whole page. Fields filled from the CSV drop off.
+                        // (naics/setAside/placeOfPerformance are the originally-null trio.)
+                        fieldsUnavailable = ["naics", "setAside", "placeOfPerformance"].filter((f) => !outcome.fieldsFilled.has(f));
+                        const filledList = [...outcome.fieldsFilled];
+                        if (filledList.length > 0) {
+                            enrichmentNotes.push(`naics/set-aside/place-of-performance for results present in today's GSA CSV snapshot were enriched from the GSA daily bulk CSV (source: gsa-csv) — filled fields this page: ${filledList.join(", ")}. set-aside here is the CSV short code (e.g. 'SBA') that matches sam_get_opportunity's setAside. Confirm real-time values (e.g. a just-amended deadline) with sam_get_opportunity.`);
+                        }
+                        else {
+                            enrichmentNotes.push("GSA-CSV enrichment ran but filled no fields on this page (the matched snapshot rows carried no non-empty naics/set-aside/place-of-performance) — values remain null; fetch sam_get_opportunity.");
+                        }
+                        if (outcome.missingCount > 0) {
+                            enrichmentNotes.push(`${outcome.missingCount} of ${data.returned} results were not in the current CSV snapshot (too new or archived) — their naics/set-aside/PoP remain null; fetch sam_get_opportunity for those noticeIds.`);
+                        }
+                        enrichmentNotes.push(`GSA CSV freshness — snapshot last-modified: ${outcome.freshness.csvLastModified ?? "unknown"}; index built: ${outcome.freshness.indexBuiltAt}; index age: ${outcome.freshness.indexAgeHours ?? "unknown"}h. The snapshot can lag the live HAL by up to ~24h.`);
+                    }
+                    else if (data.returned > 0) {
+                        // Enabled, rows exist that COULD be enriched, but the index isn't
+                        // warm yet (cold cache / background refresh in flight). Return the
+                        // normal HAL page un-enriched and disclose the pending warm — never
+                        // block on the download. Gated on returned>0: on a genuinely-empty
+                        // (returned===0) page there are NO rows to enrich, so a "retry for an
+                        // enriched page" note would be misleading — a retry cannot add rows.
+                        // That case falls through with the plain un-enriched source/notes
+                        // (the empty page is a complete, honest result).
+                        source = "sam.gov/sgs/v1 (keyless HAL) + gsa-csv (index warming)";
+                        enrichmentNotes.push("GSA-CSV enrichment pending — the CSV index is warming (a background download/build was kicked off); naics/set-aside/place-of-performance were NOT enriched this call. Retry shortly for an enriched page, or fetch sam_get_opportunity now.");
+                    }
+                }
+                const notes = [];
+                if (filtersApplied.length > 0) {
+                    notes.push("Keyless SAM search filtered server-side by the applied facets (NAICS/set-aside/place-of-performance state/keyword) — the result count reflects them. But the keyless list payload OMITS each notice's naics/setAside/placeOfPerformance VALUES (null here); call sam_get_opportunity on a noticeId to read those values.");
+                }
+                else {
+                    notes.push("naics/setAside/placeOfPerformance are null because the keyless list endpoint omits those values — call sam_get_opportunity for a notice to obtain them.");
+                }
+                if (filtersDropped.length > 0) {
+                    notes.push("The organization-name filter is NOT supported by the keyless endpoint and was ignored (results are unfiltered on organization). Set SAM_GOV_API_KEY to filter by organization, or filter client-side on the returned `agency` field.");
+                }
+                notes.push(...enrichmentNotes);
+                // freshness is surfaced structurally in `data` (the ResponseMeta type
+                // has no typed freshness field, mirroring sam_lookup_notice_fields) —
+                // present only when enrichment actually ran.
+                const dataOut = freshness !== undefined
+                    ? { ...enrichedData, freshness }
+                    : enrichedData;
+                return withMeta(dataOut, {
+                    source,
+                    keylessMode: true,
+                    truncated: r.totalRecords > data.returned,
+                    returned: data.returned,
+                    totalAvailable: r.totalRecords,
+                    filtersApplied,
+                    filtersDropped,
+                    fieldsUnavailable,
+                    notes,
+                });
+            }
+            // Keyed path: api.sam.gov honors the structured filters and populates
+            // the fields, so nothing is dropped or unavailable.
+            return withMeta(data, {
+                source: "api.sam.gov/opportunities/v2 (keyed)",
+                keylessMode: false,
+                truncated: r.totalRecords > data.returned,
+                returned: data.returned,
+                totalAvailable: r.totalRecords,
+                filtersApplied: [],
+                filtersDropped: [],
+                fieldsUnavailable: [],
+            });
+        },
+    }),
+    defineTool({
+        name: "sam_search_shaping",
+        description: "PRE-SOLICITATION shaping radar (keyless HAL). Surfaces Sources Sought / Presolicitation / Special Notices BEFORE the RFP exists — the free, real-time analogue of paid agency-forecast feeds. Closes the pre-solicitation lifecycle gap: catch a requirement while it's still shapeable (submit capabilities, influence NAICS/set-aside/PWS). Defaults to noticeType ['r','p','s']; opt into k/i/u for combined-synopsis / intent-to-bundle / J&A tells. Each notice carries noticeTypeCode (rank r/p over s), postedDate, responseDeadline + daysUntilResponse (null when no deadline — counted, not hidden), and a uiLink. HONEST KEYLESS LIMITS: naics/setAside/placeOfPerformance are null in the list rows (call sam_get_opportunity(noticeId) for those); and a responseDeadlineFrom/To window is applied CLIENT-SIDE over the fetched page (the feed ignores rdlfrom/rdlto) and disclosed in _meta. data.totalRecords is the TRUE server-side count for the type+facet filter.",
+        inputSchema: SamSearchShapingInput,
+        handler: async (input, { sam }) => {
+            // Default shaping window = Sources Sought + Presolicitation + Special
+            // Notice. These are the notice types that exist BEFORE an RFP — the
+            // whole point of the radar.
+            const noticeType = input.noticeType ?? ["r", "p", "s"];
+            const wantWindow = input.responseDeadlineFrom !== undefined ||
+                input.responseDeadlineTo !== undefined;
+            // Map noticeType → filters.ptype so the client sends the keyless
+            // `notice_type` facet (server-side filter, VERIFIED LIVE). We deliberately
+            // do NOT pass responseDeadlineFrom/To to the client — the keyless feed
+            // IGNORES rdlfrom/rdlto, so the window is applied client-side below and
+            // disclosed. activeOnly is honored by searchPublic's is_active=true.
+            const r = await sam.searchOpportunities({
+                query: input.query,
+                ncode: input.ncode,
+                organizationName: input.organizationName,
+                state: input.state,
+                setAside: input.setAside,
+                ptype: noticeType,
+                limit: input.limit ?? 25,
+            });
+            // OUTAGE HONESTY (C19). r.degraded ⇒ the keyless feed was totally down
+            // (all tiers threw), NOT a genuine "no shaping notices". Emit an
+            // explicitly-incomplete `_meta` (complete:false, totalAvailable:null,
+            // degraded source, retry note) and a null data count instead of the
+            // silent "0 pre-solicitation notices, complete" lie. noticeTypesRequested
+            // is still echoed so the caller knows what was attempted. The genuine-zero
+            // path + the client-side response-deadline window disclosure below are
+            // UNCHANGED.
+            if (r.degraded) {
+                return withMeta({
+                    totalRecords: null,
+                    returned: 0,
+                    noticeTypesRequested: noticeType,
+                    notices: [],
+                }, {
+                    source: "sam.gov/api/prod/sgs/v1/search (keyless HAL, notice_type filter) (DEGRADED — search backend unavailable)",
+                    keylessMode: true,
+                    complete: false,
+                    totalAvailable: null,
+                    returned: 0,
+                    filtersApplied: [],
+                    filtersDropped: [],
+                    fieldsUnavailable: [],
+                    notes: [
+                        r.degraded.reason +
+                            " This is a service outage, not a confirmed zero — retry.",
+                    ],
+                });
+            }
+            // Shape each keyless list row. naics/setAside/PoP are NULL in the keyless
+            // list payload (fieldsUnavailable) — NOT fabricated. noticeTypeCode
+            // (type.code) lets the AI rank r/p over s; daysUntilResponse is a whole-day
+            // count (null when no deadline — counted, not hidden).
+            const now = new Date();
+            const allNotices = r.opportunitiesData.map((o) => ({
+                noticeId: o.noticeId,
+                title: o.title,
+                noticeType: o.type ?? null, // type.value (human label)
+                noticeTypeCode: o.baseType ?? null, // type.code (r/p/s/…)
+                agency: o.fullParentPathName,
+                solicitationNumber: o.solicitationNumber,
+                postedDate: o.postedDate,
+                responseDeadline: o.responseDeadLine ?? null,
+                daysUntilResponse: daysUntilResponse(o.responseDeadLine, now),
+                naics: o.naicsCode, // null in keyless list rows
+                setAside: o.typeOfSetAside, // null in keyless list rows
+                uiLink: o.uiLink,
+            }));
+            // Response-deadline WINDOW — CLIENT-SIDE over the fetched page (the feed
+            // ignores rdlfrom/rdlto). A notice with no deadline is excluded from a
+            // windowed query. Disclosed via filtersDropped + a note below.
+            const notices = wantWindow
+                ? applyResponseDeadlineWindow(allNotices, input.responseDeadlineFrom, input.responseDeadlineTo)
+                : allNotices;
+            const data = {
+                totalRecords: r.totalRecords, // TRUE server-side count for type+facets
+                returned: notices.length,
+                noticeTypesRequested: noticeType,
+                notices,
+            };
+            // _meta honesty. filtersApplied lists what the FEED honored server-side
+            // (mirror EXACTLY searchPublic's append conditions). Always: noticeType.
+            const filtersApplied = ["noticeType"];
+            if (input.query)
+                filtersApplied.push("query");
+            if (input.ncode)
+                filtersApplied.push("ncode");
+            if ((input.setAside?.length ?? 0) > 0)
+                filtersApplied.push("setAside");
+            if (input.state)
+                filtersApplied.push("state");
+            // filtersDropped: organization-name has NO keyless param (ignored), and a
+            // requested response-deadline window is applied client-side (feed ignores
+            // rdlfrom/rdlto) — both must be disclosed so the AI never treats the page
+            // as server-filtered on them.
+            const filtersDropped = [];
+            if (input.organizationName)
+                filtersDropped.push("organizationName");
+            if (wantWindow)
+                filtersDropped.push("responseDeadline");
+            const notes = [
+                "Pre-solicitation shaping radar: notice_type is filtered SERVER-SIDE by the keyless feed (r=Sources Sought, p=Presolicitation, s=Special Notice by default; k/i/u opt-in). totalRecords is the TRUE server-side count for the type+facet filter.",
+            ];
+            if (wantWindow) {
+                notes.push("response-deadline window applied client-side over the fetched page (the keyless feed ignores rdlfrom/rdlto); widen limit or narrow via NAICS/agency for completeness. Notices with no deadline are excluded from a windowed query.");
+            }
+            if (input.organizationName) {
+                notes.push("The organization-name filter is NOT supported by the keyless endpoint and was ignored (results are unfiltered on organization). Filter client-side on the returned `agency`, or set SAM_GOV_API_KEY.");
+            }
+            notes.push("naics/setAside/placeOfPerformance are null in the keyless list rows — call sam_get_opportunity(noticeId) for per-notice NAICS/set-aside/place-of-performance.");
+            // truncated when the server has more than we returned OR a client-side
+            // deadline window trimmed the page (either way the caller isn't seeing the
+            // complete in-scope set).
+            const truncated = r.totalRecords > data.returned ||
+                (wantWindow && allNotices.length !== notices.length);
+            return withMeta(data, {
+                source: "sam.gov/api/prod/sgs/v1/search (keyless HAL, notice_type filter)",
+                keylessMode: true,
+                truncated,
+                returned: data.returned,
+                totalAvailable: r.totalRecords,
+                filtersApplied,
+                filtersDropped,
+                fieldsUnavailable: ["naics", "setAside", "placeOfPerformance"],
+                notes,
+            });
+        },
+    }),
+    defineTool({
         name: "sam_get_opportunity",
         description: "Fetch full detail for a single SAM.gov notice by 32-char hex noticeId. Returns title, agency, solicitation #, POCs, response deadline, attachments (with download URLs), inline description body. Call BEFORE drafting bid/no-bid or compliance work.",
         inputSchema: SamGetOpportunityInput,
-    },
-    {
+        handler: async (input, { sam }) => {
+            const { noticeId } = input;
+            const o = await sam.getOpportunity(noticeId);
+            if (!o)
+                return { found: false, noticeId };
+            const data = {
+                found: true,
+                noticeId: o.noticeId,
+                title: o.title,
+                agency: o.fullParentPathName,
+                solicitationNumber: o.solicitationNumber,
+                responseDeadline: o.responseDeadLine,
+                type: o.type,
+                naics: o.naicsCode,
+                setAside: o.typeOfSetAside,
+                placeOfPerformance: o.placeOfPerformance,
+                pointsOfContact: o.pointOfContact ?? [],
+                description: o.description,
+                attachments: (o.resourceLinks ?? []).map((url, idx) => ({
+                    index: idx,
+                    url,
+                })),
+                uiLink: o.uiLink,
+            };
+            // A HEALTHY notice returns EXACTLY as before (plain object → the server
+            // synthesizes a default complete:true `_meta`) — no crying wolf. ONLY
+            // when an enrichment sub-fetch DEGRADED (an outage, not a genuine empty)
+            // do we attach a degraded `_meta` disclosing that the empty field is
+            // UNKNOWN, not confirmed-absent. One failing sub-fetch does not flag the
+            // other: notes carry exactly one entry per degraded bucket.
+            if (o.enrichmentDegraded?.length) {
+                const notes = o.enrichmentDegraded.map((bucket) => bucket === "attachments"
+                    ? "The attachment list could not be fetched (a service issue) — this notice MAY have attachments not shown here; retry. This is NOT a confirmation it has none."
+                    : "The awarding-organization path could not be resolved (a service issue) — it is unavailable here, not absent.");
+                return withMeta(data, {
+                    source: sam.isKeyless ? "sam.gov (keyless)" : "api.sam.gov (keyed)",
+                    keylessMode: sam.isKeyless,
+                    complete: false,
+                    degraded: {
+                        attempted: o.enrichmentDegraded.length,
+                        succeeded: 0,
+                        failed: o.enrichmentDegraded.length,
+                    },
+                    notes,
+                });
+            }
+            return data;
+        },
+    }),
+    defineTool({
         name: "sam_fetch_description",
         description: "Return the full description / RFP body text for a notice as plain text. Useful when sam_get_opportunity returned a description URL instead of inline body, or for an LLM-friendly text dump.",
         inputSchema: SamFetchDescriptionInput,
-    },
-    {
+        handler: async (input, { sam }) => {
+            const { noticeId } = input;
+            const o = await sam.getOpportunity(noticeId);
+            if (!o)
+                return { found: false, noticeId };
+            const text = o.description
+                ? await sam.fetchOpportunityDescription(o.description)
+                : "";
+            return {
+                found: true,
+                noticeId,
+                descriptionLength: text.length,
+                description: text || "(no description body available)",
+            };
+        },
+    }),
+    defineTool({
         name: "sam_attachment_url",
         description: "Build the public download URL for an attachment resourceId. The URL returns a 303 redirect to a signed S3 URL — fetch with redirect:'follow' to get the file bytes.",
         inputSchema: SamAttachmentUrlInput,
-    },
-    {
+        handler: async (input, { sam }) => ({
+            downloadUrl: sam.publicDownloadUrl(input.resourceId),
+        }),
+    }),
+    defineTool({
+        name: "sam_fetch_attachment_text",
+        description: "Extract the TEXT of a SAM notice attachment (the actual RFP / SOW / Q&A / wage tables) by its download URL — so an AI can read the real solicitation, not just its metadata. Give it a sam_get_opportunity attachments[].url (resourceLinks). Keyless. Handles PDF (via pdfjs) + text/HTML; returns { format, text, pages, filename, sizeBytes, truncated, extracted }. HONEST: a DOCX / binary that can't be read keyless returns text:null + a note (never fabricated); a corrupt/encrypted PDF returns text:null + an extractionError note (never a crash); a DOWN fetch throws a retryable upstream_unavailable (never empty text); a 404 throws not_found. Only sam.gov / api.sam.gov URLs are fetched (SSRF hygiene). maxChars caps the text (default 200000) and truncation is disclosed.",
+        inputSchema: SamFetchAttachmentTextInput,
+        handler: (input) => fetchAttachmentText(input),
+    }),
+    defineTool({
         name: "sam_lookup_organization",
         description: "Resolve a SAM.gov federal-organization id to its canonical fullParentPathName (e.g. 'VETERANS AFFAIRS, DEPARTMENT OF.VETERANS AFFAIRS, DEPARTMENT OF.245-NETWORK CONTRACT OFFICE 5'). Use when sam_get_opportunity returned only an organizationId.",
         inputSchema: SamLookupOrgInput,
-    },
-    // ━━━ USAspending — Awards & Recipients (8) ━━━
-    {
+        handler: async (input, { sam }) => {
+            const { organizationId } = input;
+            // SamGovClient internal method — exposed via direct fetch since
+            // it's not on the public surface. Use the public sam.gov endpoint
+            // directly (already keyless).
+            const orgUrl = `https://sam.gov/api/prod/federalorganizations/v1/organizations/${encodeURIComponent(organizationId)}`;
+            let r;
+            try {
+                r = await fetch(orgUrl, {
+                    headers: { Accept: "application/hal+json" },
+                    signal: AbortSignal.timeout(10_000),
+                });
+            }
+            catch (e) {
+                // A network-level fault (DNS, connection reset, timeout) is an OUTAGE, not
+                // absence — classify as retryable rather than letting it surface as the
+                // generic `unknown` (which an agent won't retry).
+                if (e instanceof ToolErrorCarrier)
+                    throw e;
+                throw new ToolErrorCarrier({
+                    kind: "upstream_unavailable",
+                    message: `SAM federalorganizations lookup for '${organizationId}' failed: ${e.message}. This is an outage, not a missing organization. Retry.`,
+                    retryable: true,
+                    retryAfterSeconds: 30,
+                    upstreamEndpoint: "sam:federalorganizations",
+                });
+            }
+            if (!r.ok) {
+                // 404 = the organization genuinely does not exist → a real negative (the
+                // tool's found:false contract), NOT an error.
+                if (r.status === 404) {
+                    return { found: false, organizationId, status: 404 };
+                }
+                // Every OTHER non-2xx is an upstream fault, not absence — classify via the
+                // shared errorFromResponse matrix (400/403→invalid_input non-retryable,
+                // 429→rate_limited with Retry-After, 5xx→upstream_unavailable retryable;
+                // carries upstreamStatus) so a down/blocking service is NEVER read as "org
+                // not found" (the fetch-failure-as-absent masquerade).
+                throw new ToolErrorCarrier(errorFromResponse(r, "sam:federalorganizations"));
+            }
+            // This endpoint signals a NONEXISTENT org id with a 200 + EMPTY body
+            // (live-verified 2026-07-06) — a genuine absence. Read text first so an
+            // empty/degraded body never crashes `r.json()` into a mislabeled `unknown`.
+            const orgText = await r.text();
+            if (!orgText.trim()) {
+                return { found: false, organizationId, status: 200 };
+            }
+            let orgJson;
+            try {
+                orgJson = JSON.parse(orgText);
+            }
+            catch {
+                // A non-empty, non-JSON 200 (e.g. an HTML error/interstitial page from the
+                // CDN/WAF) is a DEGRADED response — do NOT fabricate found:false on garbage;
+                // surface it as schema_drift so the caller knows it's unconfirmed, not absent.
+                throw new ToolErrorCarrier({
+                    kind: "schema_drift",
+                    message: `SAM federalorganizations returned a 200 with a non-JSON body for '${organizationId}' — unexpected shape; cannot confirm whether the organization exists.`,
+                    retryable: false,
+                    upstreamEndpoint: "sam:federalorganizations",
+                });
+            }
+            const org = orgJson._embedded?.[0]?.org;
+            return {
+                found: !!org,
+                organizationId,
+                fullParentPathName: org?.fullParentPathName ?? "",
+                agencyName: org?.agencyName ?? "",
+                name: org?.name ?? "",
+                type: org?.type,
+                level: org?.level,
+            };
+        },
+    }),
+    defineTool({
+        name: "sam_lookup_notice_fields",
+        description: "BATCH-complete a sam_search_opportunities page in ONE call from the GSA daily bulk CSV (keyless). The keyless HAL list endpoint NULLS each result's naics/setAside/place-of-performance/responseDeadline/type; this tool returns those fields for 1..100 noticeIds at once (naicsCode, setAside + setAsideCode, popState/popCity/popZip/popCountry, responseDeadline, type, active, title) from a cached on-disk CSV index, instead of one sam_get_opportunity detail call per notice. OFF BY DEFAULT (no forced 226 MB download): enable by setting SAM_GOV_CSV_CACHE (a cache dir) or SAM_GOV_ENABLE_CSV=1 — when disabled the tool returns data.enabled:false + a structured 'how to enable' note (never fake data, no network). HONEST: _meta carries the CSV last-modified + index build time (freshness), a noticeId absent from the current snapshot returns found:false + nulls with an explicit 'not in current CSV snapshot' disclosure (never faked), a cold first call discloses 'index warming', and a download/parse failure is a structured retryable error (never a silent empty). setAsideCode (e.g. 'SBA') matches sam_get_opportunity's setAside; the snapshot can lag live by up to ~24h — confirm real-time-critical fields with sam_get_opportunity.",
+        inputSchema: SamLookupNoticeFieldsInput,
+        handler: (input) => gsaCsv.lookupNoticeFields(input),
+    }),
+    // ━━━ USAspending — Awards & Recipients (10) ━━━
+    defineTool({
         name: "usas_search_awards",
-        description: "Aggregate share-of-wallet on USAspending. Given an agency × NAICS × fiscal year, returns top recipients by total $ + count. Use for competitive landscape ('who wins at VA in 541512?').",
+        description: "Aggregate share-of-wallet on USAspending. Given an agency × NAICS × fiscal year, returns top recipients by total obligated $ ONLY — per-recipient award COUNTS are NOT returned by this endpoint (`awards`/`totalAwards` are null, not 0); for real per-recipient contract counts use usas_search_awards_by_recipient (its _meta.totalAvailable) or usas_get_recipient_profile. Use for competitive landscape ('who wins at VA in 541512?').",
         inputSchema: UsasFiltersBase,
-    },
-    {
+        handler: (input) => usas.searchAwards(input),
+    }),
+    defineTool({
         name: "usas_search_individual_awards",
         description: "Line-item federal contracts on USAspending. Returns specific awards (recipient + $ + sub-agency + state + description). Use AFTER usas_search_awards when the user wants 'show me the actual contracts'. Each result includes a generatedInternalId for usas_get_award_detail follow-ups.",
         inputSchema: UsasIndividualAwardsInput,
-    },
-    {
+        handler: (input) => usas.searchIndividualAwards(input),
+    }),
+    defineTool({
         name: "usas_search_subagency_spending",
         description: "Break down a parent agency's spending by sub-agency / office. Surfaces which office holds the budget (e.g. VA OI&T vs VHA, DoD vs Army vs DISA).",
         inputSchema: UsasSubAgencyInput,
-    },
-    {
+        handler: (input) => usas.searchSubAgencySpending(input),
+    }),
+    defineTool({
         name: "usas_lookup_agency",
         description: "Resolve a user-friendly agency reference ('VA', 'Veterans Affairs', 'DHS') to USAspending's canonical toptier name + 4-digit code. ALWAYS call this FIRST if the user uses an abbreviation — other USAspending tools require the canonical name.",
         inputSchema: UsasLookupAgencyInput,
-    },
-    {
+        handler: (input) => usas.lookupAgency(input.searchText),
+    }),
+    defineTool({
         name: "usas_search_awards_by_recipient",
         description: "Pull every contract a specific recipient has won within an agency × NAICS slice. Use when the user asks 'show me Booz Allen wins at VA last year' — returns line items + naicsCode + description, not aggregates.",
         inputSchema: UsasRecipientAwardsInput,
-    },
-    {
+        handler: (input) => usas.searchAwardsByRecipient(input),
+    }),
+    defineTool({
         name: "usas_search_subawards",
         description: "Enumerate subcontracts on prime awards. Use for 'who teams with Leidos at DISA' or 'show small-business subs on Accenture's DHS contracts' — surfaces the prime/sub network for teaming-map artifacts.",
         inputSchema: UsasSubawardsInput,
-    },
-    {
+        handler: (input) => usas.searchSubawards(input),
+    }),
+    defineTool({
+        name: "usas_search_recompetes",
+        description: "Recompete radar — federal contracts whose CURRENT period of performance ends inside a window around today (default -90d .. +18mo), sorted soonest-first. Use for 'what VA 541512 contracts are up for recompete in the next 18 months'. Reads the current PoP end date directly from spending_by_award (no per-award enrichment), counts (never drops) rows with missing end dates, and flags in _meta when the scan budget truncates the window (totalAvailable becomes null). Filter by agency/naics/pscCodes/setAside/minAwardValue; set includePotentialEnd for option-inclusive end dates. Public signals only — no CPARS/protest/option-intent, no composite vulnerability score.",
+        inputSchema: UsasRecompetesInput,
+        handler: (input) => usas.searchRecompetes(input),
+    }),
+    defineTool({
         name: "usas_search_expiring_contracts",
-        description: "Find federal contracts at agency × NAICS that expire within N months. Recompete radar — end-date sorted, top 10 by value. Use for 'what VA cloud contracts are up for recompete' or 'show 541512 contracts expiring in 6 months'.",
+        description: "DEPRECATED — use usas_search_recompetes. Thin backward-compatible alias: finds contracts at agency × NAICS expiring within N months and returns the legacy { contracts, searchedCount } shape. New callers should use usas_search_recompetes for the full window/pagination controls and truthful completeness metadata.",
         inputSchema: UsasExpiringInput,
-    },
-    {
+        handler: (input) => usas.searchExpiringContracts(input),
+    }),
+    defineTool({
         name: "usas_get_award_detail",
         description: "Fetch full detail for a single award by generatedInternalId (from usas_search_individual_awards). Returns period_of_performance (start/end/potential_end), base_and_all_options, set-aside type, competition extent, number_of_offers — the per-award fields the search endpoint omits.",
         inputSchema: UsasAwardDetailInput,
-    },
+        handler: (input) => usas.getAwardDetail(input.generatedInternalId),
+    }),
+    defineTool({
+        name: "usas_analyze_incumbent",
+        description: "Per-award incumbent + PUBLIC recompete-pressure analysis for ONE award (generatedInternalId). Assembles the incumbent identity, the vehicle/IDV linkage, and individual PUBLIC pressure SIGNALS — obligated-vs-ceiling consumption (pctConsumed), modification count (lower-bounded), competition extent + number of offers, set-aside, days to the current PoP end, and option-extendable days — plus, optionally, the incumbent's other awards in the same agency×NAICS. Bounded & keyless: at most 3 upstream calls (detail + 1 transactions page + 1 recipient search), no per-record fan-out. Emits pressureHints ('single_offer', 'ceiling_nearly_exhausted', 'hard_stop_no_options') as HINTS, NEVER a composite vulnerability score — CPARS/past-performance, protest history, and option-exercise intent are not public (declared in _meta.fieldsUnavailable).",
+        inputSchema: UsasAnalyzeIncumbentInput,
+        handler: (input) => usas.analyzeIncumbent(input),
+    }),
     // ━━━ USAspending — Aggregate Analysis (6) ━━━
-    {
+    defineTool({
         name: "usas_spending_over_time",
-        description: "Time-series aggregation of federal spending. Group by fiscal_year / quarter / month, filter by agency / NAICS / set-aside. Use for 'how has VA 541512 spending trended over the past 5 years' — returns yearly/quarterly/monthly $ rollups.",
+        description: "Time-series aggregation of federal CONTRACT spending (award types A/B/C/D — grants, IDVs, loans, and other obligation types are EXCLUDED, matching the other usas_search_*_spending tools; disclosed in _meta). Group by fiscal_year / quarter / month, filter by agency / NAICS / set-aside. Use for 'how has VA 541512 contract spending trended over the past 5 years' — returns yearly/quarterly/monthly $ rollups of contract obligations (grantObligations/idvObligations are null, NOT 0, to avoid implying an agency has no grant/IDV spending).",
         inputSchema: UsasSpendingOverTimeInput,
-    },
-    {
+        handler: (input) => usas.spendingOverTime(input),
+    }),
+    defineTool({
         name: "usas_search_psc_spending",
         description: "Spending broken down by Product Service Code (PSC). Use for 'what PSC categories see the most $ at DoD' — surfaces market structure beyond NAICS (e.g. PSC R425 = engineering support services).",
         inputSchema: UsasCategorySpendingInput,
-    },
-    {
+        handler: (input) => usas.searchPscSpending(input),
+    }),
+    defineTool({
         name: "usas_search_state_spending",
         description: "Spending broken down by state / territory. Use for 'where is the most federal $ flowing for NAICS 541512' — answers like 'VA $128B, MD $66B, DC $58B'.",
         inputSchema: UsasCategorySpendingInput,
-    },
-    {
+        handler: (input) => usas.searchStateSpending(input),
+    }),
+    defineTool({
         name: "usas_search_cfda_spending",
         description: "Spending broken down by CFDA grant program code. Use for grant analysis — 'top federal grant programs by $'. Note: CFDA is grants (award_type 02-05), not contracts. Use usas_search_psc_spending for contract market analysis.",
         inputSchema: UsasCfdaInput,
-    },
-    {
+        handler: (input) => usas.searchCfdaSpending(input),
+    }),
+    defineTool({
         name: "usas_search_federal_account_spending",
         description: "Spending broken down by federal account / Treasury Account Symbol (TAS). Use to map money to the actual budget line item (e.g. '036-0167 = Information Technology Systems, VA').",
         inputSchema: UsasCategorySpendingInput,
-    },
-    {
+        handler: (input) => usas.searchFederalAccountSpending(input),
+    }),
+    defineTool({
         name: "usas_search_agency_spending",
         description: "Spending broken down by awarding agency. Use for 'which agencies spend the most on NAICS 541512' — top buyers by $.",
         inputSchema: UsasAgencySpendingInput,
-    },
+        handler: (input) => usas.searchAgencySpending(input),
+    }),
     // ━━━ USAspending — Agency Profile (3) ━━━
-    {
+    defineTool({
         name: "usas_get_agency_profile",
         description: "Get full agency profile by toptier code (3-4 digits, from usas_lookup_agency). Returns mission, abbreviation, website, subtier_agency_count, congressional_justification_url.",
         inputSchema: UsasAgencyProfileInput,
-    },
-    {
+        handler: (input) => usas.getAgencyProfile(input.toptierCode),
+    }),
+    defineTool({
         name: "usas_get_agency_awards_summary",
-        description: "High-level award activity for a fiscal year — transaction_count + obligations + latest_action_date. Snapshot of agency volume.",
+        description: "High-level award activity for a fiscal year — transaction_count + obligations + latest_action_date. SCOPE: obligations/transaction_count span ALL award types (contracts, grants, direct payments incl. benefits, loans) — NOT prime contracts only. For benefit-heavy agencies (VA/SSA/HHS) this is dominated by direct benefit payments (e.g. VA FY2024 ~$238B all-awards vs ~$67B prime contracts), so do NOT read it as the contract/procurement market; for procurement-heavy agencies (DoD/DHS) it closely tracks contract spending. For contracts-only obligations use usas_spending_over_time (contractObligations) — it takes the agency canonical NAME, so resolve it from this toptierCode via usas_get_agency_profile first.",
         inputSchema: UsasAgencyAwardsInput,
-    },
-    {
+        handler: (input) => usas.getAgencyAwardsSummary(input),
+    }),
+    defineTool({
         name: "usas_get_agency_budget_function",
         description: "Budget function breakdown for an agency × fiscal year. Returns the agency's spending by program area (e.g. VA: 'Income security for veterans' $204B, 'Hospital and medical care for veterans' $126B).",
         inputSchema: UsasAgencyBudgetInput,
-    },
+        handler: (input) => usas.getAgencyBudgetFunction(input),
+    }),
     // ━━━ USAspending — Recipient Profile (2) ━━━
-    {
+    defineTool({
         name: "usas_search_recipients",
         description: "Search USAspending recipient list with parent/child/recipient hierarchy. Returns recipients with id, duns, uei, level (P=parent, C=child, R=recipient), total_amount. Use for 'find the recipient_id for Booz Allen' before usas_get_recipient_profile.",
         inputSchema: UsasSearchRecipientsInput,
-    },
-    {
+        handler: (input) => usas.searchRecipients(input),
+    }),
+    defineTool({
         name: "usas_get_recipient_profile",
         description: "Full recipient detail by recipient_id (from usas_search_recipients). Returns alternate_names (M&A history), DUNS, UEI, parent linkage, business_types, location, total_amount, total_transactions.",
         inputSchema: UsasGetRecipientInput,
-    },
-    // ━━━ USAspending — Reference / Autocomplete (4) ━━━
-    {
+        handler: (input) => usas.getRecipientProfile(input.recipientId),
+    }),
+    // ━━━ USAspending — Reference / Autocomplete (5) ━━━
+    defineTool({
         name: "usas_autocomplete_naics",
         description: "Autocomplete NAICS codes by free-text. ANTI-HALLUCINATION GUARD — call this when the user mentions a NAICS theme but no specific code (e.g. 'computer systems design' → 541512). Avoids inventing NAICS codes.",
         inputSchema: UsasAutocompleteInput,
-    },
-    {
+        handler: (input) => usas.autocompleteNaics(input),
+    }),
+    defineTool({
         name: "usas_autocomplete_recipient",
         description: "Autocomplete recipient names. ANTI-HALLUCINATION — confirm a recipient's exact USAspending-canonical legal name before searching by name. Returns up to 10 fuzzy matches with UEI/DUNS where available.",
         inputSchema: UsasAutocompleteInput,
-    },
-    {
+        handler: (input) => usas.autocompleteRecipient(input),
+    }),
+    defineTool({
         name: "usas_naics_hierarchy",
-        description: "Navigate the NAICS hierarchy (2-digit → 4-digit → 6-digit). Returns parent/child relationships + active-contract count per code. Use to explore market scope ('what's under NAICS 541' = 'Professional, Scientific, and Technical Services').",
+        description: "Navigate the NAICS hierarchy (2→4→6 digit) + active-contract count per code. No naicsFilter ⇒ the top-level 2-digit sectors. With naicsFilter=<code> ⇒ that node is in `parent` and its DIRECT children are in `hierarchy` (drill into any row where hasChildren:true by re-calling with its code). A 6-digit leaf returns hierarchy:[] with the node in `parent` (found:true); a nonexistent code returns hierarchy:[] with parent:null (found:false). Use to explore market scope (e.g. what's under NAICS 54 = Professional, Scientific, and Technical Services).",
         inputSchema: UsasNaicsHierarchyInput,
-    },
-    {
+        handler: (input) => usas.naicsHierarchy(input),
+    }),
+    defineTool({
         name: "usas_glossary",
         description: "USAspending glossary of 151 federal-spending terms. Use to confirm terminology ('what's a TAS?', 'what's an obligation vs outlay?') before answering compliance/budget questions.",
         inputSchema: UsasGlossaryInput,
-    },
-    {
+        handler: (input) => usas.glossary(input),
+    }),
+    defineTool({
         name: "usas_list_toptier_agencies",
         description: "List all toptier federal agencies with toptier_code, abbreviation, slug, current-FY obligations. Use for 'show me every cabinet department + their FY26 spending' or to find a toptier_code for usas_get_agency_*.",
         inputSchema: UsasListAgenciesInput,
-    },
-    // ━━━ Federal Register (3) ━━━
-    {
+        handler: (input) => usas.listToptierAgencies(input),
+    }),
+    // ━━━ Federal Register (4) ━━━
+    defineTool({
         name: "fed_register_search_documents",
         description: "Search Federal Register documents (proposed rules, final rules, notices, presidential documents) by query / agency / type / date range. Use for regulatory-context queries ('what new VA cybersecurity rules came out this quarter?').",
         inputSchema: FedRegSearchInput,
-    },
-    {
+        handler: (input) => fedreg.searchDocuments(input),
+    }),
+    defineTool({
         name: "fed_register_get_document",
         description: "Fetch full detail for a Federal Register document by number. Returns title, abstract, citation, publication_date, effective_on, raw_text_url (for the full body), CFR references — everything needed to ground a regulation citation.",
         inputSchema: FedRegGetDocInput,
-    },
-    {
+        handler: (input) => fedreg.getDocument(input.documentNumber),
+    }),
+    defineTool({
         name: "fed_register_list_agencies",
         description: "List all Federal Register agencies with slugs (needed for fed_register_search_documents). Use to resolve 'what's the FedReg slug for Veterans Affairs?'",
         inputSchema: FedRegListAgenciesInput,
-    },
-    // ━━━ eCFR (2) ━━━
-    {
+        handler: (input) => fedreg.listAgencies(input),
+    }),
+    defineTool({
+        name: "fed_register_public_inspection",
+        description: "Federal Register PUBLIC INSPECTION desk — documents FILED with the Office of the Federal Register but NOT YET published (a pre-publication LEADING INDICATOR, ~1-to-several days ahead of the official publication_date). mode: 'current' (all on inspection now), 'date' (a specific available_on day), 'search' (full-text over the on-inspection set). Returns per-doc leadDays (pre-publication head-start), filing_type special-vs-regular, and unflattened agencies. NOTE: a public-inspection doc is NOT the authoritative published rule (no FR citation/page yet; may change or be withdrawn) — after publication_date cross-check fed_register_get_document.",
+        inputSchema: FedRegPublicInspectionInput,
+        handler: (input) => fedreg.publicInspection(input),
+    }),
+    // ━━━ eCFR (5) ━━━
+    defineTool({
         name: "ecfr_search",
         description: "Full-text search across the entire CFR (Code of Federal Regulations). Use for compliance questions — pass titleNumber=48 for FAR (Federal Acquisition Regulation), titleNumber=2 for federal financial assistance, etc. Returns excerpt + section path + ecfrUrl.",
         inputSchema: EcfrSearchInput,
-    },
-    {
+        handler: (input) => ecfr.search(input),
+    }),
+    defineTool({
         name: "ecfr_list_titles",
         description: "List all 50 CFR titles with name + last_amended_on date. Use to discover what's in each title (Title 48 = FAR, Title 32 = National Defense, Title 14 = Aeronautics, etc.).",
         inputSchema: EcfrListTitlesInput,
-    },
+        handler: () => ecfr.listTitles(),
+    }),
+    defineTool({
+        name: "far_clause_lookup",
+        description: "Authoritative FAR/DFARS clause text + its PRESCRIPTION (the 'As prescribed in …' rule for when the clause applies), from the eCFR versioner-full endpoint (Title 48). Use this — NOT ecfr_search — for an EXACT clause number: full-text search mis-ranks '52.212-4' (returns GSAM 552.212-4 above the real FAR clause). Returns heading, revision date, clause/provision kind, regulation (FAR/DFARS/GSAM), full text, the prescribing section, and ecfrUrl. Every response carries farOverhaulRisk — a structural currency caveat that eCFR reflects only the CODIFIED FAR, so a clause may be superseded by a Revolutionary-FAR-Overhaul agency class deviation not shown here. A genuinely-absent clause returns a not_found error (never a fake empty clause). Keyless.",
+        inputSchema: FarClauseLookupInput,
+        handler: (input) => far.farClauseLookup(input),
+    }),
+    defineTool({
+        name: "far_compliance_matrix",
+        description: "Turn a solicitation's cited FAR/DFARS clause list into a proposal-ready compliance matrix (for a Section L/M response). COMPOSES far_clause_lookup over 1–25 clauses (deduped case-insensitively): each resolved row carries the clause text + prescription + regulation + a gate flag marking pass/fail award-eligibility GATES (Section 889 52.204-24/25/26, limitations on subcontracting 52.219-14, DFARS cyber 252.204-7012/7020/7021 incl. CMMC) + the farOverhaulRisk currency caveat. TRUTHFUL by construction: a clause that genuinely isn't in Title 48 (HTTP 404) goes to `unresolved`, while a clause that couldn't be fetched (eCFR down/5xx/rate-limited) goes to a SEPARATE `errored` bucket — a DOWN service is never reported as 'clause doesn't exist'; `summary.total` proves no clause is dropped. Does NOT parse the PDF solicitation to extract the clause list, and gives NO legal advice or compliance verdict. Keyless.",
+        inputSchema: FarComplianceMatrixInput,
+        handler: (input) => far.farComplianceMatrix(input),
+    }),
+    defineTool({
+        name: "far_search",
+        description: "FAR/DFARS-scoped semantic search — the 'which clauses touch topic X' front-door that feeds far_clause_lookup. COMPOSES ecfr_search but fixes its two compliance flaws: (1) it filters to FAR (Title 48 chapter 1) or DFARS (chapter 2), EXCLUDING GSAM/agency supplements (so 'limitations on subcontracting' no longer mis-ranks GSAM 552.x over FAR 52.x), and (2) it collapses eCFR's ~5-versions-per-section HISTORICAL duplicates to the CURRENT in-force version (endsOn==null). scope: far (default) | dfars | both. dedupeVersions (default true; false shows all historical rows). partsOnly restricts to given parts (e.g. [52] clause text). Returns distinct sections with regulation/section/headingPath/excerpt/score/ecfrUrl/effectiveOn/endsOn/isCurrent, distinctSections, and the farOverhaulRisk caveat. TRUTHFUL: dedupe never drops a distinct section (the raw→distinct collapse is disclosed); a kept-historical row is marked isCurrent:false; a search-endpoint outage THROWS (never a fake 0 results); totalAvailable is null (a deduped view has no clean upstream count). Keyless.",
+        inputSchema: FarSearchInput,
+        handler: (input) => far.farSearch(input),
+    }),
+    // ━━━ SBA — Size Standards (1) ━━━
+    defineTool({
+        name: "sba_size_standard",
+        description: "SBA small-business size standard for a 6-digit NAICS (keyless sba.gov naics.json). Answers 'is a firm SMALL for this NAICS?' — the gate for set-aside eligibility and for vetting a usas_search_teaming_partners candidate. Returns standardType (receipts | employees | assets [financial institutions] | receipts+assets), a normalized threshold (receipts/assets in DOLLARS — the dataset's $millions figure ×1,000,000; employees as a count), the unit, and any SBA footnote. HONESTY: the dataset carries no effective-date field, so the value is 'as published as of retrieval' (asOf) and _meta.notes flags that SBA adjusts standards periodically — re-verify at sba.gov for high-stakes eligibility. An unknown NAICS returns found:false (never a fabricated standard).",
+        inputSchema: SbaSizeStandardInput,
+        handler: (input) => sba.sizeStandard(input),
+    }),
     // ━━━ Grants.gov (2) ━━━
-    {
+    defineTool({
         name: "grants_search",
-        description: "Search Grants.gov federal grant opportunities (financial assistance, distinct from contracts on SAM.gov). Filter by keyword / CFDA / agency / opportunity number. Default status = forecasted + posted.",
+        description: "Search Grants.gov federal grant opportunities (financial assistance, distinct from contracts on SAM.gov). Filter by keyword / CFDA / agency / opportunity number. Default status = forecasted + posted. KEYWORD: Grants.gov OR-matches multi-word keywords (returns grants containing ANY word), so a multi-word keyword BROADENS results — pass ONE specific term for relevance (phrase quoting returns 0); narrow with cfda/agency/oppStatuses.",
         inputSchema: GrantsSearchInput,
-    },
-    {
+        handler: (input) => grants.searchGrants(input),
+    }),
+    defineTool({
         name: "grants_get_opportunity",
-        description: "Fetch full detail for a single grant opportunity by id. Returns description, agency, posting/response/archive dates, award_ceiling, award_floor, estimated_funding, expected_number_of_awards, applicant_types, funding_instruments, CFDA programs.",
+        description: "Fetch full detail for a single grant opportunity by id. Returns found:true with description, agency, posting/response/archive dates, award_ceiling, award_floor, estimated_funding, expected_number_of_awards, applicant_types, funding_instruments, CFDA programs. `agency` is { code, name (the REAL posting/sub-tier agency, e.g. 'Food and Nutrition Service'), department (the top-tier agency, e.g. 'Department of Agriculture'), contactName (the program officer — NOT the agency) } — Grants.gov's raw `agencyName` field is actually the contact person, so this tool sources the real agency from agencyDetails; `name` may be null if the record carries no structured agency. A NONEXISTENT id returns { found:false, opportunityId } — never a fabricated grant with empty fields (Grants.gov answers a bad id with a hollow 200, which this tool detects). Check `found` before reading the other fields.",
         inputSchema: GrantsGetInput,
-    },
+        handler: (input) => grants.getGrant(input),
+    }),
+    // ━━━ Pricing / Wage (3) ━━━
+    defineTool({
+        name: "sam_search_wage_determinations",
+        description: "Find the Service Contract Act (SCA) or Davis-Bacon (DBA) wage determination(s) governing a locality (keyless SAM SGS). Filter by coverage (sca|dba), state (2-letter, server-side), county (client-side), or WD number/title. Returns the structured WD list; follow with sam_get_wage_rates to read the rate table. NOTE: `query` matches WD number/title only, NOT occupation.",
+        inputSchema: WageSearchInput,
+        handler: (input) => pricing.searchWageDeterminations(input),
+    }),
+    defineTool({
+        name: "sam_get_wage_rates",
+        description: "Return the prevailing-wage + fringe/H&W rate table for a specific wage determination, PARSED from its plain-text document (SAM exposes no structured rate JSON), plus the Executive-Order minimum-wage floor. Distinguishes SCA (WD-wide Health & Welfare) vs DBA (per-craft fringe). Always returns parseConfidence and supports format:'parsed'|'raw'|'both' so you can read the raw text when parsing is low-confidence. Resolves the latest active revision via /history when `revision` is omitted.",
+        inputSchema: WageRatesInput,
+        handler: (input) => pricing.getWageRates(input),
+    }),
+    defineTool({
+        name: "gsa_benchmark_labor_rates",
+        description: "GSA CALC awarded ceiling-rate market band for a labor category (keyless). Returns a DISTRIBUTION (currentRate min/median/max + escalated medians) over a fetched sample, NOT a single price. CALC rates are CEILING/catalog and FULLY BURDENED (do not re-add wrap); the match count SATURATES at 10000 for broad queries (totalAvailable null then). Filter by businessSize/educationLevel(code)/experience/sin to narrow.",
+        inputSchema: BenchmarkLaborInput,
+        handler: (input) => pricing.benchmarkLaborRates(input),
+    }),
+    // ━━━ Integrity / Teaming (3) ━━━
+    defineTool({
+        name: "sam_check_exclusions",
+        description: "Keyless SAM debarment/exclusion screening. Screen a firm or individual by name (query) and/or UEI/CAGE against the SAM exclusions index (FAPIIS). Returns excluded (true iff ≥1 ACTIVE matching record), matchCount, and per-record { name, classification, uei, cage, excludingAgency, exclusionType, exclusionProgram, isActive, activation/terminationDate, samFapiisUrl }. CRITICAL: an EMPTY result means 'no matching exclusion under these terms' — it is NOT proof of general responsibility (stated in _meta.notes). A name match is not identity-proof; verify the UEI/CAGE + dates against the FAPIIS record. Requires at least one of query/uei/cage.",
+        inputSchema: CheckExclusionsInput,
+        handler: (input) => integrity.checkExclusions(input),
+    }),
+    defineTool({
+        name: "sam_integrity_lookup",
+        description: "Keyless ONE-CALL integrity screen — 'any integrity red flags on this entity?'. Composes the keyless government-wide EXCLUSION verdict (via sam_check_exclusions) with an honest pointer to the FAPIIS / Responsibility-Qualification record. Requires at least one of uei/cage/name (uei preferred; name maps to the exclusions text search). Returns { entity, exclusions:{excluded,activeCount,records}, fapiisRecords, fapiisUrl, integrityFlag }. integrityFlag is 'excluded' when ≥1 ACTIVE matching exclusion is found, else 'review_fapiis' — it NEVER returns 'clear' keylessly, because FAPIIS records (terminations for default/cause, non-responsibility determinations, self-reported criminal/civil/administrative proceedings) have NO keyless machine API, so absence of an exclusion is NOT proof of integrity. fapiisRecords is ALWAYS null (never faked; record-level retrieval needs an optional SAM Entity key) with _meta.fieldsUnavailable:['fapiisRecords']; fapiisUrl deep-links the viewable SAM page. An upstream exclusions failure surfaces as the classified error, never a fake clearance.",
+        inputSchema: IntegrityLookupInput,
+        handler: (input) => integrity.integrityLookup(input),
+    }),
+    defineTool({
+        name: "usas_search_teaming_partners",
+        description: "Small-business teaming-partner discovery by socioeconomic certification + NAICS + agency award history (keyless USAspending proxy), integrity-screened. Given a cert (enum-validated), optional naics/agency/subagency, and a lookback window, aggregates federal awardees by recipient and returns candidates ranked by agencyObligated with agencyAwardCount, mostRecentAwardDate, and sampleAwards; optionally screens the top candidates via sam_check_exclusions and drops active exclusions (excludeDebarred, default true). HONESTY: cert is AWARD-DERIVED (recorded on the firm's federal awards), NOT the SBA certification of record (which needs a keyed SAM Entity call) — verify active certification in SAM/SBS before teaming (stated in _meta). A bogus cert is rejected as invalid_input (the endpoint would silently return 0).",
+        inputSchema: TeamingPartnersInput,
+        handler: (input) => integrity.searchTeamingPartners(input),
+    }),
+    // ━━━ OFAC — Denied-Party Sanctions Screening (1) ━━━ ADR-0034
+    defineTool({
+        name: "ofac_screen_entity",
+        description: "Keyless OFAC denied-party sanctions screening — the legally-required leg that SAM exclusions does NOT cover (31 CFR ch. V, strict-liability). Screens a `name` against OFAC's published SDN + Consolidated bulk lists (primary names AND AKAs from ALT.CSV joined by ent_num AND a.k.a./f.k.a./n.k.a. aliases mined from SDN/CONS Remarks — so an alias-only party like 'BNC' for BANCO NACIONAL DE CUBA is caught). Optional post-filters: type (individual|entity|vessel|aircraft), program (e.g. CUBA/IRAN/SDGT), list (sdn|consolidated|all, default all), minMatchQuality (exact|strong|weak, default weak), limit. Returns result ('potential_matches' | 'no_name_match' — NEVER 'clear'), matchCount, and per-match { name, matchedVia (primary|aka(alt)|aka(remarks)), akaType, matchQuality, list, programs, type, entNum, ofacSearchUrl }. ★SAFETY: this is a NAME SCREEN, NOT a legal determination — a no_name_match is NOT a clearance (transliterations/variants can miss a real hit) and a weak/strong hit is a REVIEW CANDIDATE requiring human adjudication against OFAC's Sanctions List Search. Every fetch failure / SSRF reject / parse drift / floor-fail THROWS (a download failure is NEVER read as a clear). minMatchQuality/type/program only trim returned matches — result reflects existence at any quality. Snapshot freshness (publish date + cache age) rides in _meta.",
+        inputSchema: OfacScreenInput,
+        handler: (input) => ofac.screenEntity(input),
+    }),
+    // ━━━ NVD + CISA KEV — the IT/CYBER-COMPLIANCE lane (2) ━━━ ADR-0035
+    // Opens the FedRAMP/CMMC/SBOM IT-compliance lane the server lacked: NIST NVD
+    // CVE/CVSS severity JOINED with the CISA KEV mandatory-remediation catalog.
+    // Keyless (an OPTIONAL free NVD_API_KEY lifts the rate; header-only, never
+    // logged). Never-fake: a genuine totalResults:0/found:false is honest, but any
+    // 403/429/404/5xx/timeout/redirect-off-host THROWS; a KEV outage degrades
+    // kev.listed to null (never false), and a kevOnly filter during an outage THROWS.
+    defineTool({
+        name: "cve_lookup",
+        description: "Look up NIST NVD CVE records (keyless; services.nvd.nist.gov CVE API 2.0) — exact by `cveId` (CVE-YYYY-NNNN) OR search by `keyword`/`cpeName`/`cvssV3Severity`/a publication or last-modified date range — each row JOINED with its CISA KEV (Known Exploited Vulnerabilities) status. THE B2G unlock for FedRAMP/CMMC/SBOM IT-compliance: CVSS severity AND whether CISA mandates remediation by a date, in one row. Returns { results:[{ cveId, vulnStatus, rejected, published, lastModified, description, cvssMetrics:[{version,source,type,baseScore,baseSeverity,vectorString,exploitabilityScore,impactScore}], primaryCvss:{version,baseScore,baseSeverity,type}|null, cwes, references, kev }] } + honest _meta. Optional `kevOnly` (KEV-listed rows only), `resultsPerPage` (≤2000, def 50), `startIndex`. CVSS HONESTY: every metrics key matching ^cvssMetric (V2/V30/V31/V40) is surfaced as its own cvssMetrics[] element — versions are NEVER conflated and ssvcV203/non-CVSS keys are excluded; V2 baseSeverity reads from the metric level; primaryCvss is the highest-version metric, preferring type:'Primary' but FALLING BACK to the highest Secondary (a real CNA score is never dropped), null ONLY when no CVSS exists (Rejected/Awaiting) — base scores are null-never-0. KEV HONESTY: kev is {listed:true,dateAdded,dueDate,ransomware,requiredAction,catalogVersion} | {listed:false,note} | {listed:null,status:'unavailable'}; a not-listed result carries the not-in-KEV≠safe caveat (absence is NOT a clearance); if the KEV catalog cannot load, kev.listed degrades to NULL (never false) with fieldsUnavailable:['kev'], and a kevOnly filter during that outage THROWS (a KEV-membership filter is unanswerable without the catalog). PAGINATION is from NVD's EXACT totalResults, never page length. A genuine totalResults:0 is an honest found:false; a 403/429 rate breach THROWS rate_limited with the NVD_API_KEY tier disclosure; 404/5xx/timeout/off-host-redirect THROW (never a fake-empty). An OPTIONAL free NVD_API_KEY (env; https://nvd.nist.gov/developers/request-an-api-key) lifts the rate and is sent ONLY in the apiKey header — never a URL/label/_meta/log.",
+        inputSchema: CveLookupInput,
+        handler: (input) => nvd.cveLookup(input),
+    }),
+    defineTool({
+        name: "cisa_kev_lookup",
+        description: "Filter the CISA Known Exploited Vulnerabilities (KEV) catalog standalone (keyless; www.cisa.gov feed, cached) — the mandatory-remediation list carrying BINDING due-dates under BOD 22-01 / its 2026 successor BOD 26-04. Works even when NVD is rate-limited (a separate host, no key). Filters (all optional, AND-combined, client-side): `cveId` (exact KEV membership check), `vendorProject`/`product` (case-insensitive substring), `ransomwareOnly` (knownRansomwareCampaignUse === 'Known'), `addedSince`/`dueBefore` (ISO YYYY-MM-DD); `limit` (≤1000, def 100), `offset`. Returns { catalogVersion, dateReleased, count, found?, matches:[{ cveID, vendorProject, product, vulnerabilityName, dateAdded, dueDate, knownRansomwareCampaignUse, shortDescription, requiredAction, cwes, nvdUrl }] } + honest _meta. ★HONESTY: knownRansomwareCampaignUse and requiredAction are surfaced VERBATIM (never defaulted); dueDate is the CISA-mandated remediation deadline. A cveId NOT in the catalog ⇒ found:false — but the not-in-KEV≠safe caveat rides on EVERY response: KEV is a CURATED SUBSET of confirmed in-the-wild exploitation, so absence means CISA has not catalogued it, NOT that the component is unexploited/safe. A catalog download failure / floor-fail / count-drift THROWS (a truncated/near-empty catalog must never read as 'nothing is exploited') — never a fake-empty. The snapshot freshness (catalogVersion + release date + cache age) is disclosed.",
+        inputSchema: CisaKevLookupInput,
+        handler: (input) => nvd.cisaKevLookup(input),
+    }),
+    // ━━━ NPPES NPI Registry — Healthcare-Provider Vetting (1) ━━━ ADR-0036
+    defineTool({
+        name: "nppes_lookup_provider",
+        description: "Keyless CMS/HHS NPPES NPI Registry lookup — the authoritative PUBLIC registry of every US healthcare provider (individual NPI-1 + organization NPI-2), for VA/HHS/CMS subcontractor/provider/teaming due-diligence (validate an NPI, confirm taxonomy/specialty, enumeration status, practice state, org/name match). Host npiregistry.cms.hhs.gov/api (version=2.1). Mode is inferred from `number` (no mode flag). EXACT-NPI mode (`number` given): the NPI is CMS-Luhn-validated client-side (Luhn over 80840+first-9) ⇒ a typo'd NPI is invalid_input, NEVER a fake 'does not exist'; ★the wire query carries `number` (+version) ALONE — any co-supplied filter (last_name/state/…) is DROPPED from the wire and checked CLIENT-SIDE (disclosed in data.filterMatch:{field:bool} + data.filtersDropped), because NPPES AND-combines a number with filters and a mismatch would falsely zero a real active provider into found:false. SEARCH mode: required-one of { first_name, last_name, organization_name, taxonomy_description, city, postal_code } (state + enumeration_type are REFINERS ONLY — rejected alone); a trailing '*' wildcard on a name/org field needs ≥2 leading literal chars. Returns EXACT-mode { found, provider:{ number, enumerationType, active, status, basic{…individual OR org fields, null-never-fabricated…}, taxonomies[{code,desc,primary,state,license,taxonomyGroup}], addresses[{purpose,address1,city,state,postalCode,telephone,fax,countryCode}], practiceLocations[…same, SEPARATE from addresses], identifiers[], otherNames[], endpoints[], createdEpoch, lastUpdatedEpoch }, filterMatch? } OR SEARCH-mode { providers:[…] } + honest _meta. HONESTY: active = basic.status==='A' (a deactivated/absent NPI is NOT active); epochs are ms numeric STRINGS → number|null (null-never-0); addresses[] and practiceLocations[] are kept SEPARATE (a provider can practice in a state that appears ONLY in practiceLocations); NPPES exposes NO match total, so a full page ⇒ totalAvailable is a disclosed LOWER BOUND (totalIsLowerBound) + a ~1,200-row-per-query reach cap (limit ≤ 200, skip ≤ 1,000 — OUR policy, a PER-QUERY cap only; cross-query enumeration is not architecturally prevented). A genuine {result_count:0} ⇒ honest found:false/empty; a {Errors:[…]} 200 body (no results key) ⇒ THROWS invalid_input (never a fake empty); any 4xx/5xx/timeout/off-host-redirect ⇒ THROWS; result_count !== results.length ⇒ schema_drift. ★NOT a fitness/exclusion/licensure/sanctions determination — cross-check SAM exclusions + OFAC; individual (NPI-1) records may surface personal/home addresses + phone/fax verbatim with NO enrichment. The caveat + reach-cap disclosure ride EVERY response.",
+        inputSchema: NppesLookupInput,
+        handler: (input) => nppes.lookupProvider(input),
+    }),
+    // ━━━ CMS Open Payments — Healthcare Spend/Transparency (DKAN, keyless) (2) ━━━ ADR-0037
+    defineTool({
+        name: "cms_search_datasets",
+        description: "Discover CMS Open Payments datasets on the keyless DKAN DCAT metastore (openpaymentsdata.cms.gov) — the Physician Payments Sunshine Act transparency catalog (industry→physician/teaching-hospital payments, other transfers of value, ownership interests). Returns { query, results:[{ datasetId, title, description, distributions:[{index, distId, title, mediaType, downloadURL}], keyword, modified }] } + honest _meta. Feed a result's datasetId + a distribution index to cms_query_dataset (use results:false there to enumerate the column schema before pulling rows). Optional `q` (case-insensitive title/description substring), `limit` (≤100, def 20), `offset`. ★HONESTY: the DKAN metastore IGNORES limit/offset/page and returns the ENTIRE catalog in one response, so q/limit/offset are applied CLIENT-SIDE against the in-memory array and totalAvailable is the EXACT post-q catalog size (never fabricated, never null) — hasMore is computed against the KNOWN catalog length (no false-more, no dead-end offset). The flagship targets are '2025 Research Payment Data', the General-Payment, and Ownership datasets. A non-array metastore body / HTML / 5xx / timeout THROWS (never a fake empty). NOT a determination — see cms_query_dataset's caveat.",
+        inputSchema: CmsSearchDatasetsInput,
+        handler: (input) => cms.searchDatasets(input),
+    }),
+    defineTool({
+        name: "cms_query_dataset",
+        description: "Query a CMS Open Payments DKAN datastore distribution by datasetId + index (keyless; openpaymentsdata.cms.gov) — the healthcare industry-financial-relationship / COI-vetting + market-intelligence lane NPPES (provider identity) cannot answer. GET /api/1/datastore/query/{datasetId}/{index} with server-side `conditions` filters, an EXACT `count`, offset/limit pagination, and a `properties` projection. Returns { datasetId, index, results (mode), fields:[{name,type,mysqlType,description}] (from the DKAN schema), rows:[…verbatim…] } + honest _meta. A confirmed target: 2025 Research Payment Data 'f0d1de67-6852-4093-a036-c9328c256a05' index 0 (count 931959; + a recipient_state='CA' condition → 92097). ★HONESTY: `count` is the EXACT grand total (P1) → totalAvailable=count + real offset pagination (NOT a page-length lower bound); `conditions` are server-side and self-policing — a valid column narrows the count, a BAD column ⇒ HTTP 400 ⇒ invalid_input, so filtersDropped is ALWAYS empty (no silent-drop path, P4); limit ≤ 500 is the HARD API cap (a higher limit ⇒ invalid_input, no silent clamp); every column is text, so amounts (total_amount_of_payment_usdollars, …) arrive as STRINGS surfaced verbatim (a missing amount is null-never-0, P3). ★results:false = a COUNT/SCHEMA-discovery mode: no rows, pagination disabled (no livelock), but the EXACT count + every column's schema returned (count=true is ALWAYS on the wire — not a caller toggle). A genuine {count:0} ⇒ honest empty; a 400 (bad column/limit) / 404 (bad datasetId/index) / HTML (SPA/WAF) / 5xx / timeout / a missing schema anchor or non-array results (in results:true) ⇒ THROW (never a fake empty). ★SSRF: datasetId (36-char lowercase UUID) + index interpolate into the URL PATH (validated before interpolation). ★PII: Open Payments is PUBLIC transparency-BY-LAW data (in-scope per the NPPES precedent) naming physicians + amounts verbatim — bounded to targeted vetting (offset ≤ 2000 reach cap), NO enrichment, NO covered_recipient_npi→NPPES auto-join. NOT a conflict-of-interest finding / fitness / exclusion determination — cross-check SAM exclusions + OFAC + the OIG-LEIE. The caveat + reach-cap disclosure ride EVERY response.",
+        inputSchema: CmsQueryDatasetInput,
+        handler: (input) => cms.queryDataset(input),
+    }),
+    // ━━━ FAC Federal Audit Clearinghouse — Single Audit audit-risk vetting (2) ━━━ ADR-0038
+    defineTool({
+        name: "fac_search_audits",
+        description: "Search entity Single Audit summaries from the Federal Audit Clearinghouse (keyless via the api.data.gov DEMO_KEY; api.fac.gov PostgREST `general` table) — the SUBCONTRACTOR / teaming AUDIT-RISK vetting entry point (2 CFR 200 Subpart F / Single Audit Act; every entity expending ≥$750K/yr in federal awards). Structured filters (all optional, AND-combined): `auditeeUei` (12-char SAM UEI — the PRIMARY join key to SAM/USAspending/EDGAR, → auditee_uei), `auditeeState` (2-letter → auditee_state), `auditYear` (int → audit_year), `totalExpendedMin`/`totalExpendedMax` (USD → total_amount_expended gte/lte). `limit` (≤100, def 25), `offset`. Returns { audits:[{ report_id, auditee_uei, audit_year, auditee_name, auditee_ein, auditee_state, auditee_city, total_amount_expended, fac_accepted_date }] } + honest _meta. Feed a row's report_id (or the UEI) to fac_get_findings for the audit-RISK flags. ★PII: a HARDCODED select-allowlist surfaces ONLY entity + audit-summary fields and DELIBERATELY EXCLUDES the auditee's personal-contact columns (email/phone/certifying-official name) — the vetting subject is the ENTITY; there is NO caller `select`/column param. HONESTY: totalAvailable is the EXACT Content-Range total (a response header under Prefer:count=exact; a '*'/absent/non-numeric denominator ⇒ totalAvailable:null + a page-fullness hedge, NEVER 0); total_amount_expended is null-never-0 (a missing amount is null, never 0); a bad column ⇒ PostgREST 400 ⇒ invalid_input (filtersDropped is ALWAYS empty); a genuine [] ⇒ honest empty; 400/403/5xx/timeout/HTML/non-array THROW (206 = success, never a fake empty). NOT a debarment/exclusion/fitness determination — an audit finding is the auditor's opinion; cross-check SAM exclusions + OFAC. Keyless-first via DEMO_KEY (~10 req/hr shared ceiling; set DATA_GOV_API_KEY for production — never logged).",
+        inputSchema: FacSearchAuditsInput,
+        handler: (input) => fac.searchAudits(input),
+    }),
+    defineTool({
+        name: "fac_get_findings",
+        description: "Drill into the audit-RISK findings for an entity from the Federal Audit Clearinghouse (keyless via the api.data.gov DEMO_KEY; api.fac.gov PostgREST `findings` table) — the risk-detail step after fac_search_audits. At least ONE of `auditeeUei` (12-char UEI → auditee_uei) or `reportId` (→ report_id, from a fac_search_audits row) is REQUIRED (an empty query is refused, never a whole-table scan); optional `auditYear` (int), `limit` (≤100, def 50), `offset`. Returns { findings:[{ report_id, auditee_uei, audit_year, award_reference, reference_number, is_material_weakness, is_modified_opinion, is_questioned_costs, is_repeat_finding, is_significant_deficiency, is_other_findings, is_other_matters, type_requirement, prior_finding_ref_numbers, riskFlags:{materialWeakness, modifiedOpinion, questionedCosts, repeatFinding, significantDeficiency, otherFindings, otherMatters} }] } + honest _meta. ★RISK-FLAG HONESTY: the is_* flags are surfaced VERBATIM as the auditor reported them (\"Y\"/\"N\") PLUS a typed riskFlags tri-state (\"Y\"→true / \"N\"→false / blank/absent/other → null=UNKNOWN) — a null flag is NEVER rendered as false/\"no material weakness\" (the false-CLEAR class). ★EMPTY ≠ CLEAN: an empty findings list does NOT confirm a clean audit — the entity may not have filed a Single Audit (below the $750K threshold), the audit may predate FAC coverage, or the UEI may be wrong; a disclosure note fires on any empty result telling you to confirm an ACCEPTED audit exists via fac_search_audits. ★PII: a HARDCODED select-allowlist (NO caller column param) surfaces only entity + audit-risk fields — no personal contact. totalAvailable is the EXACT Content-Range total ('*'/absent ⇒ null + hedge, never 0); a bad column ⇒ 400 ⇒ invalid_input; 400/403/5xx/timeout/HTML/non-array THROW (206 = success). NOT a debarment/determination — cross-check SAM exclusions + OFAC + the specific finding text. Keyless-first via DEMO_KEY (~10 req/hr; set DATA_GOV_API_KEY — never logged).",
+        inputSchema: FacGetFindingsInput,
+        handler: (input) => fac.getFindings(input),
+    }),
+    // ━━━ GAO — Bid Protests (1) ━━━
+    defineTool({
+        name: "gao_protest_lookup",
+        description: "Recent GAO (Comptroller General) bid-protest decisions from the public Legal-Products RSS feed, enriched from each decision page (protester, contracting agency, decision date, outcome sustained/denied/dismissed/withdrawn, solicitation #, decision PDF). Filter client-side by agency/protester/solicitation/outcome, or pull one decision directly by bNumber. HONEST SCOPE: keyless covers only the RECENT feed window (~25 items) — GAO's faceted historical protest search (all years, by protester/agency/outcome/date) is WAF-blocked to bots and available only via a paid third-party API, so results are ALWAYS marked complete:false and are NOT the full protest history (see the accessNote).",
+        inputSchema: GaoProtestInput,
+        handler: (input) => gao.gaoProtestLookup(input),
+    }),
+    // ━━━ US Treasury — Fiscal Data (keyless) (4) ━━━ ADR-0002
+    defineTool({
+        name: "treasury_query_dataset",
+        description: "Escape-hatch query over 5 confirmed US Treasury Fiscal Data datasets (keyless): debt_to_penny, avg_interest_rates, mts_table_1 (Monthly Treasury Statement), rates_of_exchange, debt_outstanding. Choose `dataset` (enum — no free path), and optionally project `fields` (CSV), `filter` (CSV 'col:op:val', ops lt|lte|gt|gte|eq|in, AND-combined), and `sort` (CSV, '-' = desc), with page[size]/page[number] pagination. Returns raw rows plus a truthful `_meta` (totalAvailable = upstream total-count, offset pagination). Value/amount fields are raw upstream strings — the string \"null\"/empty means 'no value', never 0. Covers rates_of_exchange + debt_outstanding without a dedicated tool.",
+        inputSchema: TreasuryQueryDatasetInput,
+        handler: (input) => treasury.queryDataset(input),
+    }),
+    defineTool({
+        name: "treasury_debt_to_penny",
+        description: "Daily total US public debt outstanding ('Debt to the Penny', keyless Treasury Fiscal Data). Returns record_date + totalPublicDebtOutstanding, debtHeldByPublic, intragovernmentalHoldings (USD). `latest` (default true) ⇒ the single most-recent day; set latest=false with startDate/endDate (ISO YYYY-MM-DD) for a date range, newest-first. Amounts are coerced to number|null (a null amount is 'no value reported', never 0).",
+        inputSchema: TreasuryDebtToPennyInput,
+        handler: (input) => treasury.debtToPenny(input),
+    }),
+    defineTool({
+        name: "treasury_monthly_statement",
+        description: "Monthly Treasury Statement (MTS table 1, keyless): federal receipts, outlays, and deficit/surplus by month. Returns record_date, classification, grossReceipts, grossOutlays, deficitSurplus (USD, number|null). `startDate`/`endDate` (ISO YYYY-MM-DD) filter record_date (default: trailing ~12 months). By default excludeSummaryRows=true drops the fiscal-year parent/summary header rows (whose amounts are all null) via a server-side filter, so totalAvailable and rows reflect real child line-items only; set excludeSummaryRows=false to include them. Highest-value budget-analysis tool.",
+        inputSchema: TreasuryMonthlyStatementInput,
+        handler: (input) => treasury.monthlyStatement(input),
+    }),
+    defineTool({
+        name: "treasury_avg_interest_rates",
+        description: "Average interest rate the US Treasury pays by security type/description (keyless Treasury Fiscal Data). Returns record_date, securityType, securityDescription, avgInterestRatePercent (percent, number|null). `latest` (default true) returns the most-recent month's full breakdown across security types (pinned to the latest record_date, memoized 5 min); set latest=false with startDate/endDate for a range. Optional `securityType` narrows by exact security_type_desc (e.g. 'Marketable', 'Non-marketable').",
+        inputSchema: TreasuryAvgInterestRatesInput,
+        handler: (input) => treasury.avgInterestRates(input),
+    }),
+    // ━━━ SEC EDGAR — filings / XBRL facts / CIK / full-text / frames / full-index / daily-index / companyconcept (keyless) (8) ━━━ ADR-0003 / ADR-0017 / ADR-0026 / ADR-0027 / ADR-0041
+    defineTool({
+        name: "edgar_lookup_cik",
+        description: "Resolve a company ticker or name to its 10-digit SEC CIK (keyless, via SEC company_tickers.json). Input `query` (exact ticker or a title substring) ⇒ up to 50 { cik, ticker, title } matches; found:false on none. The CIK is the join key for edgar_company_filings/edgar_company_facts. NOTE: EDGAR keys on CIK, NOT SAM UEI/DUNS — there is no authoritative CIK↔UEI join.",
+        inputSchema: EdgarLookupCikInput,
+        handler: (input) => edgar.lookupCik(input),
+    }),
+    defineTool({
+        name: "edgar_company_filings",
+        description: "A company's SEC filings (keyless, from data.sec.gov submissions). Input `cikOrTicker` (CIK or resolvable ticker/name), optional `forms` (e.g. ['10-K','8-K']), `limit` (≤100, default 20), `offset`, `fullHistory` (default false), `maxShards` (1..100, default 10). Returns filings with the REAL primary-document archive URL. By default returns the recent window (up to 1 year OR 1000 filings, whichever is more); set `fullHistory:true` to ALSO fetch the older filings.files[] shards (newest-first up to `maxShards`) and assemble the COMPLETE history (recent ++ shard001..N, descending, no re-sort). HONESTY: totalAvailable = recent + Σ ALL older-shard counts (the grand total, incl un-fetched shards — never recomputed down), so a capped/failed fan-out reads complete:false; a note discloses COMPLETE vs PARTIAL-BY-CAP (RAISE maxShards for older un-fetched shards — pagination does NOT reach them) vs PARTIAL-BY-FAILURE (a 404/bad-CIK/transient shard is skipped, missing filings disclosed, never fabricated); fullHistory serializes N shard GETs through the shared EDGAR throttle gate. NOTE: EDGAR keys on CIK, NOT SAM UEI/DUNS.",
+        inputSchema: EdgarCompanyFilingsInput,
+        handler: (input) => edgar.companyFilings(input),
+    }),
+    defineTool({
+        name: "edgar_company_facts",
+        description: "Curated XBRL financial facts for a filer (keyless, from data.sec.gov companyfacts). Input `cikOrTicker`, optional `concepts` (default: 6 curated USD concepts — Revenues/RevenueFromContractWithCustomerExcludingAssessedTax, Assets, Liabilities, StockholdersEquity, NetIncomeLoss, CashAndCashEquivalentsAtCarryingValue), `unit` (default USD), `latest`. A concept absent for the filer is OMITTED (never 0); a concept present only in another unit (e.g. EPS in USD/shares) is reported under wrongUnit with a note.",
+        inputSchema: EdgarCompanyFactsInput,
+        handler: (input) => edgar.companyFacts(input),
+    }),
+    defineTool({
+        name: "edgar_full_text_search",
+        description: "Full-text search across EDGAR filings, 2001-present (keyless, efts.sec.gov). Input `q` (phrase in double-quotes for exact), optional `forms`, `startdt`/`enddt` (ISO), `ciks` (pin filings BY entities — numeric 10-digit SEC CIKs, zero-padded, exact-entity match), `entityName` (FUZZY filer-name narrowing — can match related filers, e.g. multiple 'Apple*'), `from` (offset; page size FIXED at 100 — no size param). Returns { accession, form, filingDate, entityNames, ciks, filingIndexUrl }. HONESTY: totalAvailable = the true match count, or a LOWER BOUND (totalIsLowerBound:true) when SEC reports ≥10000; a 0-result set with ciks/entityName applied is NOT proof of absence (verify the CIK via edgar_lookup_cik by name/ticker); from > 9900 is rejected (10000-result window). NOTE: EDGAR keys on CIK, NOT SAM UEI/DUNS.",
+        inputSchema: EdgarFullTextSearchInput,
+        handler: (input) => edgar.fullTextSearch(input),
+    }),
+    defineTool({
+        name: "edgar_xbrl_frames",
+        description: "Keyless cross-filer XBRL cross-section (SEC EDGAR frames, data.sec.gov). In ONE call, return EVERY filer's reported value for a single us-gaap/dei concept in a single calendar period — the complete cross-section — for peer benchmarking + distribution stats. Input `tag` (EXACT alnum concept, e.g. 'Assets'), `period` (CY2023 annual · CY2023Q1 quarterly · CY2023Q4I instant/trailing-I), optional `taxonomy` (us-gaap|dei), `unit` (default USD; EPS uses 'USD-per-shares'), `limit`/`offset` (CLIENT-SIDE window over the fully-fetched set), `includeStats`. Rows: { accn, cik, entityName, loc, end, val, start? } (start only for duration concepts). HONESTY: totalAvailable = SEC's own pts (asserted === data.length, else schema_drift THROW — no fake completeness); the whole frame is fetched upstream in one call and limit/offset is a disclosed client-side page (never a subset labeled complete); a tag/unit/period mismatch ⇒ 404 ⇒ found:false (NEVER a fabricated val:0); val is null-never-0; includeStats covers the FULL set with linear-interpolated percentiles (count===0 ⇒ all-null, never 0/NaN). taxonomy/tag/unit/period are validated path segments (enum+regex, re-checked pre-fetch) — no injection surface. NOTE: EDGAR keys on CIK, NOT SAM UEI/DUNS.",
+        inputSchema: EdgarXbrlFramesInput,
+        handler: (input) => edgar.xbrlFrames(input),
+    }),
+    defineTool({
+        name: "edgar_filing_index",
+        description: "Bulk cross-filer SEC filing index for a quarter (keyless, from the www.sec.gov EDGAR full-index master.idx). Reads the WHOLE quarter's index (every filer's every filing — CIK|Company|Form|Date|Filename, ~370K rows), FULL-SCANS it, and returns offset-paginated filings matching CLIENT-SIDE filters with the EXACT total. Input `year` (>=1993, <= current year), `quarter` (1..4); optional `formType` (exact form, e.g. '8-K'), `cik` (numeric, leading-zero-safe), `companyContains` (LITERAL case-insensitive substring), `dateFrom`/`dateTo` (ISO YYYY-MM-DD), `limit` (<=1000, def 100), `offset`. Returns { year, quarter, indexFile, returned, totalAvailable, filings:[{ cik, cikPadded, companyName, formType, dateFiled, filename, filingUrl }] }. This is the BULK-ENUMERATION primitive (the per-filer edgar tools need a CIK you already hold; this sweeps a whole quarter by form/date/company, e.g. 'every 8-K in 2024 Q1'). HONESTY: totalAvailable is the EXACT match count over the full quarter scan — never a page length, never a byte-capped subset (SEC ignores HTTP Range); a 0-match result is a genuine EXACT ZERO (complete:true), NOT a truncation; a bounds-valid but unpublished quarter returns HTTP 403 and is surfaced as an AMBIGUOUS both-causes error (quarter-not-published OR the 10 req/s rate-block), never a bare rate-limit and never a fake-empty; a non-index / all-malformed body is refused as schema_drift; a future year / bad quarter is rejected pre-fetch (invalid_input, 0 fetch). The CURRENT quarter grows daily (totalAvailable is exact AS-OF-snapshot). filingUrl is a resolvable archive URL. NOTE: EDGAR keys on CIK, NOT SAM UEI/DUNS — there is no authoritative CIK↔UEI join.",
+        inputSchema: EdgarFilingIndexInput,
+        handler: (input) => edgar.filingIndex(input),
+    }),
+    defineTool({
+        name: "edgar_daily_filing_index",
+        description: "Per-DAY cross-filer SEC filing index (keyless, from the www.sec.gov EDGAR daily-index master.YYYYMMDD.idx). The per-day sibling of edgar_filing_index (~30× smaller): reads ONE calendar day's index (every filer's every filing that day — CIK|Company|Form|Date|File Name, ~8K rows), FULL-SCANS it, and returns offset-paginated filings matching CLIENT-SIDE filters with the EXACT total. Answers the monitoring/alerting question the quarterly tool cannot ('every 8-K filed on 2024-01-03', 'watch a CIK day-by-day'). Input `date` (required ISO YYYY-MM-DD, >=1994-01-01, not future); optional `formType` (exact form, e.g. '8-K'), `cik` (numeric, leading-zero-safe), `companyContains` (LITERAL case-insensitive substring), `limit` (<=1000, def 100), `offset`. Returns { found, date, year, quarter, indexFile, returned, totalAvailable, filings:[{ cik, cikPadded, companyName, formType, dateFiled, filename, filingUrl }] }. HONESTY: totalAvailable is the EXACT match count over the full day scan — never a page length, never a byte-capped subset (SEC ignores HTTP Range). The daily-index's pervasive-403 empty model is disambiguated via the quarter's index.json existence oracle, RECENCY-AWARE: a day NEWER than the newest published index (weekend/holiday/not-yet-disseminated recent trading day) ⇒ found:false, complete:FALSE, retryable not-yet-disseminated note (NEVER a confident empty); an unlisted day INSIDE the covered range (a real weekend/holiday) ⇒ found:false, complete:true genuine-absent; a LISTED day whose .idx 403s ⇒ honest rate_limited; the oracle itself inconclusive ⇒ ambiguous both-causes upstream_unavailable. A non-real/future date is rejected pre-fetch (invalid_input, 0 fetch); a non-index / all-malformed body is refused as schema_drift. dateFiled is normalized to ISO from the compact YYYYMMDD column. NOTE: EDGAR keys on CIK, NOT SAM UEI/DUNS — there is no authoritative CIK↔UEI join.",
+        inputSchema: EdgarDailyFilingIndexInput,
+        handler: (input) => edgar.dailyFilingIndex(input),
+    }),
+    defineTool({
+        name: "edgar_company_concept",
+        description: "One filer × one XBRL concept × the COMPLETE reported time-series (keyless, from data.sec.gov companyconcept). The focused financial-TREND / entity-vetting primitive BETWEEN edgar_company_facts (many curated concepts for one filer) and edgar_xbrl_frames (one concept across ALL filers for one period) — 'track THIS filer's Assets/Revenues/NetIncomeLoss OVER TIME, and was it ever revised?'. Input `cikOrTicker` (CIK or resolvable ticker/name), `concept` (EXACT alnum XBRL tag, e.g. 'Assets'), optional `taxonomy` (us-gaap|dei|ifrs-full, def us-gaap), `unit` (CLIENT-SIDE key filter), `form`/`fy` (client-side), `canonicalOnly` (def false), `limit`/`offset`. Returns { found, cik, entityName, taxonomy, concept, label, description, unitsAvailable:[{unit,count}], rows:[{ unit, start, end, val, accn, fy, fp, form, filed, frame, canonical }] }. HONESTY: (M1) period identity is the (start,end) PAIR — every row carries `start` (null for INSTANT concepts, the ISO date for DURATION/flow concepts); the SAME `end` with a DIFFERENT `start` is a different-duration fact (a 3-month quarter vs the 12-month year), NOT a revision — a revision is only multiple rows sharing the same (start,end) with a differing accn/filed/val. DEFAULT returns ALL rows incl. the amendment/restatement history + a per-row `canonical` (frame-tagged = SEC's consolidated value); `canonicalOnly:true` dedups to one canonical row per (unit,start,end), fully disclosed, never a silent drop. Every row is unit-tagged (a USD amount is NEVER conflated with a share count); unitsAvailable discloses ALL units with their RAW counts even under a unit filter; val is null-never-0. A bad CIK/taxonomy/concept ⇒ upstream 404 ⇒ found:false (NEVER a fabricated val:0); a 5xx/timeout/non-JSON/units-shape-drift THROWS; a `unit` not present ⇒ honest empty + the available-units note (unit is CLIENT-SIDE, not a path segment). cik/taxonomy/concept are validated path segments (regex+enum, re-checked pre-fetch) — no injection surface. NOTE: EDGAR keys on CIK, NOT SAM UEI/DUNS — there is no authoritative CIK↔UEI join.",
+        inputSchema: EdgarCompanyConceptInput,
+        handler: (input) => edgar.companyConcept(input),
+    }),
+    // ━━━ Socrata / SODA — keyless SLED + E-rate open data (2) ━━━ ADR-0004
+    defineTool({
+        name: "socrata_query",
+        description: "Query rows from an allowlisted Socrata/SODA open-data portal (keyless; ~a dozen US state portals + USAC E-rate on one identical API — state spend/checkbook/contract/vendor-payment datasets). Input `domain` (curated allowlist enum — the SSRF host guard), `datasetId` (4x4, from socrata_discover_datasets), optional SoQL `select`/`where`/`order`/`q`, `limit` (≤1000, def 100), `offset`, `withTotal` (def true). HONESTY: SODA's row response has no total, so a count(*) companion supplies an exact totalAvailable; if it fails the rows still return with totalAvailable:null + a note (hasMore is then inferred from page-fill, never a false complete). Genuine-empty ⇒ complete:true/total:0; an outage/400/404 THROWS (never a fake empty). Value fields are strings.",
+        inputSchema: SocrataQueryInput,
+        handler: (input) => socrata.query(input),
+    }),
+    defineTool({
+        name: "socrata_discover_datasets",
+        description: "Find Socrata dataset 4x4 ids by keyword via the Socrata catalog (keyless, api.us.socrata.com). Input `q` (e.g. 'procurement', 'vendor payments'), optional `domain` (scope to one allowlisted portal; omit to search the whole allowlist), `limit` (≤100, def 20). Returns [{ id, name, description, domain, updatedAt, link }] + totalAvailable = the catalog resultSetSize. Feed a result's `id` to socrata_query as `datasetId`. NOTE: the federated catalog does not index every allowlisted host (e.g. USAC E-rate) — those stay queryable via socrata_query with a known 4x4.",
+        inputSchema: SocrataDiscoverDatasetsInput,
+        handler: (input) => socrata.discoverDatasets(input),
+    }),
+    // ━━━ CKAN datastore_search — keyless SLED open data (2) ━━━ ADR-0006
+    defineTool({
+        name: "ckan_query",
+        description: "Query rows from an allowlisted CKAN datastore resource (keyless; the FIRST source on the R2 DataSource port — state/city spend/checkbook/procurement/vendor tables on the identical CKAN Action API). Input `host` (curated allowlist enum — the SSRF host guard: data.ca.gov, data.virginia.gov, data.boston.gov), `resourceId` (36-char lowercase UUID, from ckan_discover_datasets), optional `q` (full-text), `filters` (constrained object {field:value} we JSON.stringify), `sort`, `limit` (≤1000, def 100), `offset`. HONESTY: CKAN's envelope carries a real result.total — the DEFAULT is an EXACT total (exact totalAvailable + hasMore); the rare estimated total (total_was_estimated:true) is disclosed via totalIsEstimated + a note and does NOT drive pagination (it can be above OR below the truth). Genuine-empty ⇒ complete:true/total:0; an outage/404/409 or success:false THROWS (never a fake empty). Values are typed per result.fields[].type.",
+        inputSchema: CkanQueryInput,
+        handler: (input) => ckan.query(input),
+    }),
+    defineTool({
+        name: "ckan_discover_datasets",
+        description: "Find CKAN datastore resource ids by keyword via package_search (keyless). Input `host` (allowlisted enum), `q` (e.g. 'procurement', 'checkbook'), `limit` (≤100, def 20). Returns per-resource rows [{ resourceId, name, datasetTitle, format, datastoreActive }] + totalAvailable = the matching DATASET count. Feed a datastoreActive:true result's `resourceId` to ckan_query (a datastoreActive:false resource is a raw file blob NOT in the datastore, not queryable).",
+        inputSchema: CkanDiscoverDatasetsInput,
+        handler: (input) => ckan.discoverDatasets(input),
+    }),
+    // ━━━ FDIC BankFind Suite — keyless institution directory + financials + failures + history + industry aggregates (5) ━━━ ADR-0028 / ADR-0029 / ADR-0030 / ADR-0031
+    defineTool({
+        name: "fdic_search_institutions",
+        description: "Search the FDIC-insured-institution directory (keyless FDIC BankFind, api.fdic.gov/banks/institutions) — a regulated-entity directory for B2G counterparty / bank due-diligence. Structured filters: `state` (2-letter, → STALP), `activeOnly` (→ ACTIVE 1/0), `cert` (→ CERT, the STABLE entity key), plus `name`/`city` matched via FDIC's case-insensitive full-text `search` param (NOT `filters` — `filters=NAME:\"chase\"` is case-sensitive exact-keyword and returns a false-empty; `search=NAME:chase` finds JPMorgan Chase etc.). `limit` (≤1000, def 100), `offset` (≤100000), `sortBy` (allowlisted enum NAME/CERT/ASSET/ESTYMD/STALP/CITY/ACTIVE), `sortOrder` (ASC/DESC). Returns { institutions:[{ name, city, state, cert, assetUSD, active, establishedDate, id }] }. HONESTY: totalAvailable is the EXACT meta.total (stable across offset — never the page length); ASSET is published in $thousands and normalized to whole USD ×1000 (null-never-0 — a real 0 stays 0, absent → null); the ONLY honest empty is meta.total:0/data:[] ⇒ complete:true/total:0, every other envelope (400 errors[]/404/non-JSON/missing meta or data) THROWS (never a fake empty); a multi-word name/city is matched per-token (disclosed); the point-in-time snapshot build time is disclosed. NOTE: FDIC keys on CERT, not SAM UEI/DUNS.",
+        inputSchema: FdicSearchInstitutionsInput,
+        handler: (input) => fdic.searchInstitutions(input),
+    }),
+    defineTool({
+        name: "fdic_institution_financials",
+        description: "Quarterly financial time-series for ONE FDIC-insured institution by certificate number (keyless FDIC BankFind, api.fdic.gov/banks/financials). Input `cert` (REQUIRED FDIC certificate number, from fdic_search_institutions), `limit` (≤1000, def 100), `offset` (≤100000), `sortBy` (allowlisted enum REPDTE/ASSET/DEP/NETINC, def REPDTE), `sortOrder` (def DESC → newest quarter first). Returns { cert, financials:[{ cert, reportDate, assetUSD, depositsUSD, netIncomeUSD, id }] } (e.g. CERT 10363 → 169 quarterly rows). HONESTY: totalAvailable is the EXACT meta.total (stable across offset — page via offset for the full history); ASSET/DEP/NETINC are published in $thousands and normalized to whole USD ×1000 (null-never-0); the ONLY honest empty is meta.total:0/data:[] ⇒ complete:true/total:0, every other envelope THROWS (never a fake empty); the snapshot build time is disclosed.",
+        inputSchema: FdicInstitutionFinancialsInput,
+        handler: (input) => fdic.institutionFinancials(input),
+    }),
+    defineTool({
+        name: "fdic_bank_failures",
+        description: "Historical FDIC-insured bank failures & assistance transactions (keyless FDIC BankFind, api.fdic.gov/banks/failures) — B2G counterparty / entity due-diligence: a failed or FDIC-assisted institution is a red flag, and CERT links a failure back to fdic_search_institutions / fdic_institution_financials. Exact-key filters: `state` (2-letter → PSTALP — NOTE the /failures state field is PSTALP, NOT STALP), `failYear` (→ FAILYR; e.g. 2023 → the 5 real 2023 failures incl. Silicon Valley Bank & First Republic Bank), `cert` (→ CERT, the STABLE entity key). `limit` (≤1000, def 100), `offset` (≤100000), `sortBy` (allowlisted enum FAILDATE/COST/QBFASSET/QBFDEP/NAME/FAILYR, def FAILDATE), `sortOrder` (def DESC → most-recent first). Returns { failures:[{ name, cert, failDate, failYear, city, state, resolutionType, resolutionFund, estimatedLossUSD, depositsUSD, assetsUSD, id }] }. NO name/city filter — FDIC's /failures `search` param is IGNORED (it returns the whole dataset), so name/city are SHOWN in each row but NOT searchable; to find a specific bank's failure, resolve its CERT via fdic_search_institutions then filter here by `cert`. HONESTY: totalAvailable is the EXACT meta.total (stable across offset — never the page length); failDate is normalized from FDIC's M/D/YYYY to ISO YYYY-MM-DD (an unrecognized value is surfaced raw + disclosed, never nulled/fabricated); COST/QBFDEP/QBFASSET are $thousands normalized to whole USD ×1000 (null-never-0 — a genuine 0 = a fully-assisted no-loss stays 0, a NEGATIVE COST = a net DIF recovery/gain not a loss, absent → null); the ONLY honest empty is meta.total:0/data:[] ⇒ complete:true/total:0, every other envelope (400 errors[]/404/non-JSON/missing meta or data) THROWS (never a fake empty); the point-in-time snapshot build time is disclosed. NOTE: FDIC keys on CERT, not SAM UEI/DUNS.",
+        inputSchema: FdicBankFailuresInput,
+        handler: (input) => fdic.bankFailures(input),
+    }),
+    defineTool({
+        name: "fdic_institution_history",
+        description: "Institution-level STRUCTURAL-CHANGE event log for FDIC-insured banks (keyless FDIC BankFind, api.fdic.gov/banks/history) — the full lineage of mergers, absorptions, consolidations, failures, name/location/charter/regulator changes, branch open/close, trust-power grants & FRS-membership changes. Completes the FDIC entity cluster (directory + financials + failures + history). Killer feature: CERT-linked MERGER LINEAGE — a merger/failure row carries the acquiring / outgoing / surviving institution's CERT + name, each linking back to fdic_search_institutions / fdic_institution_financials / fdic_bank_failures. Exact-key filters (all optional, AND-combined): `cert` (→ CERT, the STABLE entity key & PRIMARY lookup; e.g. 3510 → Bank of America's 13,794 rows), `changeCode` (→ CHANGECODE; e.g. 223 = merger, 211 = failure, 721 = branch closing, 520 = location change), `effYear` (→ EFFYEAR), `state` (2-letter → PSTALP — NOTE the /history state field is PSTALP, NOT STALP). `limit` (≤1000, def 100), `offset` (≤100000), `sortBy` (allowlisted enum EFFDATE/PROCDATE/CHANGECODE/TRANSNUM, def EFFDATE), `sortOrder` (def DESC → newest change first). Returns { history:[{ cert, instName, state, changeCode, changeDescription, effectiveDate, processDate, effYear, transNum, acquirerCert, acquirerName, outgoingCert, outgoingName, survivingCert, survivingName, id }] }. NO name/city filter — FDIC's /history `search` param returns 0 for INSTNAME (a false-empty), so names are SHOWN in each row but NOT searchable; to find a specific bank's history, resolve its CERT via fdic_search_institutions then filter here by `cert`. HONESTY: totalAvailable is the EXACT meta.total (stable across offset — never the page length); changeDescription is FDIC's OWN co-served CHANGECODE_DESC passed through verbatim (the numeric changeCode is authoritative — never a hand-map); effectiveDate/processDate are normalized from FDIC's YYYY-MM-DDT00:00:00 to ISO YYYY-MM-DD (an unrecognized value is surfaced raw + disclosed, never nulled/fabricated); the acquirer/outgoing/surviving CERTs are null on a non-merger event (null-never-0 — a real absence, never a fabricated 0; *_UNINUM's 0 sentinel is NOT surfaced); the ONLY honest empty is meta.total:0/data:[] ⇒ complete:true/total:0, every other envelope (400 errors[]/404/non-JSON/missing meta or data) THROWS (never a fake empty); the point-in-time snapshot build time is disclosed. NOTE: FDIC keys on CERT, not SAM UEI/DUNS.",
+        inputSchema: FdicInstitutionHistoryInput,
+        handler: (input) => fdic.institutionHistory(input),
+    }),
+    defineTool({
+        name: "fdic_industry_summary",
+        description: "FDIC industry & state banking-sector ANNUAL AGGREGATES — the FDIC's own roll-ups (keyless FDIC BankFind, api.fdic.gov/banks/summary). The FIRST aggregate/statistical FDIC tool (the other 4 are per-ENTITY, keyed on CERT): total assets, deposits, net income, equity & net interest income + structural counts (institutions, offices, branches, employees) for the whole US banking industry OR one state/territory in one year, split by charter class. Answers 'how big is the US (or a state's) banking industry this year, and how many institutions?' — a question the entity tools cannot express without summing thousands of rows. Exact-key filters (all optional, AND-combined): `year` (→ YEAR; e.g. 2023 → 121 rows), `state` (2-or-3-letter → STALP — NOTE the /summary state field is STALP, NOT PSTALP; accepts a jurisdiction code TX/CA/DC/GU/PR… OR a ROLL-UP code USA/US/OT/PI), `charterClass` (CB = commercial banks, SI = savings institutions; omit for both — there is NO combined row). `limit` (≤1000, def 100), `offset` (≤100000), `sortBy` (allowlisted enum YEAR/ASSET/DEP/NETINC/BANKS, def YEAR), `sortOrder` (def DESC → newest year / largest first). Returns { summary:[{ year, charterClass, charterClassCode, geography, stateCode, stateFips, scope, isRollup, institutionCount, officeCount, branchCount, employeeCount, totalAssetsUSD, totalDepositsUSD, netIncomeUSD, totalEquityUSD, netInterestIncomeUSD, id }] }. ★ROLL-UP HONESTY: each row crosses charter × geography; STALP ∈ {USA,US,OT,PI} are GEOGRAPHIC AGGREGATES (scope national_total/national_states_dc/territories_total/pacific_islands, isRollup:true), every other STALP is a jurisdiction (isRollup:false) — NEVER sum a roll-up row with jurisdiction rows or across scopes (national_total = national_states_dc + territories_total; a geography's total = its CB row + its SI row), read the national_total (USA) row directly for one national figure; a roll-up is NOT a state. ★NIM is net interest INCOME (a $ sum surfaced as netInterestIncomeUSD), NOT the margin ratio; this endpoint has NO ratio fields (ROA/ROE — derive from netIncomeUSD/totalAssetsUSD/totalEquityUSD). NO name/city filter — FDIC's /summary `search` param is ignored (returns the whole year); drill to institutions via fdic_search_institutions. HONESTY: totalAvailable is the EXACT meta.total (stable across offset — never the page length); money (ASSET/DEP/NETINC/EQ/NIM) is $thousands → whole USD ×1000 (null-never-0 — a genuine 0 like American Samoa's zero commercial banks stays 0, absent → null), counts (BANKS/OFFICES/BRANCHES/employees) pass through un-scaled (a count ×1000 is a fabrication); a non-int year is rejected pre-fetch (a malformed year is a live HTTP-200 total:0 false-empty); the ONLY honest empty is meta.total:0/data:[] ⇒ complete:true/total:0, every other envelope (400 errors[]/404/non-JSON/missing meta or data) THROWS (never a fake empty); the point-in-time snapshot build time is disclosed. NOTE: FDIC keys on CERT, not SAM UEI/DUNS.",
+        inputSchema: FdicIndustrySummaryInput,
+        handler: (input) => fdic.industrySummary(input),
+    }),
+    // ━━━ FDIC BankFind Suite — WITHIN-SOURCE DEPTH: counterparty risk ratios + branch deposits (2) ━━━ ADR-0040
+    defineTool({
+        name: "fdic_risk_ratios",
+        description: "FDIC counterparty RISK RATIOS for ONE institution by certificate number (keyless FDIC BankFind, api.fdic.gov/banks/financials) — the SOUNDNESS lane the balance-sheet tools cannot express: profitability (ROA/pretax ROA/ROE), net interest margin, efficiency ratio, asset quality (net charge-offs to loans), capital adequacy (leverage, tier-1 risk-based, total risk-based ratios) + the tier-1 capital LEVEL. Input `cert` (REQUIRED FDIC certificate number, from fdic_search_institutions), `reportDate` (optional YYYYMMDD quarter-end → REPDTE; omit for the full quarterly time-series), `limit` (≤1000, def 100), `offset` (≤100000), `sortBy` (allowlisted enum REPDTE/ROA/ROE/RBCRWAJ/EEFFR, def REPDTE), `sortOrder` (def DESC → newest quarter first). Returns { cert, ratios:[{ cert, reportDate, cblrFramework, returnOnAssetsPct, preTaxReturnOnAssetsPct, returnOnEquityPct, netInterestMarginPct, efficiencyRatioPct, netChargeOffsToLoansPct, leverageRatioPct, tier1RiskBasedCapitalRatioPct, totalRiskBasedCapitalRatioPct, tier1CapitalUSD, id }] }. ★UNITS-IN-THE-KEY: every *Pct field is an FDIC-published PERCENTAGE surfaced VERBATIM (no scaling, no recompute) — do NOT read it as a dollar amount or ×1000-scale it; tier1CapitalUSD is a DOLLAR amount (FDIC publishes it in $thousands, normalized ×1000). ★NULL-NEVER-0: a not-reported ratio is null (never 0% — a false 'no return / no capital'). ★CBLR (community-bank-leverage) banks (cblrFramework:true) do NOT report the risk-based capital ratios — FDIC returns a literal 0 for the total risk-based ratio, which this tool maps to null for BOTH tier1RiskBasedCapitalRatioPct and totalRiskBasedCapitalRatioPct (a null there is a normal framework artifact, read alongside leverageRatioPct — NOT a 0% capital red flag). No ratio is recomputed; each is exactly FDIC's published Call-Report figure. HONESTY: totalAvailable is the EXACT meta.total (stable across offset); the ONLY honest empty is meta.total:0/data:[] ⇒ complete:true/total:0, every other envelope (400 errors[]/404/non-JSON/missing meta or data) THROWS (never a fake empty); the snapshot build time is disclosed. NOTE: reported regulatory metrics, NOT a soundness rating or failure prediction; FDIC keys on CERT, not SAM UEI/DUNS.",
+        inputSchema: FdicRiskRatiosInput,
+        handler: (input) => fdic.riskRatios(input),
+    }),
+    defineTool({
+        name: "fdic_branch_deposits",
+        description: "FDIC branch-deposit footprint — the Summary of Deposits (keyless FDIC BankFind, api.fdic.gov/banks/sod): the annual June-30 branch-office deposit distribution ('where does this bank hold deposits, and how concentrated?'). Exact-key filters (all optional, AND-combined; ≥1 recommended): `cert` (→ CERT, the STABLE entity key), `state` (2-letter → STALPBR, the branch-state field, C118-quoted so Oregon is operator-safe), `year` (→ YEAR, the June-30 snapshot year). `limit` (≤1000, def 100), `offset` (≤100000), `sortBy` (allowlisted enum YEAR/DEPSUMBR, def YEAR), `sortOrder` (def DESC → newest snapshot / largest deposits first). Returns { branches:[{ cert, institutionName, branchNumber, branchName, city, state, zip, address, depositsUSD, year, id }] } (e.g. CERT 10004 → 74 branch-year rows). HONESTY: totalAvailable is the EXACT meta.total (stable across offset — never the page length); depositsUSD is DEPSUMBR published in $thousands, normalized to whole USD ×1000 (null-never-0 — a real 0 stays 0, absent → null); a bad/mistyped filter field can never reach the wire (server-side allowlist by construction — FDIC would otherwise return a silent total:0 false-empty, not an error); the ONLY honest empty is meta.total:0/data:[] ⇒ complete:true/total:0, every other envelope (400 errors[]/404/non-JSON/missing meta or data) THROWS (never a fake empty); the DISTINCT annual snapshot build time is disclosed. Branch facility data only (name/address/city/state/zip/deposits) — no personal/officer PII. NOTE: SOD is an annual June-30 snapshot; FDIC keys on CERT, not SAM UEI/DUNS.",
+        inputSchema: FdicBranchDepositsInput,
+        handler: (input) => fdic.branchDeposits(input),
+    }),
+    // ━━━ USITC Harmonized Tariff Schedule — keyless import-tariff / duty-rate lookup (1) ━━━ ADR-0039
+    defineTool({
+        name: "hts_lookup",
+        description: "Look up US import-tariff classification + duty rates from the USITC Harmonized Tariff Schedule (keyless; hts.usitc.gov/reststop/search) — the IMPORT-TARIFF / supply-chain PRICE lane a product-reseller / supply-chain bidder needs to price a hardware or commodity contract (extends the THIN Price lane with a NON-labor cost input, a sibling of gsa_benchmark_labor_rates). A single `query` serves BOTH modes: a KEYWORD (e.g. 'laptop', 'cotton shirt') OR an HTS number (e.g. '8471.30' / '8471.30.01.00') — both ride the `keyword=` search. Returns { query, lines:[{ htsno, statisticalSuffix, indent, description, units, columnOneGeneral, specialPreferential, columnTwo, additionalDuties, footnotes, quotaQuantity, effectivePeriod, status, isChapter99 }] } + honest _meta. ★DUTY-RATE HONESTY (the crux): columnOneGeneral (Column-1 General), specialPreferential (Special/preferential/FTA), and columnTwo (Column-2) are AUTHORITATIVE VERBATIM TEXT surfaced as strings — 'Free', a percentage ('35%'), a specific rate ('0.47¢/kg'), a compound/range, or null — NEVER coerced to a number (a coerced 0/NaN would fabricate a false 'duty-free'); an empty Special ('') → null = NO special-program rate published (NEVER read as Free). ★HIERARCHY (M1): a lookup returns rows across levels; the rate is stated ONCE at a shallower level (usually the 6/8-digit subheading) and inherits DOWNWARD to the blank statistical-suffix lines — to find a specific line's rate, read UP to the nearest ANCESTOR line (shallower indent, same htsno prefix) with a non-empty rate; a blank deepest line is NOT no/unknown duty. ★ADDITIONAL DUTIES (S1): the per-line `additionalDuties` is frequently null even when Section 301/232 duties apply — the real additional duty rides the Chapter-99 rows (isChapter99:true, htsno beginning '99') returned alongside the base line + the footnotes; they STACK on the base rate. ★COMPLETENESS (M2): the endpoint returns the FULL match array with NO server-side total and NO working pagination (offset is IGNORED) → totalAvailable is the EXACT served array length and paging is CLIENT-SIDE; there is no fixed cap (a single-char/common fragment can return 10,000–16,000+ rows / several MB), so `query` must be ≥3 non-whitespace chars (a 1–2 char query is rejected invalid_input before the fetch). `limit` (≤200, def 50), `offset`. A no-match ⇒ honest empty; a 404/5xx/timeout/non-array/HTML(→schema_drift) ⇒ THROWS (never a fake empty); a transient 400 on the validated query ⇒ upstream_unavailable (retryable). NOT a binding CBP classification ruling and NOT a landed-cost quote — the duty owed depends on country of origin + trade program + Section 301/232 / Chapter-99 additional duties + footnotes; confirm via CBP (CROSS / eRulings). The not-a-ruling caveat rides EVERY response.",
+        inputSchema: HtsLookupInput,
+        handler: (input) => usitc.htsLookup(input),
+    }),
+    // ━━━ BLS Public Data API v1/v2 — keyless US labor/price time series (1) ━━━ ADR-0032
+    // A NEW capability axis: the PRICING / ESCALATION layer (CPI-U & ECI EPA-clause
+    // escalation, PPI materials benchmarking, CES labor-rate context). The SECOND
+    // POST-batch getJson-port consumer (after NIH). SSRF surface = a compile-time-
+    // constant host+path; seriesids ride in the module-built POST body. Honesty crux:
+    // the "-" unavailable marker → null-never-0 with the footnote reason surfaced; a
+    // non-SUCCESS status THROWS (never a fake-empty); per-series units are labeled;
+    // the span is clamped to the tier cap + disclosed. An OPTIONAL free BLS_API_KEY
+    // rides ONLY in the POST body (v2) — never a URL/header/label/_meta/log.
+    defineTool({
+        name: "bls_timeseries",
+        description: "Fetch US Bureau of Labor Statistics time series — the PRICING / ESCALATION layer (keyless; api.bls.gov Public Data API v1, POST/JSON batch). CPI-U & ECI drive federal contract escalation / economic-price-adjustment (EPA) clauses; PPI benchmarks materials pricing; CES employment/wages give labor-rate context (next to gsa_benchmark_labor_rates + sam wage determinations). Inputs (at least one of series/seriesId REQUIRED; both combinable): `series` — a FROZEN 9-key CURATED enum (typo-proof; each carries meaning + units): cpi_u_all/cpi_u_core (CPI-U index, NSA — the escalation reference), ppi_final_demand (PPI index), eci_total_comp/eci_wages (★12-MONTH % CHANGE, NOT an index — a consumer misreads 3.4 as an index level otherwise), unemployment_rate/labor_force_participation (percent, SA), employment_total_nonfarm (thousands of persons, SA), avg_hourly_earnings (dollars/hour, SA). `seriesId` — raw BLS IDs (charclass ^[A-Z0-9]{1,20}$; the OEWS/local-area/regional passthrough; units:null for a raw ID). `startYear`/`endYear` (1900..currentYear+1; default a ~10-year window; span CLAMPED to the tier cap ~10y and disclosed). Returns { series:[{ seriesId, key, meaning, units, observations:[{ year, period, periodName, value, valueUnavailable, footnotes, latest }], observationCount, coveredRange }] } + honest _meta. HONESTY: each `value` is PARSED number|null — the BLS \"-\" unavailable marker (e.g. the 2025 lapse-in-appropriations gap) → null NEVER 0, with valueUnavailable:true + the footnote reason on the observation AND lifted into _meta.notes (a data gap is DISCLOSED, never a silent null and never a fabricated 0); a genuine \"0\" stays 0. A non-SUCCESS status THROWS (never a fake-empty): REQUEST_NOT_PROCESSED (the v1 ~25/day limit) ⇒ rate_limited with the tier disclosure; REQUEST_FAILED ⇒ upstream_unavailable/invalid_input surfacing message[]. A non-JSON 200 or a SUCCESS body missing Results.series ⇒ schema_drift. An empty data[] on SUCCESS ⇒ observations:[] + an ambiguity note (a curated key = a genuine empty range; a raw seriesId = EITHER genuine-empty OR a nonexistent/typo'd ID — verify it). Every response discloses the active tier (v1 keyless ~25/day, 25 series/query, ~10y span | v2 with a free BLS_API_KEY ~500/day) + the per-series units caveat. An OPTIONAL free BLS_API_KEY (env; https://data.bls.gov/registrationEngine/) lifts to v2 and is sent ONLY in the request body — never a URL/header/log.",
+        inputSchema: BlsTimeseriesInput,
+        handler: (input) => bls.timeseries(input),
+    }),
+    // ━━━ BLS OEWS — keyless occupational wage benchmarking (2nd BLS tool) ━━━ ADR-0033
+    // The LEVEL layer next to bls_timeseries's ESCALATION layer: mean/median annual &
+    // hourly wages + employment by SOC occupation × geography — the highest-value B2G
+    // BLS slice (labor-rate benchmarking) that bls_timeseries structurally cannot reach
+    // (OEWS IDs are 25 chars > the raw-seriesId 20-char cap). BUILDS the 25-char series
+    // ID INTERNALLY from validated structured inputs; REUSES the same POST/JSON transport
+    // + parseBlsBody status-throw + mapObservation ("-"→null-never-0) + tier/key seam.
+    defineTool({
+        name: "bls_oews_wages",
+        description: "Benchmark US occupational wages & employment from BLS OEWS (Occupational Employment & Wage Statistics) — the LEVEL layer for labor-rate benchmarking (keyless; api.bls.gov Public Data API, POST/JSON batch). The actual mean/median annual & hourly wage a labor category commands, by area — next to gsa_benchmark_labor_rates (GSA CALC), sam wage determinations, and bls_timeseries (the CPI/ECI escalation layer). OEWS series IDs are 25 chars (area×occupation×industry×datatype), EXCEEDING bls_timeseries's raw-seriesId cap, so this tool BUILDS the ID INTERNALLY from validated structured inputs. Inputs (at least one of occupation/soc REQUIRED; all arrays batch into ONE POST — the cartesian product area×occupation×datatype is capped at the active tier's series cap and refused over-cap WITH THE COUNT NAMED, never silently truncated): `occupation` — a CURATED 16-key SOC enum (typo-proof; e.g. software_developer=15-1252, civil_engineer=17-2051, management_analyst=13-1111); `soc` — raw 6-digit HYPHENLESS SOC codes for the ~830-SOC long tail (use 151252, not 15-1252); `area` — default [\"national\"]; each is \"national\", a 2-letter USPS state code (CA/TX/DC…), or a 5-digit CBSA metro code (19100 = Dallas-Fort Worth); `datatype` — default [\"annual_mean\"]: annual_mean/annual_median (dollars/year), hourly_mean/hourly_median (dollars/hour), employment (count jobs). NO year input — OEWS is ANNUAL and the API serves only the latest release; the tool requests a recent window internally and DISCLOSES the reference year. Returns { results:[{ area:{type,code,label}, occupation:{soc,key,label}, measure:{key,code,units}, value:number|null, valueUnavailable, referenceYear, referencePeriod, footnotes, seriesId }] } + honest _meta. HONESTY: (H1) OEWS is an ANNUAL point-in-time snapshot (reference May <year>, period A01), NOT monthly/current-quarter — disclosed every call; (H2) a built ID that returns empty/absent ⇒ value:null, valueUnavailable:FALSE (the occupation is not surveyed/estimated there OR the cell is suppressed for confidentiality) + the not-published note + the surfaced upstream \"Series does not exist\" message + the ID in fieldsUnavailable — NEVER a fabricated 0; a PRESENT \"-\" in-band value ⇒ null + valueUnavailable:true + footnote; (H3) each row's measure.units labels the datatype (never read an employment count as a wage); (H4) the API returns real numerics (no top-code); a non-SUCCESS status THROWS (REQUEST_NOT_PROCESSED ⇒ rate_limited with the tier disclosure; a non-JSON 200 ⇒ schema_drift). Every response discloses the active tier. An OPTIONAL free BLS_API_KEY lifts to v2 and is sent ONLY in the request body — never a URL/header/log.",
+        inputSchema: BlsOewsWagesInput,
+        handler: (input) => bls.oewsWages(input),
+    }),
+    // ━━━ BLS QCEW — county×NAICS market-size / wages / location-quotient (3rd BLS tool) ━━━ ADR-0042
+    // A SECOND, DIFFERENT, keyless, un-rate-limited BLS DOMAIN (data.bls.gov/cew — the
+    // QCEW Open Data Access CSV, NOT the rate-limited api.bls.gov/publicAPI timeseries
+    // API). Answers the market-size / competition-density question no existing tool can:
+    // establishment COUNT (market size / competitor density), county×NAICS employment,
+    // avg weekly wage (labor cost), and the LOCATION QUOTIENT (concentration vs national).
+    // Honesty crux: a suppressed employment/wage 0-sentinel → null (never 0), block/code/
+    // field-scoped (base/lq/oty each keyed on its OWN *_disclosure_code; qtrly_estabs /
+    // lq_qtrly_estabs / oty_qtrly_estabs_chg stay disclosed under 'N'). Symmetric CSV
+    // column-drift + a POST-parse quoted-header assertion → schema_drift. A per-tuple 404
+    // → honest empty. NEW gate key "bls_qcew"; NO BLS_API_KEY on this keyless path.
+    defineTool({
+        name: "bls_qcew",
+        description: "BLS QCEW (Quarterly Census of Employment & Wages) — county×NAICS MARKET-SIZE / wages / location-quotient (keyless; data.bls.gov/cew Open Data Access CSV, a SECOND un-rate-limited BLS domain — NOT the ~25/day api.bls.gov timeseries API). Answers the market-size / competition-density question no other tool can: for ONE area_fips (county/state/metro/US) OR ONE NAICS × quarter — establishment COUNT (market size / competitor density), county×NAICS employment, average weekly wage (labor cost), and the LOCATION QUOTIENT (lq_* = concentration vs the national average; >1.00 = more concentrated / higher competition density). Inputs: `mode` (REQUIRED {area,industry}); `area` (area_fips ^[0-9A-Za-z]{1,6}$ — REQUIRED path segment for mode=area, else an optional client-side narrow); `industry` (NAICS ^[0-9]{1,6}$ DIGIT-ONLY — REQUIRED path segment for mode=industry, else an optional narrow; a hyphenated 31-33 404s, use the digit aggregate); `year` (REQUIRED 1990..current), `quarter` (REQUIRED 1|2|3|4); client-side `ownership`(own_code)/`aggregationLevel`(agglvl_code)/`sizeCode`; `limit` (≤1000, def 50)/`offset`. Wire: GET data.bls.gov/cew/data/api/{year}/{quarter}/{mode}/{code}.csv. Returns { found, mode, area|industry, year, quarter, rows:[{ area_fips, own_code, industry_code, agglvl_code, size_code, base:{ disclosed, disclosureCode, qtrly_estabs, month1/2/3_emplvl, total_qtrly_wages, taxable_qtrly_wages, qtrly_contributions, avg_wkly_wage }, locationQuotient:{ disclosed, disclosureCode, lq_qtrly_estabs, lq_… }, overTheYear:{ disclosed, disclosureCode, oty_qtrly_estabs_chg, oty_…_pct_chg } }] } + honest _meta. ★DISCLOSURE-SUPPRESSION HONESTY (the crux): each row carries THREE disclosure codes (base/lq/oty), each governing its block. QCEW encodes a SUPPRESSED (confidential) employment/wage value as a literal 0 — so under 'N' the confidential emplvl/wage/avg-wkly fields map to null (WITHHELD, never a fabricated $0), while the establishment COUNT (qtrly_estabs / lq_qtrly_estabs) AND its over-the-year change (oty_qtrly_estabs_chg / _pct_chg) stay DISCLOSED (real); under '-' the WHOLE block incl. the estabs field(s) → null; under blank a genuine reported/NEGATIVE 0 SURVIVES (the disclosed federal taxable=0/contrib=0 and the oty_*_chg=0 'no change'). NEVER a blanket 0→null. A null carries disclosed:false + the raw disclosureCode; a suppression note fires whenever any page row is suppressed. HONESTY: totalAvailable is the EXACT filtered row count (fetch-once + client-side limit/offset — QCEW does not paginate; never the page length); a per-tuple HTTP 404 ⇒ honest empty (found:false, the HTML 404 body NEVER parsed as CSV); a 5xx/timeout ⇒ THROW; a 200 non-CSV / a renamed/±column header / a wrong field-count row ⇒ schema_drift THROW (symmetric drift guard). The file MIXES aggregation levels + ownerships — a do-NOT-sum-across-agglvl/ownership note rides every response. PUBLIC AGGREGATE stats (the suppression mechanism keeps small-cell data non-identifying — no PII). Keyless, un-rate-limited; NO BLS_API_KEY is read.",
+        inputSchema: BlsQcewInput,
+        handler: (input) => bls.qcew(input),
+    }),
+    // ━━━ OpenFEMA — keyless disaster declarations + emergency-assistance spend (2) ━━━ ADR-0016
+    defineTool({
+        name: "fema_search_public_assistance",
+        description: "Search FEMA Public Assistance funded projects — federal emergency-assistance spend to state/local/tribal applicants (keyless OpenFEMA, dataset PublicAssistanceFundedProjectsDetails v2, ~800k rows). Structured filters (module-built into an OData $filter; each LIVE-VERIFIED to narrow): `state` (→ stateAbbreviation), `disasterNumber`, `applicantId`, `damageCategoryCode` (e.g. 'B' = Emergency Protective Measures), `incidentType`, `minProjectAmount`/`maxProjectAmount` (projectAmount ge/le), `declaredDateFrom`/`declaredDateTo` (declarationDate ge/le). `limit` (≤1000, def 100 → $top), `offset` (→ $skip). HONESTY: the module ALWAYS sends $inlinecount=allpages so totalAvailable is the EXACT filtered total (metadata.count), never the page length; amount fields are number|null (a real 0 stays 0, absent → null); genuine-empty ⇒ complete:true/total:0; an outage/400/404 THROWS (never a fake empty).",
+        inputSchema: FemaSearchPublicAssistanceInput,
+        handler: (input) => fema.searchPublicAssistance(input),
+    }),
+    defineTool({
+        name: "fema_disaster_declarations",
+        description: "Look up FEMA disaster / emergency declarations by state, type, incident, year, or date (keyless OpenFEMA, dataset DisasterDeclarationsSummaries v2, ~70k rows). Structured filters (module-built into an OData $filter; each LIVE-VERIFIED to narrow): `state` (→ state), `incidentType` (e.g. 'Flood'), `declarationType` (DR/EM/FM), `fyDeclared`, `disasterNumber`, `declaredDateFrom`/`declaredDateTo` (declarationDate ge/le), `paProgramDeclared`/`iaProgramDeclared` (booleans). `limit` (≤1000, def 100 → $top), `offset` (→ $skip). HONESTY: the module ALWAYS sends $inlinecount=allpages so totalAvailable is the EXACT filtered total (metadata.count), never the page length; genuine-empty ⇒ complete:true/total:0; an outage/400/404 THROWS (never a fake empty). NOTE: per-dataset OData field names differ — 'state' here is the real field, whereas the public-assistance tool maps 'state' to 'stateAbbreviation'.",
+        inputSchema: FemaDisasterDeclarationsInput,
+        handler: (input) => fema.disasterDeclarations(input),
+    }),
+    // ━━━ FPDS-NG — federal contract AWARD ACTIONS (keyless ATOM) (1) ━━━ ADR-0012
+    // The FIRST XML/ATOM source (bounded, ReDoS-safe hand-parser — the far.ts/gao.ts
+    // lineage; NOT the getJson port). FPDS is the system-of-record USAspending
+    // derives from — this closes the action-level / mod-level latest-truth gap.
+    defineTool({
+        name: "fpds_search_awards",
+        description: "Search FPDS-NG federal contract AWARD ACTIONS (keyless ATOM) — the AUTHORITATIVE system-of-record for contract actions (each modification is its own transaction), the source USAspending.gov derives from (and lags 1-2 days). Structured filters ONLY, AND-combined (NO raw query — a typo'd FPDS field name is a SILENT ZERO, so the tool builds the fielded q): naics (PRINCIPAL_NAICS_CODE), vendorName, piid, departmentId, contractingAgencyName, signedDate range (from/to ISO), lastModified range, keyword. At least one filter is REQUIRED. Returns award/IDV rows { piid, modNumber, parentIdvPiid, actionType, signedDate, vendorName, vendorUei, ultimateParentUei, obligatedAmount, totalObligatedAmount, naics, psc, placeOfPerformanceState, extentCompeted, setAside, businessSize, socioeconomic, … } (content root is award OR IDV — both parse). HONESTY: page size is FIXED at 10; for >10 results totalAvailable is a LOWER BOUND (totalIsLowerBound:true; true count ∈ [total, total+9]) and you MUST paginate by pagination.hasMore (page-fullness), NEVER by totalAvailable (keyless deep-paging is capped ~200K far below the advertised total). Genuine-empty (offset 0) ⇒ complete:true/total:0 + a silent-zero disclosure; an empty page at offset>0 ⇒ totalAvailable:null/complete:false (deep-paging ceiling, ambiguous); an HTML/non-feed body or an all-null-piid page ⇒ schema_drift (never a fake empty); an outage/5xx/timeout THROWS. Amounts are number|null (a 0.00 obligation and negative de-obligations are REAL, absent ⇒ null). Prefer usas_* tools for spending rollups / sub-award graphs.",
+        inputSchema: FpdsSearchAwardsInput,
+        handler: (input) => fpds.searchAwards(input),
+    }),
+    // ━━━ NIH RePORTER v2 — keyless federal research-GRANT projects (1) ━━━ ADR-0014
+    // The R2 getJson port's FIRST non-GET consumer (POST + JSON body). A NEW axis:
+    // federal research-funding footprint by organization / UEI / state (recipient
+    // enrichment, joinable to SAM/USAspending via primary_uei). SSRF surface = a
+    // compile-time-constant URL; all filters ride in the module-built POST body.
+    // Only live-confirmed-narrowing criteria ship (M1); agency_ic_codes is excluded
+    // (silent no-op). The 15,000-record retrieval window is disclosed, not hidden.
+    defineTool({
+        name: "nih_reporter_search_projects",
+        description: "Search awarded NIH RePORTER research-GRANT projects (keyless; api.reporter.nih.gov v2, POST/JSON — the FIRST non-GET getJson-port consumer) — the NEW federal research-funding recipient-enrichment axis (who receives NIH research money, by organization / state, joinable to SAM/USAspending via primary_uei). Structured, LIVE-CONFIRMED-narrowing criteria ONLY, AND-combined in a module-built body (NO raw passthrough): orgStates (UPPERCASE 2-letter USPS enum — the SSRF + silent-zero guard; a lowercase/unknown code silently returns zeros), orgNames (≤512 each, ≤20), fiscalYears (int array 1985..currentYear+1, ≤20), limit (1..500, def 50), offset (0..14,999, def 0). Returns { projects:[{ projectNum, projectTitle, fiscalYear, awardAmount, organization:{ name, state, primaryUei, primaryDuns, ueis, duns }, principalInvestigators, contactPiName, fundingIc }] } + honest _meta. HONESTY: (M2) records are RESEARCH GRANTS, NOT procurement contracts — primary_uei joins to SAM/USAspending recipients but the award nature differs (disclosed in every _meta.notes); totalAvailable = the EXACT meta.total (NEVER the page size, NEVER a lower bound); NIH caps keyless retrieval at the first 15,000 records (offset 0..14,999) — offset ≥ 15,000 ⇒ invalid_input, and past the window the count stays exact while records are UNREACHABLE (disclosed in a note; nextOffset is never a dead-end). Disclose-not-refuse: an unscoped query still returns the first page + the exact total + a narrow-your-criteria note. agencyIcCodes is intentionally NOT a filter (NIH silently drops it — it would be a false 'applied'). Genuine-empty (total:0) ⇒ complete:true/total:0; an outage/5xx/timeout THROWS; a 400 (bad offset/limit/type) ⇒ invalid_input; a 200 body that isn't {meta,results} or a non-numeric meta.total ⇒ schema_drift (never a fake empty). awardAmount is number|null (a real $0 award is 0, an absent amount is null).",
+        inputSchema: NihSearchProjectsInput,
+        handler: (input) => nih.searchProjects(input),
+    }),
+    // ━━━ NSF Awards API — keyless federal research-GRANT awards (2) ━━━ ADR-0020
+    // Source #20. The grant-SIBLING of NIH RePORTER on a DIFFERENT agency: NSF
+    // research-grant awards with recipient / PI / UEI enrichment, strengthening the
+    // WEAK entity/recipient layer (ueiNumber/parentUeiNumber join to SAM/USAspending).
+    // SSRF surface = a compile-time-constant host+path; all filters ride in a
+    // module-built URLSearchParams from a validated whitelist. HONESTY: totalCount is
+    // EXACT below 10,000 and SATURATES at 10,000 (ES track_total_hits ⇒ totalIsLower-
+    // Bound + a note); the offset+rpp ≤ 10,000 retrieval window is clamped/disclosed;
+    // a multi-word keyword is OR-tokenized (disclosed, M1); a serviceNotification at
+    // HTTP 200 loud-fails (never a fake empty); grant≠contract in every response.
+    defineTool({
+        name: "nsf_search_awards",
+        description: "Search awarded NSF research-GRANT awards (keyless; api.nsf.gov/services/v1/awards.json) — the NEW federal research-funding recipient-enrichment axis (who receives NSF research money, by organization / UEI / PI / state, joinable to SAM/USAspending via ueiNumber/parentUeiNumber). The grant-SIBLING of nih_reporter_search_projects on a different agency. LIVE-CONFIRMED-narrowing filters ONLY, module-built into a URLSearchParams query (NO raw passthrough): keyword (free text; MULTI-WORD is OR-tokenized — 'machine learning' = machine OR learning, disclosed in _meta.notes), awardeeStateCode (UPPERCASE 2-letter USPS enum — the SSRF + silent-zero guard; a non-state typo silently returns 0), awardeeName, ueiNumber (12-char UEI — an EXACT SAM/USAspending join), parentUeiNumber (parent-org roll-up), pdPIName, dateStart/dateEnd (STRICT mm/dd/yyyy on the award ACTION date — a wrong format is silently mis-parsed), limit (1..100, def 25 → rpp), offset (0..9999). Returns { awards:[{ id, title, agency, cfdaNumber, transType, awardee:{ name, city, stateCode, ueiNumber, parentUeiNumber }, performanceSite, principalInvestigator:{ fullName, firstName, lastName, middleInitial, email, id }, coPrincipalInvestigators, programOfficer, amounts:{ fundsObligatedAmt, estimatedTotalAmt, fundsObligatedByYear }, dates, program, activeAward, historicalAward }] } (abstract EXCLUDED — use nsf_get_award) + honest _meta. HONESTY: NSF Awards are RESEARCH GRANTS, NOT procurement contracts (ueiNumber joins to SAM/USAspending but the award nature differs — disclosed every response); totalAvailable = the EXACT metadata.totalCount below 10,000 and SATURATES at 10,000 (an ES track_total_hits cap ⇒ totalIsLowerBound:true + a note — the true total is ≥10,000 and only the first 10,000 are retrievable); NSF caps keyless retrieval at offset+rpp ≤ 10,000 (offset ≥ 10,000 ⇒ invalid_input; the outgoing rpp is clamped so a page never crosses the window). fundsObligatedAmt/estimatedTotalAmt arrive as STRINGS → number|null (a real $0 is 0, absent is null). Genuine-empty (totalCount:0) ⇒ complete:true/total:0; a serviceNotification at HTTP 200 (bad param / deep offset) ⇒ invalid_input/upstream_unavailable THROWS (never a fake empty); an outage/5xx/timeout THROWS; a 200 body that isn't {response:{award,metadata}} or a non-numeric totalCount ⇒ schema_drift. Feed a row's id to nsf_get_award for the full record + abstractText.",
+        inputSchema: NsfSearchAwardsInput,
+        handler: (input) => nsf.searchAwards(input),
+    }),
+    defineTool({
+        name: "nsf_get_award",
+        description: "Fetch ONE NSF award by its numeric award id (keyless; api.nsf.gov/services/v1/awards.json). Input `awardId` (all-digit, 5..9 digits — NSF ids are 7-digit numeric, live-verified; numeric-only is injection-safe). Returns { found, award:{ …the FULL curated record INCLUDING abstractText… } } + honest _meta. A nonexistent id ⇒ a genuine empty (totalCount:0) ⇒ found:false / award:null (NEVER a fabricated record). HONESTY: NSF Awards are RESEARCH GRANTS, NOT procurement contracts (ueiNumber joins to SAM/USAspending but the award nature differs — disclosed every response); fundsObligatedAmt/estimatedTotalAmt arrive as STRINGS → number|null (a real $0 is 0, absent is null); a serviceNotification at HTTP 200 ⇒ invalid_input/upstream_unavailable THROWS; an outage/5xx ⇒ THROWS; a 200 body that isn't {response:{award,metadata}} ⇒ schema_drift (never a fabricated record).",
+        inputSchema: NsfGetAwardInput,
+        handler: (input) => nsf.getAward(input),
+    }),
+    // ━━━ ClinicalTrials.gov API v2 — keyless clinical-study registrations (2) ━━━ ADR-0021
+    // Source #21. The trial-REGISTRATION sibling of the research-GRANT sources (NIH
+    // RePORTER / NSF Awards): leadSponsor / collaborators / organization are the
+    // pharma/biotech/university/agency entities that ALSO receive federal money.
+    // SSRF surface = a compile-time host literal (CT_BASE) + a single audited getCT
+    // helper ([M2]); the single-study nctId is ^NCT\d{8}$-validated before the path
+    // is built. HONESTY: countTotal=true is ALWAYS sent (the exact filter-respecting
+    // uncapped total — a missing totalCount ⇒ schema_drift, NEVER studies.length);
+    // an OPAQUE nextPageToken cursor (terminal = token absent, passed back verbatim);
+    // funderType is a 4-value enum RE-VALIDATED IN THE HANDLER ([M1] — an invalid
+    // value silently fake-empties at HTTP 200); multi-word term/sponsor/condition is
+    // AND-tokenized (disclosed); trial≠federal-award caveat in every response.
+    defineTool({
+        name: "clinicaltrials_search_studies",
+        description: "Search federally-registered clinical-research studies with LEAD-SPONSOR / COLLABORATOR / ORGANIZATION / FUNDING-SOURCE entity enrichment (keyless; clinicaltrials.gov/api/v2/studies) — the trial-REGISTRATION axis of the research-funding entity layer (the sponsor/collaborator NAMES overlap the pharma/biotech/university/agency entities in NIH RePORTER / NSF Awards / SAM / USAspending). LIVE-CONFIRMED-narrowing filters ONLY, module-built into a URLSearchParams query (NO raw passthrough): query.term (broad free-text), sponsor (→query.spons — a fuzzy sponsor NAME search), condition (→query.cond), location (→query.locn), overallStatus (a frozen 14-value enum → filter.overallStatus), funderType (a frozen 4-value enum nih/fed/industry/other → aggFilters — the FEDERAL-funding axis), pageSize (1..1000, def 20), pageToken (the OPAQUE cursor). Returns { studies:[{ nctId, briefTitle, orgStudyId, organization:{ name, class }, leadSponsor:{ name, class }, collaborators:[{ name, class }], fundingClass, overallStatus, startDate, studyType, phases, conditions }] } (briefSummary EXCLUDED — use clinicaltrials_get_study) + honest _meta. HONESTY: countTotal=true is ALWAYS sent ⇒ totalAvailable = the EXACT filter-respecting UNCAPPED total (NEVER studies.length; a missing/non-number totalCount ⇒ schema_drift; a genuine 0 ⇒ 0, never null); pagination is an OPAQUE cursor (offset/nextOffset null; nextCursor = nextPageToken passed back verbatim as pageToken; terminal = token absent; a bad token ⇒ HTTP 400 THROWS). funderType is re-validated IN the handler — an UNLISTED value silently returns totalCount:0 at HTTP 200 (a fake-empty trap) ⇒ invalid_input pre-fetch (0 fetch); funderType is an OVERLAPPING facet (counts MUST NOT be summed). A MULTI-WORD query.term/sponsor/condition is AND-conjunctive (ALL tokens must co-occur — disclosed). A registered trial is NOT a federal award and leadSponsor.name is FREE TEXT (not a UEI) ⇒ a NOMINAL name match only (disclosed every response). Genuine-empty (totalCount:0, no token) ⇒ complete:true/total:0; a bad overallStatus/pageToken/nctId ⇒ HTTP 400/404 THROWS; an outage/5xx ⇒ THROWS (never a fake empty). Feed a row's nctId to clinicaltrials_get_study for the full record + briefSummary.",
+        inputSchema: ClinicaltrialsSearchStudiesInput,
+        handler: (input) => clinicaltrials.searchStudies(input),
+    }),
+    defineTool({
+        name: "clinicaltrials_get_study",
+        description: "Fetch ONE clinical study by its NCT id (keyless; clinicaltrials.gov/api/v2/studies/{nctId}). Input `nctId` (the form NCT followed by exactly 8 digits, e.g. NCT02403869 — validated before the path is built, injection-safe). Returns { found, nctId, study:{ …the FULL curated entity record INCLUDING briefSummary… } } + honest _meta. A nonexistent id ⇒ HTTP 404 ⇒ found:false / study:null (NEVER a fabricated record). HONESTY: a registered trial is NOT a federal award and leadSponsor.name is FREE TEXT (not a UEI) ⇒ a NOMINAL name match only (disclosed every response); a 200 body missing protocolSection ⇒ schema_drift; an outage/5xx ⇒ THROWS.",
+        inputSchema: ClinicaltrialsGetStudyInput,
+        handler: (input) => clinicaltrials.getStudy(input),
+    }),
+    // ── ADR-0024: the aggregate/statistical SIBLING (+1 tool). EXACT per-value study
+    //    counts over the WHOLE registry for whitelisted ENUM fields (/stats/field/values,
+    //    the SAME fixed host + audited getCT). HONESTY: [M1] _meta.totalAvailable/returned
+    //    count DISTINCT FIELD VALUES (not studies — a mandatory unit note); [M2] the
+    //    whole-registry scope note carries NO frozen registry size; the returned<unique⇒
+    //    truncated invariant discloses the 250-cap the instant it binds (never for v1
+    //    ENUMs); a non-ENUM shape for a whitelisted field ⇒ schema_drift; Phase is
+    //    ARRAY-valued (overlap/not-a-partition note); the facet-scoped trial≠award caveat
+    //    every response. NO free-text ⇒ no tokenization.
+    defineTool({
+        name: "clinicaltrials_facet_counts",
+        description: "Aggregate/statistical view: EXACT per-value STUDY counts over the WHOLE ClinicalTrials.gov registry for one or more whitelisted ENUM fields (keyless; clinicaltrials.gov/api/v2/stats/field/values) — the DISTRIBUTION sibling of clinicaltrials_search_studies (which gives the exact FILTERED total for a query). Input `fields`: 1..11 ENUM fields (deduped) — OverallStatus, StudyType, Phase, LeadSponsorClass (★ the funding-SOURCE-class distribution: NIH/FED/OTHER_GOV/INDUSTRY/OTHER/NETWORK/INDIV/UNKNOWN/AMBIG — richer than, and distinct from, the search tool's 4-value funderType filter), Sex, DesignAllocation, DesignPrimaryPurpose, DesignInterventionModel, DesignMasking, DesignObservationalModel, DesignTimePerspective. Module-built comma-joined into fields=<…> (NO raw passthrough). Returns { facets:[{ field, fieldPath, valueType, uniqueValuesCount, missingStudiesCount, returned, truncated, overlapping, values:[{ value, studiesCount }] }] } + honest _meta. HONESTY: each studiesCount/uniqueValuesCount is EXACT (typeof-checked to a NUMBER before num() — a non-number ⇒ schema_drift, NEVER a silent 0); a non-ENUM shape for a whitelisted field (e.g. a BOOLEAN {trueCount,falseCount}) ⇒ schema_drift (never read as empty). [M1] _meta.totalAvailable/returned count DISTINCT FIELD VALUES across the requested facet(s), NOT studies (a mandatory unit note points to facets[].values[].studiesCount / clinicaltrials_search_studies for a study count). These counts cover the ENTIRE registry and are NOT filterable — /stats/field/values rejects query.*/filter.*/countTotal/pageSize (HTTP 400) — a scope note cross-links the search tool for filtered totals. The returned<uniqueValuesCount⇒truncated invariant discloses the endpoint's hard 250-value cap the instant it binds (never for these v1 ENUM fields — all complete). Phase is ARRAY-valued (a study can carry several) ⇒ overlapping:true + a not-a-partition note (counts MUST NOT be summed); scalar fields partition the registry minus missingStudiesCount. A high missingStudiesCount ⇒ a note that the shown buckets cover a MINORITY of the registry. MANDATORY CAVEAT every response: a facet count is a distribution over trial REGISTRATIONS, NOT federal awards; LeadSponsorClass is the funding-SOURCE class, not a UEI-keyed award join. An unlisted field ⇒ invalid_input pre-fetch (0 fetch); a 404/400/5xx ⇒ THROWS (never a fake-empty distribution).",
+        inputSchema: ClinicaltrialsFacetCountsInput,
+        handler: (input) => clinicaltrials.facetCounts(input),
+    }),
+    // ━━━ EPA ECHO REST — keyless facility environmental compliance/enforcement (2) ━━━ ADR-0009
+    // A NEW capability axis: facility & competitor environmental compliance-risk
+    // screening / due diligence. KEYLESS (keylessMode:true, byte-clean init), single
+    // fixed host + three fixed service paths (the SSRF core). The two-step QueryID
+    // pagination is HIDDEN in-call (the ephemeral globally-recycled QueryID is never
+    // exposed); the 200-with-error-body failure mode is guarded FIRST (never a fake
+    // empty). M2: sic narrows (real filter), naics is dropped upstream (best-effort +
+    // disclosed in _meta.filtersDropped/notes).
+    defineTool({
+        name: "echo_search_facilities",
+        description: "Search EPA-regulated facilities by US state (+ optional sic / facilityName / majorOnly / federalOnly) with compliance/enforcement screening fields (EPA ECHO, keyless) — the NEW facility environmental compliance-risk / due-diligence axis (CAA/CWA/RCRA/SDWA violation, inspection, penalty, SNC history). Input `state` (REQUIRED enum — the SSRF + silent-zero guard), `sic` (2–4 digits, a REAL filter), `naics` (2–6 digits, BEST-EFFORT — ECHO DROPS it upstream, reported in _meta.filtersDropped + a note), `facilityName` (substring; a typo silently returns 0), `majorOnly`/`federalOnly` (bool), `limit` (≤1000, def 100), `offset` (multiple of limit). Returns { state, facilities:[…verbatim rows incl. RegistryID…], summary:{ queryRows, programCounts, totalPenalties } } + honest _meta. HONESTY: totalAvailable = the EXACT QueryRows total (NEVER the page size); a hidden two-step QueryID pagination fetches the rows (the QueryID is ephemeral/globally-recycled, never exposed); genuine-empty ⇒ complete:true/total:0; a queryset-limit overflow / bad query ⇒ invalid_input; an outage/5xx ⇒ THROWS (never a fake empty). Feed a row's RegistryID to echo_facility_report.",
+        inputSchema: EchoSearchFacilitiesInput,
+        handler: (input) => echo.searchFacilities(input),
+    }),
+    defineTool({
+        name: "echo_facility_report",
+        description: "Fetch the EPA ECHO Detailed Facility Report (DFR) for ONE facility by its FRS RegistryID (keyless) — the per-facility compliance / enforcement / inspection / permit deep-dive for competitor or acquisition-target due diligence. Input `registryId` (all-digit FRS id, 9–12 digits, from echo_search_facilities rows). Returns { registryId, report:{…verbatim compliance/enforcement/permit detail…} } + single-record _meta (complete:true, no pagination). A bad/unknown RegistryID ⇒ not_found (never a fabricated report).",
+        inputSchema: EchoFacilityReportInput,
+        handler: (input) => echo.facilityReport(input),
+    }),
+    // ━━━ api.data.gov keyed trio — Regulations.gov + Congress.gov (6) ━━━ ADR-0007/0044
+    // The project's FIRST KEYED source. The key (DATA_GOV_API_KEY, else the public
+    // DEMO_KEY) travels ONLY in the X-Api-Key header — never the URL/label/_meta.
+    // keylessMode:false (genuinely keyed); a DEMO_KEY note discloses the shared
+    // ~10 req/hr ceiling + the free-key upgrade path. ADR-0044 adds the two docket
+    // tools (the rulemaking CONTAINER + its `rin` cross-source join key).
+    defineTool({
+        name: "regulations_search_documents",
+        description: "Search Regulations.gov rulemaking DOCUMENTS (rules, proposed rules, notices) — the flagship of the api.data.gov keyed source (JSON:API; DATA_GOV_API_KEY or the shared DEMO_KEY). Input `searchTerm`/`query`, filters (agencyId, docketId, documentType, withinCommentPeriod, postedDateGe/Le YYYY-MM-DD), `sort` (def -postedDate), `pageNumber` (1..40 HARD cap), `pageSize` (5..250, def 25). Returns { documents:[{ id, documentType, title, agencyId, docketId, postedDate, commentEndDate, openForComment, withinCommentPeriod, frDocNum, objectId }] } + honest _meta. HONESTY: totalAvailable = meta.totalElements (the EXACT real total, ~millions), NOT the capped totalPages; page[number] is hard-capped at 40 (10,000-record ceiling) — at the ceiling hasMore stays true but nextOffset is null + a note says how to reach the rest (narrow filters / seek by lastModifiedDate). Genuine-empty ⇒ complete:true/total:0; an outage/4xx THROWS (never a fake empty).",
+        inputSchema: RegulationsSearchInput,
+        handler: (input) => datagov.searchDocuments(input),
+    }),
+    defineTool({
+        name: "regulations_search_comments",
+        description: "Search Regulations.gov public COMMENTS on rulemakings — the killer B2G dataset (who is lobbying which rule). Same JSON:API envelope + input shape as regulations_search_documents (searchTerm/query, agencyId, docketId, postedDateGe/Le, sort, pageNumber 1..40, pageSize 5..250) against /v4/comments. Returns { comments:[{ id, documentType, title, agencyId, docketId, postedDate, objectId }] } + honest _meta (same totalElements-exact total + 40-page/10,000-record ceiling handling as documents).",
+        inputSchema: RegulationsSearchInput,
+        handler: (input) => datagov.searchComments(input),
+    }),
+    defineTool({
+        name: "congress_search_bills",
+        description: "Search Congress.gov BILLS/legislation (api.data.gov keyed; DATA_GOV_API_KEY or DEMO_KEY). Input optional `congress` (e.g. 118), `billType` (hr/s/hjres/sjres/hconres/sconres/hres/sres — requires `congress`), `fromDateTime`/`toDateTime` (ISO-8601 with offset), `offset`, `limit` (≤250, def 20). Returns { bills:[{ congress, type, number, title, originChamber, latestAction, updateDate, url }] } + _meta with totalAvailable = pagination.count (EXACT). NOTE: /v3/bill has no keyword search, so a `query` arg is NOT applied and is disclosed in _meta.filtersDropped. Outage/4xx THROWS (never a fake empty).",
+        inputSchema: CongressSearchBillsInput,
+        handler: (input) => datagov.searchBills(input),
+    }),
+    defineTool({
+        name: "congress_get_bill",
+        description: "Fetch ONE Congress.gov bill by id via /v3/bill/{congress}/{billType}/{billNumber} (api.data.gov keyed; DATA_GOV_API_KEY or DEMO_KEY). Input `congress` (int), `billType` (enum), `billNumber` (int). Returns { bill:{…} } + single-record _meta. A nonexistent bill ⇒ not_found (never fabricated).",
+        inputSchema: CongressGetBillInput,
+        handler: (input) => datagov.getBill(input),
+    }),
+    defineTool({
+        name: "regulations_search_dockets",
+        description: "Search Regulations.gov DOCKETS — the rulemaking/nonrulemaking CONTAINER that groups every document + comment under one regulatory action (api.data.gov keyed; DATA_GOV_API_KEY or the shared DEMO_KEY). Input `searchTerm`/`query`, filters (agencyId, docketType Rulemaking/Nonrulemaking, lastModifiedDateGe/Le YYYY-MM-DD), `sort` (def -lastModifiedDate), `limit` (1..250, def 20), `pageNumber` (1..40 HARD cap). Returns { dockets:[{ docketId, title, agencyId, docketType, lastModifiedDate, objectId, id }] } + honest _meta. HONESTY: totalAvailable = meta.totalElements (the EXACT real total, ~277k), NOT the capped totalPages (a 40 sentinel — deriving a total from totalPages lies); page[number] is hard-capped at 40 (10,000-record ceiling) — at the ceiling hasMore stays true but nextOffset is null + a note on how to reach the rest (narrow filters). The API's page[size] floor is 5, so a limit<5 fetches 5 and returns the first `limit` rows client-side (disclosed; totalAvailable stays exact). NOTE: `rin` is NULL in list rows — call regulations_get_docket for a docket's rin. DEMO_KEY ~10 req/hr (every call, incl. errors, decrements) — set DATA_GOV_API_KEY for 1000/hr. Genuine-empty ⇒ complete:true/total:0; outage/4xx/429 THROWS (never a fake empty).",
+        inputSchema: RegulationsSearchDocketsInput,
+        handler: (input) => datagov.searchDockets(input),
+    }),
+    defineTool({
+        name: "regulations_get_docket",
+        description: "Fetch ONE Regulations.gov docket by id via /v4/dockets/{docketId} (api.data.gov keyed; DATA_GOV_API_KEY or DEMO_KEY) — the detail view where `rin` lives. Input `docketId` (e.g. 'BLM-2026-0001'; the ONLY path-segment value, charclass-validated — a bad id ⇒ invalid_input, 0 fetch). Returns { docket:{ docketId, title, agencyId, docketType, rin, dkAbstract, keywords, program, shortTitle, effectiveDate, modifyDate, objectId, id } } + single-record _meta (returned:1, totalAvailable:null, complete:true). HONESTY: `rin` (Regulatory Identifier Number) is the cross-source JOIN KEY to the Federal Register (fed_register_search_documents) and the Unified Agenda — null-when-absent (never '', e.g. many Nonrulemaking dockets have no assigned RIN), which is NOT a join failure. A nonexistent id ⇒ not_found (or schema_drift if the API returns a 200 error-envelope) — never a fabricated docket. DEMO_KEY ~10 req/hr; set DATA_GOV_API_KEY for 1000/hr.",
+        inputSchema: RegulationsGetDocketInput,
+        handler: (input) => datagov.getDocket(input),
+    }),
+    // ━━━ data.gov v4 Catalog API (api.gsa.gov) — CKAN-retirement replacement (1) ━━━ ADR-0046
+    // Resilience Phase 3. data.gov RETIRED the CKAN package_search endpoint in 2025;
+    // the v4 Catalog API restores federal open-dataset DISCOVERY as a NEW keyed source.
+    // A DIFFERENT host (api.gsa.gov) than the datagov trio, but the SAME api.data.gov
+    // key (X-Api-Key header, shared datagovKey.ts seam) — keylessMode:false. The v4
+    // API reports NO match count ⇒ totalAvailable is NULL (P1, never results.length);
+    // pagination is an OPAQUE `after` cursor (nextCursor passed back verbatim); the
+    // dcat.accessLevel openness field is surfaced verbatim.
+    defineTool({
+        name: "datagov_search_datasets",
+        description: "Search the data.gov DATASET CATALOG for federal open datasets across all publishing agencies (api.gsa.gov v4 Catalog API, keyed — DATA_GOV_API_KEY or the shared DEMO_KEY) — the replacement for the CKAN package_search endpoint data.gov RETIRED in 2025, restoring federal dataset DISCOVERY. Input `query` (→_q free-text), `organization` (publisher slug, e.g. 'epa-gov'), `limit` (1..100, def 20 → _size), `cursor` (the OPAQUE continuation → after). Returns { datasets:[{ id (slug), title, organization, description, accessLevel, license, landingPage, modified, lastHarvested, keywords, themes, distributions:[{ title, format }], identifier }] } + honest _meta. HONESTY: the v4 API reports NO total match count ⇒ totalAvailable is NULL (NEVER results.length, NEVER a fabricated total — a note discloses it); pagination is an OPAQUE cursor (offset/nextOffset null; nextCursor = the `after` token passed back verbatim as `cursor`; nextCursor:null / hasMore:false = last page). accessLevel is surfaced VERBATIM (public / restricted public / non-public) — the openness signal, null-when-absent (this tool DISCOVERS datasets; it does not ingest distributions). A genuine no-match (results:[], no cursor) ⇒ complete:true/returned:0; a 429 (DEMO_KEY ~10 req/hr, hit quickly) ⇒ rate_limited THROWS; a 5xx/timeout ⇒ upstream_unavailable THROWS; a 200 non-JSON / a non-array results ⇒ schema_drift (never a fake empty). DEMO_KEY ~10 req/hr shared ceiling — set DATA_GOV_API_KEY (free at api.data.gov/signup) for 1000/hr. The key rides ONLY in the X-Api-Key header (never the URL/_meta).",
+        inputSchema: DatagovSearchDatasetsInput,
+        handler: (input) => datagovCatalog.searchDatasets(input),
+    }),
+    // ━━━ GovInfo (api.govinfo.gov) — the api.data.gov keyed trio's 3rd API (3) ━━━ ADR-0010
+    // GPO-authoritative bulk publications (BILLS/PLAW/USCODE/CREC/CFR-FR editions/
+    // BUDGET/GAOREPORTS) with PDF/XML/MODS downloads + provenance. 2nd consumer of the
+    // shared api.data.gov env-key adapter (datagovKey.ts) — key ONLY in the X-Api-Key
+    // header, keylessMode:false, DEMO_KEY disclosure. The novel piece is the OPAQUE
+    // offsetMark cursor: continuation rides in _meta.nextCursor (passed back as
+    // pageMark); pagination.offset/nextOffset are null (no numeric offset). The raw
+    // upstream nextPage URL (which embeds pageSize+api_key) is NEVER surfaced.
+    defineTool({
+        name: "govinfo_list_collections",
+        description: "List the GovInfo collection catalog (GPO-authoritative publications; api.data.gov keyed — DATA_GOV_API_KEY or the shared DEMO_KEY). No input. Returns { collections:[{ collectionCode, collectionName, packageCount, granuleCount }] } + _meta (complete:true, totalAvailable = collection count). The discovery entry-point: feed a collectionCode to govinfo_search_packages. Memoized ~6h; also the validator source for search_packages' collection arg. packageCount = whole packages; granuleCount = sub-package granules (a missing count is null, never 0).",
+        inputSchema: GovinfoListCollectionsInput,
+        handler: () => govinfo.listCollections(),
+    }),
+    defineTool({
+        name: "govinfo_search_packages",
+        description: "Search GovInfo packages in a collection modified since a date (GPO-authoritative bulk publications; api.data.gov keyed). Input `collection` (uppercase code — validated against the live catalog; an unknown code ⇒ invalid_input listing valid codes, NEVER a misleading empty), `startDate`/`endDate?` (YYYY-MM-DD or ISO datetime; filters by lastModified — the record UPDATE date, NOT dateIssued — disclosed in _meta), `pageSize?` (1..1000, def 100), `pageMark?` (opaque cursor, def '*'). Returns { collection, packages:[{ packageId, title, dateIssued, lastModified, docClass, congress, packageLink }] } + cursor _meta. HONESTY: totalAvailable = count (the EXACT real total, NOT the page size); GovInfo uses an OPAQUE cursor, so pagination.offset/nextOffset are null — continue by passing _meta.nextCursor back as `pageMark` (hasMore:false / nextCursor:null = last page). The raw upstream nextPage URL is never surfaced (it embeds the key). Genuine-empty ⇒ complete:true/total:0; outage/4xx THROWS (never a fake empty). CFR/ECFR/FR collections carry a note routing to the ecfr_*/fed_register_* tools for point lookups.",
+        inputSchema: GovinfoSearchPackagesInput,
+        handler: (input) => govinfo.searchPackages(input),
+    }),
+    defineTool({
+        name: "govinfo_get_package",
+        description: "Fetch ONE GovInfo package's summary (metadata + download links txt/xml/pdf/mods/premis/zip + related links) by packageId (api.data.gov keyed). Input `packageId` (from govinfo_search_packages, e.g. 'BILLS-118hr1enr', 'PLAW-117publ58', 'CFR-2023-title1-vol1'). Returns { found:true, packageId, package:{…} } + single-record _meta (complete:true). A nonexistent packageId ⇒ found:false (HTTP 404, never a fabricated summary). Any api_key embedded in a download link is stripped key-free before the payload is surfaced.",
+        inputSchema: GovinfoGetPackageInput,
+        handler: (input) => govinfo.getPackage(input),
+    }),
+    // ━━━ US Census Geocoder — keyless territory/geospatial (2) ━━━ ADR-0023
+    // A NEW capability domain (territory/geospatial) serving the WEAK set-aside /
+    // place-of-performance layer. KEYLESS (keylessMode:true, byte-clean init), single
+    // fixed host + two fixed endpoint paths (the SSRF core — no id in the path). The
+    // layer mapper resolves each canonical geography by SUFFIX pattern (the key names
+    // ROLL: "119th Congressional Districts") and handles >1 KEY PER SUFFIX ([B1] — a
+    // historical vintage returns 111th+113th CDs with DISTINCT GEOIDs; both surfaced +
+    // a note, never silently dropped). Drift-guard scoped to 4 sentinels ([M1]); the
+    // vintage enum is the (benchmark,vintage) UNION ([M2]); GEOIDs stay strings.
+    defineTool({
+        name: "census_geocode_address",
+        description: "Resolve a one-line US address → its matched address(es) + the Census GEOGRAPHIES that drive set-aside / place-of-performance analysis (US Census Geocoder, keyless; geocoding.geo.census.gov/geocoder/geographies/onelineaddress) — the NEW territory/geospatial domain. Input `address` (≤500 chars), optional `benchmark` (default Public_AR_Current) / `vintage` (default Current_Current). Returns { matches:[{ matchedAddress, coordinates:{x,y}, tigerLineId, addressComponents, geographies:{ state, county, congressionalDistrict, censusTract, censusBlock, place, cbsaOrCsa, stateLegislativeUpper, stateLegislativeLower } }], matchCount, vintageResolved } + honest _meta. Each geography = { layerKey (the RAW vintage-versioned key, e.g. '119th Congressional Districts'), geoid (a STRING — leading zeros survive: '0102'), name }. HONESTY: genuine-empty (addressMatches:[]) ⇒ matchCount:0/complete:true (NOT an error; verify spelling + add city/state/ZIP); MULTIPLE matches are ALL surfaced (each with its own geographies) + a note; a historical vintage can return >1 layer per type (e.g. 111th+113th Congressional Districts with DISTINCT GEOIDs for a redistricted place) ⇒ BOTH surfaced (chosen + alternates[]) + a mandatory note (NEVER silently dropped); the resolved benchmark/vintage is echoed + a 'Current is a MOVING vintage' note; an invalid/missing benchmark/vintage ⇒ HTTP 400 THROWS (never a fake empty); an outage/5xx ⇒ THROWS. MANDATORY CAVEAT every response: these are a NOMINAL input, NOT an authoritative HUBZone / Opportunity-Zone / set-aside determination (those require SBA's HUBZone map / Treasury's OZ-tract list). Feed censusTract.geoid / county.geoid onward to those authoritative sources.",
+        inputSchema: CensusGeocodeAddressInput,
+        handler: (input) => census.geocodeAddress(input),
+    }),
+    defineTool({
+        name: "census_geographies_by_coordinates",
+        description: "Resolve a longitude/latitude point → the Census GEOGRAPHIES at that point, no address parsing (US Census Geocoder, keyless; geocoding.geo.census.gov/geocoder/geographies/coordinates). For a caller that already holds coordinates. Input `longitude`/`x` (required, -180..180) + `latitude`/`y` (required, -90..90) — x=longitude, y=latitude (the Census API's own names; `longitude`/`latitude` are the clearer aliases), optional `benchmark`/`vintage`. Returns { found, coordinates:{x,y}, geographies:{ state, county, congressionalDistrict, censusTract, censusBlock, place, cbsaOrCsa, stateLegislativeUpper, stateLegislativeLower }, vintageResolved } + honest _meta. HONESTY: a point outside any US Census geography (offshore / out-of-US) ⇒ geographies all null / found:false / complete:true (an honest empty geographies:{}, NOT an error); coordinate finiteness is re-guarded PRE-fetch (a non-finite x/y ⇒ invalid_input, 0 fetch); a historical vintage's >1-layer-per-type is surfaced with alternates[] + a note (same [B1] multi-key handling as the address tool); GEOIDs are STRINGS (leading zeros survive); the resolved benchmark/vintage is echoed + a moving-vintage note; a bad benchmark/vintage ⇒ HTTP 400 THROWS; an outage/5xx ⇒ THROWS. MANDATORY CAVEAT every response: these are a NOMINAL input, NOT an authoritative HUBZone / Opportunity-Zone / set-aside determination.",
+        inputSchema: CensusGeographiesByCoordinatesInput,
+        handler: (input) => census.geographiesByCoordinates(input),
+    }),
 ];
 // ─── Server bootstrap ────────────────────────────────────────────
 async function main() {
@@ -436,17 +4073,37 @@ async function main() {
     server.setRequestHandler(CallToolRequestSchema, async (req) => {
         const { name, arguments: args } = req.params;
         try {
-            const result = await runTool(name, args ?? {}, sam);
+            const raw = await runTool(name, args ?? {}, sam);
+            // A handler may return either its raw domain object OR a MetaBundle
+            // (via withMeta) carrying a partial `_meta`. Unwrap to `data` + finalize
+            // the `_meta` sibling. `data` is byte-identical either way.
+            const data = isMetaBundle(raw) ? raw.data : raw;
+            const _meta = isMetaBundle(raw)
+                ? buildMeta(raw.meta)
+                : synthesizeDefaultMeta(name, sam);
+            // Structured success envelope. Calling agent can rely on
+            // `ok: true` to know the payload is in `data`, and read `_meta`
+            // for completeness / provenance (see meta.ts).
+            const envelope = { ok: true, data, _meta };
             return {
                 content: [
-                    { type: "text", text: JSON.stringify(result, null, 2) },
+                    { type: "text", text: JSON.stringify(envelope, null, 2) },
                 ],
             };
         }
         catch (err) {
-            const message = err instanceof Error ? err.message : String(err);
+            // Structured error envelope. The agent can read `error.kind`
+            // and `error.retryable` to decide what to do next. Classification is
+            // centralized in `toToolError`, which maps a Zod input-validation failure
+            // (e.g. a value outside an enum, or a limit above the max) to a
+            // NON-retryable `invalid_input` with readable field-level issues — never a
+            // generic `unknown` carrying Zod's raw JSON dump.
+            const error = toToolError(err, name);
+            const envelope = { ok: false, error };
             return {
-                content: [{ type: "text", text: `Tool error: ${message}` }],
+                content: [
+                    { type: "text", text: JSON.stringify(envelope, null, 2) },
+                ],
                 isError: true,
             };
         }
@@ -455,171 +4112,78 @@ async function main() {
     await server.connect(transport);
     console.error(`[mcp-sam-gov] v${SERVER_VERSION} listening on stdio (${TOOLS.length} tools).`);
 }
-async function runTool(name, args, sam) {
-    switch (name) {
-        // SAM.gov
-        case "sam_search_opportunities": {
-            const input = SamSearchInput.parse(args);
-            const r = await sam.searchOpportunities({
-                ...input,
-                setAside: input.setAside,
-            });
-            return {
-                totalRecords: r.totalRecords,
-                returned: r.opportunitiesData.length,
-                opportunities: r.opportunitiesData.map((o) => ({
-                    noticeId: o.noticeId,
-                    title: o.title,
-                    agency: o.fullParentPathName,
-                    solicitationNumber: o.solicitationNumber,
-                    responseDeadline: o.responseDeadLine,
-                    naics: o.naicsCode,
-                    setAside: o.typeOfSetAside,
-                    uiLink: o.uiLink,
-                })),
-            };
-        }
-        case "sam_get_opportunity": {
-            const { noticeId } = SamGetOpportunityInput.parse(args);
-            const o = await sam.getOpportunity(noticeId);
-            if (!o)
-                return { found: false, noticeId };
-            return {
-                found: true,
-                noticeId: o.noticeId,
-                title: o.title,
-                agency: o.fullParentPathName,
-                solicitationNumber: o.solicitationNumber,
-                responseDeadline: o.responseDeadLine,
-                type: o.type,
-                naics: o.naicsCode,
-                setAside: o.typeOfSetAside,
-                placeOfPerformance: o.placeOfPerformance,
-                pointsOfContact: o.pointOfContact ?? [],
-                description: o.description,
-                attachments: (o.resourceLinks ?? []).map((url, idx) => ({
-                    index: idx,
-                    url,
-                })),
-                uiLink: o.uiLink,
-            };
-        }
-        case "sam_fetch_description": {
-            const { noticeId } = SamFetchDescriptionInput.parse(args);
-            const o = await sam.getOpportunity(noticeId);
-            if (!o)
-                return { found: false, noticeId };
-            const text = o.description
-                ? await sam.fetchOpportunityDescription(o.description)
-                : "";
-            return {
-                found: true,
-                noticeId,
-                descriptionLength: text.length,
-                description: text || "(no description body available)",
-            };
-        }
-        case "sam_attachment_url": {
-            const { resourceId } = SamAttachmentUrlInput.parse(args);
-            return { downloadUrl: sam.publicDownloadUrl(resourceId) };
-        }
-        case "sam_lookup_organization": {
-            const { organizationId } = SamLookupOrgInput.parse(args);
-            // SamGovClient internal method — exposed via direct fetch since
-            // it's not on the public surface. Use the public sam.gov endpoint
-            // directly (already keyless).
-            const r = await fetch(`https://sam.gov/api/prod/federalorganizations/v1/organizations/${encodeURIComponent(organizationId)}`, {
-                headers: { Accept: "application/hal+json" },
-                signal: AbortSignal.timeout(10_000),
-            });
-            if (!r.ok) {
-                return { found: false, organizationId, status: r.status };
-            }
-            const json = (await r.json());
-            const org = json._embedded?.[0]?.org;
-            return {
-                found: !!org,
-                organizationId,
-                fullParentPathName: org?.fullParentPathName ?? "",
-                agencyName: org?.agencyName ?? "",
-                name: org?.name ?? "",
-                type: org?.type,
-                level: org?.level,
-            };
-        }
-        // USAspending — Awards & Recipients
-        case "usas_search_awards":
-            return await usas.searchAwards(UsasFiltersBase.parse(args));
-        case "usas_search_individual_awards":
-            return await usas.searchIndividualAwards(UsasIndividualAwardsInput.parse(args));
-        case "usas_search_subagency_spending":
-            return await usas.searchSubAgencySpending(UsasSubAgencyInput.parse(args));
-        case "usas_lookup_agency":
-            return await usas.lookupAgency(UsasLookupAgencyInput.parse(args).searchText);
-        case "usas_search_awards_by_recipient":
-            return await usas.searchAwardsByRecipient(UsasRecipientAwardsInput.parse(args));
-        case "usas_search_subawards":
-            return await usas.searchSubawards(UsasSubawardsInput.parse(args));
-        case "usas_search_expiring_contracts":
-            return await usas.searchExpiringContracts(UsasExpiringInput.parse(args));
-        case "usas_get_award_detail":
-            return await usas.getAwardDetail(UsasAwardDetailInput.parse(args).generatedInternalId);
-        // USAspending — Aggregate
-        case "usas_spending_over_time":
-            return await usas.spendingOverTime(UsasSpendingOverTimeInput.parse(args));
-        case "usas_search_psc_spending":
-            return await usas.searchPscSpending(UsasCategorySpendingInput.parse(args));
-        case "usas_search_state_spending":
-            return await usas.searchStateSpending(UsasCategorySpendingInput.parse(args));
-        case "usas_search_cfda_spending":
-            return await usas.searchCfdaSpending(UsasCfdaInput.parse(args));
-        case "usas_search_federal_account_spending":
-            return await usas.searchFederalAccountSpending(UsasCategorySpendingInput.parse(args));
-        case "usas_search_agency_spending":
-            return await usas.searchAgencySpending(UsasAgencySpendingInput.parse(args));
-        // USAspending — Agency Profile
-        case "usas_get_agency_profile":
-            return await usas.getAgencyProfile(UsasAgencyProfileInput.parse(args).toptierCode);
-        case "usas_get_agency_awards_summary":
-            return await usas.getAgencyAwardsSummary(UsasAgencyAwardsInput.parse(args));
-        case "usas_get_agency_budget_function":
-            return await usas.getAgencyBudgetFunction(UsasAgencyBudgetInput.parse(args));
-        // USAspending — Recipient Profile
-        case "usas_search_recipients":
-            return await usas.searchRecipients(UsasSearchRecipientsInput.parse(args));
-        case "usas_get_recipient_profile":
-            return await usas.getRecipientProfile(UsasGetRecipientInput.parse(args).recipientId);
-        // USAspending — Reference / Autocomplete
-        case "usas_autocomplete_naics":
-            return await usas.autocompleteNaics(UsasAutocompleteInput.parse(args));
-        case "usas_autocomplete_recipient":
-            return await usas.autocompleteRecipient(UsasAutocompleteInput.parse(args));
-        case "usas_naics_hierarchy":
-            return await usas.naicsHierarchy(UsasNaicsHierarchyInput.parse(args));
-        case "usas_glossary":
-            return await usas.glossary(UsasGlossaryInput.parse(args));
-        case "usas_list_toptier_agencies":
-            return await usas.listToptierAgencies(UsasListAgenciesInput.parse(args));
-        // Federal Register
-        case "fed_register_search_documents":
-            return await fedreg.searchDocuments(FedRegSearchInput.parse(args));
-        case "fed_register_get_document":
-            return await fedreg.getDocument(FedRegGetDocInput.parse(args).documentNumber);
-        case "fed_register_list_agencies":
-            return await fedreg.listAgencies(FedRegListAgenciesInput.parse(args));
-        // eCFR
-        case "ecfr_search":
-            return await ecfr.search(EcfrSearchInput.parse(args));
-        case "ecfr_list_titles":
-            return await ecfr.listTitles();
-        // Grants.gov
-        case "grants_search":
-            return await grants.searchGrants(GrantsSearchInput.parse(args));
-        case "grants_get_opportunity":
-            return await grants.getGrant(GrantsGetInput.parse(args));
-        default:
-            throw new Error(`Unknown tool: ${name}`);
+/**
+ * Minimal truthful `_meta` for handlers that don't attach their own.
+ *
+ * Defaults to `complete:true, truncated:false` — correct for the single-record
+ * and known-complete tools (detail lookups, reference tables). List/search and
+ * two-phase tools that can be capped or drop filters should instead return
+ * `withMeta(...)` with the real completeness signals; those are migrated
+ * incrementally (A1 landed first). The source label is keyless-aware for SAM
+ * tools so provenance is honest from day one.
+ */
+function synthesizeDefaultMeta(toolName, sam) {
+    // The wage tools carry a `sam_` prefix but hit the keyless SGS/WDOL
+    // subsystems (never the keyed opportunities API), so they are always keyless.
+    const isWage = toolName === "sam_search_wage_determinations" ||
+        toolName === "sam_get_wage_rates";
+    // sam_lookup_notice_fields also carries a `sam_` prefix but is served from the
+    // keyless GSA daily CSV (never the keyed opportunities API) — always keyless.
+    const isGsaCsv = toolName === "sam_lookup_notice_fields";
+    const isSam = toolName.startsWith("sam_") && !isWage && !isGsaCsv;
+    const keylessMode = isSam ? sam.isKeyless : true;
+    let source;
+    if (isWage) {
+        source = "sam.gov wage-determinations (keyless)";
+    }
+    else if (isGsaCsv) {
+        source = "gsa.gov daily bulk CSV (keyless)";
+    }
+    else if (toolName.startsWith("gsa_")) {
+        source = "api.gsa.gov CALC v3 (keyless)";
+    }
+    else if (isSam) {
+        source = sam.isKeyless ? "sam.gov (keyless)" : "api.sam.gov (keyed)";
+    }
+    else if (toolName.startsWith("usas_")) {
+        source = "usaspending.gov/api/v2";
+    }
+    else if (toolName.startsWith("fed_register_")) {
+        source = "federalregister.gov/api/v1";
     }
+    else if (toolName.startsWith("ecfr_")) {
+        source = "ecfr.gov/api";
+    }
+    else if (toolName.startsWith("grants_")) {
+        source = "grants.gov/api";
+    }
+    else if (toolName.startsWith("sba_")) {
+        source = "sba.gov naics.json (keyless)";
+    }
+    else if (toolName.startsWith("gao_")) {
+        source = "gao.gov Legal Products RSS + decision pages (keyless)";
+    }
+    else if (toolName.startsWith("fpds_")) {
+        source = "www.fpds.gov ezSearch ATOM (FPDS-NG, keyless)";
+    }
+    else {
+        source = "unknown";
+    }
+    return buildMeta({ source, keylessMode, complete: true, truncated: false });
+}
+export async function runTool(name, args, sam) {
+    // R1 (ADR-0001) — registry dispatch. Every tool's TOOLS[] entry carries a
+    // co-located `handler`: route through it by parsing `args` with the entry's
+    // own schema, then calling the handler. Its return value flows into
+    // CallTool's existing envelope logic (isMetaBundle? buildMeta :
+    // synthesizeDefaultMeta) byte-identically. The legacy dispatch `switch` is
+    // gone (all 52 tools migrated) — an unknown name has no entry and throws.
+    const entry = TOOLS.find((t) => t.name === name);
+    if (entry?.handler) {
+        const input = entry.inputSchema.parse(args);
+        return await entry.handler(input, { sam });
+    }
+    throw new Error(`Unknown tool: ${name}`);
 }
 /**
  * Hand-rolled Zod → JSON Schema converter (subset we use).
@@ -678,8 +4242,46 @@ function zodToJsonSchema(schema) {
     }
     return { type: "string", ...(description ? { description } : {}) };
 }
-main().catch((err) => {
-    console.error("[mcp-sam-gov] FATAL:", err);
-    process.exit(1);
-});
+// Start the stdio server ONLY when run directly (node dist/server.js / the
+// mcp-sam-gov bin) — NOT when imported (the fault-injection tests import
+// runTool to exercise the REAL tool-dispatch over a mocked fetch). Preserves
+// the launch: `node dist/server.js` → argv[1] === this file → main() runs;
+// smoke-test.mjs's spawn("node", ["dist/server.js"]) is the same (a subprocess
+// whose argv[1] is dist/server.js); an `import { runTool }` sets argv[1] to the
+// importing script → no match → main() does NOT run, the server is not spawned.
+// Was this module run DIRECTLY (node dist/server.js / the mcp-sam-gov bin), or
+// merely IMPORTED (the fault-injection tests import runTool over a mocked fetch)?
+// Only the direct case starts the stdio server. Canonicalize BOTH sides through
+// realpathSync before comparing, so it holds no matter how symlinks land:
+//   - the `mcp-sam-gov` bin is a symlink to dist/server.js on Unix/macOS, and npm
+//     installs the package dir itself via a symlink — argv[1] must be realpath'd;
+//   - under `--preserve-symlinks-main`, Node keeps import.meta.url as the symlink
+//     path, so THAT side must be realpath'd too.
+// Realpath'ing both and comparing as file:// URLs (pathToFileURL normalizes
+// Windows drive-casing/slashes) makes the check robust across every real launch.
+const invokedDirectly = (() => {
+    const argv1 = process.argv[1];
+    if (!argv1)
+        return false;
+    const canonical = (p) => {
+        try {
+            return pathToFileURL(realpathSync(p)).href;
+        }
+        catch {
+            return pathToFileURL(p).href; // not a real path → best-effort raw
+        }
+    };
+    try {
+        return canonical(argv1) === canonical(fileURLToPath(import.meta.url));
+    }
+    catch {
+        return import.meta.url === pathToFileURL(argv1).href; // extreme fallback
+    }
+})();
+if (invokedDirectly) {
+    main().catch((err) => {
+        console.error("[mcp-sam-gov] FATAL:", err);
+        process.exit(1);
+    });
+}
 //# sourceMappingURL=server.js.map