tickmarkr 1.85.0 → 1.86.0

package/dist/config/config.js

@@ -2,8 +2,9 @@ import { existsSync, readFileSync } from "node:fs";
 import { homedir } from "node:os";
 import { join } from "node:path";
 import picomatch from "picomatch";
-import { parse, stringify } from "yaml";
+import { Document, parse } from "yaml";
 import { z } from "zod";
+import { CITED_MODEL_WINDOWS } from "../adapters/model-windows.js";
 import { stateDirName } from "../graph/graph.js";
 import { SHAPES, TIERS } from "../graph/schema.js";
 const TierEnum = z.enum(TIERS, {
@@ -11,6 +12,8 @@ const TierEnum = z.enum(TIERS, {
 });
 export const TIER_RANK = { cheap: 0, mid: 1, frontier: 2 };
 export const DEFAULT_DIFF_CAP = 60_000;
+// This is deliberately only a rejection guard. T20's cited table, not this floor, supplies each claim.
+export const MODEL_WINDOW_PLAUSIBILITY_FLOOR = 1_024;
 // v1.51 T1: routing.mode — a preset COMPILED INTO FLOORS at config load. The router never sees the
 // mode; it receives resolved floors only (the structural defense against the quality-silently-loses
 // class — no fourth runtime authority, no new precedence key).
@@ -39,12 +42,36 @@ export const MapEntrySchema = z.object({
     prefer: z.array(z.string()).optional(),
     escalate: z.boolean().optional(),
 });
+const DeclaredVendorSchema = z.string().trim().min(1, "vendor must be nonempty");
+const ModelOverrideSchema = z
+    .object({
+    vendor: DeclaredVendorSchema.optional(),
+    channel: z.enum(["sub", "api"]).optional(),
+})
+    .refine((override) => override.vendor !== undefined || override.channel !== undefined, {
+    message: "a model override must declare vendor, channel, or both",
+});
 export const TierEntrySchema = z.object({
-    vendor: z.string(),
+    vendor: DeclaredVendorSchema.nullable(),
     channel: z.enum(["sub", "api"]),
     models: z.record(z.string(), TierEnum),
+    // T19: keep model values scalar for existing consumers. Optional sibling metadata resolves before
+    // parent metadata in channelsFromConfig; neither provider prefixes nor adapter ids manufacture it.
+    modelOverrides: z.record(z.string(), ModelOverrideSchema).optional(),
     // v1.47 T3: optional per-model context-window sizes (tokens). Absent block ⇒ no doctor column, no plan lint.
     windows: z.record(z.string(), z.number().int().positive()).optional(),
+}).superRefine((entry, ctx) => {
+    if (entry.vendor !== null)
+        return;
+    for (const model of Object.keys(entry.models)) {
+        if (entry.modelOverrides?.[model]?.vendor)
+            continue;
+        ctx.addIssue({
+            code: "custom",
+            path: ["modelOverrides", model, "vendor"],
+            message: "vendor: null requires every configured model to declare a nonempty vendor override",
+        });
+    }
 });
 // v1.10 FLEET-06: optional routing.allow/deny fleet preference; absent blocks ⇒ byte-identical routing/discovery.
 // deny wins on conflict; presence of allow (even {} / empty arrays) activates allowlist (fail-closed).
@@ -341,6 +368,24 @@ export class ConfigError extends Error {
         this.name = "ConfigError";
     }
 }
+function assertSeededModelWindows() {
+    const cited = new Map(CITED_MODEL_WINDOWS.map((claim) => [claim.modelId, claim.window]));
+    for (const [adapter, entry] of Object.entries(DEFAULT_CONFIG.tiers)) {
+        for (const [model, window] of Object.entries(entry.windows ?? {})) {
+            const path = `tiers.${adapter}.windows.${model}`;
+            if (window < MODEL_WINDOW_PLAUSIBILITY_FLOOR) {
+                throw new ConfigError(`${path} declares window ${window}, below plausibility floor ${MODEL_WINDOW_PLAUSIBILITY_FLOOR} (non-authoritative rejection guard)`);
+            }
+            const citedWindow = cited.get(model);
+            if (citedWindow === undefined) {
+                throw new ConfigError(`${path} has no cited model-window entry in T20's vendored table`);
+            }
+            if (window !== citedWindow) {
+                throw new ConfigError(`${path} declares ${window} but does not match cited window ${citedWindow} in T20's vendored table`);
+            }
+        }
+    }
+}
 export const DEFAULT_CONFIG = {
     concurrency: 3,
     driver: "auto",
@@ -374,10 +419,13 @@ export const DEFAULT_CONFIG = {
         allowUnverifiedModels: false,
     },
     // Seed table (spec §13). New models = edit this (or your config.yaml), never code.
+    // Windows are copied rather than derived from T20's table so config load can reject disagreement
+    // between two independently owned artifacts instead of letting one side certify itself.
     tiers: {
         "claude-code": {
             vendor: "anthropic", channel: "sub",
             models: { fable: "frontier", opus: "frontier", sonnet: "mid", haiku: "cheap" },
+            windows: { fable: 1_000_000, opus: 1_000_000, sonnet: 1_000_000, haiku: 200_000 },
         },
         // ids verified against installed CLI models_cache.json (codex 0.144.0, fetched 2026-07-09);
         // gpt-5.4 / gpt-5.4-mini removed — OpenAI retirement 2026-07-23; sol=frontier / terra=mid / luna=cheap per OpenAI tier framing.
@@ -389,6 +437,7 @@ export const DEFAULT_CONFIG = {
         codex: {
             vendor: "openai", channel: "sub",
             models: { "gpt-5.6-sol": "frontier", "gpt-5.5": "frontier", "gpt-5.6-terra": "mid", "gpt-5.6-luna": "cheap" },
+            windows: { "gpt-5.6-sol": 1_050_000, "gpt-5.5": 1_050_000, "gpt-5.6-terra": 1_050_000, "gpt-5.6-luna": 1_050_000 },
         },
         // grok-4.5 (xAI, released 2026-07-08) → mid: AA Intelligence 54 (#4), Terminal-Bench 2.1 83.3%
         // (≈ GPT-5.5 83.4, Fable 5 84.3), SWE-bench Pro 64.7%, ~4.2× more token-efficient than Opus 4.8
@@ -409,6 +458,7 @@ export const DEFAULT_CONFIG = {
                 // stop burning its mid. Operator-approved 2026-07-16.
                 "composer-2.5-fast": "cheap",
             },
+            windows: { "composer-2.5": 200_000, "composer-2.5-fast": 200_000 },
         },
         // GLM-5.2 → mid per benchmark policy (2026-07): SWE-bench Pro 62.1 (> GPT-5.5 58.6), FrontierSWE 74.4 ≈ Opus 4.8;
         // no independent Terminal-Bench score → conservative mid, overlays may raise.
@@ -419,6 +469,7 @@ export const DEFAULT_CONFIG = {
         opencode: {
             vendor: "mixed", channel: "sub",
             models: { "zai-coding-plan/glm-5.2": "mid" },
+            windows: { "zai-coding-plan/glm-5.2": 1_000_000 },
         },
         // GLM-5.2 via pi/ZAI Coding Plan (sub, flat-rate). mid per benchmark policy (2026-07):
         // SWE-bench Pro 62.1, FrontierSWE 74.4 — same rationale as the opencode glm-5.2 seed above.
@@ -430,6 +481,7 @@ export const DEFAULT_CONFIG = {
         pi: {
             vendor: "zhipu", channel: "sub",
             models: { "zai/glm-5.2": "mid" },
+            windows: { "zai/glm-5.2": 1_000_000 },
         },
         // Native grok CLI (Phase 40). grok-4.5 → mid: same benchmark provenance as the retired cursor-agent
         // grok-4.5-xhigh seed above (AA 54, TB2.1 83.3%, SWE-b Pro 64.7% — researched 2026-07-10).
@@ -439,6 +491,7 @@ export const DEFAULT_CONFIG = {
         grok: {
             vendor: "xai", channel: "sub",
             models: { "grok-4.5": "mid", "grok-composer-2.5-fast": "cheap" },
+            windows: { "grok-4.5": 500_000, "grok-composer-2.5-fast": 200_000 },
         },
         // Native kimi CLI (Phase 48). kimi-code/k3 → frontier: 81.2 FrontierSWE, 88.3 Terminal-Bench 2.1,
         // top-3 across six coding benchmarks (research 2026-07-17, K3 released 2026-07-16).
@@ -451,6 +504,11 @@ export const DEFAULT_CONFIG = {
                 "kimi-code/kimi-for-coding": "mid",
                 "kimi-code/kimi-for-coding-highspeed": "cheap",
             },
+            windows: {
+                "kimi-code/k3": 1_048_576,
+                "kimi-code/kimi-for-coding": 262_144,
+                "kimi-code/kimi-for-coding-highspeed": 262_144,
+            },
         },
     },
     pricing: { cheap: 0.1, mid: 0.5, frontier: 2.5 },
@@ -464,7 +522,7 @@ export const DEFAULT_CONFIG = {
     // v1.2: workers are the real agent TUI in the pane; "print" restores the -p-rendered-in-pane path.
     visibility: { llm: "headless", keepPanes: "run", worker: "interactive", workersPerTab: 3 },
 };
-function deepMerge(base, over) {
+function deepMerge(base, over, path = []) {
     if (over === undefined || over === null)
         return base;
     if (Array.isArray(base) || Array.isArray(over) || typeof base !== "object" || typeof over !== "object" || base === null) {
@@ -473,12 +531,18 @@ function deepMerge(base, over) {
     const out = { ...base };
     for (const [k, v] of Object.entries(over)) {
         if (v === null) {
-            delete out[k]; // v1.1 tombstone: an explicit null in an overlay removes the key (e.g. stale tiers model ids)
+            // T19: null is a declared multi-vendor state only at tiers.<adapter>.vendor. Every other null
+            // remains the v1.1 overlay tombstone (for example a stale tiers model id).
+            if (path.length === 2 && path[0] === "tiers" && k === "vendor")
+                out[k] = null;
+            else
+                delete out[k];
             continue;
         }
         // OBS-75 class: merge fresh-landing objects onto {} so nested tombstone nulls are pruned even when
         // no lower layer set the key (a fleet-cleared deny writes {adapters: null} — the schema rejects raw null)
-        out[k] = k in out ? deepMerge(out[k], v) : deepMerge({}, v);
+        const childPath = [...path, k];
+        out[k] = k in out ? deepMerge(out[k], v, childPath) : deepMerge({}, v, childPath);
     }
     return out;
 }
@@ -556,7 +620,7 @@ function resolveRoutingMode(cfg, layers) {
                 lints.push(`floors.${shape}: ${val} (config floors) overrides mode ${mode} — shadowed delta: ${shape} ${dflt}→${moded}`);
             }
             if (TIER_RANK[val] < TIER_RANK[integrityMin(shape)]) {
-                lints.push(`floors.${shape}: ${val} is below integrity minimum frontier — integrity class ${INTEGRITY_FLOOR_SHAPES.join("/")} holds regardless of mode`);
+                lints.push(`floors.${shape}: ${val} is below integrity minimum frontier — integrity class ${INTEGRITY_FLOOR_SHAPES.join("/")} is advisory to explicit floors and map pins; map pins are supreme`);
             }
             continue;
         }
@@ -586,6 +650,9 @@ export function loadConfigWithMode(repoRoot, opts = {}) {
     const r = TickmarkrConfigSchema.safeParse(merged);
     if (!r.success)
         throw new ConfigError(z.prettifyError(r.error));
+    // The vendored table attests Tickmarkr's built-in seed artifact. Operator overlays remain free to
+    // declare private/future models and deliberately smaller test windows without pretending T20 cited them.
+    assertSeededModelWindows();
     return { cfg: r.data, mode: resolveRoutingMode(r.data, [globalCfg, repoCfg]) };
 }
 export function loadConfig(repoRoot, opts = {}) {
@@ -642,7 +709,7 @@ export function configTemplate(overlay) {
 #   llm: headless         # headless (default): judge/review/consult run silently | pane: visible agents
 #   keepPanes: run        # run (default): ephemeral judge/review/consult panes close when read; a merged task's worker pane closes on done; other worker panes persist until run end | attempt | forever (keep everything for debugging)
 #   worker: interactive   # interactive (default): workers run the real agent TUI | print
-#   workersPerTab: 3      # cap concurrent worker panes per WORKERS tab; overflow opens WORKERS-2 (VIS-09). Default 3
+#   workersPerTab: 3      # cap worker panes sharing one tab; gate panes ride with their task and never count. Default 3
 # scope:
 #   allowDeviations: []    # globs an operator permits out-of-scope edits into, e.g. ["package-lock.json"]
 # gates:                  # override auto-detected commands; per-shape participation may only skip LLM gates
@@ -696,9 +763,8 @@ export function configTemplate(overlay) {
     const nl = base.indexOf("\n");
     return `${base.slice(0, nl + 1)}${lines.join("\n")}\n${base.slice(nl + 1)}`;
 }
-// v1.61 seed 8: the fleet-overlay provenance cluster (harvesting, serialization, diff rendering)
-// moved to fleet-overlay.ts as a pure move — re-exported here so prior import paths keep working.
-export { FLEET_OVERLAY_KEYS, fleetEditableEquals, fleetRepoOverlayFromDelta, harvestFleetProvenance, repoOverlayYaml, serializeFleetOverlay, unifiedYamlDiff, } from "./fleet-overlay.js";
+// Fleet overlay mutation, legacy serialization, and diff rendering live together in fleet-overlay.ts.
+export { FLEET_OVERLAY_KEYS, fleetEditableEquals, fleetRepoOverlayFromDelta, renderFleetOverlayWrite, repoOverlayYaml, serializeFleetOverlay, unifiedYamlDiff, } from "./fleet-overlay.js";
 export function repoOverlayPath(repoRoot) {
     return join(repoRoot, stateDirName(repoRoot), "config.yaml");
 }
@@ -710,15 +776,12 @@ export function readOverlayFile(path) {
         throw new ConfigError(`invalid overlay at ${path}`);
     return raw;
 }
-export function fleetEditableFromConfig(cfg, provenance = {}) {
+export function fleetEditableFromConfig(cfg) {
     const tiers = {};
     for (const [adapter, entry] of Object.entries(cfg.tiers)) {
         tiers[adapter] = {};
         for (const [model, tier] of Object.entries(entry.models)) {
-            // OBS-88: notes harvested from the overlay ride the session state from load, so a later
-            // write knows about every existing note — not only the ones this session typed itself
-            const note = provenance[adapter]?.[model];
-            tiers[adapter][model] = note ? { tier, provenance: note } : { tier };
+            tiers[adapter][model] = { tier };
         }
     }
     return {
@@ -754,55 +817,24 @@ export function fleetKeyLayer(repoRoot, dotted, opts = {}) {
 export function formatFleetPrint(repoRoot, opts = {}) {
     const gdir = opts.globalDir ?? globalConfigDir();
     const effective = loadConfig(repoRoot, { globalDir: gdir });
-    const editable = fleetEditableFromConfig(effective);
-    const lines = ["# tickmarkr fleet — effective state (repo > global > defaults)"];
-    const annotate = (dotted, yaml) => `${yaml}  # ${fleetKeyLayer(repoRoot, dotted, opts)}`;
-    const denyAdapters = editable.denyAdapters;
-    const denyModels = editable.denyModels;
-    if (denyAdapters.length || denyModels.length) {
-        lines.push("routing:");
-        lines.push("  deny:");
-        if (denyAdapters.length)
-            lines.push(annotate("routing.deny.adapters", `    adapters: ${stringify(denyAdapters).trim()}`));
-        if (denyModels.length)
-            lines.push(annotate("routing.deny.models", `    models: ${stringify(denyModels).trim()}`));
-    }
-    const mapKeys = Object.keys(editable.map).sort();
-    if (mapKeys.length) {
-        if (!lines.some((l) => l === "routing:"))
-            lines.push("routing:");
-        lines.push("  map:");
-        for (const shape of mapKeys) {
-            const entry = editable.map[shape];
-            const body = stringify(entry, { indent: 4 }).trim().split("\n").map((l) => `    ${l}`).join("\n");
-            lines.push(annotate(`routing.map.${shape}`, `    ${shape}:`));
-            lines.push(body.split("\n").slice(1).join("\n"));
-        }
-    }
-    const floorKeys = Object.keys(editable.floors).sort();
-    if (floorKeys.length) {
-        if (!lines.some((l) => l === "routing:"))
-            lines.push("routing:");
-        lines.push(annotate("routing.floors", `  floors: ${stringify(Object.fromEntries(floorKeys.map((k) => [k, editable.floors[k]]))).trim().replace(/^/gm, "  ")}`));
-    }
-    const tierAdapters = Object.keys(editable.tiers).sort();
-    if (tierAdapters.length) {
-        lines.push("tiers:");
-        for (const adapter of tierAdapters) {
-            const models = editable.tiers[adapter];
-            const modelIds = Object.keys(models).sort();
-            if (!modelIds.length)
-                continue;
-            lines.push(`  ${adapter}:`);
-            lines.push("    models:");
-            for (const model of modelIds) {
-                const v = models[model];
-                if (v === null)
-                    lines.push(`      ${model}: null`);
-                else
-                    lines.push(annotate(`tiers.${adapter}.models.${model}`, `      ${model}: ${v.tier}`));
-            }
+    const doc = new Document({ routing: effective.routing, tiers: effective.tiers });
+    doc.commentBefore = " tickmarkr fleet — effective state (repo > global > defaults)";
+    const annotate = (path, dotted = path.join(".")) => {
+        const node = doc.getIn(path, true);
+        if (node && typeof node === "object" && "comment" in node) {
+            node.comment = ` ${fleetKeyLayer(repoRoot, dotted, opts)}`;
         }
+    };
+    if (effective.routing.deny?.adapters !== undefined)
+        annotate(["routing", "deny", "adapters"]);
+    if (effective.routing.deny?.models !== undefined)
+        annotate(["routing", "deny", "models"]);
+    for (const shape of Object.keys(effective.routing.map))
+        annotate(["routing", "map", shape]);
+    annotate(["routing", "floors"]);
+    for (const [adapter, entry] of Object.entries(effective.tiers)) {
+        for (const model of Object.keys(entry.models))
+            annotate(["tiers", adapter, "models", model]);
     }
-    return `${lines.join("\n")}\n`;
+    return String(doc);
 }

package/dist/config/fleet-overlay.d.ts

@@ -1,27 +1,32 @@
-import type { FleetEditable } from "./config.js";
+import type { FleetEditable, RoutingMode } from "./config.js";
 /** Fleet-owned overlay keys — the only config surface `tickmarkr fleet` may write. */
 export declare const FLEET_OVERLAY_KEYS: readonly ["routing", "tiers"];
-/** Trailing `# note` comments on per-entry lines an operator may have hand-written or a prior
- *  fleet write stamped: model tier lines (including null tombstones) and deny list items. */
-export type HarvestedProvenance = {
-    tiers: Record<string, Record<string, string>>;
-    denyAdapters: Record<string, string>;
-    denyModels: Record<string, string>;
+export type FleetOverlayWrite = {
+    initial: FleetEditable;
+    edited: FleetEditable;
+    mode?: RoutingMode;
+    steering?: {
+        initial: {
+            review?: string[];
+            consult?: string[];
+        };
+        edited: {
+            review?: string[];
+            consult?: string[];
+        };
+    };
 };
-/** OBS-88: harvest existing provenance comments from raw repo-overlay bytes at fleet-session
- *  load. yaml.parse discards comments, so before this every fleet write re-serialized the file
- *  knowing only the current session's own notes and silently stripped all prior ones — a typed
- *  benchmark-provenance note survived exactly one write. Fail-open to empty: an unreadable
- *  overlay is the loader's problem to reject, never the harvester's. */
-export declare function harvestFleetProvenance(overlayText: string): HarvestedProvenance;
-/** Harvested deny-entry notes keyed by the exact entry string, re-attached at serialize time. */
-export type FleetDenyNotes = {
-    adapters?: Record<string, string>;
-    models?: Record<string, string>;
+export type FleetFirstTouch = {
+    vendor: string;
+    channel: "sub" | "api";
 };
+export declare function fleetFirstTouchProvenance(note: string, firstTouch: FleetFirstTouch): string;
+/** Apply only fields fleet authored to the parsed YAML document. Untouched nodes retain their
+ * keys, ordering, scalar style, and comments; fresh provenance is written directly on its tier node. */
+export declare function renderFleetOverlayWrite(priorBytes: string, write: FleetOverlayWrite): string;
 /** Build the repo overlay fragment fleet would write for edits since session start. */
-export declare function fleetRepoOverlayFromDelta(initial: FleetEditable, edited: FleetEditable, existingRepo?: Record<string, unknown>): Record<string, unknown>;
-export declare function repoOverlayYaml(overlay: Record<string, unknown>, provenance?: Record<string, Record<string, string>>, denyNotes?: FleetDenyNotes): string;
-export declare function serializeFleetOverlay(overlay: Record<string, unknown>, provenance?: Record<string, Record<string, string>>, denyNotes?: FleetDenyNotes): string;
+export declare function fleetRepoOverlayFromDelta(initial: FleetEditable, edited: FleetEditable, existingRepo?: Record<string, unknown>, firstTouches?: Readonly<Record<string, FleetFirstTouch>>): Record<string, unknown>;
+export declare function repoOverlayYaml(overlay: Record<string, unknown>): string;
+export declare function serializeFleetOverlay(overlay: Record<string, unknown>): string;
 export declare function unifiedYamlDiff(before: string, after: string, label?: string): string;
 export declare function fleetEditableEquals(a: FleetEditable, b: FleetEditable): boolean;