agentlinkops 0.6.5 → 0.6.7

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "agentlinkops",
3
- "version": "0.6.5",
3
+ "version": "0.6.7",
4
4
  "description": "Connect an agent to AgentLinkOps and run backlink work with it: the connect skill (views, the three discovery verbs, evidence rules), local ledger and campaign skills, a portable CRM, the source-cited content pipeline, user-managed disavow lists, HTML reports and monitoring-event sync.",
5
5
  "author": {
6
6
  "name": "AgentLinkOps"
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "agentlinkops",
3
- "version": "0.6.5",
3
+ "version": "0.6.7",
4
4
  "description": "Connect an agent to AgentLinkOps and run backlink work with it: the connect skill (views, the three discovery verbs, evidence rules), local ledger and campaign skills, a portable CRM, the source-cited content pipeline, user-managed disavow lists, HTML reports and monitoring-event sync.",
5
5
  "author": {
6
6
  "name": "AgentLinkOps"
package/CHANGELOG.md CHANGED
@@ -3,6 +3,19 @@
3
3
  All notable changes to the `agentlinkops` package are recorded here. The version follows the
4
4
  plugin manifests in `.claude-plugin/` and `.codex-plugin/`.
5
5
 
6
+ ## 0.6.7 (2026-09-20)
7
+
8
+ - Default browser citation measurement to accountless mode. Saved session files are ignored unless `AGENTLINKOPS_BROWSER_AUTH=saved-session` is explicitly selected.
9
+ - Reject custom launch, environment, extension and Firefox-preference overrides in accountless mode. Proxy admission remains required; direct diagnostics still need explicit authorization and configuration.
10
+ - Keep anonymous login walls, blocked pages and failed answers unknown. Accountless mode does not guarantee that a provider will answer.
11
+
12
+ ## 0.6.6 (2026-09-20)
13
+
14
+ - Retain supplied browser usage receipts in a bounded local journal after each attempt and cleanup. Missing meter values remain unknown; failed writes stop the run.
15
+ - Close browser adapters after failed runs and admission errors. Cleanup uncertainty stays explicit in retained receipts.
16
+ - Check citation files against their recorded observation hashes before freezing reports or uploading evidence. Changed or unreferenced files cannot silently add samples.
17
+ - Include retained inspection-attempt status and rolling website quota in the command response schema.
18
+
6
19
  ## 0.6.5 (2026-09-20)
7
20
 
8
21
  - Browser measurement requires a configured proxy. Authorized direct diagnostics need an explicit setting; missing proxy configuration never selects direct traffic.
package/README.md CHANGED
@@ -149,14 +149,15 @@ its sample size, classified against the previous epoch as declined, grown or
149
149
  not_distinguishable, or `insufficient_data` when n is too thin to interpret.
150
150
 
151
151
  ```sh
152
- export PERPLEXITY_API_KEY=... # live engine; omit it and the mock engine runs at $0
152
+ agentlinkops citation panel --domain example.com --brand Example --out panel.json
153
153
  agentlinkops citation run panel.json --max-usd 1.00
154
154
  ```
155
155
 
156
- The panel is JSON: `targets` (domain or url, brand, aliases), `prompts`, `engines`
157
- (`mock` or `perplexity`), optional `samples` and `maxUsd`. Every run writes immutable
158
- content-addressed answer snapshots plus append-only observation and epoch rows under
159
- `.agentlinkops/citations/`. The runner reserves estimated cost before each call and retry. A reported supplier
156
+ New panels use the free mock engine. A live engine requires its own configuration and credentials;
157
+ missing credentials never select mock results. The JSON panel contains `targets`, `prompts`,
158
+ `engines`, optional `samples` and `maxUsd`. Answer snapshots live in
159
+ `.agentlinkops/citations/evidence/`; observation and epoch ledgers live at the `.agentlinkops/` root.
160
+ The runner reserves estimated cost before each call and retry. A reported supplier
160
161
  overrun stops later calls; estimates do not enforce a provider invoice cap. A budget
161
162
  abort leaves a partial-epoch receipt and exit code 2. A cell
162
163
  that declined with complete evidence exits 1, like an expected link observed absent.
@@ -166,6 +167,16 @@ For an explicitly authorized direct diagnostic, set `AGENTLINKOPS_BROWSER_EGRESS
166
167
  and leave the proxy unset. The standard URL proxy provides no actual usage meter;
167
168
  its cost remains an estimate. Browser login and live account acceptance are separate.
168
169
 
170
+ Browser epochs retain supplied usage and cleanup receipts in
171
+ `.agentlinkops/citations/egress/<epochId>.jsonl`. Unknown meter values stay null. A missing
172
+ final record means the journal is incomplete; it does not establish complete costs or cleanup.
173
+ Journals stop at 10,000 records or 8 MiB, and a failed write stops further samples.
174
+
175
+ Frozen inventories and evidence uploads check files against their retained observation hashes.
176
+ Historical hashes verify the recorded answer fields and screenshot filename; they do not prove
177
+ when an answer was captured or authenticate old screenshot pixels. Hosted uploads retain a
178
+ separate hash of the exact uploaded file bytes.
179
+
169
180
  ## Environment variables
170
181
 
171
182
  | Variable | Purpose |
package/cli/catalog.js CHANGED
@@ -14687,6 +14687,111 @@ export const CATALOG_COMMANDS = [
14687
14687
  "type": "null"
14688
14688
  }
14689
14689
  ]
14690
+ },
14691
+ "attemptSummary": {
14692
+ "type": "object",
14693
+ "properties": {
14694
+ "observedAt": {
14695
+ "type": "string"
14696
+ },
14697
+ "latestAttempt": {
14698
+ "anyOf": [
14699
+ {
14700
+ "type": "object",
14701
+ "properties": {
14702
+ "id": {
14703
+ "type": "string",
14704
+ "pattern": "^[a-f0-9]{64}$"
14705
+ },
14706
+ "admittedAt": {
14707
+ "type": "string"
14708
+ },
14709
+ "status": {
14710
+ "type": "string",
14711
+ "enum": [
14712
+ "in_progress",
14713
+ "unsettled",
14714
+ "completed"
14715
+ ]
14716
+ },
14717
+ "receiptId": {
14718
+ "anyOf": [
14719
+ {
14720
+ "type": "string",
14721
+ "pattern": "^[a-f0-9]{64}$"
14722
+ },
14723
+ {
14724
+ "type": "null"
14725
+ }
14726
+ ]
14727
+ },
14728
+ "leaseExpiresAt": {
14729
+ "type": [
14730
+ "string",
14731
+ "null"
14732
+ ]
14733
+ }
14734
+ },
14735
+ "required": [
14736
+ "id",
14737
+ "admittedAt",
14738
+ "status",
14739
+ "receiptId",
14740
+ "leaseExpiresAt"
14741
+ ],
14742
+ "additionalProperties": false
14743
+ },
14744
+ {
14745
+ "type": "null"
14746
+ }
14747
+ ]
14748
+ },
14749
+ "unsettledCount": {
14750
+ "type": "integer",
14751
+ "minimum": 0,
14752
+ "maximum": 9007199254740991
14753
+ }
14754
+ },
14755
+ "required": [
14756
+ "observedAt",
14757
+ "latestAttempt",
14758
+ "unsettledCount"
14759
+ ],
14760
+ "additionalProperties": false
14761
+ },
14762
+ "websiteQuota": {
14763
+ "type": "object",
14764
+ "properties": {
14765
+ "windowStart": {
14766
+ "type": "string"
14767
+ },
14768
+ "observedAt": {
14769
+ "type": "string"
14770
+ },
14771
+ "admitted": {
14772
+ "type": "integer",
14773
+ "minimum": 0,
14774
+ "maximum": 9007199254740991
14775
+ },
14776
+ "limit": {
14777
+ "type": "integer",
14778
+ "exclusiveMinimum": 0,
14779
+ "maximum": 9007199254740991
14780
+ },
14781
+ "remaining": {
14782
+ "type": "integer",
14783
+ "minimum": 0,
14784
+ "maximum": 9007199254740991
14785
+ }
14786
+ },
14787
+ "required": [
14788
+ "windowStart",
14789
+ "observedAt",
14790
+ "admitted",
14791
+ "limit",
14792
+ "remaining"
14793
+ ],
14794
+ "additionalProperties": false
14690
14795
  }
14691
14796
  },
14692
14797
  "required": [
package/cli/citation.js CHANGED
@@ -28,8 +28,8 @@ const USAGE = `agentlinkops citation — AI citation watches in this repository
28
28
  tallies land under .agentlinkops/citations/
29
29
  agentlinkops citation run PANEL.json --engine chatgpt:web-own-browser
30
30
  measure the consumer UI through Camoufox
31
- (dedicated account, personal cadence; evidence
32
- includes a screenshot of every answer)
31
+ (accountless by default; saved browser sessions
32
+ are not loaded)
33
33
  agentlinkops citation sweep PANEL.json --max-usd USD --out INVENTORY.json [--dir DIR]
34
34
  run a bounded epoch and freeze its retained evidence
35
35
  agentlinkops citation inventory PANEL.json --out INVENTORY.json [--epoch ID] [--dir DIR]
@@ -52,8 +52,8 @@ const USAGE = `agentlinkops citation — AI citation watches in this repository
52
52
  agentlinkops citation alert PANEL.json [--webhook URL --report REPORT.html]
53
53
  check the latest epoch for declines;
54
54
  logs an alert line if any cell declined
55
- agentlinkops citation login ENGINE open a browser, log in once, save the session
56
- (the engine adapter uses it on every run)
55
+ agentlinkops citation login ENGINE optional legacy saved-session setup
56
+ (not needed for accountless measurement)
57
57
  agentlinkops citation logout ENGINE remove a saved session
58
58
  agentlinkops citation sessions list saved engine sessions
59
59
  agentlinkops citation sync [--dir DIR] [--evidence] push local epoch rows to the hosted
@@ -65,8 +65,11 @@ const USAGE = `agentlinkops citation — AI citation watches in this repository
65
65
 
66
66
  Panel (JSON): targets (domain or url, brand, aliases), prompts, engines
67
67
  ([{ "engine": "mock" | "perplexity", "model": "…" }]), optional samples and maxUsd.
68
- Live engines need their configured credentials or a usable browser session; the mock
69
- engine runs everything at zero cost. Every run states its estimated spend, and the
68
+ API engines need their configured credentials. Browser measurement defaults to
69
+ AGENTLINKOPS_BROWSER_AUTH=accountless and never loads saved sessions in that mode.
70
+ Optional legacy saved-session mode requires explicit AGENTLINKOPS_BROWSER_AUTH=saved-session.
71
+ Anonymous login walls or failed answers remain unknown; access is not guaranteed.
72
+ The mock engine runs everything at zero cost. Every run states its estimated spend, and the
70
73
  cap reserves estimated cost before each call. A supplier charge above that estimate
71
74
  is recorded as an overrun and stops subsequent calls. Browser measurement requires
72
75
  AGENTLINKOPS_PROXY_URL. Authorized direct diagnostics need
package/cli/client.js CHANGED
@@ -2,7 +2,7 @@ import {restResponseContract} from '../shared/response-contract.js';
2
2
  // A small HTTP client for the cloud. Scoped keys only; no human session ever reaches here.
3
3
  // DP-0036-T13: the CLI names itself so the usage counters can tell it from other REST callers.
4
4
  // The version follows the plugin manifests; bump it with them.
5
- export const CLI_VERSION = '0.6.5';
5
+ export const CLI_VERSION = '0.6.7';
6
6
  export const CLI_USER_AGENT = `agentlinkops-cli/${CLI_VERSION}`;
7
7
  export class CloudError extends Error {
8
8
  constructor(code, status, details = null) {
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "agentlinkops",
3
- "version": "0.6.5",
3
+ "version": "0.6.7",
4
4
  "description": "A backlink and AI citation ledger that lives in your repository: local link and citation checks with dated evidence, supplier-export imports and optional sync to the hosted AgentLinkOps service.",
5
5
  "author": "AgentLinkOps (https://agentlinkops.com)",
6
6
  "license": "Apache-2.0",
@@ -12,7 +12,9 @@ This command defaults to the mock engine. Its suggested questions are writing pr
12
12
 
13
13
  Use `--competitors competitors.json` to add domains to check beside your own. The file is an array of objects with `domain`, `brand` and optional `aliases` fields. Each target gets separate results for each question. A target missing from one answer is only missing from that sample; it does not establish absence from the engine.
14
14
 
15
- Choose a live engine explicitly after checking access and budget. For an existing browser setup, use `--engine chatgpt:web-own-browser` or `--engine grok:web-own-browser`. `--engine google-aio` requires your configured supplier access. This workflow does not create accounts or buy services.
15
+ Choose a live engine explicitly after checking access and budget. For accountless browser measurement, use `--engine chatgpt:web-own-browser` or `--engine grok:web-own-browser`. `--engine google-aio` requires your configured supplier access. This workflow does not create accounts or buy services.
16
+
17
+ Starting with CLI 0.6.7, browser authentication defaults to `AGENTLINKOPS_BROWSER_AUTH=accountless`. This mode does not load saved browser sessions, even if session files already exist. A login wall, blocked page or failed answer remains unknown; anonymous access is not guaranteed. Use accountless measurement for this workflow. The optional legacy `AGENTLINKOPS_BROWSER_AUTH=saved-session` mode is separate and requires an explicit opt-in; logging in is not a prerequisite for the default workflow.
16
18
 
17
19
  Starting with CLI 0.6.5, browser measurement requires `AGENTLINKOPS_PROXY_URL`. Missing, failed or quarantined proxy admission never selects direct traffic. An explicitly authorized direct diagnostic requires `AGENTLINKOPS_BROWSER_EGRESS=direct-diagnostic` with no proxy configured. The standard URL proxy has no usage meter; estimated costs are not a provider invoice cap.
18
20
 
@@ -10,5 +10,7 @@ export const indexSettingsResponse=z.object({projectId:id,enabled:z.boolean(),da
10
10
  export const indexObservation=z.looseObject({id:digest,observationId:id,checkedAt:time,recordedAt:time,expiresAt:time,evidenceSha256:digest,eventId:digest.nullable(),ignored:z.string().nullable(),version:z.literal(1),url:z.url(),lane:z.enum(['gsc','licensed_site']),backend:z.string(),source_key:z.string(),tier:z.enum(['indexed','not_indexed','likely_indexed','not_found_in_site_query','unknown']),confidence:z.enum(['google_index_snapshot','search_result_sample','unknown']),reason:z.string(),live_page_test:z.literal(false),submission_requested:z.literal(false)});
11
11
  export const indexReceiptResponse=z.object({observation:indexObservation,rawResponse:z.unknown(),evidenceAvailable:z.boolean()}).passthrough();
12
12
  export const indexInspectResponse=indexReceiptResponse.extend({replayed:z.boolean(),providerError:z.enum(['PROVIDER_PERMISSION_LOST','PROVIDER_RATE_LIMITED','PROVIDER_UNAVAILABLE','PROVIDER_RESPONSE_INVALID']).nullable().optional()});
13
- export const indexHistoryResponse=z.object({observations:z.array(indexObservation),nextCursor:digest.nullable()}).passthrough();
13
+ export const indexAttemptSummary=z.object({observedAt:time,latestAttempt:z.object({id:digest,admittedAt:time,status:z.enum(['in_progress','unsettled','completed']),receiptId:digest.nullable(),leaseExpiresAt:time.nullable()}).nullable(),unsettledCount:z.number().int().nonnegative()});
14
+ export const indexWebsiteQuota=z.object({windowStart:time,observedAt:time,admitted:z.number().int().nonnegative(),limit:z.number().int().positive(),remaining:z.number().int().nonnegative()});
15
+ export const indexHistoryResponse=z.object({observations:z.array(indexObservation),nextCursor:digest.nullable(),attemptSummary:indexAttemptSummary.optional(),websiteQuota:indexWebsiteQuota.optional()}).passthrough();
14
16
  export const indexEventsResponse=z.object({events:z.array(z.looseObject({id:digest,receiptId:digest,type:z.enum(['index.deindexed','index.reindexed']),createdAt:time,url:z.url(),source_key:z.string(),before:z.enum(['indexed','not_indexed']),after:z.enum(['indexed','not_indexed']),observation_ids:z.array(id)}))}).passthrough();
@@ -16,6 +16,6 @@ If the user authorizes outreach, use their existing email tool and honor their s
16
16
 
17
17
  When a placement is earned, save source and destination URLs and its relationship to the opportunity. Use connected monitoring tools when requested; local campaign work remains useful without cloud access. Do not promise rankings, response rates, exhaustive competitor coverage, or improvements unsupported by evidence.
18
18
 
19
- For citation panels, read [the panel authoring reference](../../references/citation-panels.md). Gather real audience questions with the user's available sources and record where each came from. `agentlinkops citation panel` drafts editable JSON with a mock engine by default. Suggested questions are not evidence of search demand. Choose a live engine only after checking the user's access and budget; keep citations, mentions, competitors and locale context separate in the results.
19
+ For citation panels, read [the panel authoring reference](../../references/citation-panels.md). Gather real audience questions with the user's available sources and record where each came from. `agentlinkops citation panel` drafts editable JSON with a mock engine by default. Suggested questions are not evidence of search demand. Choose a live engine only after checking the user's access and budget; keep citations, mentions, competitors and locale context separate in the results. Starting with CLI 0.6.7, browser citation measurement defaults to accountless and does not load saved sessions. Do not ask the user to log in as a prerequisite. Anonymous failures remain unknown. The panel reference keeps required proxy admission separate from an explicitly authorized direct diagnostic.
20
20
 
21
21
  Use the [campaign handoff recipe](../../references/recipes/qualified-campaign-handoff.md) when preparing records and reviewed drafts for the person's CRM or sender. It defines checkpoints and later readback before treating an uncertain write as complete.
@@ -9,7 +9,7 @@ records. This page is the complete agent reference: the same text prints from
9
9
  full once per session, then use `describe` for any command's exact schema.
10
10
 
11
11
  <!-- generated:skill-version begin (npm run generate:tools) -->
12
- Pack version 0.6.3. 125 commands in 13 toolsets.
12
+ Pack version 0.6.5. 125 commands in 13 toolsets.
13
13
  <!-- generated:skill-version end -->
14
14
 
15
15
  ## Install and connect (once per machine)
@@ -61,6 +61,18 @@ export function createBrowserEngine({ engineName, env = process.env, onEvent = (
61
61
  if (!BROWSER_PROVIDERS.includes(engineName)) {
62
62
  throw new EngineError(`unknown browser engine: ${engineName} (known: ${BROWSER_PROVIDERS.join(', ')})`);
63
63
  }
64
+ const authentication = env.AGENTLINKOPS_BROWSER_AUTH ?? 'accountless';
65
+ if (!['accountless', 'saved-session'].includes(authentication)) {
66
+ throw Object.assign(new EngineError('Unsupported browser authentication policy.'), { code: 'BROWSER_AUTH_POLICY_INVALID' });
67
+ }
68
+ // The vendored resolver reads process.env directly; arbitrary launch/profile
69
+ // and extension overrides must not reintroduce ambient identity.
70
+ const assertAccountlessLaunch = () => {
71
+ if (authentication === 'accountless' && ['CAMOUFOX_EXTRA_LAUNCH_JSON', 'CAMOUFOX_ENV_JSON', 'CAMOUFOX_ADDONS', 'CAMOUFOX_ARGS', 'CAMOUFOX_FIREFOX_USER_PREFS_JSON'].some(key => env[key] || process.env[key])) {
72
+ throw Object.assign(new EngineError('Accountless measurement does not accept custom browser launch or extension overrides.'), { code: 'BROWSER_AUTH_POLICY_INVALID' });
73
+ }
74
+ };
75
+ assertAccountlessLaunch();
64
76
  const identity = { engine: engineName, provider: 'web-own-browser', model: engineName };
65
77
  const sessionDir = runtime.sessionDir ?? join(homedir(), '.agentlinkops', 'citations', 'sessions');
66
78
  const load = runtime.loadModule ?? lazy;
@@ -110,6 +122,7 @@ export function createBrowserEngine({ engineName, env = process.env, onEvent = (
110
122
  const proxy = admission.proxy;
111
123
  await event(proxy?'egress: proxy-required':'egress: direct-diagnostic');
112
124
 
125
+ assertAccountlessLaunch();
113
126
  const options = await resolveCamoufoxLaunchOptions({
114
127
  provider: engineName,
115
128
  proxy: proxy ?? undefined,
@@ -123,6 +136,7 @@ export function createBrowserEngine({ engineName, env = process.env, onEvent = (
123
136
  const launchOpts = { ...options, executablePath: options.executablePath };
124
137
  if (proxy) launchOpts.proxy = proxy;
125
138
  admission.assertReady();
139
+ assertAccountlessLaunch();
126
140
  const browser = await fw.launch(launchOpts);
127
141
 
128
142
  browserHandle = {
@@ -139,6 +153,8 @@ export function createBrowserEngine({ engineName, env = process.env, onEvent = (
139
153
  }
140
154
 
141
155
  async function loadStorageState() {
156
+ // Accountless measurement never opens session files, even when they exist.
157
+ if (authentication === 'accountless') return undefined;
142
158
  try {
143
159
  const { readFile } = await load('node:fs/promises');
144
160
  return JSON.parse(await readFile(join(sessionDir, `${engineName}.json`), 'utf8'));
@@ -362,7 +378,7 @@ export function createBrowserEngine({ engineName, env = process.env, onEvent = (
362
378
  },
363
379
  };
364
380
  const sample=engine.run.bind(engine),closeBrowser=engine.close.bind(engine);
365
- const policyCodes=new Set(['PROXY_REQUIRED','PROXY_POLICY_INVALID','PROXY_POLICY_CONFLICT','PROXY_CONFIGURATION_INVALID','PROXY_ACQUIRE_FAILED','PROXY_SUPPLIER_INVALID','PROXY_LEASE_INVALID','PROXY_LEASE_EXPIRED','PROXY_QUARANTINE_FAILED','PROXY_RELEASE_FAILED','PROXY_RECEIPT_FAILED','PROXY_CUSTODY_UNAVAILABLE','BROWSER_CLEANUP_UNCONFIRMED','BROWSER_RUN_IN_PROGRESS']);
381
+ const policyCodes=new Set(['BROWSER_AUTH_POLICY_INVALID','PROXY_REQUIRED','PROXY_POLICY_INVALID','PROXY_POLICY_CONFLICT','PROXY_CONFIGURATION_INVALID','PROXY_ACQUIRE_FAILED','PROXY_SUPPLIER_INVALID','PROXY_LEASE_INVALID','PROXY_LEASE_EXPIRED','PROXY_QUARANTINE_FAILED','PROXY_RELEASE_FAILED','PROXY_RECEIPT_FAILED','PROXY_CUSTODY_UNAVAILABLE','BROWSER_CLEANUP_UNCONFIRMED','BROWSER_RUN_IN_PROGRESS']);
366
382
  const safeError=cause=>{
367
383
  const code=policyCodes.has(cause?.code)?cause.code:'BROWSER_RUN_FAILED';
368
384
  const error=Object.assign(new EngineError(code==='BROWSER_RUN_FAILED'?'Browser sample failed.':'Browser admission or cleanup failed.',{retriable:code==='BROWSER_RUN_FAILED'&&cause?.retriable===true}),{code});
@@ -384,6 +400,7 @@ export function createBrowserEngine({ engineName, env = process.env, onEvent = (
384
400
  if(running)throw Object.assign(new EngineError('A browser sample is already running.'),{code:'BROWSER_RUN_IN_PROGRESS'});
385
401
  running=true;let result,error,reason=null,admitted=false;
386
402
  try{
403
+ assertAccountlessLaunch();
387
404
  if(admission.expired)await retire('lease_expiry');
388
405
  await admission.begin();admitted=true;
389
406
  result=await sample(input);
@@ -0,0 +1,80 @@
1
+ // Supplemental operational evidence; never part of citation-envelope hashes.
2
+ import { open, mkdir } from 'node:fs/promises';
3
+ import { join } from 'node:path';
4
+ import { EngineError } from './adapter.js';
5
+ export const EGRESS_LIMITS = Object.freeze({ rows: 10000, bytes: 8 * 1024 * 1024 });
6
+ const fail = code => Object.assign(new EngineError('Egress evidence could not be persisted; the epoch stopped.'), { code });
7
+ const codes = new Set(['PROXY_REQUIRED','PROXY_POLICY_INVALID','PROXY_POLICY_CONFLICT','PROXY_CONFIGURATION_INVALID','PROXY_ACQUIRE_FAILED','PROXY_SUPPLIER_INVALID','PROXY_LEASE_INVALID','PROXY_LEASE_EXPIRED','PROXY_QUARANTINED','PROXY_QUARANTINE_FAILED','PROXY_RELEASE_FAILED','PROXY_RECEIPT_FAILED','PROXY_CUSTODY_UNAVAILABLE','BROWSER_CLEANUP_UNCONFIRMED','BROWSER_RUN_IN_PROGRESS','BROWSER_RUN_FAILED']);
8
+ const safeCode = error => error ? codes.has(error.code) ? error.code : 'ENGINE_ERROR' : null;
9
+ const enumeration = (value, values) => { if (!values.includes(value)) throw fail('EGRESS_RECEIPT_INVALID'); return value; };
10
+ const count = value => { if (!Number.isSafeInteger(value) || value < 0) throw fail('EGRESS_RECEIPT_INVALID'); return value; };
11
+ function sanitize(row) {
12
+ if (!row || row.version !== 1 || count(row.sequence) < 1) throw fail('EGRESS_RECEIPT_INVALID');
13
+ const observedAt = typeof row.observedAt === 'string' && row.observedAt.length <= 35 ? new Date(row.observedAt) : null;
14
+ if (!observedAt || !Number.isFinite(observedAt.getTime())) throw fail('EGRESS_RECEIPT_INVALID');
15
+ if (row.leaseHash !== null && !/^[a-f0-9]{64}$/.test(row.leaseHash ?? '')) throw fail('EGRESS_RECEIPT_INVALID');
16
+ const meterState = enumeration(row.meterState, ['measured','bytes_only','unavailable','not_applicable','incomplete']);
17
+ const bytes = row.bytes === null ? null : count(row.bytes);
18
+ const costMicrousd = row.costMicrousd === null ? null : count(row.costMicrousd);
19
+ if ((meterState === 'measured' && (bytes === null || costMicrousd === null)) || (meterState === 'bytes_only' && (bytes === null || costMicrousd !== null)) || (['unavailable','not_applicable'].includes(meterState) && (bytes !== null || costMicrousd !== null))) throw fail('EGRESS_RECEIPT_INVALID');
20
+ return { version: 1, sequence: row.sequence,
21
+ egress: enumeration(row.egress, ['proxy-required','direct-diagnostic']),
22
+ phase: enumeration(row.phase, ['sample','cleanup','failure_cleanup','lease_expiry']),
23
+ outcome: enumeration(row.outcome, ['observed','unknown','failed','closed','cleanup_unconfirmed']),
24
+ reason: enumeration(row.reason, [null,'blocked','rate_limited','connection_error','cleanup_unconfirmed']),
25
+ leaseHash: row.leaseHash, observedAt: observedAt.toISOString(), meterState, bytes, costMicrousd };
26
+ }
27
+
28
+ export async function createEgressJournal({ dir, epochId, engines, identityHashes = new Map(), limits = EGRESS_LIMITS }) {
29
+ if (typeof epochId !== 'string' || !/^[A-Za-z0-9_-]{1,100}$/.test(epochId)) throw fail('EGRESS_JOURNAL_UNAVAILABLE');
30
+ const tracked = new Map(engines.map((engine, index) => [engine, { index, retained: [] }]));
31
+ if (!tracked.size) return null;
32
+ if (!Number.isSafeInteger(limits.rows) || limits.rows < 1 || !Number.isSafeInteger(limits.bytes) || limits.bytes < 1) throw fail('EGRESS_LIMIT_INVALID');
33
+ // A reusable adapter may retain earlier epochs in memory. Establish the cursor
34
+ // before this epoch's first request; never count or rewrite those older deltas.
35
+ for (const [engine, state] of tracked) {
36
+ try {
37
+ const prior = engine.getEgressReceipts?.() ?? [];
38
+ if (!Array.isArray(prior) || prior.length > limits.rows) throw fail('EGRESS_RECEIPT_LIMIT');
39
+ state.retained = prior.map((row, index) => {
40
+ const value = sanitize(row);
41
+ if (value.sequence !== index + 1) throw fail('EGRESS_RECEIPT_INVALID');
42
+ return JSON.stringify(value);
43
+ });
44
+ } catch { throw fail('EGRESS_RECEIPT_UNAVAILABLE'); }
45
+ }
46
+ let file;
47
+ try { await mkdir(join(dir, 'citations/egress'), { recursive: true }); file = await open(join(dir, 'citations/egress', `${epochId}.jsonl`), 'wx', 0o600); }
48
+ catch { throw fail('EGRESS_JOURNAL_UNAVAILABLE'); }
49
+ let rows = 0, bytes = 0;
50
+ async function append(row) {
51
+ const text = JSON.stringify(row) + '\n', size = Buffer.byteLength(text);
52
+ if (rows + 1 > limits.rows || bytes + size > limits.bytes) throw fail('EGRESS_RECEIPT_LIMIT');
53
+ try { await file.writeFile(text); await file.sync(); }
54
+ catch { throw fail('EGRESS_WRITE_FAILED'); }
55
+ rows++; bytes += size;
56
+ }
57
+ try { await append({ type: 'start', version: 1, epochId, engines: [...tracked].map(([engine, state]) => ({ engineIndex: state.index, firstSequence: state.retained.length + 1, identityHash: /^[a-f0-9]{64}$/.test(identityHashes.get(engine) ?? '') ? identityHashes.get(engine) : null })), completeness: 'open', costBasis: 'supplier_meter_only', totals: null }); }
58
+ catch (error) { await file.close().catch(() => {}); throw error; }
59
+ return {
60
+ async capture(engine, { phase, error = null, cellHash = null, runIndex = null, attempt = null } = {}) {
61
+ const state = tracked.get(engine); if (!state) return;
62
+ if (cellHash !== null && !/^[a-f0-9]{64}$/.test(cellHash)) throw fail('EGRESS_RECEIPT_INVALID');
63
+ if (runIndex !== null) count(runIndex);
64
+ if (attempt !== null) count(attempt);
65
+ let receipts;
66
+ try { receipts = engine.getEgressReceipts?.() ?? []; }
67
+ catch { throw fail('EGRESS_RECEIPT_UNAVAILABLE'); }
68
+ if (!Array.isArray(receipts) || receipts.length > limits.rows || receipts.length < state.retained.length) throw fail('EGRESS_RECEIPT_LIMIT');
69
+ for (let i = 0; i < receipts.length; i++) {
70
+ const receipt = sanitize(receipts[i]), serialized = JSON.stringify(receipt);
71
+ if (receipt.sequence !== i + 1 || (i < state.retained.length && state.retained[i] !== serialized)) throw fail('EGRESS_RECEIPT_INVALID');
72
+ if (i < state.retained.length) continue;
73
+ await append({ type: 'receipt', engineIndex: state.index, receipt }); state.retained.push(serialized);
74
+ }
75
+ await append({ type: 'status', engineIndex: state.index, phase: enumeration(phase, ['attempt','cleanup']), outcome: error ? 'error' : 'returned', code: safeCode(error), cellHash, runIndex, attempt, receiptsAvailable: typeof engine.getEgressReceipts === 'function' });
76
+ },
77
+ async finish({ failed, cleanupUncertain, evidenceComplete }) { await append({ type: 'finish', outcome: failed ? 'failed' : 'returned', cleanupUncertain, evidenceComplete, totals: null }); },
78
+ async close() { try { await file.close(); } catch { throw fail('EGRESS_WRITE_FAILED'); } },
79
+ };
80
+ }
@@ -0,0 +1,51 @@
1
+ import { createHash } from 'node:crypto';
2
+ const hash = value => createHash('sha256').update(value).digest('hex');
3
+
4
+ function suppliedFieldsPreserved(raw, parsed) {
5
+ if (raw === null || typeof raw !== 'object') return Object.is(raw, parsed);
6
+ if (Array.isArray(raw)) return Array.isArray(parsed) && raw.length === parsed.length && raw.every((value, index) => suppliedFieldsPreserved(value, parsed[index]));
7
+ return parsed !== null && typeof parsed === 'object' && !Array.isArray(parsed) &&
8
+ Object.keys(raw).every(key => Object.hasOwn(parsed, key) && suppliedFieldsPreserved(raw[key], parsed[key]));
9
+ }
10
+
11
+ // Historical files were written by JSON.stringify. Retain whitespace and key
12
+ // order compatibility, but reject duplicate keys and alternate encodings that
13
+ // JSON.parse would discard before semantic reconstruction. Quoted whitespace
14
+ // remains part of its token. This is not a replacement JSON grammar parser.
15
+ function canonicalEncoding(text, raw) {
16
+ return (text.match(/"(?:\\[\s\S]|[^"\\])*"|[^\s]/g) ?? []).join('') === JSON.stringify(raw);
17
+ }
18
+
19
+ // The v1 runner hashes its construction-order object BEFORE schema parsing and
20
+ // before adding screenshot_file. Do not rename or rewrite those historical files.
21
+ export function verifyLocalEvidenceDigest({ envelope, text, digest }) {
22
+ if (!envelope || typeof text !== 'string' || !/^[a-f0-9]{64}$/.test(digest ?? '')) return false;
23
+ try {
24
+ const raw = JSON.parse(text);
25
+ if (!suppliedFieldsPreserved(raw, envelope) || !canonicalEncoding(text, raw)) return false;
26
+ } catch { return false; }
27
+ if (envelope.screenshot_file && envelope.screenshot_file !== `${digest}.screenshot.png`) return false;
28
+ if (hash(text) === digest) return true;
29
+ const original = {
30
+ schema_version: envelope.schema_version, epoch_id: envelope.epoch_id,
31
+ cell_id: envelope.cell_id, run_index: envelope.run_index, prompt: envelope.prompt,
32
+ engine_identity: envelope.engine_identity,
33
+ ...(envelope.locale_context ? { locale_context: envelope.locale_context } : {}),
34
+ provider_model_version: envelope.provider_model_version,
35
+ answer: envelope.answer, citations: envelope.citations, fan_out: envelope.fan_out,
36
+ usage: envelope.usage, cost_estimate_usd: envelope.cost_estimate_usd, at: envelope.at,
37
+ ...(envelope.provenance ? { provenance: envelope.provenance } : {}),
38
+ ...(envelope.browser_context ? { browser_context: envelope.browser_context } : {}),
39
+ ...(envelope.failure ? { failure: envelope.failure } : {}),
40
+ };
41
+ // Token counts default to zero at schema parse. Try only those lossless zero
42
+ // defaults, not arbitrary omitted evidence or alternative citation contents.
43
+ for (const omitInput of [false, true]) for (const omitOutput of [false, true]) {
44
+ if ((omitInput && envelope.usage.input_tokens !== 0) || (omitOutput && envelope.usage.output_tokens !== 0)) continue;
45
+ const usage = { ...envelope.usage };
46
+ if (omitInput) delete usage.input_tokens;
47
+ if (omitOutput) delete usage.output_tokens;
48
+ if (hash(JSON.stringify({ ...original, usage })) === digest) return true;
49
+ }
50
+ return false;
51
+ }
@@ -1,20 +1,45 @@
1
1
  import { readFile, readdir, lstat, realpath } from 'node:fs/promises';
2
2
  import { join, resolve } from 'node:path';
3
3
  import { createHash } from 'node:crypto';
4
- import { citationPanelSchema, evidenceEnvelopeSchema, targetKey } from './contract.js';
4
+ import { citationPanelSchema, evidenceEnvelopeSchema, observationRowSchema, targetKey } from './contract.js';
5
5
  import { panelEpochs } from './report.js';
6
+ import { verifyLocalEvidenceDigest } from './evidence-integrity.js';
6
7
  import { analyzeRetainedAnswer } from './match.js';
7
8
  import { parseOpportunityRecord, opportunityRecordId, ELIGIBILITY_GATES } from '../../shared/opportunity-record.js';
8
9
  const hash = value => createHash('sha256').update(value).digest('hex');
9
10
 
11
+ async function observationReferences(root, epochId, cells) {
12
+ const path = join(root, 'citations-observations.jsonl');
13
+ let text;
14
+ try {
15
+ if (!(await lstat(path)).isFile() || !(await realpath(path)).startsWith(root + '/')) throw new Error('Unsafe observation ledger');
16
+ text = await readFile(path, 'utf8');
17
+ } catch (error) { if (error.code === 'ENOENT') return new Map(); throw error; }
18
+ const references = new Map();
19
+ for (const line of text.split('\n').filter(line => line.trim())) {
20
+ let raw;
21
+ try { raw = JSON.parse(line); } catch { throw new Error('Invalid observation ledger JSON'); }
22
+ if (raw.epoch_id !== epochId || !cells.some(cell => cell.cell_id === raw.cell_id)) continue;
23
+ const parsed = observationRowSchema.safeParse(raw);
24
+ if (!parsed.success || !/^[a-f0-9]{64}$/.test(parsed.data.evidence_sha256)) throw new Error('Invalid retained observation reference');
25
+ const row = parsed.data, identity = `${row.cell_id}|${row.run_index}`;
26
+ const previous = references.get(identity);
27
+ if (previous && JSON.stringify(previous) !== JSON.stringify(row)) throw new Error('Conflicting retained observation identity');
28
+ references.set(identity, row);
29
+ }
30
+ return references;
31
+ }
32
+
10
33
  export async function freezeCitationInventory({ dir, panel: input, epochId }) {
11
34
  const panel = citationPanelSchema.parse(input);
12
35
  const epochs = await panelEpochs({ dir, panel });
13
36
  const cells = epochId ? epochs.find(rows => rows[0]?.epoch_id === epochId) : epochs[0];
14
37
  if (!cells) throw new Error('No matching panel epoch to freeze');
38
+ if (cells.some(cell => !Number.isSafeInteger(cell.n) || cell.n < 0 || !Number.isSafeInteger(cell.unknowns) || cell.unknowns < 0 || !Number.isSafeInteger(cell.n + cell.unknowns))) throw new Error('Invalid retained epoch sample counts');
15
39
  epochId = cells[0].epoch_id;
16
40
  if (!/^[A-Za-z0-9_.-]+$/.test(epochId) || ['.', '..'].includes(epochId)) throw new Error('Invalid epoch reference');
17
41
  const root = await realpath(resolve(dir)); const evidenceDir = join(root, 'citations/evidence', epochId);
42
+ const references = await observationReferences(root, epochId, cells);
18
43
  let files = [];
19
44
  try {
20
45
  if (!(await lstat(evidenceDir)).isDirectory() || !(await realpath(evidenceDir)).startsWith(root + '/')) throw new Error('Unsafe evidence directory');
@@ -28,21 +53,26 @@ export async function freezeCitationInventory({ dir, panel: input, epochId }) {
28
53
  let envelope, text;
29
54
  try { text = await readFile(path, 'utf8'); envelope = evidenceEnvelopeSchema.parse(JSON.parse(text)); }
30
55
  catch { invalid++; continue; }
31
- if (envelope.epoch_id !== epochId || !cells.some(cell => cell.cell_id === envelope.cell_id)) continue;
56
+ const cell = cells.find(cell => cell.cell_id === envelope.cell_id);
57
+ if (envelope.epoch_id !== epochId || !cell) { invalid++; continue; }
32
58
  const identity = `${envelope.cell_id}|${envelope.run_index}`;
59
+ const reference = references.get(identity), digest = file.slice(0, -5);
60
+ if (!reference || reference.evidence_sha256 !== digest || envelope.run_index >= cell.n + cell.unknowns ||
61
+ !verifyLocalEvidenceDigest({ envelope, text, digest })) { invalid++; continue; }
33
62
  if (identities.has(identity)) throw new Error('Duplicate observation identity in retained evidence'); identities.add(identity);
34
63
  const [, promptId, key] = envelope.cell_id.split('|');
35
64
  const target = panel.targets.find(target => targetKey(target) === key);
36
65
  if (!target) continue;
37
66
  const prompt = panel.prompts.find(prompt => { const id = prompt.id ?? hash(prompt.text).slice(0, 10); return id === promptId || promptId?.startsWith(`${id}@`); });
38
- const analysis = envelope.engine_identity.startsWith('google-aio:') && !envelope.provenance ? { mentioned: null, cited: null, outcome: 'unknown' } : analyzeRetainedAnswer(target, envelope);
67
+ if (!prompt || prompt.text !== envelope.prompt || envelope.engine_identity !== envelope.cell_id.split('|')[0]) { invalid++; continue; }
68
+ const analysis = reference.outcome === 'unknown' || (envelope.engine_identity.startsWith('google-aio:') && !envelope.provenance) ? { mentioned: null, cited: null, outcome: 'unknown' } : analyzeRetainedAnswer(target, envelope);
39
69
  observations.push({ cell_id: envelope.cell_id, run_index: envelope.run_index, engine_identity: envelope.engine_identity,
40
70
  prompt_id: promptId, prompt: envelope.prompt, headline_eligible: !prompt?.branded, prompt_provenance: prompt?.provenance ?? null,
41
71
  intent_cluster: prompt?.intent_cluster ?? null, target: key, observed_at: envelope.at,
42
72
  provider_model_version: envelope.provider_model_version, locale_context: envelope.locale_context ?? null,
43
73
  mentioned: analysis.mentioned, cited: analysis.cited, outcome: analysis.outcome,
44
74
  citations: analysis.outcome === 'unknown' ? [] : envelope.citations,
45
- evidence_sha256: hash(text), fixture: envelope.engine_identity.startsWith('mock:') });
75
+ evidence_sha256: hash(text), local_evidence_sha256: digest, fixture: envelope.engine_identity.startsWith('mock:') });
46
76
  }
47
77
  const expected = cells.reduce((count, cell) => count + (cell.n ?? 0) + (cell.unknowns ?? 0), 0);
48
78
  const payload = { schema_version: 1, kind: 'frozen_citation_inventory', epoch_id: epochId, frozen_at: new Date().toISOString(),
@@ -14,6 +14,7 @@ import { citationPanelSchema, engineIdentity, cellId, targetKey, panelLocaleCont
14
14
  import { classifyOutcome, analyzeRetainedAnswer } from './match.js';
15
15
  import { wilsonInterval, smoothedRate, classifyEpoch, tallyOutcomes, confidenceSequence, compareEpochs, changepointPosterior } from './stats.js';
16
16
  import { EngineError } from './adapter.js';
17
+ import { createEgressJournal } from './egress-receipts.js';
17
18
 
18
19
  const sha256 = (text) => createHash('sha256').update(text).digest('hex');
19
20
  const nowIso = () => new Date().toISOString();
@@ -50,157 +51,178 @@ async function readPriorEpochs(dir) {
50
51
  * Retriable failures receive one retry; exhausted attempts remain unknown.
51
52
  */
52
53
  export async function runEpoch(panelInput, options = {}) {
53
- const panel = citationPanelSchema.parse(panelInput);
54
- const dir = options.dir ?? '.agentlinkops';
55
- const now = options.now ?? new Date();
56
- const samples = panel.samples ?? CITATION_LIMITS.defaultSamples;
57
- const maxUsd = panel.maxUsd ?? CITATION_LIMITS.defaultMaxUsd;
58
- const delayMs = options.delayMs ?? 0; // mock/tests: 0; live callers pass pacing
54
+ // Supplied adapters transfer epoch cleanup responsibility to this runner.
55
+ const uniqueEngines = [...new Set(options.engines?.values() ?? [])];
56
+ let journal, primaryError;
57
+ try {
58
+ const panel = citationPanelSchema.parse(panelInput);
59
+ const dir = options.dir ?? '.agentlinkops';
60
+ const now = options.now ?? new Date();
61
+ const samples = panel.samples ?? CITATION_LIMITS.defaultSamples;
62
+ const maxUsd = panel.maxUsd ?? CITATION_LIMITS.defaultMaxUsd;
63
+ const delayMs = options.delayMs ?? 0; // mock/tests: 0; live callers pass pacing
59
64
 
60
- const epochId = `${now.toISOString().slice(0, 10)}-${sha256(JSON.stringify({ panel, at: now.toISOString() })).slice(0, 10)}`;
61
- const evidenceRoot = join(dir, CITATION_FILES.evidenceDir, epochId);
62
- await mkdir(evidenceRoot, { recursive: true });
65
+ const epochId = `${now.toISOString().slice(0, 10)}-${sha256(JSON.stringify({ panel, at: now.toISOString() })).slice(0, 10)}`;
66
+ const evidenceRoot = join(dir, CITATION_FILES.evidenceDir, epochId);
67
+ await mkdir(evidenceRoot, { recursive: true });
63
68
 
64
- const cells = [];
65
- for (const engineSpec of panel.engines) {
66
- const identity = engineIdentity(engineSpec);
67
- const engine = options.engines?.get(identity);
68
- const context = panelLocaleContext(panel, engineSpec);
69
- if ((context.locale !== 'en-US' || context.country !== 'US') && !['mock', 'google-aio'].includes(engineSpec.engine)) throw new EngineError('this engine has only an unverified en-US/US context; use mock or a configured supplier for locale variants');
70
- if (!engine) throw new EngineError(`no engine instance provided for ${identity}`);
71
- for (const prompt of panel.prompts) {
72
- for (const target of panel.targets) {
73
- cells.push({ identity, prompt, target, engine, context, id: cellId(identity, localizedPromptId(prompt.id ?? sha256(prompt.text).slice(0, 10), context), targetKey(target)) });
69
+ const cells = [];
70
+ for (const engineSpec of panel.engines) {
71
+ const identity = engineIdentity(engineSpec);
72
+ const engine = options.engines?.get(identity);
73
+ const context = panelLocaleContext(panel, engineSpec);
74
+ if ((context.locale !== 'en-US' || context.country !== 'US') && !['mock', 'google-aio'].includes(engineSpec.engine)) throw new EngineError('this engine has only an unverified en-US/US context; use mock or a configured supplier for locale variants');
75
+ if (!engine) throw new EngineError(`no engine instance provided for ${identity}`);
76
+ for (const prompt of panel.prompts) {
77
+ for (const target of panel.targets) {
78
+ cells.push({ identity, prompt, target, engine, context, id: cellId(identity, localizedPromptId(prompt.id ?? sha256(prompt.text).slice(0, 10), context), targetKey(target)) });
79
+ }
74
80
  }
75
81
  }
76
- }
77
82
 
78
- let spent = 0;
79
- let aborted = null;
80
- const observationRows = [];
81
- const outcomesByCell = new Map(cells.map((c) => [c.id, []]));
83
+ let spent = 0;
84
+ let aborted = null;
85
+ const observationRows = [];
86
+ const outcomesByCell = new Map(cells.map((c) => [c.id, []]));
82
87
 
83
- outer:
84
- for (const cell of cells) {
85
- for (let runIndex = 0; runIndex < samples; runIndex += 1) {
86
- const estimate = cell.engine.estimateCostUsd?.() ?? 0;
87
- if (!Number.isFinite(estimate) || estimate < 0) throw new EngineError('engine cost estimate must be a finite nonnegative number');
88
- let run = null, lastCause = null, attemptCost = 0;
89
- for (let attempt = 0; attempt < 2 && !run; attempt += 1) {
90
- // Reserve each attempt, including retries, before calling a supplier.
91
- if (spent + estimate > maxUsd) {
92
- aborted = { reason: 'budget', spent, nextEstimate: estimate, maxUsd };
93
- break;
88
+ const browserEngines = uniqueEngines.filter(engine => typeof engine.getEgressReceipts === 'function' || cells.some(cell => cell.engine === engine && cell.identity.includes(':web-own-browser')));
89
+ journal = await createEgressJournal({ dir, epochId, engines: browserEngines, identityHashes: new Map(cells.map(cell => [cell.engine, sha256(cell.identity)])) });
90
+ outer:
91
+ for (const cell of cells) {
92
+ for (let runIndex = 0; runIndex < samples; runIndex += 1) {
93
+ const estimate = cell.engine.estimateCostUsd?.() ?? 0;
94
+ if (!Number.isFinite(estimate) || estimate < 0) throw new EngineError('engine cost estimate must be a finite nonnegative number');
95
+ let run = null, lastCause = null, attemptCost = 0;
96
+ for (let attempt = 0; attempt < 2 && !run; attempt += 1) {
97
+ // Reserve each attempt, including retries, before calling a supplier.
98
+ if (spent + estimate > maxUsd) {
99
+ aborted = { reason: 'budget', spent, nextEstimate: estimate, maxUsd };
100
+ break;
101
+ }
102
+ spent += estimate;
103
+ attemptCost += estimate;
104
+ let attemptError;
105
+ try {
106
+ run = await cell.engine.run({ prompt: cell.prompt.text, runIndex, localeContext: cell.context, maxCostUsd: estimate || maxUsd - spent });
107
+ const actual = run.costEstimateUsd ?? estimate;
108
+ if (!Number.isFinite(actual) || actual < 0) throw new EngineError('engine returned an invalid cost');
109
+ spent += actual - estimate;
110
+ attemptCost += actual - estimate;
111
+ if (spent > maxUsd) aborted = { reason: 'supplier_cost_exceeded', spent, maxUsd };
112
+ } catch (cause) {
113
+ attemptError = cause;
114
+ run = null;
115
+ lastCause = cause;
116
+ // Failed or timed-out requests may already be billed. Keep the
117
+ // reservation unless the supplier reports a greater charge.
118
+ const charged = Number.isFinite(cause?.costEstimateUsd) ? Math.max(estimate, cause.costEstimateUsd) : estimate;
119
+ spent += charged - estimate;
120
+ attemptCost += charged - estimate;
121
+ if (spent > maxUsd) aborted = { reason: 'supplier_cost_exceeded', spent, maxUsd };
122
+ if (!(cause instanceof EngineError) || (!cause.retriable && !cause.unknown)) throw cause;
123
+ if (!cause.retriable || aborted) break;
124
+ if (options.err && attempt === 0) options.err(`retrying observation (${cell.id}): ${cause.message}`);
125
+ } finally {
126
+ await journal?.capture(cell.engine, { phase: 'attempt', error: attemptError, cellHash: sha256(cell.id), runIndex, attempt });
127
+ }
94
128
  }
95
- spent += estimate;
96
- attemptCost += estimate;
97
- try {
98
- run = await cell.engine.run({ prompt: cell.prompt.text, runIndex, localeContext: cell.context, maxCostUsd: estimate || maxUsd - spent });
99
- const actual = run.costEstimateUsd ?? estimate;
100
- if (!Number.isFinite(actual) || actual < 0) throw new EngineError('engine returned an invalid cost');
101
- spent += actual - estimate;
102
- attemptCost += actual - estimate;
103
- if (spent > maxUsd) aborted = { reason: 'supplier_cost_exceeded', spent, maxUsd };
104
- } catch (cause) {
105
- run = null;
106
- lastCause = cause;
107
- // Failed or timed-out requests may already be billed. Keep the
108
- // reservation unless the supplier reports a greater charge.
109
- const charged = Number.isFinite(cause?.costEstimateUsd) ? Math.max(estimate, cause.costEstimateUsd) : estimate;
110
- spent += charged - estimate;
111
- attemptCost += charged - estimate;
112
- if (spent > maxUsd) aborted = { reason: 'supplier_cost_exceeded', spent, maxUsd };
113
- if (!(cause instanceof EngineError) || (!cause.retriable && !cause.unknown)) throw cause;
114
- if (!cause.retriable || aborted) break;
115
- if (options.err && attempt === 0) options.err(`retrying observation (${cell.id}): ${cause.message}`);
129
+ if (!run) {
130
+ if (!lastCause && aborted) break outer;
131
+ if (options.err) options.err(`unknown observation (${cell.id}): ${lastCause?.message ?? 'unspecified'}`);
132
+ run = { engine: cell.engine.identity?.engine ?? 'unknown', model: cell.engine.identity?.model ?? 'unknown', answer: '', citations: [], unknown: true,
133
+ provenance: lastCause?.provenance, failure: { code: lastCause?.code ?? 'ENGINE_UNAVAILABLE', message: lastCause?.message ?? 'Engine unavailable' } };
116
134
  }
117
- }
118
- if (!run) {
119
- if (!lastCause && aborted) break outer;
120
- if (options.err) options.err(`unknown observation (${cell.id}): ${lastCause?.message ?? 'unspecified'}`);
121
- run = { engine: cell.engine.identity?.engine ?? 'unknown', model: cell.engine.identity?.model ?? 'unknown', answer: '', citations: [], unknown: true,
122
- provenance: lastCause?.provenance, failure: { code: lastCause?.code ?? 'ENGINE_UNAVAILABLE', message: lastCause?.message ?? 'Engine unavailable' } };
123
- }
124
- run.costEstimateUsd = attemptCost;
135
+ run.costEstimateUsd = attemptCost;
125
136
 
126
- const envelope = {
127
- schema_version: CITATIONS_VERSION,
128
- epoch_id: epochId, cell_id: cell.id, run_index: runIndex, prompt: cell.prompt.text,
129
- engine_identity: cell.identity,
130
- locale_context: cell.context,
131
- provider_model_version: run.providerModelVersion ?? `${run.engine}:${run.model}`,
132
- answer: run.answer, citations: run.citations, fan_out: run.fanOut ?? [],
133
- usage: run.usage ?? { input_tokens: 0, output_tokens: 0 },
134
- cost_estimate_usd: run.costEstimateUsd ?? 0,
135
- at: nowIso(),
136
- ...(run.provenance ? { provenance: run.provenance } : {}),
137
- ...(run.browserContext ? { browser_context: run.browserContext } : {}),
138
- ...(run.failure ? { failure: run.failure } : {}),
139
- };
140
- const digest = sha256(JSON.stringify(envelope));
141
- // The screenshot is a sibling of the envelope: same content-addressed
142
- // directory, named by the envelope digest it belongs to. An observation
143
- // of a rendered surface carries the pixels of that surface.
144
- if (run.screenshotPng) {
145
- const shotName = `${digest}.screenshot.png`;
146
- await writeFile(join(evidenceRoot, shotName), run.screenshotPng);
147
- envelope.screenshot_file = shotName;
148
- }
149
- await writeFile(join(evidenceRoot, `${digest}.json`), JSON.stringify(evidenceEnvelopeSchema.parse(envelope)), 'utf8');
137
+ const envelope = {
138
+ schema_version: CITATIONS_VERSION,
139
+ epoch_id: epochId, cell_id: cell.id, run_index: runIndex, prompt: cell.prompt.text,
140
+ engine_identity: cell.identity,
141
+ locale_context: cell.context,
142
+ provider_model_version: run.providerModelVersion ?? `${run.engine}:${run.model}`,
143
+ answer: run.answer, citations: run.citations, fan_out: run.fanOut ?? [],
144
+ usage: run.usage ?? { input_tokens: 0, output_tokens: 0 },
145
+ cost_estimate_usd: run.costEstimateUsd ?? 0,
146
+ at: nowIso(),
147
+ ...(run.provenance ? { provenance: run.provenance } : {}),
148
+ ...(run.browserContext ? { browser_context: run.browserContext } : {}),
149
+ ...(run.failure ? { failure: run.failure } : {}),
150
+ };
151
+ const digest = sha256(JSON.stringify(envelope));
152
+ // The screenshot is a sibling of the envelope: same content-addressed
153
+ // directory, named by the envelope digest it belongs to. An observation
154
+ // of a rendered surface carries the pixels of that surface.
155
+ if (run.screenshotPng) {
156
+ const shotName = `${digest}.screenshot.png`;
157
+ await writeFile(join(evidenceRoot, shotName), run.screenshotPng);
158
+ envelope.screenshot_file = shotName;
159
+ }
160
+ await writeFile(join(evidenceRoot, `${digest}.json`), JSON.stringify(evidenceEnvelopeSchema.parse(envelope)), 'utf8');
150
161
 
151
- const tiered = run.unknown ? { outcome: 'unknown', citedUrls: [], verify: 'unknown' } : classifyOutcome(cell.target, run, options.verifyFetch ?? null);
152
- const analysis = analyzeRetainedAnswer(cell.target, run);
153
- outcomesByCell.get(cell.id).push(tiered.outcome);
154
- observationRows.push(observationRowSchema.parse({
155
- schema_version: CITATIONS_VERSION, epoch_id: epochId, cell_id: cell.id, run_index: runIndex,
156
- outcome: tiered.outcome, brand_mentioned: analysis.mentioned, target_cited: analysis.cited, cited_urls: tiered.citedUrls, verify: tiered.verify,
157
- evidence_sha256: digest, cost_estimate_usd: run.costEstimateUsd ?? 0, at: nowIso(),
158
- }));
159
- // Duck engines need extra pacing: their anonymous rate limit (~3-4 per exit)
160
- // benefits from a cooldown between observations even with exit rotation.
161
- if (aborted) break outer;
162
- const cellDelay = cell.identity.startsWith('duck') ? Math.max(delayMs, 2000) : delayMs;
163
- if (cellDelay > 0) await new Promise((r) => setTimeout(r, cellDelay));
162
+ const tiered = run.unknown ? { outcome: 'unknown', citedUrls: [], verify: 'unknown' } : classifyOutcome(cell.target, run, options.verifyFetch ?? null);
163
+ const analysis = analyzeRetainedAnswer(cell.target, run);
164
+ outcomesByCell.get(cell.id).push(tiered.outcome);
165
+ observationRows.push(observationRowSchema.parse({
166
+ schema_version: CITATIONS_VERSION, epoch_id: epochId, cell_id: cell.id, run_index: runIndex,
167
+ outcome: tiered.outcome, brand_mentioned: analysis.mentioned, target_cited: analysis.cited, cited_urls: tiered.citedUrls, verify: tiered.verify,
168
+ evidence_sha256: digest, cost_estimate_usd: run.costEstimateUsd ?? 0, at: nowIso(),
169
+ }));
170
+ // Duck engines need extra pacing: their anonymous rate limit (~3-4 per exit)
171
+ // benefits from a cooldown between observations even with exit rotation.
172
+ if (aborted) break outer;
173
+ const cellDelay = cell.identity.startsWith('duck') ? Math.max(delayMs, 2000) : delayMs;
174
+ if (cellDelay > 0) await new Promise((r) => setTimeout(r, cellDelay));
175
+ }
164
176
  }
165
- }
166
177
 
167
- const prior = await readPriorEpochs(dir);
168
- const epochRows = [];
169
- for (const cell of cells) {
170
- const outcomes = outcomesByCell.get(cell.id);
171
- const { k, n, unknowns, mentioned } = tallyOutcomes(outcomes);
172
- const ci = wilsonInterval(k, n);
173
- const baselineRow = prior.filter((r) => r.cell_id === cell.id && r.n >= CITATION_LIMITS.minSamplesForInterpretation).at(-1);
174
- const baseline = baselineRow ? { k: baselineRow.k, n: baselineRow.n } : null;
175
- const priorCount = prior.filter(row => row.cell_id === cell.id).length;
176
- const pairAlpha = priorCount > 0 ? 0.05 / (priorCount * (priorCount + 1)) : 0.05;
177
- const { run_length_posterior, ...changepoint } = changepointPosterior(outcomes);
178
- const statistics = { scope: 'fixed-epoch-supplemental',
179
- confidence_sequence: confidenceSequence(outcomes, { alpha: 0.05, comparisons: cells.length }).at(-1) ?? null,
180
- epoch_comparison: baseline ? compareEpochs({ k, n }, baseline, { alpha: pairAlpha, comparisons: cells.length }) : null,
181
- changepoint,
182
- };
183
- const row = epochRowSchema.parse({
184
- schema_version: CITATIONS_VERSION, epoch_id: epochId, cell_id: cell.id,
185
- k, n, unknowns, mentioned,
186
- rate: smoothedRate(k, n), ci_low: ci?.[0] ?? null, ci_high: ci?.[1] ?? null,
187
- classification: classifyEpoch({ k, n }, baseline, CITATION_LIMITS), statistics,
188
- baseline: baselineRow ? { epoch_id: baselineRow.epoch_id, rate: baselineRow.rate, ci_low: baselineRow.ci_low, ci_high: baselineRow.ci_high } : undefined,
189
- spent_estimate_usd: Number(spent.toFixed(6)), at: nowIso(),
190
- });
191
- epochRows.push(row);
192
- // Incremental flush: every completed cell's row is on disk immediately, so
193
- // a killed epoch loses nothing — the lesson of the lost 3x3 tallies.
194
- await writeFile(join(dir, CITATION_FILES.epochs), JSON.stringify(row) + '\n', { flag: 'a' });
195
- }
178
+ const prior = await readPriorEpochs(dir);
179
+ const epochRows = [];
180
+ for (const cell of cells) {
181
+ const outcomes = outcomesByCell.get(cell.id);
182
+ const { k, n, unknowns, mentioned } = tallyOutcomes(outcomes);
183
+ const ci = wilsonInterval(k, n);
184
+ const baselineRow = prior.filter((r) => r.cell_id === cell.id && r.n >= CITATION_LIMITS.minSamplesForInterpretation).at(-1);
185
+ const baseline = baselineRow ? { k: baselineRow.k, n: baselineRow.n } : null;
186
+ const priorCount = prior.filter(row => row.cell_id === cell.id).length;
187
+ const pairAlpha = priorCount > 0 ? 0.05 / (priorCount * (priorCount + 1)) : 0.05;
188
+ const { run_length_posterior, ...changepoint } = changepointPosterior(outcomes);
189
+ const statistics = { scope: 'fixed-epoch-supplemental',
190
+ confidence_sequence: confidenceSequence(outcomes, { alpha: 0.05, comparisons: cells.length }).at(-1) ?? null,
191
+ epoch_comparison: baseline ? compareEpochs({ k, n }, baseline, { alpha: pairAlpha, comparisons: cells.length }) : null,
192
+ changepoint,
193
+ };
194
+ const row = epochRowSchema.parse({
195
+ schema_version: CITATIONS_VERSION, epoch_id: epochId, cell_id: cell.id,
196
+ k, n, unknowns, mentioned,
197
+ rate: smoothedRate(k, n), ci_low: ci?.[0] ?? null, ci_high: ci?.[1] ?? null,
198
+ classification: classifyEpoch({ k, n }, baseline, CITATION_LIMITS), statistics,
199
+ baseline: baselineRow ? { epoch_id: baselineRow.epoch_id, rate: baselineRow.rate, ci_low: baselineRow.ci_low, ci_high: baselineRow.ci_high } : undefined,
200
+ spent_estimate_usd: Number(spent.toFixed(6)), at: nowIso(),
201
+ });
202
+ epochRows.push(row);
203
+ // Incremental flush: every completed cell's row is on disk immediately, so
204
+ // a killed epoch loses nothing — the lesson of the lost 3x3 tallies.
205
+ await writeFile(join(dir, CITATION_FILES.epochs), JSON.stringify(row) + '\n', { flag: 'a' });
206
+ }
196
207
 
197
- await writeFile(join(dir, CITATION_FILES.observations), observationRows.map((r) => JSON.stringify(r)).join('\n') + (observationRows.length ? '\n' : ''), { flag: 'a' });
208
+ await writeFile(join(dir, CITATION_FILES.observations), observationRows.map((r) => JSON.stringify(r)).join('\n') + (observationRows.length ? '\n' : ''), { flag: 'a' });
198
209
 
199
- // Browser engines hold a browser and maybe an Xvfb display for the whole
200
- // epoch; release both once the tallies are written.
201
- for (const engine of new Set(cells.map((c) => c.engine))) {
202
- try { await engine.close?.(); } catch { /* cleanup is best-effort */ }
210
+ return { epochId, cells: epochRows, spentEstimateUsd: Number(spent.toFixed(6)), aborted, samples, maxUsd };
211
+ } catch (error) {
212
+ primaryError = error;
213
+ throw error;
214
+ } finally {
215
+ let persistenceError, cleanupError;
216
+ for (const engine of uniqueEngines) {
217
+ let error;
218
+ try { await engine.close?.(); } catch (cause) { error = cause; cleanupError ??= cause; }
219
+ try { await journal?.capture(engine, { phase: 'cleanup', error }); }
220
+ catch (cause) { persistenceError ??= cause; }
221
+ }
222
+ try { await journal?.finish({ failed: Boolean(primaryError || persistenceError || cleanupError), cleanupUncertain: Boolean(cleanupError), evidenceComplete: !persistenceError && !String(primaryError?.code ?? '').startsWith('EGRESS_') }); }
223
+ catch (cause) { persistenceError ??= cause; }
224
+ try { await journal?.close(); } catch (cause) { persistenceError ??= cause; }
225
+ if (persistenceError) throw persistenceError;
226
+ if (!primaryError && cleanupError && journal) throw Object.assign(new EngineError('Browser cleanup could not be confirmed; inspect the egress journal.'), { code: 'BROWSER_CLEANUP_UNCONFIRMED' });
203
227
  }
204
-
205
- return { epochId, cells: epochRows, spentEstimateUsd: Number(spent.toFixed(6)), aborted, samples, maxUsd };
206
228
  }
@@ -8,6 +8,7 @@
8
8
  // truth; the hosted view is a projection.
9
9
  import { readFile } from 'node:fs/promises';
10
10
  import { join, resolve } from 'node:path';
11
+ import { verifyLocalEvidenceDigest } from './evidence-integrity.js';
11
12
 
12
13
  export async function syncEpochs({ dir, api, workspaceId, projectId }) {
13
14
  if (!api || !workspaceId || !projectId) {
@@ -83,7 +84,7 @@ export async function syncEvidence({ dir, api, projectId, cliVersion }) {
83
84
  try { text = await readFile(join(root, 'citations-observations.jsonl'), 'utf8'); }
84
85
  catch (error) { if (error.code !== 'ENOENT') throw error; return { synced: 0, total: 0, skipped: 0, errors: 0, reasons: {} }; }
85
86
  const rows = text.split('\n').filter(line => line.trim()).map(line => JSON.parse(line));
86
- const records = []; const reasons = {}; const seen = new Set();
87
+ const records = []; const reasons = {}; const seen = new Map();
87
88
  const skip = reason => { reasons[reason] = (reasons[reason] ?? 0) + 1; };
88
89
  const hash = bytes => createHash('sha256').update(bytes).digest('hex');
89
90
  async function safeFile(path) {
@@ -95,13 +96,19 @@ export async function syncEvidence({ dir, api, projectId, cliVersion }) {
95
96
  for (const row of rows) {
96
97
  if (!/^[A-Za-z0-9_.-]+$/.test(row.epoch_id ?? '') || ['.', '..'].includes(row.epoch_id) || !/^[a-f0-9]{64}$/.test(row.evidence_sha256 ?? '')) throw new Error('Malformed local evidence reference');
97
98
  const path = join(root, 'citations/evidence', row.epoch_id, `${row.evidence_sha256}.json`);
98
- if (seen.has(path)) continue; seen.add(path);
99
+ const identity = JSON.stringify([row.epoch_id, row.cell_id, row.run_index]);
100
+ if (seen.has(path)) {
101
+ if (seen.get(path) !== identity) throw new Error('Conflicting local evidence reference');
102
+ continue;
103
+ }
104
+ seen.set(path, identity);
99
105
  let file;
100
106
  try { file = await safeFile(path); } catch (error) { if (error.code === 'ENOENT') { skip('missing_envelope'); continue; } throw error; }
101
107
  if ((await lstat(file)).size > limits.envelopeBytes) throw new Error('Evidence envelope exceeds the upload limit');
102
108
  const envelopeJson = await readFile(file, 'utf8');
103
109
  const envelope = evidenceEnvelopeSchema.parse(JSON.parse(envelopeJson));
104
110
  if (envelope.epoch_id !== row.epoch_id || envelope.cell_id !== row.cell_id || envelope.run_index !== row.run_index) throw new Error('Evidence identity does not match its observation');
111
+ if (!verifyLocalEvidenceDigest({ envelope, text: envelopeJson, digest: row.evidence_sha256 })) throw new Error('Evidence digest does not match its retained observation');
105
112
  if (envelope.engine_identity.startsWith('google-aio:') && !envelope.provenance) { skip('legacy_aio_without_provenance'); continue; }
106
113
  const record = { envelopeJson, sha256: hash(envelopeJson), cliVersion, localEvidenceSha256: row.evidence_sha256 };
107
114
  if (envelope.screenshot_file) {