@jinn-network/client 0.1.9 → 0.2.0

package/dist/eval/paired.js

@@ -1,93 +1,3 @@
-/**
- * Paired (matched-design) comparison for the held-out exam.
- *
- * The exam scores the SAME slate against the before- and after-checkpoints, so
- * the two result sets are MATCHED, not independent. The marginal test in
- * `wilson.ts` (`compareRates`) compares two INDEPENDENT Wilson intervals and
- * calls a delta trustworthy only when they are disjoint. On a matched design
- * that is statistically wasteful: it carries the full between-instance
- * difficulty variance ("this bug is just hard") in BOTH intervals, which swamps
- * a real, consistent within-instance improvement. Example: a learner that flips
- * 3 hard instances fail→pass and regresses none moves 4/10→7/10 — two Wilson
- * intervals ([17,69]% vs [40,89]%) overlap → "within-noise", even though every
- * observed change was an improvement.
- *
- * McNemar's test is the textbook test for matched binary data: it looks ONLY at
- * the discordant pairs (instances that flipped) and asks whether the flips are
- * asymmetric beyond chance. This is NOT a weaker bar — it is the CORRECT test
- * for the design, and it still gates the claim on significance (exact two-sided
- * p < alpha) AND on the improvement direction (more fail→pass than pass→fail).
- * It is reported ALONGSIDE the conservative marginal verdict, never replacing it
- * — so the exam is strengthened, never weakened (DR-2026-06-02-b §2a).
- *
- * R=1 (one run per instance) ships here: McNemar exact on per-instance flips.
- * R>1 (per-instance pass RATES → Wilcoxon signed-rank / paired bootstrap) layers
- * on once `eval_results` supports appended runs (DR-2026-06-02-b §2b).
- */
-/**
- * Exact two-sided McNemar p-value for discordant counts `b` (improvements) and
- * `c` (regressions). Under H0 each discordant pair is an independent fair coin,
- * so the count of one direction is Binomial(n=b+c, 0.5); the two-sided p is
- * `2 * P(X <= min(b,c))`, capped at 1. No `b==c==0` division (returns 1: no
- * evidence). Computed iteratively (term ratio) so it is exact and overflow-free
- * for any slate size.
- */
-export function mcnemarExact(b, c) {
-    const n = b + c;
-    if (n === 0)
-        return 1;
-    const k = Math.min(b, c);
-    // sum_{i=0}^{k} C(n,i) * 0.5^n, via prob_0 = 0.5^n, prob_i = prob_{i-1}*(n-i+1)/i.
-    let term = Math.pow(0.5, n); // i = 0
-    let cdf = term;
-    for (let i = 1; i <= k; i++) {
-        term = (term * (n - i + 1)) / i;
-        cdf += term;
-    }
-    return Math.min(1, 2 * cdf);
-}
-/**
- * Paired McNemar comparison of two matched per-instance result sets. An instance
- * contributes a pair ONLY when it is scorable (passed !== null, not unscorable)
- * in BOTH arms — an unscorable/missing side is excluded (honest: a disk-skip or
- * harness failure is never coerced into a flip). `alpha` defaults to 0.05.
- */
-export function comparePaired(before, after, opts = {}) {
-    const alpha = opts.alpha ?? 0.05;
-    const scorable = (r) => !r.unscorable && r.passed !== null;
-    const beforeById = new Map(before.map((r) => [r.instance_id, r]));
-    let improved = 0;
-    let regressed = 0;
-    let concordantPass = 0;
-    let concordantFail = 0;
-    let pairs = 0;
-    let excluded = 0;
-    const seen = new Set();
-    for (const a of after) {
-        seen.add(a.instance_id);
-        const b = beforeById.get(a.instance_id);
-        if (!b || !scorable(b) || !scorable(a)) {
-            excluded++;
-            continue;
-        }
-        pairs++;
-        const wasPass = b.passed === true;
-        const nowPass = a.passed === true;
-        if (!wasPass && nowPass)
-            improved++;
-        else if (wasPass && !nowPass)
-            regressed++;
-        else if (wasPass && nowPass)
-            concordantPass++;
-        else
-            concordantFail++;
-    }
-    // before-only instances (absent from after) are excluded too.
-    for (const b of before)
-        if (!seen.has(b.instance_id))
-            excluded++;
-    const pValue = mcnemarExact(improved, regressed);
-    const verdict = pValue < alpha && improved > regressed ? 'trustworthy' : 'within-noise';
-    return { pairs, improved, regressed, concordantPass, concordantFail, excluded, pValue, verdict };
-}
+/** Compatibility export for the core-owned matched-evaluation helpers. */
+export { comparePaired, mcnemarExact, } from '../vendor/@jinn-network/core/dist/index.js';
 //# sourceMappingURL=paired.js.map

package/dist/eval/paired.js.map

@@ -1 +1 @@
-{"version":3,"file":"paired.js","sourceRoot":"","sources":["../../src/eval/paired.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AA6BH;;;;;;;GAOG;AACH,MAAM,UAAU,YAAY,CAAC,CAAS,EAAE,CAAS;IAC/C,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IAChB,IAAI,CAAC,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IACtB,MAAM,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IACzB,mFAAmF;IACnF,IAAI,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,QAAQ;IACrC,IAAI,GAAG,GAAG,IAAI,CAAC;IACf,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC5B,IAAI,GAAG,CAAC,IAAI,GAAG,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;QAChC,GAAG,IAAI,IAAI,CAAC;IACd,CAAC;IACD,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC;AAC9B,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,aAAa,CAC3B,MAA8B,EAC9B,KAA6B,EAC7B,OAA2B,EAAE;IAE7B,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC;IACjC,MAAM,QAAQ,GAAG,CAAC,CAAc,EAAW,EAAE,CAAC,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,MAAM,KAAK,IAAI,CAAC;IACjF,MAAM,UAAU,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAElE,IAAI,QAAQ,GAAG,CAAC,CAAC;IACjB,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,cAAc,GAAG,CAAC,CAAC;IACvB,IAAI,cAAc,GAAG,CAAC,CAAC;IACvB,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,IAAI,QAAQ,GAAG,CAAC,CAAC;IAEjB,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;IAC/B,KAAK,MAAM,CAAC,IAAI,KAAK,EAAE,CAAC;QACtB,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC;QACxB,MAAM,CAAC,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC;QACxC,IAAI,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC;YACvC,QAAQ,EAAE,CAAC;YACX,SAAS;QACX,CAAC;QACD,KAAK,EAAE,CAAC;QACR,MAAM,OAAO,GAAG,CAAC,CAAC,MAAM,KAAK,IAAI,CAAC;QAClC,MAAM,OAAO,GAAG,CAAC,CAAC,MAAM,KAAK,IAAI,CAAC;QAClC,IAAI,CAAC,OAAO,IAAI,OAAO;YAAE,QAAQ,EAAE,CAAC;aAC/B,IAAI,OAAO,IAAI,CAAC,OAAO;YAAE,SAAS,EAAE,CAAC;aACrC,IAAI,OAAO,IAAI,OAAO;YAAE,cAAc,EAAE,CAAC;;YACzC,cAAc,EAAE,CAAC;IACxB,CAAC;IACD,8DAA8D;IAC9D,KAAK,MAAM,CAAC,IAAI,MAAM;QAAE,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,WAAW,CAAC;YAAE,QAAQ,EAAE,CAAC;IAEjE,MAAM,MAAM,GAAG,YAAY,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC;IACjD,MAAM,OAAO,GACX,MAAM,GAAG,KAAK,IAAI,QAAQ,GAAG,SAAS,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,cAAc,CAAC;IAE1E,OAAO,EAAE,KAAK,EAAE,QAAQ,EAAE,SAAS,EAAE,cAAc,EAAE,cAAc,EAAE,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;AACnG,CAAC"}
+{"version":3,"file":"paired.js","sourceRoot":"","sources":["../../src/eval/paired.ts"],"names":[],"mappings":"AAAA,0EAA0E;AAC1E,OAAO,EACL,aAAa,EACb,YAAY,GAIb,MAAM,oBAAoB,CAAC"}

package/dist/eval/resolve-slate-tasks.js

@@ -14,7 +14,7 @@
  * propagates loudly (a missing slate instance is a hard error, never a silent
  * drop).
  */
-import { SweRebenchV2TaskSchema } from '../vendor/@jinn-network/sdk/dist/solvernets/swe-rebench-v2.js';
+import { SweRebenchV2LanguageSchema } from '../vendor/@jinn-network/sdk/dist/solvernets/swe-rebench-v2.js';
 export async function resolveSlateTasks(args) {
     const tasks = [...args.poolTasks].sort((a, b) => a.instance_id.localeCompare(b.instance_id));
     const out = [];
@@ -38,7 +38,7 @@ export async function resolveSlateTasks(args) {
  * way the generator does, so the object stays schema-valid.
  */
 function buildTask(poolTask, hf_dataset, hf_split, row) {
-    const language = SweRebenchV2TaskSchema.shape.language.safeParse(poolTask.language);
+    const language = SweRebenchV2LanguageSchema.safeParse(poolTask.language);
     return {
         schemaVersion: 'swe-rebench-v2.v1',
         instance_id: poolTask.instance_id,

package/dist/eval/resolve-slate-tasks.js.map

@@ -1 +1 @@
-{"version":3,"file":"resolve-slate-tasks.js","sourceRoot":"","sources":["../../src/eval/resolve-slate-tasks.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH,OAAO,EAAE,sBAAsB,EAAyB,MAAM,6CAA6C,CAAC;AAS5G,MAAM,CAAC,KAAK,UAAU,iBAAiB,CAAC,IAWvC;IACC,MAAM,KAAK,GAAG,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,CAAC,aAAa,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC;IAC7F,MAAM,GAAG,GAAwB,EAAE,CAAC;IACpC,KAAK,MAAM,QAAQ,IAAI,KAAK,EAAE,CAAC;QAC7B,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,YAAY,CAAC;YAC1C,UAAU,EAAE,IAAI,CAAC,UAAU;YAC3B,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,WAAW,EAAE,QAAQ,CAAC,WAAW;SAClC,CAAC,CAAC;QACH,GAAG,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,SAAS,CAAC,QAAQ,EAAE,IAAI,CAAC,UAAU,EAAE,IAAI,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC;IACpF,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAED;;;;;;;;GAQG;AACH,SAAS,SAAS,CAChB,QAAkB,EAClB,UAAkB,EAClB,QAAgB,EAChB,GAAU;IAEV,MAAM,QAAQ,GAAG,sBAAsB,CAAC,KAAK,CAAC,QAAQ,CAAC,SAAS,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACpF,OAAO;QACL,aAAa,EAAE,mBAAmB;QAClC,WAAW,EAAE,QAAQ,CAAC,WAAW;QACjC,IAAI,EAAE,GAAG,CAAC,IAAI;QACd,WAAW,EAAE,QAAQ,CAAC,WAAW,IAAI,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC;QACnD,QAAQ,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,QAAQ;QACrD,iBAAiB,EAAE,QAAQ,CAAC,iBAAiB,IAAI,EAAE;QACnD,SAAS,EAAE,QAAQ,CAAC,SAAS,IAAI,EAAE;QACnC,UAAU;QACV,QAAQ;QACR,aAAa,EAAE,CAAC;QAChB,WAAW,EAAE,QAAQ,CAAC,OAAO,CAAC,GAAG,EAAE,GAAG,CAAC;KACxC,CAAC;AACJ,CAAC"}
+{"version":3,"file":"resolve-slate-tasks.js","sourceRoot":"","sources":["../../src/eval/resolve-slate-tasks.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH,OAAO,EAAE,0BAA0B,EAAiD,MAAM,6CAA6C,CAAC;AASxI,MAAM,CAAC,KAAK,UAAU,iBAAiB,CAAC,IAWvC;IACC,MAAM,KAAK,GAAG,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,CAAC,aAAa,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC;IAC7F,MAAM,GAAG,GAAwB,EAAE,CAAC;IACpC,KAAK,MAAM,QAAQ,IAAI,KAAK,EAAE,CAAC;QAC7B,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,YAAY,CAAC;YAC1C,UAAU,EAAE,IAAI,CAAC,UAAU;YAC3B,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,WAAW,EAAE,QAAQ,CAAC,WAAW;SAClC,CAAC,CAAC;QACH,GAAG,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,SAAS,CAAC,QAAQ,EAAE,IAAI,CAAC,UAAU,EAAE,IAAI,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC;IACpF,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAED;;;;;;;;GAQG;AACH,SAAS,SAAS,CAChB,QAAkB,EAClB,UAAkB,EAClB,QAAgB,EAChB,GAAU;IAEV,MAAM,QAAQ,GAAG,0BAA0B,CAAC,SAAS,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACzE,OAAO;QACL,aAAa,EAAE,mBAAmB;QAClC,WAAW,EAAE,QAAQ,CAAC,WAAW;QACjC,IAAI,EAAE,GAAG,CAAC,IAAI;QACd,WAAW,EAAE,QAAQ,CAAC,WAAW,IAAI,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC;QACnD,QAAQ,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,QAAQ;QACrD,iBAAiB,EAAE,QAAQ,CAAC,iBAAiB,IAAI,EAAE;QACnD,SAAS,EAAE,QAAQ,CAAC,SAAS,IAAI,EAAE;QACnC,UAAU;QACV,QAAQ;QACR,aAAa,EAAE,CAAC;QAChB,WAAW,EAAE,QAAQ,CAAC,OAAO,CAAC,GAAG,EAAE,GAAG,CAAC;KACxC,CAAC;AACJ,CAAC"}

package/dist/eval/screen-discovery.d.ts

@@ -1,5 +1,5 @@
 /**
- * Held-out screening's authoritative "already trained-on" exclusion source (#986).
+ * Held-out screening's "already trained-on" exclusion boundary (#986).
  *
  * Returns the set of swe-rebench-v2 `instance_id`s that have been ATTEMPTED on
  * the network for a SolverNet — any verdict envelope, passed OR failed,
@@ -7,13 +7,10 @@
  * learner trained on it; holding it out later would make a trained-checkpoint
  * pass count as memorization, not generalization.
  *
- * Why the indexer and not the local generator-state: the local ledger only
- * reflects THIS box's posting and can be stale (a different active generator —
- * e.g. a hosted operator — posts to its own ledger). The indexer's
- * `verdictEnvelopeMeta` is the cross-operator, current record. (The on-chain
- * task/attempt tables carry no instance_id; only the indexer's IPFS enrichment
- * resolves it — same backing as `DiscoveryAPI.getInstanceSuccessCounts`, minus
- * the `actualPassed: true` filter so failed attempts count too.)
+ * The local ledger only reflects THIS box and can be stale, while the current
+ * indexer `verdictEnvelopeMeta` is permissionless and shape-parsed. Neither is
+ * authoritative enough for an exam exclusion. This function therefore aborts
+ * until the indexer materializes canonical bridge-grade verdict projections.
  *
  * Throws on indexer failure — callers MUST abort rather than screen against an
  * unknown attempted set, because a missing exclusion can silently contaminate

package/dist/eval/screen-discovery.js

@@ -1,5 +1,5 @@
 /**
- * Held-out screening's authoritative "already trained-on" exclusion source (#986).
+ * Held-out screening's "already trained-on" exclusion boundary (#986).
  *
  * Returns the set of swe-rebench-v2 `instance_id`s that have been ATTEMPTED on
  * the network for a SolverNet — any verdict envelope, passed OR failed,
@@ -7,65 +7,23 @@
  * learner trained on it; holding it out later would make a trained-checkpoint
  * pass count as memorization, not generalization.
  *
- * Why the indexer and not the local generator-state: the local ledger only
- * reflects THIS box's posting and can be stale (a different active generator —
- * e.g. a hosted operator — posts to its own ledger). The indexer's
- * `verdictEnvelopeMeta` is the cross-operator, current record. (The on-chain
- * task/attempt tables carry no instance_id; only the indexer's IPFS enrichment
- * resolves it — same backing as `DiscoveryAPI.getInstanceSuccessCounts`, minus
- * the `actualPassed: true` filter so failed attempts count too.)
+ * The local ledger only reflects THIS box and can be stale, while the current
+ * indexer `verdictEnvelopeMeta` is permissionless and shape-parsed. Neither is
+ * authoritative enough for an exam exclusion. This function therefore aborts
+ * until the indexer materializes canonical bridge-grade verdict projections.
  *
  * Throws on indexer failure — callers MUST abort rather than screen against an
  * unknown attempted set, because a missing exclusion can silently contaminate
  * the exam (the whole point of held-out discipline).
  */
-const ATTEMPTED_QUERY = `
-query InstanceAttempted($cid: String!, $limit: Int!, $after: String) {
-  verdictEnvelopeMetas(
-    where: {
-      solverNetManifestCid: $cid,
-      solverType_starts_with: "swe-rebench-v2",
-      enrichmentStatus: "ok",
-      instanceId_not: ""
-    },
-    limit: $limit,
-    after: $after,
-    orderBy: "enrichedAtBlock",
-    orderDirection: "asc"
-  ) {
-    items { instanceId }
-    pageInfo { hasNextPage endCursor }
-  }
-}`;
 export async function fetchAttemptedInstanceIds(discoveryUrl, manifestCid, fetchImpl = fetch) {
-    const gqlUrl = discoveryUrl.endsWith('/graphql') ? discoveryUrl : `${discoveryUrl.replace(/\/$/, '')}/graphql`;
-    const ids = new Set();
-    let cursor = null;
-    const MAX_PAGES = 20;
-    const PAGE_LIMIT = 1000;
-    for (let page = 0; page < MAX_PAGES; page++) {
-        const res = await fetchImpl(gqlUrl, {
-            method: 'POST',
-            headers: { 'content-type': 'application/json' },
-            body: JSON.stringify({ query: ATTEMPTED_QUERY, variables: { cid: manifestCid, limit: PAGE_LIMIT, after: cursor } }),
-        });
-        if (!res.ok) {
-            throw new Error(`held-out screening: indexer attempted-ids query failed (HTTP ${res.status}) at ${gqlUrl}`);
-        }
-        const json = (await res.json());
-        if (json.errors) {
-            throw new Error(`held-out screening: indexer attempted-ids query errors: ${JSON.stringify(json.errors).slice(0, 300)}`);
-        }
-        const conn = json.data?.verdictEnvelopeMetas;
-        for (const item of conn?.items ?? []) {
-            if (item.instanceId)
-                ids.add(item.instanceId);
-        }
-        const pageInfo = conn?.pageInfo;
-        if (!pageInfo?.hasNextPage || !pageInfo.endCursor)
-            break;
-        cursor = pageInfo.endCursor;
-    }
-    return ids;
+    void discoveryUrl;
+    void manifestCid;
+    void fetchImpl;
+    // A permissionless verdict projection can name any victim manifest and
+    // instance, which would poison the held-out exclusion set. Screening must
+    // abort—not assume an empty set—until the indexer exposes canonical
+    // publisher-Safe/signature/hash/original-task-bound verdict rows.
+    throw new Error('held-out screening: authenticated attempted-instance projection unavailable');
 }
 //# sourceMappingURL=screen-discovery.js.map

package/dist/eval/screen-discovery.js.map

@@ -1 +1 @@
-{"version":3,"file":"screen-discovery.js","sourceRoot":"","sources":["../../src/eval/screen-discovery.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAEH,MAAM,eAAe,GAAG;;;;;;;;;;;;;;;;;EAiBtB,CAAC;AAYH,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAC7C,YAAoB,EACpB,WAAmB,EACnB,YAA0B,KAAK;IAE/B,MAAM,MAAM,GAAG,YAAY,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,GAAG,YAAY,CAAC,OAAO,CAAC,KAAK,EAAE,EAAE,CAAC,UAAU,CAAC;IAC/G,MAAM,GAAG,GAAG,IAAI,GAAG,EAAU,CAAC;IAC9B,IAAI,MAAM,GAAkB,IAAI,CAAC;IACjC,MAAM,SAAS,GAAG,EAAE,CAAC;IACrB,MAAM,UAAU,GAAG,IAAI,CAAC;IAExB,KAAK,IAAI,IAAI,GAAG,CAAC,EAAE,IAAI,GAAG,SAAS,EAAE,IAAI,EAAE,EAAE,CAAC;QAC5C,MAAM,GAAG,GAAG,MAAM,SAAS,CAAC,MAAM,EAAE;YAClC,MAAM,EAAE,MAAM;YACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;YAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,eAAe,EAAE,SAAS,EAAE,EAAE,GAAG,EAAE,WAAW,EAAE,KAAK,EAAE,UAAU,EAAE,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC;SACpH,CAAC,CAAC;QACH,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;YACZ,MAAM,IAAI,KAAK,CAAC,gEAAgE,GAAG,CAAC,MAAM,QAAQ,MAAM,EAAE,CAAC,CAAC;QAC9G,CAAC;QACD,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAAkB,CAAC;QACjD,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;YAChB,MAAM,IAAI,KAAK,CAAC,2DAA2D,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;QAC1H,CAAC;QACD,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,oBAAoB,CAAC;QAC7C,KAAK,MAAM,IAAI,IAAI,IAAI,EAAE,KAAK,IAAI,EAAE,EAAE,CAAC;YACrC,IAAI,IAAI,CAAC,UAAU;gBAAE,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;QAChD,CAAC;QACD,MAAM,QAAQ,GAAG,IAAI,EAAE,QAAQ,CAAC;QAChC,IAAI,CAAC,QAAQ,EAAE,WAAW,IAAI,CAAC,QAAQ,CAAC,SAAS;YAAE,MAAM;QACzD,MAAM,GAAG,QAAQ,CAAC,SAAS,CAAC;IAC9B,CAAC;IAED,OAAO,GAAG,CAAC;AACb,CAAC"}
+{"version":3,"file":"screen-discovery.js","sourceRoot":"","sources":["../../src/eval/screen-discovery.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAEH,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAC7C,YAAoB,EACpB,WAAmB,EACnB,YAA0B,KAAK;IAE/B,KAAK,YAAY,CAAC;IAClB,KAAK,WAAW,CAAC;IACjB,KAAK,SAAS,CAAC;IACf,uEAAuE;IACvE,0EAA0E;IAC1E,oEAAoE;IACpE,kEAAkE;IAClE,MAAM,IAAI,KAAK,CACb,6EAA6E,CAC9E,CAAC;AACJ,CAAC"}

package/dist/events/types.d.ts

@@ -5,7 +5,7 @@
  * markers) and consumed by /v1/events SSE + /v1/events/recent JSON endpoints
  * served to the operator SPA. See docs/superpowers/specs/2026-05-01-operator-local-app-design.md.
  */
-import { z } from 'zod';
+import { z } from 'zod/v3';
 export declare const StructuredEventKindSchema: z.ZodEnum<["intent", "reward", "fleet", "system", "error", "log"]>;
 export type StructuredEventKind = z.infer<typeof StructuredEventKindSchema>;
 export declare const StructuredEventSchema: z.ZodObject<{

package/dist/events/types.js

@@ -5,7 +5,7 @@
  * markers) and consumed by /v1/events SSE + /v1/events/recent JSON endpoints
  * served to the operator SPA. See docs/superpowers/specs/2026-05-01-operator-local-app-design.md.
  */
-import { z } from 'zod';
+import { z } from 'zod/v3';
 export const StructuredEventKindSchema = z.enum([
     'intent',
     'reward',

package/dist/events/types.js.map

@@ -1 +1 @@
-{"version":3,"file":"types.js","sourceRoot":"","sources":["../../src/events/types.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AACH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,MAAM,CAAC,MAAM,yBAAyB,GAAG,CAAC,CAAC,IAAI,CAAC;IAC9C,QAAQ;IACR,QAAQ;IACR,OAAO;IACP,QAAQ;IACR,OAAO;IACP,KAAK;CACN,CAAC,CAAC;AAGH,MAAM,CAAC,MAAM,qBAAqB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC5C,aAAa,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC;IAC3B,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE;IACd,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE;IACd,IAAI,EAAE,yBAAyB;IAC/B,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE;IACnB,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAChC,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC7B,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAChC,OAAO,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,QAAQ,EAAE;CAC1C,CAAC,CAAC"}
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../../src/events/types.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AACH,OAAO,EAAE,CAAC,EAAE,MAAM,QAAQ,CAAC;AAE3B,MAAM,CAAC,MAAM,yBAAyB,GAAG,CAAC,CAAC,IAAI,CAAC;IAC9C,QAAQ;IACR,QAAQ;IACR,OAAO;IACP,QAAQ;IACR,OAAO;IACP,KAAK;CACN,CAAC,CAAC;AAGH,MAAM,CAAC,MAAM,qBAAqB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC5C,aAAa,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC;IAC3B,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE;IACd,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE;IACd,IAAI,EAAE,yBAAyB;IAC/B,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE;IACnB,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAChC,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC7B,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAChC,OAAO,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,QAAQ,EAAE;CAC1C,CAAC,CAAC"}

package/dist/harnesses/auth-source.d.ts

@@ -0,0 +1,51 @@
+import type { Harness } from './types.js';
+/**
+ * Tilde-abbreviate an absolute path for operator display when it sits under the
+ * real home dir; otherwise return it unchanged. Used by harness `getAuthSource`
+ * implementations so the dashboard shows `~/.hermes/.env` rather than a full
+ * `/Users/...` path.
+ */
+export declare function displayPath(absolutePath: string): string;
+export type HarnessAuthSource = {
+    sourceKind: 'file';
+    sourcePath: string;
+    absolutePath: string;
+    envKey: string;
+    docAnchor: string;
+    /**
+     * When true the credential file is structured (e.g. codex `auth.json`)
+     * rather than a flat `KEY=value` dotenv. The resolver then reports
+     * `loaded` from file existence alone and masks the suffix to null — it
+     * never parses a key line out of the structured body.
+     */
+    credentialIsJson?: boolean;
+} | {
+    sourceKind: 'env';
+    envKey: string;
+    docAnchor: string;
+} | {
+    sourceKind: 'session';
+    docAnchor: string;
+};
+export interface HarnessAuthStatusEntry {
+    harnessName: string;
+    sourceKind: 'file' | 'env' | 'session' | 'none';
+    sourcePath?: string;
+    envKey?: string;
+    keySuffix: string | null;
+    lastModified: string | null;
+    state: 'loaded' | 'missing' | 'unknown';
+    docAnchor?: string;
+}
+export interface HarnessAuthStatusResponse {
+    harnesses: HarnessAuthStatusEntry[];
+}
+export interface ResolveOptions {
+    /** Env bag for env-kind sources. Defaults to process.env. */
+    env?: NodeJS.ProcessEnv;
+}
+/**
+ * Resolve one harness's safe auth status. Iteration-safe: any error collapses
+ * to a session-shaped `unknown` entry rather than throwing.
+ */
+export declare function resolveHarnessAuthStatus(harness: Harness, opts?: ResolveOptions): Promise<HarnessAuthStatusEntry>;

package/dist/harnesses/auth-source.js

@@ -0,0 +1,155 @@
+/**
+ * Shared resolver for the operator dashboard's "Harness auth status" panel (#564).
+ *
+ * A harness declares WHERE its credential lives via `getAuthSource()`. This
+ * helper owns the unsafe part: stat the file for mtime, read the credential,
+ * keep ONLY the last-4 suffix, discard the full value. It NEVER returns full
+ * key bytes. See docs/runbooks/rotating-harness-keys.md.
+ */
+import { stat, readFile } from 'node:fs/promises';
+import { homedir } from 'node:os';
+/**
+ * Tilde-abbreviate an absolute path for operator display when it sits under the
+ * real home dir; otherwise return it unchanged. Used by harness `getAuthSource`
+ * implementations so the dashboard shows `~/.hermes/.env` rather than a full
+ * `/Users/...` path.
+ */
+export function displayPath(absolutePath) {
+    const home = homedir();
+    return absolutePath.startsWith(home) ? absolutePath.replace(home, '~') : absolutePath;
+}
+const MIN_SUFFIX_LEN = 8;
+/** Last-4 suffix, or null when the value is too short / empty to safely show. */
+function safeSuffix(value) {
+    if (value.length < MIN_SUFFIX_LEN)
+        return null;
+    return value.slice(-4);
+}
+/**
+ * Extract the value of a single `KEY=value` line from a dotenv-style file body.
+ * Matches the named key at the start of a line only — a value that merely
+ * mentions the key name elsewhere is ignored. Strips surrounding quotes.
+ */
+function readEnvValueFromFile(body, envKey) {
+    for (const rawLine of body.split(/\r?\n/)) {
+        const line = rawLine.trimStart();
+        if (!line.startsWith(`${envKey}=`))
+            continue;
+        let value = line.slice(envKey.length + 1).trim();
+        if ((value.startsWith('"') && value.endsWith('"')) ||
+            (value.startsWith("'") && value.endsWith("'"))) {
+            value = value.slice(1, -1);
+        }
+        return value;
+    }
+    return undefined;
+}
+/**
+ * Resolve one harness's safe auth status. Iteration-safe: any error collapses
+ * to a session-shaped `unknown` entry rather than throwing.
+ */
+export async function resolveHarnessAuthStatus(harness, opts = {}) {
+    if (!harness.getAuthSource) {
+        return {
+            harnessName: harness.name,
+            sourceKind: 'none',
+            keySuffix: null,
+            lastModified: null,
+            state: 'unknown',
+        };
+    }
+    let source;
+    try {
+        source = await harness.getAuthSource();
+    }
+    catch {
+        return {
+            harnessName: harness.name,
+            sourceKind: 'session',
+            keySuffix: null,
+            lastModified: null,
+            state: 'unknown',
+        };
+    }
+    if (source.sourceKind === 'session') {
+        return {
+            harnessName: harness.name,
+            sourceKind: 'session',
+            keySuffix: null,
+            lastModified: null,
+            state: 'unknown',
+            docAnchor: source.docAnchor,
+        };
+    }
+    if (source.sourceKind === 'env') {
+        const env = opts.env ?? process.env;
+        const value = env[source.envKey]?.trim() ?? '';
+        return {
+            harnessName: harness.name,
+            sourceKind: 'env',
+            envKey: source.envKey,
+            keySuffix: safeSuffix(value),
+            lastModified: null,
+            state: value.length > 0 ? 'loaded' : 'missing',
+            docAnchor: source.docAnchor,
+        };
+    }
+    // file-kind
+    let mtime = null;
+    let body;
+    try {
+        const st = await stat(source.absolutePath);
+        mtime = st.mtime.toISOString();
+        body = await readFile(source.absolutePath, 'utf8');
+    }
+    catch {
+        return {
+            harnessName: harness.name,
+            sourceKind: 'file',
+            sourcePath: source.sourcePath,
+            envKey: source.envKey,
+            keySuffix: null,
+            lastModified: null,
+            state: 'missing',
+            docAnchor: source.docAnchor,
+        };
+    }
+    // Structured credential files (e.g. codex auth.json): report `loaded` from
+    // existence + mtime, but never parse a key out — suffix stays masked.
+    if (source.credentialIsJson) {
+        return {
+            harnessName: harness.name,
+            sourceKind: 'file',
+            sourcePath: source.sourcePath,
+            envKey: source.envKey,
+            keySuffix: null,
+            lastModified: mtime,
+            state: 'loaded',
+            docAnchor: source.docAnchor,
+        };
+    }
+    const value = readEnvValueFromFile(body, source.envKey)?.trim() ?? '';
+    if (value.length === 0) {
+        return {
+            harnessName: harness.name,
+            sourceKind: 'file',
+            sourcePath: source.sourcePath,
+            envKey: source.envKey,
+            keySuffix: null,
+            lastModified: null,
+            state: 'missing',
+            docAnchor: source.docAnchor,
+        };
+    }
+    return {
+        harnessName: harness.name,
+        sourceKind: 'file',
+        sourcePath: source.sourcePath,
+        envKey: source.envKey,
+        keySuffix: safeSuffix(value),
+        lastModified: mtime,
+        state: 'loaded',
+        docAnchor: source.docAnchor,
+    };
+}
+//# sourceMappingURL=auth-source.js.map

package/dist/harnesses/auth-source.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"auth-source.js","sourceRoot":"","sources":["../../src/harnesses/auth-source.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AACH,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAClD,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAGlC;;;;;GAKG;AACH,MAAM,UAAU,WAAW,CAAC,YAAoB;IAC9C,MAAM,IAAI,GAAG,OAAO,EAAE,CAAC;IACvB,OAAO,YAAY,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,OAAO,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC;AACxF,CAAC;AA0CD,MAAM,cAAc,GAAG,CAAC,CAAC;AAEzB,iFAAiF;AACjF,SAAS,UAAU,CAAC,KAAa;IAC/B,IAAI,KAAK,CAAC,MAAM,GAAG,cAAc;QAAE,OAAO,IAAI,CAAC;IAC/C,OAAO,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;AACzB,CAAC;AAED;;;;GAIG;AACH,SAAS,oBAAoB,CAAC,IAAY,EAAE,MAAc;IACxD,KAAK,MAAM,OAAO,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,CAAC;QAC1C,MAAM,IAAI,GAAG,OAAO,CAAC,SAAS,EAAE,CAAC;QACjC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,GAAG,MAAM,GAAG,CAAC;YAAE,SAAS;QAC7C,IAAI,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QACjD,IACE,CAAC,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;YAC9C,CAAC,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,EAC9C,CAAC;YACD,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;QAC7B,CAAC;QACD,OAAO,KAAK,CAAC;IACf,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAOD;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,wBAAwB,CAC5C,OAAgB,EAChB,OAAuB,EAAE;IAEzB,IAAI,CAAC,OAAO,CAAC,aAAa,EAAE,CAAC;QAC3B,OAAO;YACL,WAAW,EAAE,OAAO,CAAC,IAAI;YACzB,UAAU,EAAE,MAAM;YAClB,SAAS,EAAE,IAAI;YACf,YAAY,EAAE,IAAI;YAClB,KAAK,EAAE,SAAS;SACjB,CAAC;IACJ,CAAC;IAED,IAAI,MAAyB,CAAC;IAC9B,IAAI,CAAC;QACH,MAAM,GAAG,MAAM,OAAO,CAAC,aAAa,EAAE,CAAC;IACzC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO;YACL,WAAW,EAAE,OAAO,CAAC,IAAI;YACzB,UAAU,EAAE,SAAS;YACrB,SAAS,EAAE,IAAI;YACf,YAAY,EAAE,IAAI;YAClB,KAAK,EAAE,SAAS;SACjB,CAAC;IACJ,CAAC;IAED,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;QACpC,OAAO;YACL,WAAW,EAAE,OAAO,CAAC,IAAI;YACzB,UAAU,EAAE,SAAS;YACrB,SAAS,EAAE,IAAI;YACf,YAAY,EAAE,IAAI;YAClB,KAAK,EAAE,SAAS;YAChB,SAAS,EAAE,MAAM,CAAC,SAAS;SAC5B,CAAC;IACJ,CAAC;IAED,IAAI,MAAM,CAAC,UAAU,KAAK,KAAK,EAAE,CAAC;QAChC,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,IAAI,OAAO,CAAC,GAAG,CAAC;QACpC,MAAM,KAAK,GAAG,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;QAC/C,OAAO;YACL,WAAW,EAAE,OAAO,CAAC,IAAI;YACzB,UAAU,EAAE,KAAK;YACjB,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,SAAS,EAAE,UAAU,CAAC,KAAK,CAAC;YAC5B,YAAY,EAAE,IAAI;YAClB,KAAK,EAAE,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,SAAS;YAC9C,SAAS,EAAE,MAAM,CAAC,SAAS;SAC5B,CAAC;IACJ,CAAC;IAED,YAAY;IACZ,IAAI,KAAK,GAAkB,IAAI,CAAC;IAChC,IAAI,IAAY,CAAC;IACjB,IAAI,CAAC;QACH,MAAM,EAAE,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,YAAY,CAAC,CAAC;QAC3C,KAAK,GAAG,EAAE,CAAC,KAAK,CAAC,WAAW,EAAE,CAAC;QAC/B,IAAI,GAAG,MAAM,QAAQ,CAAC,MAAM,CAAC,YAAY,EAAE,MAAM,CAAC,CAAC;IACrD,CAAC;IAAC,MAAM,CAAC;QACP,OAAO;YACL,WAAW,EAAE,OAAO,CAAC,IAAI;YACzB,UAAU,EAAE,MAAM;YAClB,UAAU,EAAE,MAAM,CAAC,UAAU;YAC7B,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,SAAS,EAAE,IAAI;YACf,YAAY,EAAE,IAAI;YAClB,KAAK,EAAE,SAAS;YAChB,SAAS,EAAE,MAAM,CAAC,SAAS;SAC5B,CAAC;IACJ,CAAC;IAED,2EAA2E;IAC3E,sEAAsE;IACtE,IAAI,MAAM,CAAC,gBAAgB,EAAE,CAAC;QAC5B,OAAO;YACL,WAAW,EAAE,OAAO,CAAC,IAAI;YACzB,UAAU,EAAE,MAAM;YAClB,UAAU,EAAE,MAAM,CAAC,UAAU;YAC7B,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,SAAS,EAAE,IAAI;YACf,YAAY,EAAE,KAAK;YACnB,KAAK,EAAE,QAAQ;YACf,SAAS,EAAE,MAAM,CAAC,SAAS;SAC5B,CAAC;IACJ,CAAC;IAED,MAAM,KAAK,GAAG,oBAAoB,CAAC,IAAI,EAAE,MAAM,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;IACtE,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,OAAO;YACL,WAAW,EAAE,OAAO,CAAC,IAAI;YACzB,UAAU,EAAE,MAAM;YAClB,UAAU,EAAE,MAAM,CAAC,UAAU;YAC7B,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,SAAS,EAAE,IAAI;YACf,YAAY,EAAE,IAAI;YAClB,KAAK,EAAE,SAAS;YAChB,SAAS,EAAE,MAAM,CAAC,SAAS;SAC5B,CAAC;IACJ,CAAC;IAED,OAAO;QACL,WAAW,EAAE,OAAO,CAAC,IAAI;QACzB,UAAU,EAAE,MAAM;QAClB,UAAU,EAAE,MAAM,CAAC,UAAU;QAC7B,MAAM,EAAE,MAAM,CAAC,MAAM;QACrB,SAAS,EAAE,UAAU,CAAC,KAAK,CAAC;QAC5B,YAAY,EAAE,KAAK;QACnB,KAAK,EAAE,QAAQ;QACf,SAAS,EAAE,MAAM,CAAC,SAAS;KAC5B,CAAC;AACJ,CAAC"}

package/dist/harnesses/engine/backfill-failed-deliveries.d.ts

@@ -0,0 +1,59 @@
+/**
+ * Administrative backfill (#506): scan FAILED task_runs for ones whose
+ * delivery transaction actually landed on-chain and reclassify them as
+ * COMPLETE.
+ *
+ * Background: `insertArtifact()` used to throw on databases still carrying
+ * the legacy `artifacts.desired_state_id NOT NULL` column (fixed in #511).
+ * The engine's error path (`_runTransition` catch → `_classifyAndMarkTerminal`
+ * → `markFailed()`) turned that throw into a FAILED run even though the
+ * on-chain delivery — the thing that actually matters for reward eligibility
+ * — had already succeeded. This backfill corrects the historical DB rows
+ * left behind by that bug; it does not change any live engine behavior.
+ *
+ * Scope: a FAILED row is only a reclassification candidate when its
+ * `failureReason` carries the `artifacts.desired_state_id` NOT NULL
+ * constraint signature (see `DESIRED_STATE_ID_FAILURE_SIGNATURE` below).
+ * A `deliveryTxHash` + successful receipt alone is NOT sufficient — the
+ * daemon persists `deliveryTxHash` after `deliverToMarketplace` lands but
+ * *before* `claimDelivery` runs (crash-recovery design, see
+ * `client/src/harnesses/engine/delivery.ts`), so a row that genuinely failed
+ * at the `claimDelivery` step also carries a `deliveryTxHash` with a
+ * successful receipt — reclassifying it would hide a real failure. The
+ * receipt check below is a secondary confirmation, applied only after the
+ * failure-reason signature already narrows the candidate set to the #506
+ * bug.
+ *
+ * Caveat: because the failure in every #506 case happened at the
+ * `insertArtifact()` step, reclassified rows have no corresponding
+ * `artifacts` row and no ERC-8004 on-chain metadata anchor — that write is
+ * exactly what failed. This backfill does not attempt to reconstruct either;
+ * it only corrects `task_runs.state`.
+ */
+import type { PublicClient } from 'viem';
+import { TaskRunPersistence } from './persistence.js';
+export interface BackfillFailedDeliveriesDeps {
+    persistence: TaskRunPersistence;
+    /** Only `.getTransactionReceipt` is used. */
+    publicClient: PublicClient;
+    /** When true, report what would be reclassified without writing to the DB. Default false. */
+    dryRun?: boolean;
+}
+export interface BackfillFailedDeliveriesResult {
+    /** Rows reclassified FAILED → COMPLETE (or that would be, under dryRun). */
+    reclassified: Array<{
+        requestId: string;
+        originalFailureReason: string | null;
+    }>;
+    /** Request IDs left FAILED because they aren't #506 candidates, or their delivery didn't succeed. */
+    skipped: Array<{
+        requestId: string;
+        reason: string;
+    }>;
+    /** Request IDs whose receipt lookup errored — left FAILED, not retried within this run. */
+    failed: Array<{
+        requestId: string;
+        error: string;
+    }>;
+}
+export declare function backfillFailedDeliveries(deps: BackfillFailedDeliveriesDeps): Promise<BackfillFailedDeliveriesResult>;

package/dist/harnesses/engine/backfill-failed-deliveries.js

@@ -0,0 +1,94 @@
+/**
+ * Administrative backfill (#506): scan FAILED task_runs for ones whose
+ * delivery transaction actually landed on-chain and reclassify them as
+ * COMPLETE.
+ *
+ * Background: `insertArtifact()` used to throw on databases still carrying
+ * the legacy `artifacts.desired_state_id NOT NULL` column (fixed in #511).
+ * The engine's error path (`_runTransition` catch → `_classifyAndMarkTerminal`
+ * → `markFailed()`) turned that throw into a FAILED run even though the
+ * on-chain delivery — the thing that actually matters for reward eligibility
+ * — had already succeeded. This backfill corrects the historical DB rows
+ * left behind by that bug; it does not change any live engine behavior.
+ *
+ * Scope: a FAILED row is only a reclassification candidate when its
+ * `failureReason` carries the `artifacts.desired_state_id` NOT NULL
+ * constraint signature (see `DESIRED_STATE_ID_FAILURE_SIGNATURE` below).
+ * A `deliveryTxHash` + successful receipt alone is NOT sufficient — the
+ * daemon persists `deliveryTxHash` after `deliverToMarketplace` lands but
+ * *before* `claimDelivery` runs (crash-recovery design, see
+ * `client/src/harnesses/engine/delivery.ts`), so a row that genuinely failed
+ * at the `claimDelivery` step also carries a `deliveryTxHash` with a
+ * successful receipt — reclassifying it would hide a real failure. The
+ * receipt check below is a secondary confirmation, applied only after the
+ * failure-reason signature already narrows the candidate set to the #506
+ * bug.
+ *
+ * Caveat: because the failure in every #506 case happened at the
+ * `insertArtifact()` step, reclassified rows have no corresponding
+ * `artifacts` row and no ERC-8004 on-chain metadata anchor — that write is
+ * exactly what failed. This backfill does not attempt to reconstruct either;
+ * it only corrects `task_runs.state`.
+ */
+import { TaskRunState } from './state.js';
+/**
+ * Substring present in every `failureReason` produced by the #506 bug.
+ * The raw SqliteError message is exactly
+ * `NOT NULL constraint failed: artifacts.desired_state_id`; `markFailed`
+ * persists it verbatim when invoked from `_runTransition`'s 'transition'
+ * context, or prefixed with `recovery: ` when invoked from the 'recovery'
+ * context (see `_classifyAndMarkTerminal` in engine.ts). This substring
+ * survives both forms.
+ */
+const DESIRED_STATE_ID_FAILURE_SIGNATURE = 'artifacts.desired_state_id';
+export async function backfillFailedDeliveries(deps) {
+    const { persistence, publicClient, dryRun = false } = deps;
+    const result = { reclassified: [], skipped: [], failed: [] };
+    const failedRuns = persistence.getByState(TaskRunState.FAILED);
+    for (const run of failedRuns) {
+        if (!run.failureReason || !run.failureReason.includes(DESIRED_STATE_ID_FAILURE_SIGNATURE)) {
+            result.skipped.push({
+                requestId: run.requestId,
+                reason: 'failure reason does not match the desired_state_id constraint signature',
+            });
+            continue;
+        }
+        if (!run.deliveryTxHash) {
+            result.skipped.push({ requestId: run.requestId, reason: 'no deliveryTxHash recorded' });
+            continue;
+        }
+        let receipt;
+        try {
+            receipt = await publicClient.getTransactionReceipt({
+                hash: run.deliveryTxHash,
+            });
+        }
+        catch (err) {
+            result.failed.push({
+                requestId: run.requestId,
+                error: err instanceof Error ? err.message : String(err),
+            });
+            continue;
+        }
+        if (receipt.status !== 'success') {
+            result.skipped.push({
+                requestId: run.requestId,
+                reason: `delivery tx did not succeed (status=${receipt.status})`,
+            });
+            continue;
+        }
+        if (dryRun) {
+            result.reclassified.push({ requestId: run.requestId, originalFailureReason: run.failureReason });
+            continue;
+        }
+        const wrote = persistence.reclassifyFailedAsComplete(run.requestId);
+        if (wrote) {
+            result.reclassified.push({ requestId: run.requestId, originalFailureReason: run.failureReason });
+        }
+        else {
+            result.skipped.push({ requestId: run.requestId, reason: 'row state changed concurrently' });
+        }
+    }
+    return result;
+}
+//# sourceMappingURL=backfill-failed-deliveries.js.map