@jinn-network/client 0.1.8 → 0.1.9-canary.144d87d2

package/dist/eval/screen.js

@@ -0,0 +1,159 @@
+import { HELD_OUT_SLATE_SCHEMA_VERSION, hashHeldOutSlateArtifact, } from '../solver-types/_swe-rebench-v2-held-out-slate.js';
+/** Stratification / diversity key: the org prefix of an instance_id
+ *  (`tobymao__sqlglot-4661` → `tobymao`). Derivable without an HF fetch. */
+export function repoOf(task) {
+    const idx = task.instance_id.indexOf('__');
+    return idx === -1 ? task.instance_id : task.instance_id.slice(0, idx);
+}
+/**
+ * Order candidates round-robin across repos so the first N base-fails span
+ * repos rather than clumping in alphabetically-early ones. Deterministic:
+ * instances sort by instance_id within each repo group; repo groups iterate in
+ * sorted repo order.
+ */
+export function stratifyByRepo(pool) {
+    const groups = new Map();
+    for (const task of pool) {
+        const repo = repoOf(task);
+        (groups.get(repo) ?? groups.set(repo, []).get(repo)).push(task);
+    }
+    const repos = [...groups.keys()].sort((a, b) => a.localeCompare(b));
+    for (const repo of repos) {
+        groups.get(repo).sort((a, b) => a.instance_id.localeCompare(b.instance_id));
+    }
+    const out = [];
+    let added = true;
+    for (let i = 0; added; i++) {
+        added = false;
+        for (const repo of repos) {
+            const g = groups.get(repo);
+            if (i < g.length) {
+                out.push(g[i]);
+                added = true;
+            }
+        }
+    }
+    return out;
+}
+/**
+ * The expensive part: gradeable → base R-runs (early-stop on first pass) →
+ * prover (only if base is 0/R). Pure measurement; no selection/caps. This is what
+ * gets cached for resumability.
+ */
+async function measureCandidate(task, deps, R) {
+    const none = { gradeable: false, basePasses: 0, baseRuns: 0, baseUnscorable: false, proverRan: false, proverPassed: null };
+    if (!(await deps.ensureGradeable(task)))
+        return none;
+    let basePasses = 0;
+    let baseUnscorable = false;
+    let baseUnscorableReason;
+    let r = 0;
+    for (; r < R; r++) {
+        const run = await deps.runBaseFrozen(task);
+        if (run.passed === null) {
+            baseUnscorable = true;
+            baseUnscorableReason = run.unscorableReason;
+            break;
+        }
+        if (run.passed) {
+            basePasses++;
+            break;
+        }
+    }
+    const baseRuns = r + (baseUnscorable || basePasses > 0 ? 1 : 0);
+    if (baseUnscorable) {
+        return { gradeable: true, basePasses: 0, baseRuns, baseUnscorable: true, ...(baseUnscorableReason ? { baseUnscorableReason } : {}), proverRan: false, proverPassed: null };
+    }
+    if (basePasses > 0) {
+        return { gradeable: true, basePasses, baseRuns, baseUnscorable: false, proverRan: false, proverPassed: null };
+    }
+    // Base reliably fails (0/R) → layer 3: prover (existence proof of headroom).
+    const prover = await deps.runProverFrozen(task);
+    return {
+        gradeable: true, basePasses: 0, baseRuns, baseUnscorable: false, proverRan: true, proverPassed: prover.passed,
+        ...(prover.passed === null && prover.unscorableReason ? { proverUnscorableReason: prover.unscorableReason } : {}),
+    };
+}
+/**
+ * Partition a candidate stream into the held-out exam vs the rest, applying the
+ * three filter layers cheapest-first. `candidates` MUST already be ordered (use
+ * {@link stratifyByRepo}); selection order is the iteration order and is frozen.
+ *
+ * Resumable: if `deps.getCachedMeasurement` is provided, an already-measured
+ * candidate replays from cache (no inference, no budget cost), so re-running the
+ * same command resumes — the `maxCandidates` budget bounds only NEW measurements
+ * per invocation, letting a long screen proceed in budget-sized chunks. The
+ * selection decision (caps, held-out) is always recomputed fresh, so the cached
+ * measurements stay valid even if `heldOutCount`/`perRepoCap` change.
+ */
+export async function screenBaseFailures(candidates, deps, opts) {
+    const log = deps.log ?? (() => { });
+    const heldOut = [];
+    const screened = [];
+    const perRepo = new Map();
+    let liveMeasured = 0;
+    for (const task of candidates) {
+        if (heldOut.length >= opts.heldOutCount)
+            break;
+        const repo = repoOf(task);
+        const base = { instance_id: task.instance_id, repo, basePasses: 0, proverPassed: null };
+        // Measure (from cache, or live — bounded by the per-invocation budget).
+        let m = deps.getCachedMeasurement?.(task.instance_id);
+        if (!m) {
+            if (liveMeasured >= opts.maxCandidates)
+                break; // budget bounds NEW (inference-spending) measurements
+            m = await measureCandidate(task, deps, opts.R);
+            liveMeasured += 1;
+            deps.recordMeasurement?.(task.instance_id, m);
+        }
+        // Decide from the measurement (always fresh; cap/diversity not cached).
+        if (!m.gradeable) {
+            screened.push({ ...base, baseRuns: 0, gradeable: false, heldOut: false, reason: 'not-gradeable' });
+            continue;
+        }
+        if (m.baseUnscorable) {
+            if (m.baseUnscorableReason)
+                log(`[screen] ${task.instance_id} base-unscorable: ${m.baseUnscorableReason}`);
+            screened.push({ ...base, baseRuns: m.baseRuns, gradeable: true, heldOut: false, reason: 'base-unscorable', ...(m.baseUnscorableReason ? { unscorableReason: m.baseUnscorableReason } : {}) });
+            continue;
+        }
+        if (m.basePasses > 0) {
+            screened.push({ ...base, baseRuns: m.baseRuns, basePasses: m.basePasses, gradeable: true, heldOut: false, reason: 'base-passes' });
+            continue;
+        }
+        // Base 0/R → prover outcome (layer 3).
+        if (m.proverPassed !== true) {
+            if (m.proverPassed === null && m.proverUnscorableReason)
+                log(`[screen] ${task.instance_id} prover-unscorable: ${m.proverUnscorableReason}`);
+            screened.push({
+                ...base, baseRuns: m.baseRuns, gradeable: true, proverPassed: m.proverPassed, heldOut: false, reason: 'no-headroom',
+                ...(m.proverPassed === null && m.proverUnscorableReason ? { unscorableReason: m.proverUnscorableReason } : {}),
+            });
+            continue;
+        }
+        if ((perRepo.get(repo) ?? 0) >= opts.perRepoCap) {
+            screened.push({ ...base, baseRuns: m.baseRuns, gradeable: true, proverPassed: true, heldOut: false, reason: 'per-repo-cap' });
+            continue;
+        }
+        perRepo.set(repo, (perRepo.get(repo) ?? 0) + 1);
+        heldOut.push({ instance_id: task.instance_id, repo, baseRuns: m.baseRuns });
+        screened.push({ ...base, baseRuns: m.baseRuns, gradeable: true, proverPassed: true, heldOut: true, reason: 'held-out' });
+        log(`[screen] held out ${task.instance_id} (${heldOut.length}/${opts.heldOutCount})`);
+    }
+    return { heldOut, screened };
+}
+const V2_SLATE_COMMENT = 'BASELINE-FAILURE REGRESSION BENCHMARK (issue #986). Screened: gradeable at the current ' +
+    'evalSemanticsVersion AND base claude-code/Haiku frozen fails 0/R (R≥3) AND a stronger Codex/GPT-5.5 ' +
+    'prover passes ≥1 (proven headroom). Baseline 0% by construction. Held out from the generator train ' +
+    'stream via the active-slate-version union. Content-addressed; scores comparable WITHIN this version only.';
+export function buildV2SlateFile(instanceIds, generatedAt) {
+    const artifact = {
+        schemaVersion: HELD_OUT_SLATE_SCHEMA_VERSION,
+        solverType: 'swe-rebench-v2.v1',
+        version: 'v2',
+        generatedAt,
+        instanceIds: [...instanceIds].sort((a, b) => a.localeCompare(b)),
+    };
+    return { comment: V2_SLATE_COMMENT, ...artifact, hash: hashHeldOutSlateArtifact(artifact) };
+}
+//# sourceMappingURL=screen.js.map

package/dist/eval/screen.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"screen.js","sourceRoot":"","sources":["../../src/eval/screen.ts"],"names":[],"mappings":"AACA,OAAO,EACL,6BAA6B,EAC7B,wBAAwB,GAEzB,MAAM,mDAAmD,CAAC;AAE3D;4EAC4E;AAC5E,MAAM,UAAU,MAAM,CAAC,IAAc;IACnC,MAAM,GAAG,GAAG,IAAI,CAAC,WAAW,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;IAC3C,OAAO,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;AACxE,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,cAAc,CAAC,IAAgB;IAC7C,MAAM,MAAM,GAAG,IAAI,GAAG,EAAsB,CAAC;IAC7C,KAAK,MAAM,IAAI,IAAI,IAAI,EAAE,CAAC;QACxB,MAAM,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC;QAC1B,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,MAAM,CAAC,GAAG,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,IAAI,CAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACnE,CAAC;IACD,MAAM,KAAK,GAAG,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,CAAC;IACpE,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,CAAC,GAAG,CAAC,IAAI,CAAE,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,CAAC,aAAa,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC;IAC/E,CAAC;IACD,MAAM,GAAG,GAAe,EAAE,CAAC;IAC3B,IAAI,KAAK,GAAG,IAAI,CAAC;IACjB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,EAAE,CAAC;QAC3B,KAAK,GAAG,KAAK,CAAC;QACd,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACzB,MAAM,CAAC,GAAG,MAAM,CAAC,GAAG,CAAC,IAAI,CAAE,CAAC;YAC5B,IAAI,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC;gBACjB,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAE,CAAC,CAAC;gBAChB,KAAK,GAAG,IAAI,CAAC;YACf,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAgFD;;;;GAIG;AACH,KAAK,UAAU,gBAAgB,CAAC,IAAc,EAAE,IAAgB,EAAE,CAAS;IACzE,MAAM,IAAI,GAAsB,EAAE,SAAS,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,EAAE,QAAQ,EAAE,CAAC,EAAE,cAAc,EAAE,KAAK,EAAE,SAAS,EAAE,KAAK,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC;IAC9I,IAAI,CAAC,CAAC,MAAM,IAAI,CAAC,eAAe,CAAC,IAAI,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IAErD,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,cAAc,GAAG,KAAK,CAAC;IAC3B,IAAI,oBAAwC,CAAC;IAC7C,IAAI,CAAC,GAAG,CAAC,CAAC;IACV,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAClB,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,aAAa,CAAC,IAAI,CAAC,CAAC;QAC3C,IAAI,GAAG,CAAC,MAAM,KAAK,IAAI,EAAE,CAAC;YAAC,cAAc,GAAG,IAAI,CAAC;YAAC,oBAAoB,GAAG,GAAG,CAAC,gBAAgB,CAAC;YAAC,MAAM;QAAC,CAAC;QACvG,IAAI,GAAG,CAAC,MAAM,EAAE,CAAC;YAAC,UAAU,EAAE,CAAC;YAAC,MAAM;QAAC,CAAC;IAC1C,CAAC;IACD,MAAM,QAAQ,GAAG,CAAC,GAAG,CAAC,cAAc,IAAI,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAChE,IAAI,cAAc,EAAE,CAAC;QACnB,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC,EAAE,QAAQ,EAAE,cAAc,EAAE,IAAI,EAAE,GAAG,CAAC,oBAAoB,CAAC,CAAC,CAAC,EAAE,oBAAoB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,SAAS,EAAE,KAAK,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC;IAC7K,CAAC;IACD,IAAI,UAAU,GAAG,CAAC,EAAE,CAAC;QACnB,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,UAAU,EAAE,QAAQ,EAAE,cAAc,EAAE,KAAK,EAAE,SAAS,EAAE,KAAK,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC;IAChH,CAAC;IACD,6EAA6E;IAC7E,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,eAAe,CAAC,IAAI,CAAC,CAAC;IAChD,OAAO;QACL,SAAS,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC,EAAE,QAAQ,EAAE,cAAc,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,EAAE,YAAY,EAAE,MAAM,CAAC,MAAM;QAC7G,GAAG,CAAC,MAAM,CAAC,MAAM,KAAK,IAAI,IAAI,MAAM,CAAC,gBAAgB,CAAC,CAAC,CAAC,EAAE,sBAAsB,EAAE,MAAM,CAAC,gBAAgB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAClH,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACtC,UAAsB,EACtB,IAAgB,EAChB,IAAgB;IAEhB,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;IACnC,MAAM,OAAO,GAA4B,EAAE,CAAC;IAC5C,MAAM,QAAQ,GAAwB,EAAE,CAAC;IACzC,MAAM,OAAO,GAAG,IAAI,GAAG,EAAkB,CAAC;IAC1C,IAAI,YAAY,GAAG,CAAC,CAAC;IAErB,KAAK,MAAM,IAAI,IAAI,UAAU,EAAE,CAAC;QAC9B,IAAI,OAAO,CAAC,MAAM,IAAI,IAAI,CAAC,YAAY;YAAE,MAAM;QAC/C,MAAM,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC;QAC1B,MAAM,IAAI,GAAG,EAAE,WAAW,EAAE,IAAI,CAAC,WAAW,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC,EAAE,YAAY,EAAE,IAAsB,EAAE,CAAC;QAE1G,wEAAwE;QACxE,IAAI,CAAC,GAAG,IAAI,CAAC,oBAAoB,EAAE,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;QACtD,IAAI,CAAC,CAAC,EAAE,CAAC;YACP,IAAI,YAAY,IAAI,IAAI,CAAC,aAAa;gBAAE,MAAM,CAAC,sDAAsD;YACrG,CAAC,GAAG,MAAM,gBAAgB,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC;YAC/C,YAAY,IAAI,CAAC,CAAC;YAClB,IAAI,CAAC,iBAAiB,EAAE,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;QAChD,CAAC;QAED,wEAAwE;QACxE,IAAI,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC;YACjB,QAAQ,CAAC,IAAI,CAAC,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,CAAC,EAAE,SAAS,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,eAAe,EAAE,CAAC,CAAC;YACnG,SAAS;QACX,CAAC;QACD,IAAI,CAAC,CAAC,cAAc,EAAE,CAAC;YACrB,IAAI,CAAC,CAAC,oBAAoB;gBAAE,GAAG,CAAC,YAAY,IAAI,CAAC,WAAW,qBAAqB,CAAC,CAAC,oBAAoB,EAAE,CAAC,CAAC;YAC3G,QAAQ,CAAC,IAAI,CAAC,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,CAAC,CAAC,QAAQ,EAAE,SAAS,EAAE,IAAI,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,iBAAiB,EAAE,GAAG,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,EAAE,gBAAgB,EAAE,CAAC,CAAC,oBAAoB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;YAC9L,SAAS;QACX,CAAC;QACD,IAAI,CAAC,CAAC,UAAU,GAAG,CAAC,EAAE,CAAC;YACrB,QAAQ,CAAC,IAAI,CAAC,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,CAAC,CAAC,QAAQ,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE,SAAS,EAAE,IAAI,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,aAAa,EAAE,CAAC,CAAC;YACnI,SAAS;QACX,CAAC;QACD,uCAAuC;QACvC,IAAI,CAAC,CAAC,YAAY,KAAK,IAAI,EAAE,CAAC;YAC5B,IAAI,CAAC,CAAC,YAAY,KAAK,IAAI,IAAI,CAAC,CAAC,sBAAsB;gBAAE,GAAG,CAAC,YAAY,IAAI,CAAC,WAAW,uBAAuB,CAAC,CAAC,sBAAsB,EAAE,CAAC,CAAC;YAC5I,QAAQ,CAAC,IAAI,CAAC;gBACZ,GAAG,IAAI,EAAE,QAAQ,EAAE,CAAC,CAAC,QAAQ,EAAE,SAAS,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,CAAC,YAAY,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,aAAa;gBACnH,GAAG,CAAC,CAAC,CAAC,YAAY,KAAK,IAAI,IAAI,CAAC,CAAC,sBAAsB,CAAC,CAAC,CAAC,EAAE,gBAAgB,EAAE,CAAC,CAAC,sBAAsB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC/G,CAAC,CAAC;YACH,SAAS;QACX,CAAC;QACD,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,IAAI,IAAI,CAAC,UAAU,EAAE,CAAC;YAChD,QAAQ,CAAC,IAAI,CAAC,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,CAAC,CAAC,QAAQ,EAAE,SAAS,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,cAAc,EAAE,CAAC,CAAC;YAC9H,SAAS;QACX,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QAChD,OAAO,CAAC,IAAI,CAAC,EAAE,WAAW,EAAE,IAAI,CAAC,WAAW,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC;QAC5E,QAAQ,CAAC,IAAI,CAAC,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,CAAC,CAAC,QAAQ,EAAE,SAAS,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,UAAU,EAAE,CAAC,CAAC;QACzH,GAAG,CAAC,qBAAqB,IAAI,CAAC,WAAW,KAAK,OAAO,CAAC,MAAM,IAAI,IAAI,CAAC,YAAY,GAAG,CAAC,CAAC;IACxF,CAAC;IAED,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC;AAC/B,CAAC;AAUD,MAAM,gBAAgB,GACpB,yFAAyF;IACzF,sGAAsG;IACtG,qGAAqG;IACrG,2GAA2G,CAAC;AAE9G,MAAM,UAAU,gBAAgB,CAAC,WAAqB,EAAE,WAAmB;IACzE,MAAM,QAAQ,GAAyB;QACrC,aAAa,EAAE,6BAA6B;QAC5C,UAAU,EAAE,mBAAmB;QAC/B,OAAO,EAAE,IAAI;QACb,WAAW;QACX,WAAW,EAAE,CAAC,GAAG,WAAW,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC;KACjE,CAAC;IACF,OAAO,EAAE,OAAO,EAAE,gBAAgB,EAAE,GAAG,QAAQ,EAAE,IAAI,EAAE,wBAAwB,CAAC,QAAQ,CAAC,EAAE,CAAC;AAC9F,CAAC"}

package/dist/eval/slope.d.ts

@@ -0,0 +1,29 @@
+/**
+ * Ordinary least-squares slope of resolved-rate vs cycle index for the
+ * train-arm slope measurement (issue #822, AC#1).
+ *
+ * The train-arm e2e evaluates a checkpoint against the held-out slate (#817)
+ * at intervals via the eval orchestrator (#818), collecting one
+ * `{ cycleIndex, rate }` point per interval (`rate` = passed / scorable, the
+ * Wilson point estimate). The slope of the least-squares fit is the headline
+ * "is the learner improving across the training sequence" number.
+ *
+ * It is deliberately a thin helper over the closed-form OLS slope
+ * (`cov(x,y) / var(x)`); the per-point confidence intervals come from
+ * `wilson.ts` — this module does NOT reimplement them. The slope sign alone is
+ * never a verdict at small N: a flat or slightly negative slope is "within
+ * noise", which the e2e surfaces via the §4.1 honesty caveat.
+ */
+export interface RatePoint {
+    /** Training cycle index the eval ran at (0 = baseline, before any training). */
+    cycleIndex: number;
+    /** Observed resolved rate at that interval (passed / scorable, in [0, 1]). */
+    rate: number;
+}
+/**
+ * Least-squares slope of `rate` regressed on `cycleIndex`. Returns 0 for fewer
+ * than two points (no line to fit) and for a degenerate fit where every x is
+ * identical (zero variance — division would be NaN). A flat sequence yields
+ * exactly 0.
+ */
+export declare function leastSquaresSlope(points: RatePoint[]): number;

package/dist/eval/slope.js

@@ -0,0 +1,46 @@
+/**
+ * Ordinary least-squares slope of resolved-rate vs cycle index for the
+ * train-arm slope measurement (issue #822, AC#1).
+ *
+ * The train-arm e2e evaluates a checkpoint against the held-out slate (#817)
+ * at intervals via the eval orchestrator (#818), collecting one
+ * `{ cycleIndex, rate }` point per interval (`rate` = passed / scorable, the
+ * Wilson point estimate). The slope of the least-squares fit is the headline
+ * "is the learner improving across the training sequence" number.
+ *
+ * It is deliberately a thin helper over the closed-form OLS slope
+ * (`cov(x,y) / var(x)`); the per-point confidence intervals come from
+ * `wilson.ts` — this module does NOT reimplement them. The slope sign alone is
+ * never a verdict at small N: a flat or slightly negative slope is "within
+ * noise", which the e2e surfaces via the §4.1 honesty caveat.
+ */
+/**
+ * Least-squares slope of `rate` regressed on `cycleIndex`. Returns 0 for fewer
+ * than two points (no line to fit) and for a degenerate fit where every x is
+ * identical (zero variance — division would be NaN). A flat sequence yields
+ * exactly 0.
+ */
+export function leastSquaresSlope(points) {
+    const n = points.length;
+    if (n < 2)
+        return 0;
+    let sumX = 0;
+    let sumY = 0;
+    for (const { cycleIndex, rate } of points) {
+        sumX += cycleIndex;
+        sumY += rate;
+    }
+    const meanX = sumX / n;
+    const meanY = sumY / n;
+    let cov = 0;
+    let varX = 0;
+    for (const { cycleIndex, rate } of points) {
+        const dx = cycleIndex - meanX;
+        cov += dx * (rate - meanY);
+        varX += dx * dx;
+    }
+    if (varX === 0)
+        return 0;
+    return cov / varX;
+}
+//# sourceMappingURL=slope.js.map

package/dist/eval/slope.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"slope.js","sourceRoot":"","sources":["../../src/eval/slope.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AASH;;;;;GAKG;AACH,MAAM,UAAU,iBAAiB,CAAC,MAAmB;IACnD,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;IACxB,IAAI,CAAC,GAAG,CAAC;QAAE,OAAO,CAAC,CAAC;IACpB,IAAI,IAAI,GAAG,CAAC,CAAC;IACb,IAAI,IAAI,GAAG,CAAC,CAAC;IACb,KAAK,MAAM,EAAE,UAAU,EAAE,IAAI,EAAE,IAAI,MAAM,EAAE,CAAC;QAC1C,IAAI,IAAI,UAAU,CAAC;QACnB,IAAI,IAAI,IAAI,CAAC;IACf,CAAC;IACD,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,CAAC;IACvB,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,CAAC;IACvB,IAAI,GAAG,GAAG,CAAC,CAAC;IACZ,IAAI,IAAI,GAAG,CAAC,CAAC;IACb,KAAK,MAAM,EAAE,UAAU,EAAE,IAAI,EAAE,IAAI,MAAM,EAAE,CAAC;QAC1C,MAAM,EAAE,GAAG,UAAU,GAAG,KAAK,CAAC;QAC9B,GAAG,IAAI,EAAE,GAAG,CAAC,IAAI,GAAG,KAAK,CAAC,CAAC;QAC3B,IAAI,IAAI,EAAE,GAAG,EAAE,CAAC;IAClB,CAAC;IACD,IAAI,IAAI,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IACzB,OAAO,GAAG,GAAG,IAAI,CAAC;AACpB,CAAC"}

package/dist/eval/train-sequence.d.ts

@@ -0,0 +1,35 @@
+/**
+ * Train-sequence builder with the no-train/test-overlap guard for the
+ * train-arm slope e2e (issue #822, AC#2).
+ *
+ * The learner-full-cycle e2e drives `runCycle` DIRECTLY (it does not post tasks
+ * through the generator), so the generator's `excludeHeldOutSlate` train-stream
+ * chokepoint (#817) is BYPASSED. This builder is therefore the load-bearing
+ * AC#2 mechanism for the e2e: it selects the N distinct training instances from
+ * the pool with the held-out slate excluded, and asserts the resulting
+ * sequence is disjoint from the slate (fail-loud, never a silent drop).
+ *
+ * It reuses `excludeHeldOutSlate` from the #817 primitive rather than
+ * reimplementing the exclusion. Selection is deterministic (instance-id sorted)
+ * so a given pool yields a stable sequence across runs.
+ */
+import type { PoolTask } from '../solver-types/_swe-rebench-v2-pool.js';
+/** Thrown when a chosen training sequence intersects the held-out slate (AC#2). */
+export declare class TrainTestOverlapError extends Error {
+    readonly overlap: string[];
+    constructor(overlap: string[]);
+}
+/** Assert a set of training ids is disjoint from the slate, else throw loud. */
+export declare function assertNoOverlap(trainIds: string[], slateIds: Set<string>): void;
+export declare function buildTrainSequence(args: {
+    pool: PoolTask[];
+    slateIds: Set<string>;
+    /** Number of distinct training tasks (= N training cycles). */
+    count: number;
+    /**
+     * Optional explicit, hand-picked instance_ids (in order). When set, the
+     * builder still runs the no-overlap guard and resolves each id against the
+     * pool — used to fail-loud on a hand-edited sequence that overlaps the slate.
+     */
+    explicitIds?: string[];
+}): PoolTask[];

package/dist/eval/train-sequence.js

@@ -0,0 +1,59 @@
+/**
+ * Train-sequence builder with the no-train/test-overlap guard for the
+ * train-arm slope e2e (issue #822, AC#2).
+ *
+ * The learner-full-cycle e2e drives `runCycle` DIRECTLY (it does not post tasks
+ * through the generator), so the generator's `excludeHeldOutSlate` train-stream
+ * chokepoint (#817) is BYPASSED. This builder is therefore the load-bearing
+ * AC#2 mechanism for the e2e: it selects the N distinct training instances from
+ * the pool with the held-out slate excluded, and asserts the resulting
+ * sequence is disjoint from the slate (fail-loud, never a silent drop).
+ *
+ * It reuses `excludeHeldOutSlate` from the #817 primitive rather than
+ * reimplementing the exclusion. Selection is deterministic (instance-id sorted)
+ * so a given pool yields a stable sequence across runs.
+ */
+import { excludeHeldOutSlate } from '../solver-types/_swe-rebench-v2-held-out-slate.js';
+/** Thrown when a chosen training sequence intersects the held-out slate (AC#2). */
+export class TrainTestOverlapError extends Error {
+    overlap;
+    constructor(overlap) {
+        super(`train/test overlap: training sequence includes held-out slate instance(s) ` +
+            `${overlap.join(', ')} — refusing to train on the eval slate (AC#2). ` +
+            `The slate must stay out-of-sample for the slope to mean anything.`);
+        this.overlap = overlap;
+        this.name = 'TrainTestOverlapError';
+    }
+}
+/** Assert a set of training ids is disjoint from the slate, else throw loud. */
+export function assertNoOverlap(trainIds, slateIds) {
+    const overlap = trainIds.filter((id) => slateIds.has(id));
+    if (overlap.length > 0)
+        throw new TrainTestOverlapError(overlap);
+}
+export function buildTrainSequence(args) {
+    const eligible = excludeHeldOutSlate(args.pool, args.slateIds);
+    const byId = new Map(eligible.map((t) => [t.instance_id, t]));
+    if (args.explicitIds) {
+        // Guard the hand-picked sequence against the slate BEFORE resolving, so a
+        // slate-overlapping id is a TrainTestOverlapError, not a "not eligible".
+        assertNoOverlap(args.explicitIds, args.slateIds);
+        return args.explicitIds.map((id) => {
+            const task = byId.get(id);
+            if (!task) {
+                throw new Error(`explicit training instance ${id} not in the eligible pool`);
+            }
+            return task;
+        });
+    }
+    // Deterministic selection: instance-id sorted, first `count`.
+    const sorted = [...eligible].sort((a, b) => a.instance_id.localeCompare(b.instance_id));
+    if (sorted.length < args.count) {
+        throw new Error(`train sequence needs ${args.count} distinct tasks but only ${sorted.length} eligible ` +
+            `(pool size ${args.pool.length} minus ${args.slateIds.size} held-out slate instance(s))`);
+    }
+    const picked = sorted.slice(0, args.count);
+    assertNoOverlap(picked.map((t) => t.instance_id), args.slateIds);
+    return picked;
+}
+//# sourceMappingURL=train-sequence.js.map

package/dist/eval/train-sequence.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"train-sequence.js","sourceRoot":"","sources":["../../src/eval/train-sequence.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAGH,OAAO,EAAE,mBAAmB,EAAE,MAAM,mDAAmD,CAAC;AAExF,mFAAmF;AACnF,MAAM,OAAO,qBAAsB,SAAQ,KAAK;IAClB;IAA5B,YAA4B,OAAiB;QAC3C,KAAK,CACH,4EAA4E;YAC1E,GAAG,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,iDAAiD;YACtE,mEAAmE,CACtE,CAAC;QALwB,YAAO,GAAP,OAAO,CAAU;QAM3C,IAAI,CAAC,IAAI,GAAG,uBAAuB,CAAC;IACtC,CAAC;CACF;AAED,gFAAgF;AAChF,MAAM,UAAU,eAAe,CAAC,QAAkB,EAAE,QAAqB;IACvE,MAAM,OAAO,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC;IAC1D,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC;QAAE,MAAM,IAAI,qBAAqB,CAAC,OAAO,CAAC,CAAC;AACnE,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,IAWlC;IACC,MAAM,QAAQ,GAAG,mBAAmB,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC;IAC/D,MAAM,IAAI,GAAG,IAAI,GAAG,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAE9D,IAAI,IAAI,CAAC,WAAW,EAAE,CAAC;QACrB,0EAA0E;QAC1E,yEAAyE;QACzE,eAAe,CAAC,IAAI,CAAC,WAAW,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC;QACjD,OAAO,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE;YACjC,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;YAC1B,IAAI,CAAC,IAAI,EAAE,CAAC;gBACV,MAAM,IAAI,KAAK,CAAC,8BAA8B,EAAE,2BAA2B,CAAC,CAAC;YAC/E,CAAC;YACD,OAAO,IAAI,CAAC;QACd,CAAC,CAAC,CAAC;IACL,CAAC;IAED,8DAA8D;IAC9D,MAAM,MAAM,GAAG,CAAC,GAAG,QAAQ,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,CAAC,aAAa,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC;IACxF,IAAI,MAAM,CAAC,MAAM,GAAG,IAAI,CAAC,KAAK,EAAE,CAAC;QAC/B,MAAM,IAAI,KAAK,CACb,wBAAwB,IAAI,CAAC,KAAK,4BAA4B,MAAM,CAAC,MAAM,YAAY;YACrF,cAAc,IAAI,CAAC,IAAI,CAAC,MAAM,UAAU,IAAI,CAAC,QAAQ,CAAC,IAAI,8BAA8B,CAC3F,CAAC;IACJ,CAAC;IACD,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC;IAC3C,eAAe,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,CAAC,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC;IACjE,OAAO,MAAM,CAAC;AAChB,CAAC"}

package/dist/eval/wilson.d.ts

@@ -0,0 +1,45 @@
+/**
+ * Wilson score interval + resolved-rate comparison for the `jinn eval`
+ * held-out checkpoint orchestrator (issue #818, AC#2).
+ *
+ * The Wilson score interval is a binomial proportion confidence interval that
+ * behaves well at the extremes (p=0, p=1) and for small n — unlike the naive
+ * normal-approximation interval. We write it small (no stats dependency, per
+ * repo convention): the formula is ~10 lines.
+ *
+ * Per log/decisions/2026-05-28-rl-eval-measurement.md §4: this is v1-simple.
+ * Only *large* deltas are trustworthy — we encode that as "the child and
+ * parent intervals do not overlap." No seed control, no multi-run averaging.
+ */
+export interface Interval {
+    /** Observed point estimate, passed / scorable (0 when scorable=0). */
+    p: number;
+    /** Lower bound, clamped to [0, 1]. */
+    lo: number;
+    /** Upper bound, clamped to [0, 1]. */
+    hi: number;
+}
+/**
+ * Wilson score interval for `passed` successes out of `scorable` trials.
+ * `scorable === 0` returns a degenerate `{ p: 0, lo: 0, hi: 0 }` (no NaN).
+ */
+export declare function wilsonInterval(passed: number, scorable: number, z?: number): Interval;
+export type RateVerdict = 'trustworthy' | 'within-noise';
+export interface RateComparison {
+    child: Interval;
+    parent: Interval;
+    /** child.p − parent.p (point-estimate difference, can be negative). */
+    delta: number;
+    /**
+     * 'trustworthy' iff the two Wilson intervals do NOT overlap; otherwise
+     * 'within-noise'. v1-simple: only disjoint intervals justify a claim.
+     */
+    verdict: RateVerdict;
+}
+export declare function compareRates(child: {
+    passed: number;
+    scorable: number;
+}, parent: {
+    passed: number;
+    scorable: number;
+}): RateComparison;

package/dist/eval/wilson.js

@@ -0,0 +1,48 @@
+/**
+ * Wilson score interval + resolved-rate comparison for the `jinn eval`
+ * held-out checkpoint orchestrator (issue #818, AC#2).
+ *
+ * The Wilson score interval is a binomial proportion confidence interval that
+ * behaves well at the extremes (p=0, p=1) and for small n — unlike the naive
+ * normal-approximation interval. We write it small (no stats dependency, per
+ * repo convention): the formula is ~10 lines.
+ *
+ * Per log/decisions/2026-05-28-rl-eval-measurement.md §4: this is v1-simple.
+ * Only *large* deltas are trustworthy — we encode that as "the child and
+ * parent intervals do not overlap." No seed control, no multi-run averaging.
+ */
+/** Two-sided z for a 95% interval (1.96 ≈ Φ⁻¹(0.975)). */
+const DEFAULT_Z = 1.96;
+/**
+ * Wilson score interval for `passed` successes out of `scorable` trials.
+ * `scorable === 0` returns a degenerate `{ p: 0, lo: 0, hi: 0 }` (no NaN).
+ */
+export function wilsonInterval(passed, scorable, z = DEFAULT_Z) {
+    if (scorable === 0)
+        return { p: 0, lo: 0, hi: 0 };
+    const n = scorable;
+    const p = passed / n;
+    const z2 = z * z;
+    const denom = 1 + z2 / n;
+    const centre = p + z2 / (2 * n);
+    const margin = z * Math.sqrt((p * (1 - p)) / n + z2 / (4 * n * n));
+    const lo = (centre - margin) / denom;
+    const hi = (centre + margin) / denom;
+    return {
+        p,
+        lo: Math.max(0, lo),
+        hi: Math.min(1, hi),
+    };
+}
+export function compareRates(child, parent) {
+    const c = wilsonInterval(child.passed, child.scorable);
+    const p = wilsonInterval(parent.passed, parent.scorable);
+    const disjoint = c.lo > p.hi || p.lo > c.hi;
+    return {
+        child: c,
+        parent: p,
+        delta: c.p - p.p,
+        verdict: disjoint ? 'trustworthy' : 'within-noise',
+    };
+}
+//# sourceMappingURL=wilson.js.map

package/dist/eval/wilson.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"wilson.js","sourceRoot":"","sources":["../../src/eval/wilson.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,0DAA0D;AAC1D,MAAM,SAAS,GAAG,IAAI,CAAC;AAWvB;;;GAGG;AACH,MAAM,UAAU,cAAc,CAAC,MAAc,EAAE,QAAgB,EAAE,IAAY,SAAS;IACpF,IAAI,QAAQ,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,CAAC;IAClD,MAAM,CAAC,GAAG,QAAQ,CAAC;IACnB,MAAM,CAAC,GAAG,MAAM,GAAG,CAAC,CAAC;IACrB,MAAM,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;IACjB,MAAM,KAAK,GAAG,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;IACzB,MAAM,MAAM,GAAG,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAChC,MAAM,MAAM,GAAG,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;IACnE,MAAM,EAAE,GAAG,CAAC,MAAM,GAAG,MAAM,CAAC,GAAG,KAAK,CAAC;IACrC,MAAM,EAAE,GAAG,CAAC,MAAM,GAAG,MAAM,CAAC,GAAG,KAAK,CAAC;IACrC,OAAO;QACL,CAAC;QACD,EAAE,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,CAAC;QACnB,EAAE,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,CAAC;KACpB,CAAC;AACJ,CAAC;AAgBD,MAAM,UAAU,YAAY,CAC1B,KAA2C,EAC3C,MAA4C;IAE5C,MAAM,CAAC,GAAG,cAAc,CAAC,KAAK,CAAC,MAAM,EAAE,KAAK,CAAC,QAAQ,CAAC,CAAC;IACvD,MAAM,CAAC,GAAG,cAAc,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,QAAQ,CAAC,CAAC;IACzD,MAAM,QAAQ,GAAG,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,EAAE,CAAC;IAC5C,OAAO;QACL,KAAK,EAAE,CAAC;QACR,MAAM,EAAE,CAAC;QACT,KAAK,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QAChB,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,cAAc;KACnD,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/engine/canonical-json.js

@@ -22,9 +22,11 @@
  * Used for manifest signing: produce a deterministic byte string that two
  * independent parties can reproduce from the same object graph.
  */
-import { createRequire } from 'node:module';
-const require = createRequire(import.meta.url);
-const canonicalize = require('canonicalize');
+// canonicalize v3 is ESM-only (no CJS entry); the client is `"type": "module"`
+// so a native ESM default import is the correct interop (replaces the v2-era
+// createRequire shim). v3 is packaging-only vs v2 — its RFC 8785 output is
+// byte-identical, so existing content hashes / manifest signatures are stable.
+import canonicalize from 'canonicalize';
 /**
  * Recursively replace NaN / ±Infinity with null so that canonicalize does not
  * throw — matching the JSON.stringify behaviour that the rest of the codebase

package/dist/harnesses/engine/canonical-json.js.map

@@ -1 +1 @@
-{"version":3,"file":"canonical-json.js","sourceRoot":"","sources":["../../../src/harnesses/engine/canonical-json.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAC5C,MAAM,OAAO,GAAG,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;AAC/C,MAAM,YAAY,GAAG,OAAO,CAAC,cAAc,CAA2C,CAAC;AAEvF;;;;;GAKG;AACH,SAAS,eAAe,CAAC,KAAc;IACrC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IAC/D,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC,GAAG,CAAC,eAAe,CAAC,CAAC;IAC5D,IAAI,KAAK,KAAK,IAAI,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QAChD,mEAAmE;QACnE,qEAAqE;QACrE,sEAAsE;QACtE,WAAW;QACX,MAAM,MAAM,GAAI,KAAoC,CAAC,MAAM,CAAC;QAC5D,IAAI,OAAO,MAAM,KAAK,UAAU,EAAE,CAAC;YACjC,OAAO,eAAe,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC;QAC7C,CAAC;QACD,MAAM,GAAG,GAA4B,EAAE,CAAC;QACxC,KAAK,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC;YAAE,GAAG,CAAC,CAAC,CAAC,GAAG,eAAe,CAAC,CAAC,CAAC,CAAC;QACxE,OAAO,GAAG,CAAC;IACb,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,KAAc;IAC1C,MAAM,OAAO,GAAG,eAAe,CAAC,KAAK,CAAC,CAAC;IAEvC,MAAM,MAAM,GAAG,YAAY,CAAC,OAAO,CAAC,CAAC;IACrC,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;QACzB,wEAAwE;QACxE,uEAAuE;QACvE,kCAAkC;QAClC,OAAO,MAAM,CAAC;IAChB,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC"}
+{"version":3,"file":"canonical-json.js","sourceRoot":"","sources":["../../../src/harnesses/engine/canonical-json.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,+EAA+E;AAC/E,6EAA6E;AAC7E,2EAA2E;AAC3E,+EAA+E;AAC/E,OAAO,YAAY,MAAM,cAAc,CAAC;AAExC;;;;;GAKG;AACH,SAAS,eAAe,CAAC,KAAc;IACrC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IAC/D,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC,GAAG,CAAC,eAAe,CAAC,CAAC;IAC5D,IAAI,KAAK,KAAK,IAAI,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QAChD,mEAAmE;QACnE,qEAAqE;QACrE,sEAAsE;QACtE,WAAW;QACX,MAAM,MAAM,GAAI,KAAoC,CAAC,MAAM,CAAC;QAC5D,IAAI,OAAO,MAAM,KAAK,UAAU,EAAE,CAAC;YACjC,OAAO,eAAe,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC;QAC7C,CAAC;QACD,MAAM,GAAG,GAA4B,EAAE,CAAC;QACxC,KAAK,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC;YAAE,GAAG,CAAC,CAAC,CAAC,GAAG,eAAe,CAAC,CAAC,CAAC,CAAC;QACxE,OAAO,GAAG,CAAC;IACb,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,KAAc;IAC1C,MAAM,OAAO,GAAG,eAAe,CAAC,KAAK,CAAC,CAAC;IAEvC,MAAM,MAAM,GAAG,YAAY,CAAC,OAAO,CAAC,CAAC;IACrC,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;QACzB,wEAAwE;QACxE,uEAAuE;QACvE,kCAAkC;QAClC,OAAO,MAAM,CAAC;IAChB,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC"}

package/dist/harnesses/engine/engine.d.ts

@@ -70,6 +70,10 @@ export interface JoinedSolverNetsView {
     } | undefined;
     /** Enumerate all joined manifest CIDs (used for digest-based filtering). */
     manifestCids(): string[];
+    /** Add/replace one joined entry live (used by the hot-apply join applier, #1037). */
+    set(manifestCid: string, entry: {
+        roles: Array<'solver' | 'evaluator'>;
+    }): void;
 }
 /**
  * Build a `JoinedSolverNetsView` from the raw operator-config block.
@@ -82,6 +86,17 @@ export declare function joinedSolverNetsViewFromConfig(joined: Record<string, {
     manifestCid: string;
     roles: Array<'solver' | 'evaluator'>;
 }> | undefined): JoinedSolverNetsView | undefined;
+/**
+ * Mutable `JoinedSolverNetsView` for the running daemon. Unlike
+ * `joinedSolverNetsViewFromConfig` (boot snapshot), the applier
+ * (`daemon/join-applier.ts`, #1037) keeps a handle and calls `set()` when a
+ * join is hot-applied, so the engine's per-task eligibility check sees the new
+ * cid on its next call without a restart.
+ */
+export declare function createMutableJoinedSolverNetsView(initial: Record<string, {
+    manifestCid: string;
+    roles: Array<'solver' | 'evaluator'>;
+}> | undefined): JoinedSolverNetsView;
 /**
  * Resolves a launched SolverNet manifest by IPFS CID.
  *
@@ -576,6 +591,15 @@ export declare class TaskEngine {
      *   TCAttemptAlreadyFinalized, …). We mark the row RACE_LOST and emit a
      *   `kind=race_lost` activity event so operators can audit prunes
      *   without inflating the FAILED counter (#896).
+     * - A transport-transient error (e.g. `AllRpcsFailedError` — every provider
+     *   in the L2 fallback chain failed at once) on a task whose delivery window
+     *   is still open: leave the row in its current in-flight state so the next
+     *   tick re-drives it once the RPCs recover, and emit a `tick_error` (warn)
+     *   event instead of inflating the FAILED counter. Without this the daemon
+     *   stamped the row FAILED, dropping it from `getInFlight()` permanently, so
+     *   L2 work went silent until a manual restart (#912). Past-window transient
+     *   errors still terminalize to avoid churning on work that can no longer
+     *   settle.
      * - Everything else: existing markFailed behaviour. When invoked from
      *   recovery, `contextLabel === 'recovery'` so the failure_reason
      *   carries the `recovery:` prefix the original code path used.