@jinn-network/client 0.1.9 → 0.2.0

package/dist/pilot/arm-homes.js

@@ -0,0 +1,104 @@
+/** Build one isolated jinn-agent home per pilot arm.
+ *
+ * The 2026-07-10 pilot ran all arms against the shared ~/.jinn-agent, so every
+ * arm's system prompt carried the full installed skills catalog — the arms were
+ * not experimentally distinct. Isolation is per-arm homes: identical built-ins
+ * and credentials, differing ONLY in which distilled skill sets are installed.
+ * `run-pilot.ts` routes an arm to its home via `arm.jinnAgentHome` →
+ * `JINN_AGENT_HOME` (see `envForArm`).
+ *
+ * Deliberately NOT copied: `.skills_prompt_snapshot.json` (the skills-manifest
+ * cache — a stale copy would defeat the isolation), `sessions/`, `state.db*`,
+ * `logs/`, `cache*`.
+ *
+ * Known risk: each home holds a copy of `auth.json`, so all arms share one
+ * OAuth credential; a token refresh in one home can stale the others. Run the
+ * cheap isolation probe first and keep runs short.
+ */
+import { cpSync, existsSync, mkdirSync, readdirSync, readFileSync, rmSync } from 'node:fs';
+import { join } from 'node:path';
+/**
+ * Fail-loud isolation gate, run before any solve is spawned. Arms whose skill
+ * loadouts differ MUST each run in their own agent home whose filesystem
+ * actually realizes that loadout — otherwise every arm sees the same shared
+ * catalog and the run is arm-invariant (the 2026-07-10 trap: a 27-attempt
+ * "comparison" of one condition to itself). Same-loadout arms are one
+ * experimental condition and need no isolation.
+ */
+export function assertArmIsolation(arms) {
+    const loadouts = new Set(arms.map((arm) => JSON.stringify([...arm.skills].sort())));
+    if (loadouts.size <= 1)
+        return;
+    const missingHome = arms.filter((arm) => !arm.jinnAgentHome);
+    if (missingHome.length > 0) {
+        throw new Error(`arms differ in skill loadout but ${missingHome.map((a) => `'${a.name}'`).join(', ')} ` +
+            `have no jinnAgentHome — without per-arm isolated homes every arm sees the same shared ` +
+            `skills catalog and the comparison is void. Build homes with scripts/build-pilot-arm-homes.ts.`);
+    }
+    const homes = arms.map((arm) => arm.jinnAgentHome);
+    if (new Set(homes).size !== homes.length) {
+        throw new Error('arms differ in skill loadout but share a jinnAgentHome — homes must be distinct per arm');
+    }
+    for (const arm of arms) {
+        const skillsDir = join(arm.jinnAgentHome, 'skills');
+        const installed = existsSync(skillsDir)
+            ? readdirSync(skillsDir, { withFileTypes: true }).filter((e) => e.isDirectory()).map((e) => e.name)
+            : [];
+        for (const skill of arm.skills) {
+            if (!installed.includes(skill)) {
+                throw new Error(`arm '${arm.name}' home ${arm.jinnAgentHome} is missing its skill '${skill}' — rebuild the arm homes`);
+            }
+        }
+        const allowed = new Set(arm.skills);
+        for (const name of installed) {
+            if (!allowed.has(name) && isDistilledSkill(join(skillsDir, name))) {
+                throw new Error(`arm '${arm.name}' home ${arm.jinnAgentHome} contains distilled skill '${name}' outside its loadout — ` +
+                    `the arm is contaminated; rebuild the arm homes`);
+            }
+        }
+    }
+}
+/** Top-level home files carried into every arm home (skipped if absent). */
+const HOME_FILES = ['auth.json', 'config.yaml', '.env', 'SOUL.md', 'models_dev_cache.json'];
+function isDistilledSkill(skillDir) {
+    const manifest = join(skillDir, 'SKILL.md');
+    if (!existsSync(manifest))
+        return false;
+    return /schema:\s*jinn\.skill\.v1/.test(readFileSync(manifest, 'utf-8'));
+}
+export function buildArmHomes(opts) {
+    const { armsFile, sourceDir, destDir } = opts;
+    const sourceSkillsDir = join(sourceDir, 'skills');
+    const installed = new Set(existsSync(sourceSkillsDir)
+        ? readdirSync(sourceSkillsDir, { withFileTypes: true }).filter((e) => e.isDirectory()).map((e) => e.name)
+        : []);
+    for (const skill of new Set(armsFile.flatMap((arm) => arm.skills))) {
+        if (!installed.has(skill)) {
+            throw new Error(`arm skill '${skill}' is not installed in ${sourceSkillsDir}`);
+        }
+    }
+    // Built-ins = every installed skill that is NOT Jinn distillate. Distillate
+    // is recognized by its jinn.skill.v1 frontmatter rather than by the arms
+    // union, so a rerun with a reduced arm set never misclassifies leftover
+    // distilled skills as built-ins.
+    const builtins = [...installed].filter((name) => !isDistilledSkill(join(sourceSkillsDir, name))).sort();
+    const out = [];
+    for (const arm of armsFile) {
+        const home = join(destDir, arm.name);
+        rmSync(home, { recursive: true, force: true });
+        mkdirSync(home, { recursive: true });
+        for (const file of HOME_FILES) {
+            const src = join(sourceDir, file);
+            if (existsSync(src))
+                cpSync(src, join(home, file));
+        }
+        const homeSkills = join(home, 'skills');
+        mkdirSync(homeSkills, { recursive: true });
+        for (const skill of [...builtins, ...arm.skills]) {
+            cpSync(join(sourceSkillsDir, skill), join(homeSkills, skill), { recursive: true });
+        }
+        out.push({ name: arm.name, skills: arm.skills, jinnAgentHome: home });
+    }
+    return out;
+}
+//# sourceMappingURL=arm-homes.js.map

package/dist/pilot/arm-homes.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"arm-homes.js","sourceRoot":"","sources":["../../src/pilot/arm-homes.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AAC3F,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAYjC;;;;;;;GAOG;AACH,MAAM,UAAU,kBAAkB,CAAC,IAAmB;IACpD,MAAM,QAAQ,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,GAAG,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IACpF,IAAI,QAAQ,CAAC,IAAI,IAAI,CAAC;QAAE,OAAO;IAE/B,MAAM,WAAW,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC;IAC7D,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC3B,MAAM,IAAI,KAAK,CACb,oCAAoC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG;YACvF,wFAAwF;YACxF,+FAA+F,CAChG,CAAC;IACJ,CAAC;IACD,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,aAAc,CAAC,CAAC;IACpD,IAAI,IAAI,GAAG,CAAC,KAAK,CAAC,CAAC,IAAI,KAAK,KAAK,CAAC,MAAM,EAAE,CAAC;QACzC,MAAM,IAAI,KAAK,CAAC,yFAAyF,CAAC,CAAC;IAC7G,CAAC;IAED,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,aAAc,EAAE,QAAQ,CAAC,CAAC;QACrD,MAAM,SAAS,GAAG,UAAU,CAAC,SAAS,CAAC;YACrC,CAAC,CAAC,WAAW,CAAC,SAAS,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;YACnG,CAAC,CAAC,EAAE,CAAC;QACP,KAAK,MAAM,KAAK,IAAI,GAAG,CAAC,MAAM,EAAE,CAAC;YAC/B,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;gBAC/B,MAAM,IAAI,KAAK,CAAC,QAAQ,GAAG,CAAC,IAAI,UAAU,GAAG,CAAC,aAAa,0BAA0B,KAAK,2BAA2B,CAAC,CAAC;YACzH,CAAC;QACH,CAAC;QACD,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;QACpC,KAAK,MAAM,IAAI,IAAI,SAAS,EAAE,CAAC;YAC7B,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,gBAAgB,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC,EAAE,CAAC;gBAClE,MAAM,IAAI,KAAK,CACb,QAAQ,GAAG,CAAC,IAAI,UAAU,GAAG,CAAC,aAAa,8BAA8B,IAAI,0BAA0B;oBACvG,gDAAgD,CACjD,CAAC;YACJ,CAAC;QACH,CAAC;IACH,CAAC;AACH,CAAC;AAED,4EAA4E;AAC5E,MAAM,UAAU,GAAG,CAAC,WAAW,EAAE,aAAa,EAAE,MAAM,EAAE,SAAS,EAAE,uBAAuB,CAAC,CAAC;AAE5F,SAAS,gBAAgB,CAAC,QAAgB;IACxC,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,EAAE,UAAU,CAAC,CAAC;IAC5C,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC;QAAE,OAAO,KAAK,CAAC;IACxC,OAAO,2BAA2B,CAAC,IAAI,CAAC,YAAY,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC,CAAC;AAC3E,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,IAI7B;IACC,MAAM,EAAE,QAAQ,EAAE,SAAS,EAAE,OAAO,EAAE,GAAG,IAAI,CAAC;IAC9C,MAAM,eAAe,GAAG,IAAI,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAC;IAClD,MAAM,SAAS,GAAG,IAAI,GAAG,CACvB,UAAU,CAAC,eAAe,CAAC;QACzB,CAAC,CAAC,WAAW,CAAC,eAAe,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;QACzG,CAAC,CAAC,EAAE,CACP,CAAC;IAEF,KAAK,MAAM,KAAK,IAAI,IAAI,GAAG,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,EAAE,CAAC;QACnE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC;YAC1B,MAAM,IAAI,KAAK,CAAC,cAAc,KAAK,yBAAyB,eAAe,EAAE,CAAC,CAAC;QACjF,CAAC;IACH,CAAC;IACD,4EAA4E;IAC5E,yEAAyE;IACzE,wEAAwE;IACxE,iCAAiC;IACjC,MAAM,QAAQ,GAAG,CAAC,GAAG,SAAS,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,gBAAgB,CAAC,IAAI,CAAC,eAAe,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IAExG,MAAM,GAAG,GAAkB,EAAE,CAAC;IAC9B,KAAK,MAAM,GAAG,IAAI,QAAQ,EAAE,CAAC;QAC3B,MAAM,IAAI,GAAG,IAAI,CAAC,OAAO,EAAE,GAAG,CAAC,IAAI,CAAC,CAAC;QACrC,MAAM,CAAC,IAAI,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QAC/C,SAAS,CAAC,IAAI,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACrC,KAAK,MAAM,IAAI,IAAI,UAAU,EAAE,CAAC;YAC9B,MAAM,GAAG,GAAG,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;YAClC,IAAI,UAAU,CAAC,GAAG,CAAC;gBAAE,MAAM,CAAC,GAAG,EAAE,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC,CAAC;QACrD,CAAC;QACD,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;QACxC,SAAS,CAAC,UAAU,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAC3C,KAAK,MAAM,KAAK,IAAI,CAAC,GAAG,QAAQ,EAAE,GAAG,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC;YACjD,MAAM,CAAC,IAAI,CAAC,eAAe,EAAE,KAAK,CAAC,EAAE,IAAI,CAAC,UAAU,EAAE,KAAK,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACrF,CAAC;QACD,GAAG,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,GAAG,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;IACxE,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC"}

package/dist/pilot/cost.d.ts

@@ -0,0 +1,19 @@
+import type { SolveTokens } from './solve.js';
+export interface RateTable {
+    inputPerM: number;
+    outputPerM: number;
+    cacheReadPerM?: number;
+}
+export declare const DEEPSEEK_V4_FLASH_RATES: RateTable;
+/** gpt-5.4-mini published API rate (used to price Codex-subscription token
+ *  counts at a representative metered cost — the sub bills flat, but the gate
+ *  measures what a metered user would pay). $0.75/M in, $4.50/M out. */
+export declare const GPT_5_4_MINI_RATES: RateTable;
+/** USD for one solve from provider-actual token counts. `output_tokens` already
+ *  INCLUDES `reasoning_tokens` (OpenAI `completion_tokens` semantics, verified
+ *  empirically on the Codex backend), so reasoning is NOT added again — it is
+ *  surfaced separately only for reporting. Likewise `input_tokens` is inclusive
+ *  of `cache_read_tokens`: when a cacheReadPerM rate is supplied, cache-read
+ *  tokens are re-priced at that (discounted) rate and the remaining (fresh)
+ *  input at inputPerM; otherwise all input is inputPerM. */
+export declare function solveCostUsd(tokens: SolveTokens, rates: RateTable): number;

package/dist/pilot/cost.js

@@ -0,0 +1,31 @@
+export const DEEPSEEK_V4_FLASH_RATES = { inputPerM: 0.09, outputPerM: 0.18 };
+/** gpt-5.4-mini published API rate (used to price Codex-subscription token
+ *  counts at a representative metered cost — the sub bills flat, but the gate
+ *  measures what a metered user would pay). $0.75/M in, $4.50/M out. */
+export const GPT_5_4_MINI_RATES = { inputPerM: 0.75, outputPerM: 4.5 };
+/** USD for one solve from provider-actual token counts. `output_tokens` already
+ *  INCLUDES `reasoning_tokens` (OpenAI `completion_tokens` semantics, verified
+ *  empirically on the Codex backend), so reasoning is NOT added again — it is
+ *  surfaced separately only for reporting. Likewise `input_tokens` is inclusive
+ *  of `cache_read_tokens`: when a cacheReadPerM rate is supplied, cache-read
+ *  tokens are re-priced at that (discounted) rate and the remaining (fresh)
+ *  input at inputPerM; otherwise all input is inputPerM. */
+export function solveCostUsd(tokens, rates) {
+    const per = (n, rate) => n * rate * 1e-6;
+    const outputCost = per(tokens.outputTokens, rates.outputPerM);
+    if (typeof rates.cacheReadPerM === 'number') {
+        // The fresh/cache split only matters when a discounted cacheReadPerM rate is
+        // priced. Here the `input ⊇ cache_read` invariant must hold — silently
+        // clamping fresh to 0 would hide a bad token export and mis-price the solve,
+        // so fail loud instead. (The no-cache-rate path below is untouched: it prices
+        // all input at inputPerM and never reads cacheReadTokens.)
+        if (tokens.cacheReadTokens > tokens.inputTokens) {
+            throw new Error(`solveCostUsd: cache_read_tokens (${tokens.cacheReadTokens}) > input_tokens (${tokens.inputTokens}) ` +
+                `violates the input ⊇ cache_read invariant when a cacheReadPerM rate is priced`);
+        }
+        const fresh = tokens.inputTokens - tokens.cacheReadTokens;
+        return per(fresh, rates.inputPerM) + per(tokens.cacheReadTokens, rates.cacheReadPerM) + outputCost;
+    }
+    return per(tokens.inputTokens, rates.inputPerM) + outputCost;
+}
+//# sourceMappingURL=cost.js.map

package/dist/pilot/cost.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"cost.js","sourceRoot":"","sources":["../../src/pilot/cost.ts"],"names":[],"mappings":"AAGA,MAAM,CAAC,MAAM,uBAAuB,GAAc,EAAE,SAAS,EAAE,IAAI,EAAE,UAAU,EAAE,IAAI,EAAE,CAAC;AACxF;;wEAEwE;AACxE,MAAM,CAAC,MAAM,kBAAkB,GAAc,EAAE,SAAS,EAAE,IAAI,EAAE,UAAU,EAAE,GAAG,EAAE,CAAC;AAElF;;;;;;4DAM4D;AAC5D,MAAM,UAAU,YAAY,CAAC,MAAmB,EAAE,KAAgB;IAChE,MAAM,GAAG,GAAG,CAAC,CAAS,EAAE,IAAY,EAAU,EAAE,CAAC,CAAC,GAAG,IAAI,GAAG,IAAI,CAAC;IACjE,MAAM,UAAU,GAAG,GAAG,CAAC,MAAM,CAAC,YAAY,EAAE,KAAK,CAAC,UAAU,CAAC,CAAC;IAC9D,IAAI,OAAO,KAAK,CAAC,aAAa,KAAK,QAAQ,EAAE,CAAC;QAC5C,6EAA6E;QAC7E,uEAAuE;QACvE,6EAA6E;QAC7E,8EAA8E;QAC9E,2DAA2D;QAC3D,IAAI,MAAM,CAAC,eAAe,GAAG,MAAM,CAAC,WAAW,EAAE,CAAC;YAChD,MAAM,IAAI,KAAK,CACb,oCAAoC,MAAM,CAAC,eAAe,qBAAqB,MAAM,CAAC,WAAW,IAAI;gBACrG,+EAA+E,CAChF,CAAC;QACJ,CAAC;QACD,MAAM,KAAK,GAAG,MAAM,CAAC,WAAW,GAAG,MAAM,CAAC,eAAe,CAAC;QAC1D,OAAO,GAAG,CAAC,KAAK,EAAE,KAAK,CAAC,SAAS,CAAC,GAAG,GAAG,CAAC,MAAM,CAAC,eAAe,EAAE,KAAK,CAAC,aAAa,CAAC,GAAG,UAAU,CAAC;IACrG,CAAC;IACD,OAAO,GAAG,CAAC,MAAM,CAAC,WAAW,EAAE,KAAK,CAAC,SAAS,CAAC,GAAG,UAAU,CAAC;AAC/D,CAAC"}

package/dist/pilot/instance.d.ts

@@ -0,0 +1,26 @@
+/**
+ * Solve-side pilot instance fields. The grader's `HfRow` (swe-rebench-v2-evaluator/hf-fetcher.ts)
+ * omits `base_commit` and `problem_statement` — this parses those out of the raw HF row.
+ */
+import { type FetchHfWithRetryOptions } from '../harnesses/impls/swe-rebench-v2-evaluator/hf-fetcher.js';
+export interface PilotInstance {
+    instance_id: string;
+    repo: string;
+    base_commit: string;
+    problem_statement: string;
+    /** SWE-rebench-V2 acceptance spec (the API contract the solution must satisfy).
+     *  Optional — fed to BOTH arms when present so it never biases the comparison;
+     *  it is the *spec*, not the hidden test (feeding the test would be cheating). */
+    interface?: string;
+    hf_dataset: string;
+    hf_split: string;
+}
+export declare function parsePilotInstanceRow(row: Record<string, unknown>, ctx: {
+    hf_dataset: string;
+    hf_split: string;
+}): PilotInstance;
+export declare function fetchPilotRawRow(ref: {
+    instance_id: string;
+    hf_dataset: string;
+    hf_split: string;
+}, retryOptions?: FetchHfWithRetryOptions): Promise<Record<string, unknown>>;

package/dist/pilot/instance.js

@@ -0,0 +1,53 @@
+/**
+ * Solve-side pilot instance fields. The grader's `HfRow` (swe-rebench-v2-evaluator/hf-fetcher.ts)
+ * omits `base_commit` and `problem_statement` — this parses those out of the raw HF row.
+ */
+import { fetchHfWithRetry, } from '../harnesses/impls/swe-rebench-v2-evaluator/hf-fetcher.js';
+export function parsePilotInstanceRow(row, ctx) {
+    const s = (k) => {
+        const v = row[k];
+        if (typeof v !== 'string' || !v)
+            throw new Error(`pilot instance row missing '${k}'`);
+        return v;
+    };
+    const iface = typeof row['interface'] === 'string' && row['interface'] ? row['interface'] : undefined;
+    return {
+        instance_id: s('instance_id'),
+        repo: s('repo'),
+        base_commit: s('base_commit'),
+        problem_statement: s('problem_statement'),
+        ...(iface ? { interface: iface } : {}),
+        hf_dataset: ctx.hf_dataset,
+        hf_split: ctx.hf_split,
+    };
+}
+export async function fetchPilotRawRow(ref, retryOptions = {}) {
+    const pageSize = 100;
+    let offset = 0;
+    const maxRows = 1000;
+    while (offset < maxRows) {
+        const url = new URL('https://datasets-server.huggingface.co/rows');
+        url.searchParams.set('dataset', ref.hf_dataset);
+        url.searchParams.set('config', 'default');
+        url.searchParams.set('split', ref.hf_split);
+        url.searchParams.set('offset', String(offset));
+        url.searchParams.set('length', String(pageSize));
+        const res = await fetchHfWithRetry(url.toString(), retryOptions);
+        if (!res.ok) {
+            throw Object.assign(new Error(`HF datasets-server returned ${res.status} for ${ref.hf_dataset}/${ref.hf_split}`), { httpStatus: res.status });
+        }
+        const body = (await res.json());
+        const rows = (body.rows ?? []).map((wrapper) => wrapper.row ?? {});
+        if (rows.length === 0)
+            break;
+        for (const row of rows) {
+            if (row['instance_id'] === ref.instance_id)
+                return row;
+        }
+        if (rows.length < pageSize)
+            break;
+        offset += rows.length;
+    }
+    throw new Error(`instance_id ${ref.instance_id} not found in ${ref.hf_dataset}/${ref.hf_split}`);
+}
+//# sourceMappingURL=instance.js.map

package/dist/pilot/instance.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"instance.js","sourceRoot":"","sources":["../../src/pilot/instance.ts"],"names":[],"mappings":"AAAA;;;GAGG;AACH,OAAO,EACL,gBAAgB,GAEjB,MAAM,2DAA2D,CAAC;AAenE,MAAM,UAAU,qBAAqB,CACnC,GAA4B,EAC5B,GAA6C;IAE7C,MAAM,CAAC,GAAG,CAAC,CAAS,EAAU,EAAE;QAC9B,MAAM,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;QACjB,IAAI,OAAO,CAAC,KAAK,QAAQ,IAAI,CAAC,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,+BAA+B,CAAC,GAAG,CAAC,CAAC;QACtF,OAAO,CAAC,CAAC;IACX,CAAC,CAAC;IACF,MAAM,KAAK,GAAG,OAAO,GAAG,CAAC,WAAW,CAAC,KAAK,QAAQ,IAAI,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC,CAAE,GAAG,CAAC,WAAW,CAAY,CAAC,CAAC,CAAC,SAAS,CAAC;IAClH,OAAO;QACL,WAAW,EAAE,CAAC,CAAC,aAAa,CAAC;QAC7B,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC;QACf,WAAW,EAAE,CAAC,CAAC,aAAa,CAAC;QAC7B,iBAAiB,EAAE,CAAC,CAAC,mBAAmB,CAAC;QACzC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACtC,UAAU,EAAE,GAAG,CAAC,UAAU;QAC1B,QAAQ,EAAE,GAAG,CAAC,QAAQ;KACvB,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,gBAAgB,CACpC,GAAkE,EAClE,eAAwC,EAAE;IAE1C,MAAM,QAAQ,GAAG,GAAG,CAAC;IACrB,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,MAAM,OAAO,GAAG,IAAI,CAAC;IACrB,OAAO,MAAM,GAAG,OAAO,EAAE,CAAC;QACxB,MAAM,GAAG,GAAG,IAAI,GAAG,CAAC,6CAA6C,CAAC,CAAC;QACnE,GAAG,CAAC,YAAY,CAAC,GAAG,CAAC,SAAS,EAAE,GAAG,CAAC,UAAU,CAAC,CAAC;QAChD,GAAG,CAAC,YAAY,CAAC,GAAG,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC;QAC1C,GAAG,CAAC,YAAY,CAAC,GAAG,CAAC,OAAO,EAAE,GAAG,CAAC,QAAQ,CAAC,CAAC;QAC5C,GAAG,CAAC,YAAY,CAAC,GAAG,CAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC;QAC/C,GAAG,CAAC,YAAY,CAAC,GAAG,CAAC,QAAQ,EAAE,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC;QACjD,MAAM,GAAG,GAAG,MAAM,gBAAgB,CAAC,GAAG,CAAC,QAAQ,EAAE,EAAE,YAAY,CAAC,CAAC;QACjE,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;YACZ,MAAM,MAAM,CAAC,MAAM,CACjB,IAAI,KAAK,CAAC,+BAA+B,GAAG,CAAC,MAAM,QAAQ,GAAG,CAAC,UAAU,IAAI,GAAG,CAAC,QAAQ,EAAE,CAAC,EAC5F,EAAE,UAAU,EAAE,GAAG,CAAC,MAAM,EAAE,CAC3B,CAAC;QACJ,CAAC;QACD,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAAwD,CAAC;QACvF,MAAM,IAAI,GAAG,CAAC,IAAI,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,GAAG,IAAI,EAAE,CAAC,CAAC;QACnE,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC;YAAE,MAAM;QAC7B,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;YACvB,IAAI,GAAG,CAAC,aAAa,CAAC,KAAK,GAAG,CAAC,WAAW;gBAAE,OAAO,GAAG,CAAC;QACzD,CAAC;QACD,IAAI,IAAI,CAAC,MAAM,GAAG,QAAQ;YAAE,MAAM;QAClC,MAAM,IAAI,IAAI,CAAC,MAAM,CAAC;IACxB,CAAC;IACD,MAAM,IAAI,KAAK,CAAC,eAAe,GAAG,CAAC,WAAW,iBAAiB,GAAG,CAAC,UAAU,IAAI,GAAG,CAAC,QAAQ,EAAE,CAAC,CAAC;AACnG,CAAC"}

package/dist/pilot/pipeline.d.ts

@@ -0,0 +1,17 @@
+/** Concurrency primitives for the pilot's solve/grade pipeline. Solves fan
+ *  out across instances (network/API-bound); grading stays strictly serial —
+ *  each eval pulls a multi-GB Docker image and runs under amd64 emulation,
+ *  and the eval runner's disk-floor check + prune are not concurrency-safe. */
+/** Run `fn` over every item with at most `limit` calls in flight. A rejection
+ *  from `fn` propagates to the returned promise — callers that need per-item
+ *  containment must catch inside `fn`. */
+export declare function mapWithConcurrency<T>(items: readonly T[], limit: number, fn: (item: T, index: number) => Promise<void>): Promise<void>;
+/** Strictly-serial FIFO job queue. A rejecting job never blocks later jobs;
+ *  `drain()` waits for every pushed job to settle and then rethrows the first
+ *  error. Pushing after `drain()` has settled is unsupported. */
+export declare class SerialTaskQueue {
+    private tail;
+    private readonly errors;
+    push(job: () => Promise<void>): void;
+    drain(): Promise<void>;
+}

package/dist/pilot/pipeline.js

@@ -0,0 +1,35 @@
+/** Concurrency primitives for the pilot's solve/grade pipeline. Solves fan
+ *  out across instances (network/API-bound); grading stays strictly serial —
+ *  each eval pulls a multi-GB Docker image and runs under amd64 emulation,
+ *  and the eval runner's disk-floor check + prune are not concurrency-safe. */
+/** Run `fn` over every item with at most `limit` calls in flight. A rejection
+ *  from `fn` propagates to the returned promise — callers that need per-item
+ *  containment must catch inside `fn`. */
+export async function mapWithConcurrency(items, limit, fn) {
+    const concurrency = Math.max(1, Math.min(limit, items.length));
+    let next = 0;
+    await Promise.all(Array.from({ length: concurrency }, async () => {
+        while (next < items.length) {
+            const index = next++;
+            await fn(items[index], index);
+        }
+    }));
+}
+/** Strictly-serial FIFO job queue. A rejecting job never blocks later jobs;
+ *  `drain()` waits for every pushed job to settle and then rethrows the first
+ *  error. Pushing after `drain()` has settled is unsupported. */
+export class SerialTaskQueue {
+    tail = Promise.resolve();
+    errors = [];
+    push(job) {
+        this.tail = this.tail.then(job).catch((err) => {
+            this.errors.push(err);
+        });
+    }
+    async drain() {
+        await this.tail;
+        if (this.errors.length > 0)
+            throw this.errors[0];
+    }
+}
+//# sourceMappingURL=pipeline.js.map

package/dist/pilot/pipeline.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"pipeline.js","sourceRoot":"","sources":["../../src/pilot/pipeline.ts"],"names":[],"mappings":"AAAA;;;+EAG+E;AAE/E;;0CAE0C;AAC1C,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACtC,KAAmB,EACnB,KAAa,EACb,EAA6C;IAE7C,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC;IAC/D,IAAI,IAAI,GAAG,CAAC,CAAC;IACb,MAAM,OAAO,CAAC,GAAG,CACf,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,WAAW,EAAE,EAAE,KAAK,IAAI,EAAE;QAC7C,OAAO,IAAI,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC;YAC3B,MAAM,KAAK,GAAG,IAAI,EAAE,CAAC;YACrB,MAAM,EAAE,CAAC,KAAK,CAAC,KAAK,CAAE,EAAE,KAAK,CAAC,CAAC;QACjC,CAAC;IACH,CAAC,CAAC,CACH,CAAC;AACJ,CAAC;AAED;;iEAEiE;AACjE,MAAM,OAAO,eAAe;IAClB,IAAI,GAAkB,OAAO,CAAC,OAAO,EAAE,CAAC;IAC/B,MAAM,GAAc,EAAE,CAAC;IAExC,IAAI,CAAC,GAAwB;QAC3B,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,GAAY,EAAE,EAAE;YACrD,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACxB,CAAC,CAAC,CAAC;IACL,CAAC;IAED,KAAK,CAAC,KAAK;QACT,MAAM,IAAI,CAAC,IAAI,CAAC;QAChB,IAAI,IAAI,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC;YAAE,MAAM,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;IACnD,CAAC;CACF"}

package/dist/pilot/repo.d.ts

@@ -0,0 +1,37 @@
+/** Git helpers for the pilot orchestrator, factored out so their fail-loud
+ *  behaviour is unit-testable with an injected runner (the orchestrator's own
+ *  I/O is validated by dry-run + smoke, per the pilot plan). */
+export type CmdRunner = (cmd: string, args: string[], opts?: {
+    cwd?: string;
+}) => Promise<{
+    stdout: string;
+    stderr: string;
+    exitCode: number;
+}>;
+/** A git base-checkout step (clone/checkout) exited non-zero. The base repo is
+ *  unusable, so the INSTANCE is ungradeable — it MUST NOT silently proceed into
+ *  an empty repo where both arms score `passed:false` (that mis-attributes an
+ *  environment failure to the agent and understates the resolve rate). Caught at
+ *  run-pilot's per-instance boundary, which skips the instance and continues. */
+export declare class GitStepError extends Error {
+    readonly step: 'clone' | 'checkout';
+    readonly exitCode: number;
+    readonly detail: string;
+    constructor(step: 'clone' | 'checkout', exitCode: number, detail: string);
+}
+/** Clone `repo` into `baseDir` and check out `baseCommit`, FAIL-LOUD on a
+ *  non-zero exit from either step. A 404'd/renamed repo or an unfetchable commit
+ *  must throw here rather than leave an empty/wrong checkout that both arms then
+ *  "solve" into empty patches. */
+export declare function prepareBaseCheckout(run: CmdRunner, repo: string, baseCommit: string, baseDir: string): Promise<void>;
+/** Recover the working-tree patch, INCLUDING new (untracked) files: `git add -A`
+ *  stages everything, then `git diff --cached` emits the full diff. A plain
+ *  `git diff` misses untracked files, so a fix that ADDS a file would look like
+ *  an empty patch (a spurious not-resolved). */
+export declare function recoverPatch(run: CmdRunner, cwd: string): Promise<string>;
+/**
+ * Create an isolated checkout directory. Durable runs keep live work beneath
+ * `<out>/work` so agent path protections see an operator-owned location;
+ * legacy runs without `--out` retain the OS-temp fallback.
+ */
+export declare function createPilotWorkDir(outDir: string | undefined, prefix: string): Promise<string>;

package/dist/pilot/repo.js

@@ -0,0 +1,65 @@
+/** Git helpers for the pilot orchestrator, factored out so their fail-loud
+ *  behaviour is unit-testable with an injected runner (the orchestrator's own
+ *  I/O is validated by dry-run + smoke, per the pilot plan). */
+import { mkdir, mkdtemp } from 'node:fs/promises';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import { stripTestPathHunks } from '../harnesses/impls/learner/restoration-patch.js';
+/** A git base-checkout step (clone/checkout) exited non-zero. The base repo is
+ *  unusable, so the INSTANCE is ungradeable — it MUST NOT silently proceed into
+ *  an empty repo where both arms score `passed:false` (that mis-attributes an
+ *  environment failure to the agent and understates the resolve rate). Caught at
+ *  run-pilot's per-instance boundary, which skips the instance and continues. */
+export class GitStepError extends Error {
+    step;
+    exitCode;
+    detail;
+    constructor(step, exitCode, detail) {
+        super(`git ${step} failed (exit ${exitCode})${detail ? `: ${detail}` : ''}`);
+        this.step = step;
+        this.exitCode = exitCode;
+        this.detail = detail;
+        this.name = 'GitStepError';
+    }
+}
+/** Clone `repo` into `baseDir` and check out `baseCommit`, FAIL-LOUD on a
+ *  non-zero exit from either step. A 404'd/renamed repo or an unfetchable commit
+ *  must throw here rather than leave an empty/wrong checkout that both arms then
+ *  "solve" into empty patches. */
+export async function prepareBaseCheckout(run, repo, baseCommit, baseDir) {
+    const clone = await run('git', ['clone', `https://github.com/${repo}.git`, baseDir]);
+    if (clone.exitCode !== 0)
+        throw new GitStepError('clone', clone.exitCode, clone.stderr.trim().slice(0, 300));
+    let checkout = await run('git', ['checkout', baseCommit], { cwd: baseDir });
+    if (checkout.exitCode !== 0) {
+        // Upstream history rewrites leave dataset base commits unreachable from
+        // any ref — absent from a plain clone ("unable to read tree") yet still
+        // served by GitHub to an explicit fetch-by-sha. Fetch, then retry once.
+        const fetch = await run('git', ['fetch', 'origin', baseCommit], { cwd: baseDir });
+        if (fetch.exitCode === 0) {
+            checkout = await run('git', ['checkout', baseCommit], { cwd: baseDir });
+        }
+        if (checkout.exitCode !== 0)
+            throw new GitStepError('checkout', checkout.exitCode, checkout.stderr.trim().slice(0, 300));
+    }
+}
+/** Recover the working-tree patch, INCLUDING new (untracked) files: `git add -A`
+ *  stages everything, then `git diff --cached` emits the full diff. A plain
+ *  `git diff` misses untracked files, so a fix that ADDS a file would look like
+ *  an empty patch (a spurious not-resolved). */
+export async function recoverPatch(run, cwd) {
+    await run('git', ['add', '-A'], { cwd });
+    const diff = await run('git', ['diff', '--cached'], { cwd });
+    return stripTestPathHunks(diff.stdout);
+}
+/**
+ * Create an isolated checkout directory. Durable runs keep live work beneath
+ * `<out>/work` so agent path protections see an operator-owned location;
+ * legacy runs without `--out` retain the OS-temp fallback.
+ */
+export async function createPilotWorkDir(outDir, prefix) {
+    const root = outDir ? join(outDir, 'work') : tmpdir();
+    await mkdir(root, { recursive: true });
+    return mkdtemp(join(root, prefix));
+}
+//# sourceMappingURL=repo.js.map

package/dist/pilot/repo.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"repo.js","sourceRoot":"","sources":["../../src/pilot/repo.ts"],"names":[],"mappings":"AAAA;;gEAEgE;AAEhE,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,kBAAkB,CAAC;AAClD,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,kBAAkB,EAAE,MAAM,iDAAiD,CAAC;AAQrF;;;;iFAIiF;AACjF,MAAM,OAAO,YAAa,SAAQ,KAAK;IAE1B;IACA;IACA;IAHX,YACW,IAA0B,EAC1B,QAAgB,EAChB,MAAc;QAEvB,KAAK,CAAC,OAAO,IAAI,iBAAiB,QAAQ,IAAI,MAAM,CAAC,CAAC,CAAC,KAAK,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QAJpE,SAAI,GAAJ,IAAI,CAAsB;QAC1B,aAAQ,GAAR,QAAQ,CAAQ;QAChB,WAAM,GAAN,MAAM,CAAQ;QAGvB,IAAI,CAAC,IAAI,GAAG,cAAc,CAAC;IAC7B,CAAC;CACF;AAED;;;kCAGkC;AAClC,MAAM,CAAC,KAAK,UAAU,mBAAmB,CACvC,GAAc,EACd,IAAY,EACZ,UAAkB,EAClB,OAAe;IAEf,MAAM,KAAK,GAAG,MAAM,GAAG,CAAC,KAAK,EAAE,CAAC,OAAO,EAAE,sBAAsB,IAAI,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC;IACrF,IAAI,KAAK,CAAC,QAAQ,KAAK,CAAC;QAAE,MAAM,IAAI,YAAY,CAAC,OAAO,EAAE,KAAK,CAAC,QAAQ,EAAE,KAAK,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;IAC7G,IAAI,QAAQ,GAAG,MAAM,GAAG,CAAC,KAAK,EAAE,CAAC,UAAU,EAAE,UAAU,CAAC,EAAE,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC,CAAC;IAC5E,IAAI,QAAQ,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;QAC5B,wEAAwE;QACxE,wEAAwE;QACxE,wEAAwE;QACxE,MAAM,KAAK,GAAG,MAAM,GAAG,CAAC,KAAK,EAAE,CAAC,OAAO,EAAE,QAAQ,EAAE,UAAU,CAAC,EAAE,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC,CAAC;QAClF,IAAI,KAAK,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;YACzB,QAAQ,GAAG,MAAM,GAAG,CAAC,KAAK,EAAE,CAAC,UAAU,EAAE,UAAU,CAAC,EAAE,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC,CAAC;QAC1E,CAAC;QACD,IAAI,QAAQ,CAAC,QAAQ,KAAK,CAAC;YAAE,MAAM,IAAI,YAAY,CAAC,UAAU,EAAE,QAAQ,CAAC,QAAQ,EAAE,QAAQ,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;IAC3H,CAAC;AACH,CAAC;AAED;;;gDAGgD;AAChD,MAAM,CAAC,KAAK,UAAU,YAAY,CAAC,GAAc,EAAE,GAAW;IAC5D,MAAM,GAAG,CAAC,KAAK,EAAE,CAAC,KAAK,EAAE,IAAI,CAAC,EAAE,EAAE,GAAG,EAAE,CAAC,CAAC;IACzC,MAAM,IAAI,GAAG,MAAM,GAAG,CAAC,KAAK,EAAE,CAAC,MAAM,EAAE,UAAU,CAAC,EAAE,EAAE,GAAG,EAAE,CAAC,CAAC;IAC7D,OAAO,kBAAkB,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;AACzC,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CAAC,MAA0B,EAAE,MAAc;IACjF,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;IACtD,MAAM,KAAK,CAAC,IAAI,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACvC,OAAO,OAAO,CAAC,IAAI,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC,CAAC;AACrC,CAAC"}

package/dist/pilot/resume.d.ts

@@ -0,0 +1,101 @@
+import type { HfRow } from '../harnesses/impls/swe-rebench-v2-evaluator/index.js';
+import type { PilotInstance } from './instance.js';
+import type { SolveTokens } from './solve.js';
+import type { PilotReport, SolveOutcome } from './tally.js';
+export interface PilotInstanceRef {
+    instance_id: string;
+    hf_dataset: string;
+    hf_split: string;
+}
+export interface PilotArmConfig {
+    name: string;
+    skills: string[];
+    jinnAgentHome?: string;
+}
+export interface PilotSemanticArmConfig {
+    name: string;
+    skills: string[];
+}
+export interface PilotSemanticConfig {
+    instances: PilotInstanceRef[];
+    repeats: number;
+    arms: PilotSemanticArmConfig[];
+    maxTurns: number;
+    gradeTimeoutMs: number;
+    /** Dry-run and real records must never share a durable store: dry-run
+     *  writes synthetic `graded` records that would silently pollute a real
+     *  report. Absent (legacy manifests) normalizes to 'real', so a dry run
+     *  against a pre-mode store fails closed. */
+    mode?: 'dry-run' | 'real';
+    /** Whether every arm's solve prompt carries the neutral "check your
+     *  available skills" line. Semantic: it changes solve behavior, so nudged
+     *  and un-nudged attempts must not share a store. Absent (legacy
+     *  manifests) normalizes to false. */
+    skillsNudge?: boolean;
+    provider?: string;
+    model?: string;
+    taskSource?: string;
+    slateHash?: string;
+}
+interface LegacyPilotSemanticConfig extends Omit<PilotSemanticConfig, 'arms'> {
+    arms?: PilotSemanticArmConfig[];
+    skills?: string[];
+    armBJinnAgentHome?: string;
+}
+export interface PilotManifest {
+    schema: 'jinn.pilot.manifest.v1';
+    generatedAt: string;
+    semanticConfig: PilotSemanticConfig;
+    attemptCount: number;
+}
+export interface FrozenPilotInstance {
+    ref: PilotInstanceRef;
+    instance: PilotInstance;
+    hfRow: HfRow;
+}
+export interface PilotInstancesFile {
+    schema: 'jinn.pilot.instances.v1';
+    instances: FrozenPilotInstance[];
+}
+export interface PilotAttemptSpec {
+    instance_id: string;
+    arm: string;
+    repeat: number;
+}
+export type PilotAttemptStatus = 'graded' | 'ungradeable' | 'solve-error' | 'grade-error';
+export interface PilotAttemptRecord extends PilotAttemptSpec {
+    schema: 'jinn.pilot.attempt.v1';
+    status: PilotAttemptStatus;
+    passed: boolean | null;
+    costUsd: number;
+    tokens?: SolveTokens;
+    sessionId?: string;
+    patchRelPath?: string;
+    error?: string;
+    tokenError?: string;
+}
+export declare function attemptKey(spec: PilotAttemptSpec): string;
+export declare function attemptRecordFileName(spec: PilotAttemptSpec): string;
+export declare function buildAttemptSpecs(config: PilotSemanticConfig): PilotAttemptSpec[];
+export declare function normalizePilotSemanticConfig(config: PilotSemanticConfig | LegacyPilotSemanticConfig): PilotSemanticConfig;
+export declare function buildPilotManifest(config: PilotSemanticConfig, opts?: {
+    generatedAt?: string;
+}): PilotManifest;
+export declare function instanceRefKey(ref: PilotInstanceRef): string;
+export declare function assertCompatiblePilotManifest(manifest: PilotManifest, requested: PilotSemanticConfig): void;
+export declare function writePilotManifest(outDir: string, manifest: PilotManifest): void;
+export declare function loadPilotManifest(outDir: string): PilotManifest | null;
+export declare function writeFrozenInstances(outDir: string, instances: FrozenPilotInstance[]): void;
+export declare function loadFrozenInstances(outDir: string): FrozenPilotInstance[] | null;
+export declare function clearPilotOutput(outDir: string): void;
+export declare function writePatch(outDir: string, spec: PilotAttemptSpec, patch: string): string;
+export declare function writeAttemptRecord(outDir: string, record: PilotAttemptRecord): void;
+export declare function loadAttemptRecords(outDir: string): Map<string, PilotAttemptRecord>;
+export declare function selectRunnableAttempts(specs: PilotAttemptSpec[], records: Map<string, PilotAttemptRecord>, opts: {
+    retryErrors: boolean;
+    maxNewSolves: number;
+}): PilotAttemptSpec[];
+export declare function recordsToOutcomes(records: Iterable<PilotAttemptRecord>): SolveOutcome[];
+export declare function orderedRecordsForSpecs(specs: PilotAttemptSpec[], records: Map<string, PilotAttemptRecord>): PilotAttemptRecord[];
+export declare function writePilotReport(outDir: string, report: PilotReport, outcomes: SolveOutcome[]): void;
+export {};