dsh-continual-evolve 0.11.0 → 0.12.1

package/lib/consolidate.js

@@ -1,180 +0,0 @@
-import { ARCHIVED_AT_KEY, CONFLICT_HINT_KEY, MERGED_FROM_KEY, isArchived } from "./types.js";
-import { getUsageCount } from "./usage.js";
-/** Zero-use entries at least this old are stale candidates (30d, matching the injection recency half-life scale). */
-export const STALE_MIN_AGE_MS = 30 * 24 * 60 * 60 * 1000;
-const KINDS = ["prompt", "memory", "skill", "subagent"];
-/**
- * Parse one conflictHint metadata value. Returns undefined for anything that
- * is not exactly `<kind>:<id>:<score>` with a known kind and a score in
- * [0, 1] — foreign or legacy junk must never crash the scan.
- */
-export function parseConflictHint(value) {
-    if (typeof value !== "string") {
-        return undefined;
-    }
-    const parts = value.split(":");
-    if (parts.length !== 3) {
-        return undefined;
-    }
-    const [kind, id, rawScore] = parts;
-    if (!kind || !id || !rawScore || !KINDS.includes(kind)) {
-        return undefined;
-    }
-    const score = Number(rawScore);
-    if (!Number.isFinite(score) || score < 0 || score > 1) {
-        return undefined;
-    }
-    return { kind: kind, id, score };
-}
-/**
- * Entries stamped by the write-time conflict guard whose target still exists
- * and is still active. The hinted (newer) entry is the archive candidate —
- * the pointed-to original stays as the live copy.
- */
-export function findConflictPairs(state) {
-    const candidates = [];
-    for (const kind of KINDS) {
-        for (const entry of Object.values(state.entries[kind])) {
-            if (isArchived(entry))
-                continue;
-            const hint = parseConflictHint(entry.metadata[CONFLICT_HINT_KEY]);
-            if (!hint || hint.kind !== kind)
-                continue;
-            const target = state.entries[hint.kind]?.[hint.id];
-            if (!target || isArchived(target))
-                continue;
-            candidates.push({
-                kind,
-                id: entry.id,
-                title: entry.title,
-                reason: `near-duplicate of ${hint.id} 「${target.title}」 (${Math.round(hint.score * 100)}%) — keep the original`,
-                mergeInto: { kind: hint.kind, id: hint.id, title: target.title },
-            });
-        }
-    }
-    return candidates;
-}
-/**
- * Active global entries never injected in any session and untouched for at
- * least `minAgeMs`: prime staleness candidates. Operates on whatever state
- * it is handed (callers pass the global store).
- */
-export function findStaleEntries(state, store, now, minAgeMs = STALE_MIN_AGE_MS) {
-    const candidates = [];
-    for (const kind of KINDS) {
-        for (const entry of Object.values(state.entries[kind])) {
-            if (isArchived(entry))
-                continue;
-            if (getUsageCount(store, kind, entry.id) !== 0)
-                continue;
-            const updatedAt = Date.parse(entry.updated_at);
-            if (Number.isNaN(updatedAt) || now - updatedAt < minAgeMs)
-                continue;
-            candidates.push({
-                kind,
-                id: entry.id,
-                title: entry.title,
-                reason: `0 injections since ${entry.updated_at.slice(0, 10)} (stale)`,
-            });
-        }
-    }
-    return candidates;
-}
-/**
- * Append a source entry's content to a survivor with an attributed divider,
- * so the survivor's reader can see which part came from which entry.
- */
-export function mergeContent(target, source, sourceRef, dateIso) {
-    const body = source.trim();
-    if (body.length === 0) {
-        return target;
-    }
-    return `${target.trimEnd()}\n\n---\n[Merged from ${sourceRef} on ${dateIso.slice(0, 10)} — near-duplicate consolidated]\n${body}`;
-}
-/** Existing `<kind>:<id>` strings of a target's {@link MERGED_FROM_KEY} trail. */
-function existingMergedFrom(metadata) {
-    const value = metadata[MERGED_FROM_KEY];
-    return Array.isArray(value) ? value.filter((item) => typeof item === "string") : [];
-}
-/**
- * Merge both scans (conflict reason wins on overlap), deduped by kind:id,
- * and build the batch archive edits. Each edit preserves the entry's full
- * content and existing metadata — only ARCHIVED_AT_KEY is added — so
- * `/evolve unarchive` restores everything intact.
- *
- * With `opts.mergeDuplicates`, conflict-pair candidates additionally merge
- * their content INTO the pointed-to survivor (an attributed `mergeContent`
- * section + a `mergedFrom` provenance stamp) before archiving — the #11
- * candidate-c tier the 2026-08-28 ecosystem review promoted. Per-target
- * accumulation composes multiple hints into one survivor (update edits
- * replace content/metadata wholesale, so parallel merge edits would clobber
- * each other).
- *
- * @param now Epoch ms used for the archivedAt stamp (injected for tests).
- */
-export function planConsolidation(state, store, now = Date.now(), opts) {
-    const byKey = new Map();
-    for (const candidate of [...findConflictPairs(state), ...findStaleEntries(state, store, now, opts?.minAgeMs)]) {
-        const key = `${candidate.kind}:${candidate.id}`;
-        if (!byKey.has(key)) {
-            byKey.set(key, candidate);
-        }
-    }
-    const candidates = [...byKey.values()];
-    const dateIso = new Date(now).toISOString();
-    // A merge survivor must not itself be an archive candidate in the same
-    // batch: its archive edit is built from the ORIGINAL state snapshot, so it
-    // would overwrite the merge edit's content/mergedFrom wholesale and then
-    // archive the survivor (review audit 2026-08-28 S2). Survivors stay live;
-    // their merge lands, the source archives.
-    const mergeTargets = new Set(candidates.filter((candidate) => candidate.mergeInto).map((candidate) => `${candidate.mergeInto.kind}:${candidate.mergeInto.id}`));
-    const archiveCandidates = candidates.filter((candidate) => !mergeTargets.has(`${candidate.kind}:${candidate.id}`));
-    const edits = archiveCandidates.map((candidate) => {
-        const entry = state.entries[candidate.kind][candidate.id];
-        const metadata = { ...entry?.metadata, [ARCHIVED_AT_KEY]: dateIso };
-        return {
-            action: "update",
-            kind: candidate.kind,
-            id: candidate.id,
-            title: candidate.title,
-            content: entry?.content ?? "",
-            metadata,
-        };
-    });
-    if (opts?.mergeDuplicates) {
-        const merges = new Map();
-        for (const candidate of candidates) {
-            if (!candidate.mergeInto)
-                continue;
-            const source = state.entries[candidate.kind][candidate.id];
-            const target = state.entries[candidate.mergeInto.kind]?.[candidate.mergeInto.id];
-            if (!source || !target)
-                continue;
-            const key = `${candidate.mergeInto.kind}:${candidate.mergeInto.id}`;
-            const acc = merges.get(key) ??
-                {
-                    kind: candidate.mergeInto.kind,
-                    id: candidate.mergeInto.id,
-                    title: target.title,
-                    content: target.content,
-                    mergedFrom: existingMergedFrom(target.metadata),
-                    metadata: target.metadata,
-                };
-            acc.content = mergeContent(acc.content, source.content ?? "", `${candidate.kind}:${candidate.id}`, dateIso);
-            acc.mergedFrom.push(`${candidate.kind}:${candidate.id}`);
-            merges.set(key, acc);
-        }
-        for (const acc of merges.values()) {
-            edits.unshift({
-                action: "update",
-                kind: acc.kind,
-                id: acc.id,
-                title: acc.title,
-                content: acc.content,
-                metadata: { ...acc.metadata, [MERGED_FROM_KEY]: acc.mergedFrom },
-            });
-        }
-    }
-    return { candidates, edits };
-}
-//# sourceMappingURL=consolidate.js.map

package/lib/copy.d.ts

@@ -1,37 +0,0 @@
-/**
- * Backend user-visible copy for the evolution confirmation dialogs, in both
- * shipped languages. Labels stay literal per language (the parsers match
- * them); guidance lives in `description`.
- *
- * Language is never hardcoded at a call site: every builder takes an
- * explicit `lang`, and dialog entry points resolve it per call
- * (explicit config upstream, else the durable DSH client preference,
- * else `en`) when the caller does not supply one.
- */
-import type { RecordLanguage } from "./record-language.js";
-export interface ConfirmOption {
-    label: string;
-    description: string;
-}
-export interface ConfirmCopy {
-    question: string;
-    options: ConfirmOption[];
-}
-/** Maximum edit-summary chars carried into a dialog; longer text truncates. */
-export declare const COPY_WHAT_MAX_CHARS = 300;
-/**
- * Persistent-scope write approval (`approve-global-evolve`).
- *
- * @param scope Target persistent scope.
- * @param what Edit summary shown in the dialog body.
- * @param lang Dialog language.
- */
-export declare function approvalCopy(scope: "project" | "global", what: string, lang: RecordLanguage): ConfirmCopy;
-/**
- * Single-entry wrap-up archive confirm (`evolve-wrapup-archive-review`).
- *
- * @param title Entry title shown in the headline.
- * @param lang Dialog language.
- */
-export declare function wrapupArchiveCopy(title: string, lang: RecordLanguage): ConfirmCopy;
-//# sourceMappingURL=copy.d.ts.map

package/lib/copy.js

@@ -1,58 +0,0 @@
-/** Maximum edit-summary chars carried into a dialog; longer text truncates. */
-export const COPY_WHAT_MAX_CHARS = 300;
-function compactWhat(what) {
-    return what.length > COPY_WHAT_MAX_CHARS ? `${what.slice(0, COPY_WHAT_MAX_CHARS)}…` : what;
-}
-/**
- * Persistent-scope write approval (`approve-global-evolve`).
- *
- * @param scope Target persistent scope.
- * @param what Edit summary shown in the dialog body.
- * @param lang Dialog language.
- */
-export function approvalCopy(scope, what, lang) {
-    const body = compactWhat(what);
-    if (lang === "zh") {
-        const storeLabel = scope === "project" ? "本项目跨会话 store" : "跨会话全局 store";
-        return {
-            question: `写入${storeLabel}？\n${body}\n${scope === "project" ? "影响：仅本项目会话可见，可回滚。" : "影响：所有会话可见，可回滚。"}`,
-            options: [
-                { label: "批准", description: scope === "project" ? "写入，仅本项目会话可见" : "写入，所有会话可见" },
-                { label: "拒绝", description: "不写入，本次跳过" },
-            ],
-        };
-    }
-    const storeLabel = scope === "project" ? "project cross-session store" : "global cross-session store";
-    return {
-        question: `Write to the ${storeLabel}?\n${body}\n${scope === "project" ? "Impact: visible to this project's sessions only, reversible." : "Impact: visible to all sessions, reversible."}`,
-        options: [
-            { label: "Approve", description: scope === "project" ? "Write, visible to this project only" : "Write, visible to all sessions" },
-            { label: "Decline", description: "Skip this write" },
-        ],
-    };
-}
-/**
- * Single-entry wrap-up archive confirm (`evolve-wrapup-archive-review`).
- *
- * @param title Entry title shown in the headline.
- * @param lang Dialog language.
- */
-export function wrapupArchiveCopy(title, lang) {
-    if (lang === "zh") {
-        return {
-            question: `wrapup 确认归档：条目「${title}」\n未被全局覆盖且源自真实对话。归档后不再注入，数据保留、可恢复。`,
-            options: [
-                { label: "归档", description: "隐藏但可恢复" },
-                { label: "保留", description: "继续注入" },
-            ],
-        };
-    }
-    return {
-        question: `Wrap-up archive confirm: entry "${title}"\nNot covered globally and distilled from real conversation. Archiving stops injection; data stays restorable.`,
-        options: [
-            { label: "Archive", description: "Hide but restorable" },
-            { label: "Keep", description: "Keep injecting" },
-        ],
-    };
-}
-//# sourceMappingURL=copy.js.map

package/lib/declines.d.ts

@@ -1,103 +0,0 @@
-/** Cap on remembered declines; oldest-first pruning keeps the file small. */
-export declare const MAX_DECLINED_MEMORY_LEDGER = 50;
-/** Cap on stored content tokens per entry; enough for coverage, never prose. */
-export declare const MAX_DECLINED_ENTRY_TOKENS = 128;
-/** Minimum coverage of an entry's tokens by the checkpoint to pre-suppress. */
-export declare const DECLINED_PRECHECK_COVERAGE = 0.85;
-/** Minimum overlapping tokens; tiny entries must never match vacuously. */
-export declare const DECLINED_PRECHECK_MIN_TOKENS = 5;
-/** One declined project/global batch, keyed by its content fingerprint. */
-export interface DeclinedMemoryBatch {
-    /** Persistent scope that was declined. Local never asks, so never lands here. */
-    scope: "project" | "global";
-    /** sha256-16 over the normalized batch (same cut as evaluate.ts/project.ts). */
-    fingerprint: string;
-    /** First edit title (audit readability only, never matched on). */
-    title: string;
-    /** Deduplicated content tokens for checkpoint containment checks. */
-    tokens: string[];
-    /** ISO timestamp of the human rejection. */
-    declinedAt: string;
-}
-/** Minimal edit shape needed for fingerprinting (avoids importing the agent). */
-export interface FingerprintableEdit {
-    action: string;
-    id?: string;
-    title?: string;
-    content?: string;
-}
-/** Full path of the decline ledger file. */
-export declare function declinedMemoryPath(baseDir: string): string;
-/**
- * Fingerprint one scope batch. Normalization is whitespace-only on purpose:
- * case folding could conflate distinct facts (e.g. env var names), and a
- * missed match only costs one more popup (fail-closed).
- *
- * @param scope - persistent scope the batch targets.
- * @param edits - the scope batch in proposal order (order is significant).
- * @returns 16-char hex fingerprint.
- */
-export declare function fingerprintMemoryBatch(scope: "project" | "global", edits: readonly FingerprintableEdit[]): string;
-/**
- * Load the decline ledger. Missing or malformed files read as empty so a
- * broken ledger degrades to the status quo (ask anyway), never to a crash
- * on the approval path.
- *
- * @param baseDir - engine base directory.
- * @returns validated ledger entries (possibly empty).
- */
-export declare function loadDeclinedMemory(baseDir: string): DeclinedMemoryBatch[];
-/**
- * Whether this exact batch was declined before.
- *
- * @param entries - loaded ledger.
- * @param scope - persistent scope of the pending batch.
- * @param fingerprint - fingerprint of the pending batch.
- */
-export declare function isDeclinedRepeat(entries: readonly DeclinedMemoryBatch[], scope: "project" | "global", fingerprint: string): boolean;
-/**
- * Record one human decline, pruning oldest-first past the cap. The batch
- * edits supply fingerprint material, match tokens, and the secret screen:
- * a secret-shaped decline is still declined, but never lands in the ledger
- * (it was never engine-screened, so it must not be persisted anywhere).
- *
- * @param baseDir - engine base directory.
- * @param scope - persistent scope that was declined.
- * @param edits - the declined scope batch in proposal order.
- * @param title - first edit title, kept for audit readability.
- * @returns the ledger file path, or undefined when skipped by the secret screen.
- * @throws On filesystem write failure; callers on the approval path should
- * contain it (suppression is an optimization, never load-bearing).
- */
-export declare function recordDeclinedMemory(baseDir: string, scope: "project" | "global", edits: readonly FingerprintableEdit[], title: string): string | undefined;
-/**
- * Deduplicated content tokens of one batch, capped. Shared CJK-aware
- * tokenizer with the manifest ranking, so both sides of the containment
- * check speak the same token language.
- *
- * @param edits - batch edits.
- * @returns at most MAX_DECLINED_ENTRY_TOKENS distinct tokens.
- */
-export declare function declinedContentTokens(edits: readonly FingerprintableEdit[]): string[];
-/** One checkpoint containment hit: which entry matched and how strongly. */
-export interface DeclinedCheckpointHit {
-    /** Ledger entry whose content the checkpoint re-covers. */
-    entry: DeclinedMemoryBatch;
-    /** Fraction of the entry's tokens present in the checkpoint. */
-    coverage: number;
-    /** Overlapping token count. */
-    hits: number;
-}
-/**
- * Match a pending checkpoint against declined content (entry ⊆ checkpoint).
- * Conservative by design: high coverage plus an absolute token floor, so
- * tiny or vague declines can never suppress anything.
- *
- * @param entries - loaded ledger.
- * @param checkpointTokens - tokens of the pending checkpoint trajectory.
- * @returns the strongest hit, or undefined when nothing clears the floor.
- */
-/** Convenience: load the declined ledger and match one trajectory slice. */
-export declare function matchDeclinedTrajectory(baseDir: string, trajectory: string): DeclinedCheckpointHit | undefined;
-export declare function matchDeclinedCheckpoint(entries: readonly DeclinedMemoryBatch[], checkpointTokens: readonly string[]): DeclinedCheckpointHit | undefined;
-//# sourceMappingURL=declines.d.ts.map

package/lib/declines.js

@@ -1,179 +0,0 @@
-/**
- * Cross-session ledger of declined memory proposals (the decline loop).
- *
- * In-memory `GateState.memoryDecisions` dies on restart and its decision
- * cursor is per-checkpoint, so a declined project/global batch is re-asked
- * in the next session. This file persists batch fingerprints plus content
- * tokens under `<baseDir>/evolve/declined-memory.json`: exact batches are
- * suppressed at approval time, and checkpoints overlapping declined content
- * are suppressed before the agent is even invoked. Token matching is
- * containment (entry ⊆ checkpoint) with a conservative floor; rephrasing
- * below the floor still asks (documented limitation, not silent loss).
- */
-import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
-import { createHash } from "node:crypto";
-import { join } from "node:path";
-import { secretLeakReason } from "./promotion.js";
-import { tokenize } from "./search.js";
-import { EVOLVE_DIR } from "./store.js";
-/** Cap on remembered declines; oldest-first pruning keeps the file small. */
-export const MAX_DECLINED_MEMORY_LEDGER = 50;
-/** Cap on stored content tokens per entry; enough for coverage, never prose. */
-export const MAX_DECLINED_ENTRY_TOKENS = 128;
-/** Minimum coverage of an entry's tokens by the checkpoint to pre-suppress. */
-export const DECLINED_PRECHECK_COVERAGE = 0.85;
-/** Minimum overlapping tokens; tiny entries must never match vacuously. */
-export const DECLINED_PRECHECK_MIN_TOKENS = 5;
-const DECLINED_MEMORY_FILE = "declined-memory.json";
-/** Full path of the decline ledger file. */
-export function declinedMemoryPath(baseDir) {
-    return join(baseDir, EVOLVE_DIR, DECLINED_MEMORY_FILE);
-}
-/**
- * Fingerprint one scope batch. Normalization is whitespace-only on purpose:
- * case folding could conflate distinct facts (e.g. env var names), and a
- * missed match only costs one more popup (fail-closed).
- *
- * @param scope - persistent scope the batch targets.
- * @param edits - the scope batch in proposal order (order is significant).
- * @returns 16-char hex fingerprint.
- */
-export function fingerprintMemoryBatch(scope, edits) {
-    const parts = [scope];
-    for (const edit of edits) {
-        parts.push([edit.action, edit.id ?? "", edit.title ?? "", edit.content ?? ""].join("|"));
-    }
-    const material = parts.join("\n").replace(/\s+/g, " ").trim();
-    return createHash("sha256").update(material, "utf8").digest("hex").slice(0, 16);
-}
-/**
- * Load the decline ledger. Missing or malformed files read as empty so a
- * broken ledger degrades to the status quo (ask anyway), never to a crash
- * on the approval path.
- *
- * @param baseDir - engine base directory.
- * @returns validated ledger entries (possibly empty).
- */
-export function loadDeclinedMemory(baseDir) {
-    const path = declinedMemoryPath(baseDir);
-    if (!existsSync(path))
-        return [];
-    try {
-        const raw = JSON.parse(readFileSync(path, "utf8"));
-        if (!Array.isArray(raw))
-            return [];
-        return raw
-            .filter((entry) => {
-            if (typeof entry !== "object" || entry === null)
-                return false;
-            const candidate = entry;
-            return ((candidate.scope === "project" || candidate.scope === "global") &&
-                typeof candidate.fingerprint === "string");
-        })
-            .map((entry) => ({
-            ...entry,
-            tokens: Array.isArray(entry.tokens) ? entry.tokens.filter((token) => typeof token === "string") : [],
-        }));
-    }
-    catch {
-        return [];
-    }
-}
-/**
- * Whether this exact batch was declined before.
- *
- * @param entries - loaded ledger.
- * @param scope - persistent scope of the pending batch.
- * @param fingerprint - fingerprint of the pending batch.
- */
-export function isDeclinedRepeat(entries, scope, fingerprint) {
-    return entries.some((entry) => entry.scope === scope && entry.fingerprint === fingerprint);
-}
-/**
- * Record one human decline, pruning oldest-first past the cap. The batch
- * edits supply fingerprint material, match tokens, and the secret screen:
- * a secret-shaped decline is still declined, but never lands in the ledger
- * (it was never engine-screened, so it must not be persisted anywhere).
- *
- * @param baseDir - engine base directory.
- * @param scope - persistent scope that was declined.
- * @param edits - the declined scope batch in proposal order.
- * @param title - first edit title, kept for audit readability.
- * @returns the ledger file path, or undefined when skipped by the secret screen.
- * @throws On filesystem write failure; callers on the approval path should
- * contain it (suppression is an optimization, never load-bearing).
- */
-export function recordDeclinedMemory(baseDir, scope, edits, title) {
-    const screenable = edits.map((edit) => `${edit.title ?? ""}\n${edit.content ?? ""}`).join("\n");
-    if (secretLeakReason(screenable) !== undefined)
-        return undefined;
-    const entries = loadDeclinedMemory(baseDir);
-    entries.push({
-        scope,
-        fingerprint: fingerprintMemoryBatch(scope, edits),
-        title,
-        tokens: declinedContentTokens(edits),
-        declinedAt: new Date().toISOString(),
-    });
-    entries.sort((left, right) => (left.declinedAt < right.declinedAt ? -1 : 1));
-    while (entries.length > MAX_DECLINED_MEMORY_LEDGER)
-        entries.shift();
-    const path = declinedMemoryPath(baseDir);
-    mkdirSync(join(baseDir, EVOLVE_DIR), { recursive: true });
-    writeFileSync(path, `${JSON.stringify(entries, null, 2)}\n`, "utf8");
-    return path;
-}
-/**
- * Deduplicated content tokens of one batch, capped. Shared CJK-aware
- * tokenizer with the manifest ranking, so both sides of the containment
- * check speak the same token language.
- *
- * @param edits - batch edits.
- * @returns at most MAX_DECLINED_ENTRY_TOKENS distinct tokens.
- */
-export function declinedContentTokens(edits) {
-    const seen = new Set();
-    for (const edit of edits) {
-        for (const token of tokenize(`${edit.title ?? ""} ${edit.content ?? ""}`)) {
-            seen.add(token);
-            if (seen.size >= MAX_DECLINED_ENTRY_TOKENS)
-                return [...seen];
-        }
-    }
-    return [...seen];
-}
-/**
- * Match a pending checkpoint against declined content (entry ⊆ checkpoint).
- * Conservative by design: high coverage plus an absolute token floor, so
- * tiny or vague declines can never suppress anything.
- *
- * @param entries - loaded ledger.
- * @param checkpointTokens - tokens of the pending checkpoint trajectory.
- * @returns the strongest hit, or undefined when nothing clears the floor.
- */
-/** Convenience: load the declined ledger and match one trajectory slice. */
-export function matchDeclinedTrajectory(baseDir, trajectory) {
-    return matchDeclinedCheckpoint(loadDeclinedMemory(baseDir), tokenize(trajectory));
-}
-export function matchDeclinedCheckpoint(entries, checkpointTokens) {
-    const pending = new Set(checkpointTokens);
-    let best;
-    for (const entry of entries) {
-        if (entry.tokens.length === 0)
-            continue;
-        let hits = 0;
-        for (const token of entry.tokens) {
-            if (pending.has(token))
-                hits += 1;
-        }
-        if (hits < DECLINED_PRECHECK_MIN_TOKENS)
-            continue;
-        const coverage = hits / entry.tokens.length;
-        if (coverage < DECLINED_PRECHECK_COVERAGE)
-            continue;
-        if (best === undefined || coverage > best.coverage)
-            best = { entry, coverage, hits };
-    }
-    return best;
-}
-//# sourceMappingURL=declines.js.map

package/lib/evaluate.d.ts

@@ -1,74 +0,0 @@
-/**
- * The evaluation matrix runner: executes every case × run as a TWO-STAGE
- * structured-output subagent pair, with the provider/model frozen to the
- * calling agent's own route.
- *
- * Stage 1 — executor: a fresh child performs the case task with its tools
- * and records CONCRETE EVIDENCE of what it did and found. It NEVER sees the
- * rubric (gap A1, evaluator/scorer separation): the agent under test cannot
- * optimize its behavior toward the grading criteria, and cannot grade its
- * own execution.
- *
- * Stage 2 — reviewer: an INDEPENDENT child grades the executor's evidence
- * strictly against the rubric. The rubric plaintext is decrypted in the host
- * and flows ONLY into this reviewer branch; the executor branch never
- * touches it. A separate model instance grading someone else's evidence
- * removes the self-serving bias of self-scoring.
- *
- * Failure-cell protocol (gap A2): a unit that cannot produce a score
- * (decrypt error, executor crash, reviewer crash, protocol error) returns a
- * cell with `status: "failed"` — NEVER a zero — so aggregation can exclude
- * it and the acceptance rule can reject rounds with too many failures instead
- * of silently averaging a 0 into the mean.
- *
- * Each cell carries the executor child's session id (gap A4): the score can
- * be drilled back to the exact session transcript that produced the evidence.
- *
- * Uses the host-plane `subagents` service (available in every profile) with
- * the native `outputSchema` structured-output seam: the provider validates
- * each child's reply against its cell schema, so the host never parses model
- * text for evaluations. (The workflow engine was rejected because the web
- * profile keeps it in a per-agent isolated realm a host plugin cannot
- * resolve.)
- */
-import type { Context } from "@deepseek-ai/cordis";
-import type { Agent } from "@deepseek-ai/dsh-agent";
-import type { BenchmarkCase, CellScore } from "./benchmark.js";
-export interface EvaluateOptions {
-    cases: readonly BenchmarkCase[];
-    runs: number;
-    passThreshold: number;
-    /** Serialized harness state under test (the candidate's guidance). */
-    harnessOverview: string;
-    label: string;
-    /** AES-256 key for the encrypted rubric envelopes (see src/rubric.ts). */
-    rubricKey?: Buffer;
-    signal?: AbortSignal;
-}
-export interface EvaluationOutcome {
-    label: string;
-    cells: CellScore[];
-    stopReason: string;
-}
-/** How many evaluation units may run concurrently (bounded subagent fan-out). */
-export declare const DEFAULT_EVALUATION_CONCURRENCY = 4;
-/** Evidence handed to the reviewer is capped so the grading call stays bounded. */
-export declare const MAX_EVIDENCE_CHARS = 8000;
-export declare function evaluateState(ctx: Context, agent: Agent, options: EvaluateOptions): Promise<EvaluationOutcome>;
-export interface ExecutorResult {
-    caseId: string;
-    run: number;
-    evidence: string;
-}
-/**
- * Deterministic hash of a case's statement + rubric envelope (gap A3):
- * a 16-char SHA-256 prefix, hex-encoded. Used to detect material changes
- * between reference and candidate evaluation runs (see
- * `score.flagMaterialDrift`).
- */
-export declare function caseHash(caseItem: BenchmarkCase): string;
-/** Validate a provider-validated executor result; returns undefined when malformed. */
-export declare function normalizeExecutor(value: unknown, caseId: string, run: number): ExecutorResult | undefined;
-/** Validate a provider-validated structured cell; returns undefined when malformed. */
-export declare function normalizeCell(value: unknown, caseId: string, run: number, passThreshold: number): CellScore | undefined;
-//# sourceMappingURL=evaluate.d.ts.map