@dsh-enhanced/assistant-evaluation 0.1.23 → 0.1.30

package/lib/benchmark/strategy-meter.js

@@ -0,0 +1,236 @@
+import { normalizeTokenUsage, tokenUsageCost, tokenUsageReserveCost } from './usage.js';
+const MAX = 1_000_000_000;
+const valid = (value) => typeof value === 'number' && Number.isSafeInteger(value) && value >= 0 && value <= MAX;
+const fail = (reason) => { throw new Error(`strategy benchmark meter: ${reason}`); };
+const frozen = (value) => Object.freeze(value);
+const sum = (a, b) => { if (a > MAX - b)
+    fail('budget overflow'); return a + b; };
+function clockBound(value, signal, deadline) {
+    return new Promise((resolve, reject) => {
+        const abort = () => done(() => reject(new Error('strategy benchmark meter: cancelled or timed out')));
+        const timer = setTimeout(abort, Math.max(0, deadline - Date.now()));
+        timer.unref?.();
+        const done = (complete) => { clearTimeout(timer); signal.removeEventListener('abort', abort); complete(); };
+        signal.addEventListener('abort', abort, { once: true });
+        value.then(result => done(() => resolve(result)), error => done(() => reject(error)));
+        if (signal.aborted)
+            abort();
+    });
+}
+function metadata(ctx, options) {
+    // GenerateOptions.sessionId covers auxiliary calls; currentInitiator cross-checks native AgentLoop calls.
+    const agent = ctx.get('agents')?.currentInitiator?.();
+    const sessionId = options.sessionId === undefined ? (typeof agent?.session?.id === 'string' ? agent.session.id : null) : String(options.sessionId);
+    return { sessionId, agentId: typeof agent?.id === 'string' ? agent.id : null };
+}
+function checkInput(input) {
+    const { budget, model, binding } = input;
+    if (!valid(input.modelCalls) || input.modelCalls < 1 || input.modelCalls > 10_000 || !valid(input.maxOutputTokens) || input.maxOutputTokens < 1
+        || !valid(budget.durationMs) || budget.durationMs < 1 || !valid(budget.inputTokens) || !valid(budget.outputTokens)
+        || !valid(budget.toolCalls) || (budget.costUsdMicros !== null && !valid(budget.costUsdMicros))
+        || (model.inputLimitMode ?? 'upper-bound') !== 'upper-bound' || (model.outputLimitMode ?? 'provider') !== 'provider'
+        || model.maxOutputTokens !== input.maxOutputTokens || input.maxOutputTokens > budget.outputTokens || typeof binding.inputTokenUpperBound !== 'function')
+        fail('invalid enforced observation configuration');
+    if ((model.inputUsdMicrosPerMillionTokens === null) !== (model.outputUsdMicrosPerMillionTokens === null))
+        fail('incomplete tariff');
+    if (budget.costUsdMicros !== null && [model.inputUsdMicrosPerMillionTokens, model.outputUsdMicrosPerMillionTokens, model.cacheReadUsdMicrosPerMillionTokens, model.cacheWriteUsdMicrosPerMillionTokens].some(rate => rate === null || rate === undefined))
+        fail('priced budget needs complete tariff');
+}
+/**
+ * Install before any cell work. It rejects estimate/observed modes at setup;
+ * there is deliberately no best-effort observation-only fallback in v1.
+ */
+export function installStrategyBenchmarkMeter(ctx, input) {
+    checkInput(input);
+    const budget = frozen({ ...input.budget });
+    const model = frozen({ ...input.model });
+    const countInput = input.binding.inputTokenUpperBound.bind(input.binding);
+    const maxCalls = input.modelCalls;
+    const outputCap = input.maxOutputTokens;
+    const deadline = Date.now() + budget.durationMs;
+    const controller = new AbortController();
+    const meterSignal = AbortSignal.any([input.signal, controller.signal]);
+    const deadlineTimer = setTimeout(() => controller.abort(new Error('strategy benchmark meter timed out')), budget.durationMs);
+    deadlineTimer.unref?.();
+    let live = true;
+    let calls = 0;
+    let tools = 0;
+    let activeTools = 0;
+    let measuredInput = 0;
+    let measuredOutput = 0;
+    let toolRejected = false;
+    let measuredCost = budget.costUsdMicros === null ? null : 0;
+    let heldCalls = 0;
+    let heldInput = 0;
+    let heldOutput = 0;
+    let heldCost = budget.costUsdMicros === null ? null : 0;
+    const traces = [];
+    const attemptCap = maxCalls + 1;
+    const nativeAgents = new Map();
+    const cancelNative = () => { for (const agent of nativeAgents.values())
+        try {
+            agent.cancel({ kind: 'hook', reason: 'strategy-benchmark-meter-cancelled' });
+        }
+        catch { } };
+    meterSignal.addEventListener('abort', cancelNative, { once: true });
+    function reject(trace, reason) {
+        trace.phase = 'rejected';
+        trace.reason = reason;
+        trace.completedAt = Date.now();
+        return fail(reason);
+    }
+    const stream = async function* (options, next) {
+        const requestSignal = options.signal === undefined ? meterSignal : AbortSignal.any([meterSignal, options.signal]);
+        if (traces.length >= attemptCap)
+            fail('request attempt cap exceeded');
+        const trace = { id: traces.length + 1, ...metadata(ctx, options), startedAt: Date.now(), completedAt: null, phase: 'preflight', dispatched: false, reservedInputTokens: 0, reservedOutputTokens: 0, reservedCostUsdMicros: null, usage: null, reason: null };
+        traces.push(trace);
+        if (trace.sessionId !== null) {
+            const agent = ctx.get('agents')?.get?.(trace.sessionId);
+            if (typeof agent?.cancel === 'function')
+                nativeAgents.set(trace.sessionId, agent);
+        }
+        if (!live)
+            reject(trace, 'disposed');
+        if (meterSignal.aborted || options.signal?.aborted || Date.now() > deadline)
+            reject(trace, 'cancelled');
+        if (options.provider !== model.provider || options.model !== model.model || options.maxTokens !== outputCap || options.temperature !== (model.temperature ?? undefined))
+            reject(trace, 'request-contract');
+        let upper;
+        try {
+            upper = await clockBound(Promise.resolve(countInput(options)), requestSignal, deadline);
+        }
+        catch {
+            return reject(trace, 'input-bound');
+        }
+        // The awaited counter may have crossed a cancellation, disposal, or route-change boundary.
+        if (!live)
+            reject(trace, 'disposed');
+        if (meterSignal.aborted || options.signal?.aborted || Date.now() > deadline)
+            reject(trace, 'cancelled');
+        if (options.provider !== model.provider || options.model !== model.model || options.maxTokens !== outputCap || options.temperature !== (model.temperature ?? undefined) || !valid(upper) || upper > budget.inputTokens)
+            return reject(trace, 'input-bound');
+        const reserveCost = tokenUsageReserveCost(upper, outputCap, model);
+        if (budget.costUsdMicros !== null && reserveCost === null)
+            reject(trace, 'cost');
+        if (calls + heldCalls >= maxCalls || sum(measuredInput, heldInput) > budget.inputTokens - upper
+            || sum(measuredOutput, heldOutput) > budget.outputTokens - outputCap
+            || (budget.costUsdMicros !== null && (measuredCost === null || heldCost === null || reserveCost === null || measuredCost + heldCost > budget.costUsdMicros - reserveCost)))
+            reject(trace, 'shared-budget');
+        heldCalls++;
+        heldInput += upper;
+        heldOutput += outputCap;
+        if (heldCost !== null && reserveCost !== null)
+            heldCost += reserveCost;
+        trace.phase = 'reserved';
+        trace.reservedInputTokens = upper;
+        trace.reservedOutputTokens = outputCap;
+        trace.reservedCostUsdMicros = reserveCost;
+        let usage;
+        let finished = false;
+        let iterator;
+        try {
+            // GenerateOptions is deep-frozen by native AgentLoop. We never mutate it:
+            // cancellation reaches native adapters by cancelling the session's actual Agent.
+            iterator = next()[Symbol.asyncIterator]();
+            trace.dispatched = true;
+            trace.phase = 'streaming';
+            while (true) {
+                const item = await clockBound(iterator.next(), requestSignal, deadline);
+                if (item.done)
+                    break;
+                if (finished)
+                    reject(trace, 'stream');
+                if (item.value.type === 'usage') {
+                    if (usage !== undefined || finished)
+                        reject(trace, 'usage');
+                    usage = { ...item.value.usage };
+                }
+                if (item.value.type === 'finish') {
+                    if (finished || usage === undefined || !['stop', 'tool-calls', 'max-tokens'].includes(item.value.reason.kind))
+                        reject(trace, 'stream');
+                    finished = true;
+                }
+                yield item.value;
+            }
+            if (!live)
+                reject(trace, 'disposed');
+            if (requestSignal.aborted || !finished)
+                reject(trace, 'cancelled');
+            const normalized = usage === undefined ? undefined : normalizeTokenUsage(usage);
+            if (!normalized)
+                return reject(trace, 'usage');
+            if (normalized.inputTokens > upper || normalized.outputTokens > outputCap)
+                return reject(trace, 'usage');
+            const cost = tokenUsageCost(normalized, model);
+            if (budget.costUsdMicros !== null && (cost === null || measuredCost === null || cost > budget.costUsdMicros - measuredCost))
+                reject(trace, 'cost');
+            heldCalls--;
+            heldInput -= upper;
+            heldOutput -= outputCap;
+            if (heldCost !== null && reserveCost !== null)
+                heldCost -= reserveCost;
+            calls++;
+            measuredInput = sum(measuredInput, normalized.inputTokens);
+            measuredOutput = sum(measuredOutput, normalized.outputTokens);
+            if (measuredCost !== null && cost !== null)
+                measuredCost = sum(measuredCost, cost);
+            trace.usage = frozen({ ...normalized });
+            trace.phase = 'settled';
+            trace.completedAt = Date.now();
+        }
+        catch (error) {
+            // Reservation intentionally remains held: a thrown/aborted/late stream cannot prove no bill.
+            if (trace.phase !== 'settled') {
+                trace.phase = 'retained';
+                trace.reason = meterSignal.aborted ? 'cancelled' : live ? 'stream' : 'disposed';
+                trace.completedAt = Date.now();
+            }
+            throw error;
+        }
+        finally {
+            try {
+                void iterator?.return?.().catch(() => { });
+            }
+            catch { }
+            if (trace.phase === 'reserved' || trace.phase === 'streaming') {
+                trace.phase = 'retained';
+                trace.reason = live ? 'stream' : 'disposed';
+                trace.completedAt = Date.now();
+            }
+        }
+    };
+    ctx.on('llm/stream', stream);
+    ctx.on('tools/execute', async (execution, next) => {
+        if (!live || meterSignal.aborted || Date.now() > deadline || execution.signal.aborted || tools >= budget.toolCalls) {
+            toolRejected = true;
+            fail('tool-budget');
+        }
+        if (execution.agent !== undefined)
+            nativeAgents.set(String(execution.agent.id), execution.agent);
+        tools++;
+        activeTools++; // Tool failure is still an attempted external action.
+        try {
+            return await next();
+        }
+        finally {
+            activeTools--;
+        }
+    });
+    const snapshot = () => frozen({ observationMode: 'enforced-upper-bound-provider-output', budget,
+        modelCalls: calls, toolCalls: tools, activeToolCalls: activeTools, inputTokens: measuredInput, outputTokens: measuredOutput, costUsdMicros: measuredCost,
+        heldModelCalls: heldCalls, heldInputTokens: heldInput, heldOutputTokens: heldOutput, heldCostUsdMicros: heldCost,
+        traces: frozen(traces.map(trace => frozen({ ...trace, usage: trace.usage === null ? null : frozen({ ...trace.usage }) }))) });
+    const assertComplete = () => {
+        const value = snapshot();
+        if (meterSignal.aborted || !live || Date.now() >= deadline || value.modelCalls === 0 || toolRejected || activeTools !== 0 || value.heldModelCalls !== 0 || value.traces.some(trace => trace.phase !== 'settled')
+            || value.modelCalls > maxCalls || value.toolCalls > budget.toolCalls || value.inputTokens > budget.inputTokens || value.outputTokens > budget.outputTokens
+            || (budget.costUsdMicros !== null && (value.costUsdMicros === null || value.costUsdMicros > budget.costUsdMicros)))
+            fail('incomplete benchmark meter measurement');
+    };
+    const dispose = () => { if (!live)
+        return; live = false; clearTimeout(deadlineTimer); controller.abort(new Error('strategy benchmark meter disposed')); cancelNative(); };
+    ctx.effect(() => () => dispose(), 'assistant-evaluation.strategy-benchmark-meter');
+    return frozen({ signal: meterSignal, snapshot, assertComplete, dispose });
+}
+//# sourceMappingURL=strategy-meter.js.map

package/lib/benchmark/strategy-meter.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"strategy-meter.js","sourceRoot":"","sources":["../../src/benchmark/strategy-meter.ts"],"names":[],"mappings":"AASA,OAAO,EAAE,mBAAmB,EAAE,cAAc,EAAE,qBAAqB,EAA6B,MAAM,YAAY,CAAA;AAkDlH,MAAM,GAAG,GAAG,aAAa,CAAA;AACzB,MAAM,KAAK,GAAG,CAAC,KAAc,EAAmB,EAAE,CAAC,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,KAAK,IAAI,CAAC,IAAI,KAAK,IAAI,GAAG,CAAA;AACzI,MAAM,IAAI,GAAG,CAAC,MAAc,EAAS,EAAE,GAAG,MAAM,IAAI,KAAK,CAAC,6BAA6B,MAAM,EAAE,CAAC,CAAA,CAAC,CAAC,CAAA;AAClG,MAAM,MAAM,GAAG,CAAI,KAAQ,EAAe,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAA;AACjE,MAAM,GAAG,GAAG,CAAC,CAAS,EAAE,CAAS,EAAU,EAAE,GAAG,IAAI,CAAC,GAAG,GAAG,GAAG,CAAC;IAAE,IAAI,CAAC,iBAAiB,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,CAAA,CAAC,CAAC,CAAA;AACxG,SAAS,UAAU,CAAI,KAAiB,EAAE,MAAmB,EAAE,QAAgB;IAC7E,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACrC,MAAM,KAAK,GAAG,GAAG,EAAE,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,kDAAkD,CAAC,CAAC,CAAC,CAAA;QACrG,MAAM,KAAK,GAAG,UAAU,CAAC,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC,CAAA;QACnE,KAAK,CAAC,KAAK,EAAE,EAAE,CAAA;QACf,MAAM,IAAI,GAAG,CAAC,QAAoB,EAAQ,EAAE,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,mBAAmB,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAA,CAAC,CAAC,CAAA;QAC5H,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAA;QACvD,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,EAAE,KAAK,CAAC,EAAE,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAA;QACrF,IAAI,MAAM,CAAC,OAAO;YAAE,KAAK,EAAE,CAAA;IAC7B,CAAC,CAAC,CAAA;AACJ,CAAC;AACD,SAAS,QAAQ,CAAC,GAAY,EAAE,OAAwB;IACtD,0GAA0G;IAC1G,MAAM,KAAK,GAAI,GAAG,CAAC,GAAG,CAAC,QAAiB,CAAsD,EAAE,gBAAgB,EAAE,EAA8D,CAAA;IAChL,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,KAAK,SAAS,CAAC,CAAC,CAAC,CAAC,OAAO,KAAK,EAAE,OAAO,EAAE,EAAE,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAA;IAClJ,OAAO,EAAE,SAAS,EAAE,OAAO,EAAE,OAAO,KAAK,EAAE,EAAE,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAA;AAChF,CAAC;AACD,SAAS,UAAU,CAAC,KAAkC;IACpD,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,GAAG,KAAK,CAAA;IACxC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,UAAU,CAAC,IAAI,KAAK,CAAC,UAAU,GAAG,CAAC,IAAI,KAAK,CAAC,UAAU,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,eAAe,CAAC,IAAI,KAAK,CAAC,eAAe,GAAG,CAAC;WAC1I,CAAC,KAAK,CAAC,MAAM,CAAC,UAAU,CAAC,IAAI,MAAM,CAAC,UAAU,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,WAAW,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,YAAY,CAAC;WAC/G,CAAC,KAAK,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,MAAM,CAAC,aAAa,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC;WAC3F,CAAC,KAAK,CAAC,cAAc,IAAI,aAAa,CAAC,KAAK,aAAa,IAAI,CAAC,KAAK,CAAC,eAAe,IAAI,UAAU,CAAC,KAAK,UAAU;WACjH,KAAK,CAAC,eAAe,KAAK,KAAK,CAAC,eAAe,IAAI,KAAK,CAAC,eAAe,GAAG,MAAM,CAAC,YAAY,IAAI,OAAO,OAAO,CAAC,oBAAoB,KAAK,UAAU;QAAE,IAAI,CAAC,4CAA4C,CAAC,CAAA;IAC7M,IAAI,CAAC,KAAK,CAAC,8BAA8B,KAAK,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,+BAA+B,KAAK,IAAI,CAAC;QAAE,IAAI,CAAC,mBAAmB,CAAC,CAAA;IACnI,IAAI,MAAM,CAAC,aAAa,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,8BAA8B,EAAE,KAAK,CAAC,+BAA+B,EAAE,KAAK,CAAC,kCAAkC,EAAE,KAAK,CAAC,mCAAmC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,KAAK,IAAI,IAAI,IAAI,KAAK,SAAS,CAAC;QAAE,IAAI,CAAC,qCAAqC,CAAC,CAAA;AACxS,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,6BAA6B,CAAC,GAAY,EAAE,KAAkC;IAC5F,UAAU,CAAC,KAAK,CAAC,CAAA;IACjB,MAAM,MAAM,GAAG,MAAM,CAAC,EAAE,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,CAAA;IAC1C,MAAM,KAAK,GAAG,MAAM,CAAC,EAAE,GAAG,KAAK,CAAC,KAAK,EAAE,CAAC,CAAA;IACxC,MAAM,UAAU,GAAG,KAAK,CAAC,OAAO,CAAC,oBAAqB,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAA;IAC1E,MAAM,QAAQ,GAAG,KAAK,CAAC,UAAU,CAAA;IACjC,MAAM,SAAS,GAAG,KAAK,CAAC,eAAe,CAAA;IACvC,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,MAAM,CAAC,UAAU,CAAA;IAC/C,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAA;IACxC,MAAM,WAAW,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,MAAM,EAAE,UAAU,CAAC,MAAM,CAAC,CAAC,CAAA;IACtE,MAAM,aAAa,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,CAAC,IAAI,KAAK,CAAC,oCAAoC,CAAC,CAAC,EAAE,MAAM,CAAC,UAAU,CAAC,CAAA;IAC5H,aAAa,CAAC,KAAK,EAAE,EAAE,CAAA;IACvB,IAAI,IAAI,GAAG,IAAI,CAAC;IAAC,IAAI,KAAK,GAAG,CAAC,CAAC;IAAC,IAAI,KAAK,GAAG,CAAC,CAAC;IAAC,IAAI,WAAW,GAAG,CAAC,CAAC;IAAC,IAAI,aAAa,GAAG,CAAC,CAAC;IAAC,IAAI,cAAc,GAAG,CAAC,CAAC;IAAC,IAAI,YAAY,GAAG,KAAK,CAAA;IAC3I,IAAI,YAAY,GAAkB,MAAM,CAAC,aAAa,KAAK,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAA;IAC1E,IAAI,SAAS,GAAG,CAAC,CAAC;IAAC,IAAI,SAAS,GAAG,CAAC,CAAC;IAAC,IAAI,UAAU,GAAG,CAAC,CAAC;IAAC,IAAI,QAAQ,GAAkB,MAAM,CAAC,aAAa,KAAK,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAA;IAChI,MAAM,MAAM,GAA8V,EAAE,CAAA;IAC5W,MAAM,UAAU,GAAG,QAAQ,GAAG,CAAC,CAAA;IAC/B,MAAM,YAAY,GAAG,IAAI,GAAG,EAAqE,CAAA;IACjG,MAAM,YAAY,GAAG,GAAS,EAAE,GAAG,KAAK,MAAM,KAAK,IAAI,YAAY,CAAC,MAAM,EAAE;QAAE,IAAI,CAAC;YAAC,KAAK,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,oCAAoC,EAAE,CAAC,CAAA;QAAC,CAAC;QAAC,MAAM,CAAC,CAAA,CAAC,CAAC,CAAC,CAAA;IAC7K,WAAW,CAAC,gBAAgB,CAAC,OAAO,EAAE,YAAY,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAA;IACnE,SAAS,MAAM,CAAC,KAA4B,EAAE,MAA2B;QACvE,KAAK,CAAC,KAAK,GAAG,UAAU,CAAC;QAAC,KAAK,CAAC,MAAM,GAAG,MAAM,CAAC;QAAC,KAAK,CAAC,WAAW,GAAG,IAAI,CAAC,GAAG,EAAE,CAAA;QAC/E,OAAO,IAAI,CAAC,MAAM,CAAC,CAAA;IACrB,CAAC;IACD,MAAM,MAAM,GAAG,KAAK,SAAS,CAAC,EAAE,OAAwB,EAAE,IAAsC;QAC9F,MAAM,aAAa,GAAG,OAAO,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,WAAW,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,CAAA;QACjH,IAAI,MAAM,CAAC,MAAM,IAAI,UAAU;YAAE,IAAI,CAAC,8BAA8B,CAAC,CAAA;QACrE,MAAM,KAAK,GAA0B,EAAE,EAAE,EAAE,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,GAAG,QAAQ,CAAC,GAAG,EAAE,OAAO,CAAC,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,EAAE,WAAW,EAAE,IAAI,EAAE,KAAK,EAAE,WAAW,EAAE,UAAU,EAAE,KAAK,EAAE,mBAAmB,EAAE,CAAC,EAAE,oBAAoB,EAAE,CAAC,EAAE,qBAAqB,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,CAAA;QACnR,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAA;QAClB,IAAI,KAAK,CAAC,SAAS,KAAK,IAAI,EAAE,CAAC;YAC7B,MAAM,KAAK,GAAI,GAAG,CAAC,GAAG,CAAC,QAAiB,CAA+C,EAAE,GAAG,EAAE,CAAC,KAAK,CAAC,SAAS,CAA2E,CAAA;YACzL,IAAI,OAAO,KAAK,EAAE,MAAM,KAAK,UAAU;gBAAE,YAAY,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,EAAE,KAAkE,CAAC,CAAA;QAChJ,CAAC;QACD,IAAI,CAAC,IAAI;YAAE,MAAM,CAAC,KAAK,EAAE,UAAU,CAAC,CAAA;QACpC,IAAI,WAAW,CAAC,OAAO,IAAI,OAAO,CAAC,MAAM,EAAE,OAAO,IAAI,IAAI,CAAC,GAAG,EAAE,GAAG,QAAQ;YAAE,MAAM,CAAC,KAAK,EAAE,WAAW,CAAC,CAAA;QACvG,IAAI,OAAO,CAAC,QAAQ,KAAK,KAAK,CAAC,QAAQ,IAAI,OAAO,CAAC,KAAK,KAAK,KAAK,CAAC,KAAK,IAAI,OAAO,CAAC,SAAS,KAAK,SAAS,IAAI,OAAO,CAAC,WAAW,KAAK,CAAC,KAAK,CAAC,WAAW,IAAI,SAAS,CAAC;YAAE,MAAM,CAAC,KAAK,EAAE,kBAAkB,CAAC,CAAA;QAC1M,IAAI,KAAa,CAAA;QACjB,IAAI,CAAC;YAAC,KAAK,GAAG,MAAM,UAAU,CAAC,OAAO,CAAC,OAAO,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,EAAE,aAAa,EAAE,QAAQ,CAAC,CAAA;QAAC,CAAC;QAAC,MAAM,CAAC;YAAC,OAAO,MAAM,CAAC,KAAK,EAAE,aAAa,CAAC,CAAA;QAAC,CAAC;QAC7I,2FAA2F;QAC3F,IAAI,CAAC,IAAI;YAAE,MAAM,CAAC,KAAK,EAAE,UAAU,CAAC,CAAA;QACpC,IAAI,WAAW,CAAC,OAAO,IAAI,OAAO,CAAC,MAAM,EAAE,OAAO,IAAI,IAAI,CAAC,GAAG,EAAE,GAAG,QAAQ;YAAE,MAAM,CAAC,KAAK,EAAE,WAAW,CAAC,CAAA;QACvG,IAAI,OAAO,CAAC,QAAQ,KAAK,KAAK,CAAC,QAAQ,IAAI,OAAO,CAAC,KAAK,KAAK,KAAK,CAAC,KAAK,IAAI,OAAO,CAAC,SAAS,KAAK,SAAS,IAAI,OAAO,CAAC,WAAW,KAAK,CAAC,KAAK,CAAC,WAAW,IAAI,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,MAAM,CAAC,WAAW;YAAE,OAAO,MAAM,CAAC,KAAK,EAAE,aAAa,CAAC,CAAA;QAC3P,MAAM,WAAW,GAAG,qBAAqB,CAAC,KAAK,EAAE,SAAS,EAAE,KAAK,CAAC,CAAA;QAClE,IAAI,MAAM,CAAC,aAAa,KAAK,IAAI,IAAI,WAAW,KAAK,IAAI;YAAE,MAAM,CAAC,KAAK,EAAE,MAAM,CAAC,CAAA;QAChF,IAAI,KAAK,GAAG,SAAS,IAAI,QAAQ,IAAI,GAAG,CAAC,aAAa,EAAE,SAAS,CAAC,GAAG,MAAM,CAAC,WAAW,GAAG,KAAK;eAC1F,GAAG,CAAC,cAAc,EAAE,UAAU,CAAC,GAAG,MAAM,CAAC,YAAY,GAAG,SAAS;eACjE,CAAC,MAAM,CAAC,aAAa,KAAK,IAAI,IAAI,CAAC,YAAY,KAAK,IAAI,IAAI,QAAQ,KAAK,IAAI,IAAI,WAAW,KAAK,IAAI,IAAI,YAAY,GAAG,QAAQ,GAAG,MAAM,CAAC,aAAa,GAAG,WAAW,CAAC,CAAC;YAAE,MAAM,CAAC,KAAK,EAAE,eAAe,CAAC,CAAA;QAC5M,SAAS,EAAE,CAAC;QAAC,SAAS,IAAI,KAAK,CAAC;QAAC,UAAU,IAAI,SAAS,CAAC;QAAC,IAAI,QAAQ,KAAK,IAAI,IAAI,WAAW,KAAK,IAAI;YAAE,QAAQ,IAAI,WAAW,CAAA;QAChI,KAAK,CAAC,KAAK,GAAG,UAAU,CAAC;QAAC,KAAK,CAAC,mBAAmB,GAAG,KAAK,CAAC;QAAC,KAAK,CAAC,oBAAoB,GAAG,SAAS,CAAC;QAAC,KAAK,CAAC,qBAAqB,GAAG,WAAW,CAAA;QAC9I,IAAI,KAA6B,CAAC;QAAC,IAAI,QAAQ,GAAG,KAAK,CAAC;QAAC,IAAI,QAAgD,CAAA;QAC7G,IAAI,CAAC;YACH,0EAA0E;YAC1E,iFAAiF;YACjF,QAAQ,GAAG,IAAI,EAAE,CAAC,MAAM,CAAC,aAAa,CAAC,EAAE,CAAC;YAAC,KAAK,CAAC,UAAU,GAAG,IAAI,CAAC;YAAC,KAAK,CAAC,KAAK,GAAG,WAAW,CAAA;YAC7F,OAAO,IAAI,EAAE,CAAC;gBACZ,MAAM,IAAI,GAAG,MAAM,UAAU,CAAC,QAAQ,CAAC,IAAI,EAAE,EAAE,aAAa,EAAE,QAAQ,CAAC,CAAA;gBACvE,IAAI,IAAI,CAAC,IAAI;oBAAE,MAAK;gBACpB,IAAI,QAAQ;oBAAE,MAAM,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAA;gBACrC,IAAI,IAAI,CAAC,KAAK,CAAC,IAAI,KAAK,OAAO,EAAE,CAAC;oBAChC,IAAI,KAAK,KAAK,SAAS,IAAI,QAAQ;wBAAE,MAAM,CAAC,KAAK,EAAE,OAAO,CAAC,CAAA;oBAC3D,KAAK,GAAG,EAAE,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,CAAA;gBACjC,CAAC;gBACD,IAAI,IAAI,CAAC,KAAK,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;oBAAC,IAAI,QAAQ,IAAI,KAAK,KAAK,SAAS,IAAI,CAAC,CAAC,MAAM,EAAE,YAAY,EAAE,YAAY,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC;wBAAE,MAAM,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC;oBAAC,QAAQ,GAAG,IAAI,CAAA;gBAAC,CAAC;gBAC7L,MAAM,IAAI,CAAC,KAAK,CAAA;YAClB,CAAC;YACD,IAAI,CAAC,IAAI;gBAAE,MAAM,CAAC,KAAK,EAAE,UAAU,CAAC,CAAA;YACpC,IAAI,aAAa,CAAC,OAAO,IAAI,CAAC,QAAQ;gBAAE,MAAM,CAAC,KAAK,EAAE,WAAW,CAAC,CAAA;YAClE,MAAM,UAAU,GAAG,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,mBAAmB,CAAC,KAAK,CAAC,CAAA;YAC/E,IAAI,CAAC,UAAU;gBAAE,OAAO,MAAM,CAAC,KAAK,EAAE,OAAO,CAAC,CAAA;YAC9C,IAAI,UAAU,CAAC,WAAW,GAAG,KAAK,IAAI,UAAU,CAAC,YAAY,GAAG,SAAS;gBAAE,OAAO,MAAM,CAAC,KAAK,EAAE,OAAO,CAAC,CAAA;YACxG,MAAM,IAAI,GAAG,cAAc,CAAC,UAAU,EAAE,KAAK,CAAC,CAAA;YAC9C,IAAI,MAAM,CAAC,aAAa,KAAK,IAAI,IAAI,CAAC,IAAI,KAAK,IAAI,IAAI,YAAY,KAAK,IAAI,IAAI,IAAI,GAAG,MAAM,CAAC,aAAa,GAAG,YAAY,CAAC;gBAAE,MAAM,CAAC,KAAK,EAAE,MAAM,CAAC,CAAA;YAClJ,SAAS,EAAE,CAAC;YAAC,SAAS,IAAI,KAAK,CAAC;YAAC,UAAU,IAAI,SAAS,CAAC;YAAC,IAAI,QAAQ,KAAK,IAAI,IAAI,WAAW,KAAK,IAAI;gBAAE,QAAQ,IAAI,WAAW,CAAA;YAChI,KAAK,EAAE,CAAC;YAAC,aAAa,GAAG,GAAG,CAAC,aAAa,EAAE,UAAU,CAAC,WAAW,CAAC,CAAC;YAAC,cAAc,GAAG,GAAG,CAAC,cAAc,EAAE,UAAU,CAAC,YAAY,CAAC,CAAA;YAClI,IAAI,YAAY,KAAK,IAAI,IAAI,IAAI,KAAK,IAAI;gBAAE,YAAY,GAAG,GAAG,CAAC,YAAY,EAAE,IAAI,CAAC,CAAA;YAClF,KAAK,CAAC,KAAK,GAAG,MAAM,CAAC,EAAE,GAAG,UAAU,EAAE,CAAC,CAAC;YAAC,KAAK,CAAC,KAAK,GAAG,SAAS,CAAC;YAAC,KAAK,CAAC,WAAW,GAAG,IAAI,CAAC,GAAG,EAAE,CAAA;QAClG,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,6FAA6F;YAC7F,IAAI,KAAK,CAAC,KAAK,KAAK,SAAS,EAAE,CAAC;gBAAC,KAAK,CAAC,KAAK,GAAG,UAAU,CAAC;gBAAC,KAAK,CAAC,MAAM,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,UAAU,CAAC;gBAAC,KAAK,CAAC,WAAW,GAAG,IAAI,CAAC,GAAG,EAAE,CAAA;YAAC,CAAC;YAC5K,MAAM,KAAK,CAAA;QACb,CAAC;gBAAS,CAAC;YACT,IAAI,CAAC;gBAAC,KAAK,QAAQ,EAAE,MAAM,EAAE,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAA;YAAC,CAAC;YAAC,MAAM,CAAC,CAAA,CAAC;YAC1D,IAAI,KAAK,CAAC,KAAK,KAAK,UAAU,IAAI,KAAK,CAAC,KAAK,KAAK,WAAW,EAAE,CAAC;gBAAC,KAAK,CAAC,KAAK,GAAG,UAAU,CAAC;gBAAC,KAAK,CAAC,MAAM,GAAG,IAAI,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,UAAU,CAAC;gBAAC,KAAK,CAAC,WAAW,GAAG,IAAI,CAAC,GAAG,EAAE,CAAA;YAAC,CAAC;QAC1K,CAAC;IACH,CAAC,CAAA;IACD,GAAG,CAAC,EAAE,CAAC,YAAY,EAAE,MAAM,CAAC,CAAA;IAC5B,GAAG,CAAC,EAAE,CAAC,eAAe,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,EAAE,EAAE;QAChD,IAAI,CAAC,IAAI,IAAI,WAAW,CAAC,OAAO,IAAI,IAAI,CAAC,GAAG,EAAE,GAAG,QAAQ,IAAI,SAAS,CAAC,MAAM,CAAC,OAAO,IAAI,KAAK,IAAI,MAAM,CAAC,SAAS,EAAE,CAAC;YAAC,YAAY,GAAG,IAAI,CAAC;YAAC,IAAI,CAAC,aAAa,CAAC,CAAA;QAAC,CAAC;QAChK,IAAI,SAAS,CAAC,KAAK,KAAK,SAAS;YAAE,YAAY,CAAC,GAAG,CAAC,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,EAAE,CAAC,EAAE,SAAS,CAAC,KAAK,CAAC,CAAA;QAChG,KAAK,EAAE,CAAC;QAAC,WAAW,EAAE,CAAA,CAAC,sDAAsD;QAC7E,IAAI,CAAC;YAAC,OAAO,MAAM,IAAI,EAAE,CAAA;QAAC,CAAC;gBAAS,CAAC;YAAC,WAAW,EAAE,CAAA;QAAC,CAAC;IACvD,CAAC,CAAC,CAAA;IACF,MAAM,QAAQ,GAAG,GAA6C,EAAE,CAAC,MAAM,CAAC,EAAE,eAAe,EAAE,sCAAsC,EAAE,MAAM;QACvI,UAAU,EAAE,KAAK,EAAE,SAAS,EAAE,KAAK,EAAE,eAAe,EAAE,WAAW,EAAE,WAAW,EAAE,aAAa,EAAE,YAAY,EAAE,cAAc,EAAE,aAAa,EAAE,YAAY;QACxJ,cAAc,EAAE,SAAS,EAAE,eAAe,EAAE,SAAS,EAAE,gBAAgB,EAAE,UAAU,EAAE,iBAAiB,EAAE,QAAQ;QAChH,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,MAAM,CAAC,EAAE,GAAG,KAAK,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,KAAK,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,EAAE,GAAG,KAAK,CAAC,KAAK,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAA;IAC/H,MAAM,cAAc,GAAG,GAAS,EAAE;QAChC,MAAM,KAAK,GAAG,QAAQ,EAAE,CAAA;QACxB,IAAI,WAAW,CAAC,OAAO,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,GAAG,EAAE,IAAI,QAAQ,IAAI,KAAK,CAAC,UAAU,KAAK,CAAC,IAAI,YAAY,IAAI,WAAW,KAAK,CAAC,IAAI,KAAK,CAAC,cAAc,KAAK,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,KAAK,KAAK,SAAS,CAAC;eAC3M,KAAK,CAAC,UAAU,GAAG,QAAQ,IAAI,KAAK,CAAC,SAAS,GAAG,MAAM,CAAC,SAAS,IAAI,KAAK,CAAC,WAAW,GAAG,MAAM,CAAC,WAAW,IAAI,KAAK,CAAC,YAAY,GAAG,MAAM,CAAC,YAAY;eACvJ,CAAC,MAAM,CAAC,aAAa,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,aAAa,KAAK,IAAI,IAAI,KAAK,CAAC,aAAa,GAAG,MAAM,CAAC,aAAa,CAAC,CAAC;YAAE,IAAI,CAAC,wCAAwC,CAAC,CAAA;IACtK,CAAC,CAAA;IACD,MAAM,OAAO,GAAG,GAAS,EAAE,GAAG,IAAI,CAAC,IAAI;QAAE,OAAO,CAAC,IAAI,GAAG,KAAK,CAAC,CAAC,YAAY,CAAC,aAAa,CAAC,CAAC,CAAC,UAAU,CAAC,KAAK,CAAC,IAAI,KAAK,CAAC,mCAAmC,CAAC,CAAC,CAAC,CAAC,YAAY,EAAE,CAAA,CAAC,CAAC,CAAA;IAC9K,GAAG,CAAC,MAAM,CAAC,GAAG,EAAE,CAAC,GAAG,EAAE,CAAC,OAAO,EAAE,EAAE,+CAA+C,CAAC,CAAA;IAClF,OAAO,MAAM,CAAC,EAAE,MAAM,EAAE,WAAW,EAAE,QAAQ,EAAE,cAAc,EAAE,OAAO,EAAE,CAAC,CAAA;AAC3E,CAAC"}

package/lib/benchmark/strategy-owner.d.ts

@@ -0,0 +1,71 @@
+import { Context } from '@deepseek-ai/cordis';
+import { LlmAdapter } from '@deepseek-ai/dsh-llm';
+import type { DeliveryOwnerLineage, InboundEnvelope, OutboundIntent, OwnerRouteValidationReceipt } from '@dsh-enhanced/assistant-delivery/types';
+import { type PolicyRule } from '@dsh-enhanced/assistant-policy';
+export interface BenchmarkStrategyOwnerOptions {
+    /** Fresh Context owned by one benchmark cell.  The helper installs its Host services into it. */
+    ctx: Context;
+    /** Isolated candidate workspace, never a user profile directory. */
+    workspace: string;
+    /** Separate per-cell durable state root, outside the candidate workspace. */
+    stateRoot: string;
+    /** Stable plan/cell identity; it becomes only local channel identifiers. */
+    cellId: string;
+    provider: string;
+    model: string;
+    maxOutputTokens: number;
+    persona: string;
+    /** Extra fixed policy rules, normally the two allowed Goal tools and goal actions. */
+    policyRules: readonly PolicyRule[];
+    /** Global tools visible to this Delivery owner.  Defaults to the two Goal tools. */
+    allowedToolNames?: readonly string[];
+    agentPreset?: string;
+    /** Bounded wait for the native goal driver, disabled for plain owner runtimes. */
+    goalContinuationTimeoutMs?: number;
+}
+export interface BenchmarkStrategyOwnerScope {
+    readonly workspace: string;
+    readonly preset: string;
+    readonly principalId: string;
+    readonly owner: Readonly<DeliveryOwnerLineage>;
+    readonly route: Readonly<OwnerRouteValidationReceipt>;
+}
+export interface BenchmarkStrategyOwnerRuntime {
+    readonly ctx: Context;
+    readonly workspace: string;
+    readonly stateRoot: string;
+    /** Fresh, private state directory for this exact runtime; retained for evidence after shutdown. */
+    readonly runtimeRoot: string;
+    /** Configured local identity; route validity still requires ownerScope after inbound. */
+    readonly principalId: string;
+    readonly ownerRouteId: string;
+    readonly principal: Readonly<InboundEnvelope['principal']>;
+    readonly conversation: Readonly<InboundEnvelope['conversation']>;
+    readonly outbound: readonly Readonly<OutboundIntent>[];
+    /** Register the sole exact model route after the caller installs its outer meter. */
+    installModel(adapter: LlmAdapter): Promise<void>;
+    /** Pair the synthetic owner before the one public inbound message is admitted. */
+    pairOwner(): Readonly<DeliveryOwnerLineage>;
+    /** Admit exactly one synthetic public owner message through Delivery. */
+    sendPublicInbound(text: string): Promise<Readonly<{
+        inboxId: string;
+        duplicate: boolean;
+        status: string;
+    }>>;
+    /** Drain actual Delivery work and, when supplied, actual Goals work. */
+    waitForQuiescence(goals?: Readonly<{
+        whenIdle(): Promise<void>;
+    }>): Promise<void>;
+    /** Read only public Delivery owner-route evidence after the owner binding exists. */
+    ownerScope(): Readonly<BenchmarkStrategyOwnerScope>;
+    shutdown(): Promise<void>;
+}
+/**
+ * Compose the real DSH/Delivery stack needed by a native strategy cell.
+ *
+ * Goals, Verifier, the in-process Subagent provider, the outer meter, and the
+ * model adapter deliberately remain caller-owned: they differ by cell or arm
+ * and must be installed before `sendPublicInbound()`.
+ */
+export declare function createBenchmarkStrategyOwnerRuntime(input: BenchmarkStrategyOwnerOptions): Promise<BenchmarkStrategyOwnerRuntime>;
+//# sourceMappingURL=strategy-owner.d.ts.map

package/lib/benchmark/strategy-owner.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"strategy-owner.d.ts","sourceRoot":"","sources":["../../src/benchmark/strategy-owner.ts"],"names":[],"mappings":"AAWA,OAAO,EAAE,OAAO,EAAE,MAAM,qBAAqB,CAAA;AAG7C,OAAO,EAAE,UAAU,EAAc,MAAM,sBAAsB,CAAA;AAM7D,OAAO,KAAK,EAAmB,oBAAoB,EAAE,eAAe,EAAE,cAAc,EAAE,2BAA2B,EAAE,MAAM,wCAAwC,CAAA;AACjK,OAAO,EAA0B,KAAK,UAAU,EAAE,MAAM,gCAAgC,CAAA;AA0CxF,MAAM,WAAW,6BAA6B;IAC5C,iGAAiG;IACjG,GAAG,EAAE,OAAO,CAAA;IACZ,oEAAoE;IACpE,SAAS,EAAE,MAAM,CAAA;IACjB,6EAA6E;IAC7E,SAAS,EAAE,MAAM,CAAA;IACjB,4EAA4E;IAC5E,MAAM,EAAE,MAAM,CAAA;IACd,QAAQ,EAAE,MAAM,CAAA;IAChB,KAAK,EAAE,MAAM,CAAA;IACb,eAAe,EAAE,MAAM,CAAA;IACvB,OAAO,EAAE,MAAM,CAAA;IACf,sFAAsF;IACtF,WAAW,EAAE,SAAS,UAAU,EAAE,CAAA;IAClC,oFAAoF;IACpF,gBAAgB,CAAC,EAAE,SAAS,MAAM,EAAE,CAAA;IACpC,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,kFAAkF;IAClF,yBAAyB,CAAC,EAAE,MAAM,CAAA;CACnC;AAED,MAAM,WAAW,2BAA2B;IAC1C,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;IAC1B,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAA;IACvB,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAA;IAC5B,QAAQ,CAAC,KAAK,EAAE,QAAQ,CAAC,oBAAoB,CAAC,CAAA;IAC9C,QAAQ,CAAC,KAAK,EAAE,QAAQ,CAAC,2BAA2B,CAAC,CAAA;CACtD;AAED,MAAM,WAAW,6BAA6B;IAC5C,QAAQ,CAAC,GAAG,EAAE,OAAO,CAAA;IACrB,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;IAC1B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;IAC1B,mGAAmG;IACnG,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAA;IAC5B,yFAAyF;IACzF,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAA;IAC5B,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAA;IAC7B,QAAQ,CAAC,SAAS,EAAE,QAAQ,CAAC,eAAe,CAAC,WAAW,CAAC,CAAC,CAAA;IAC1D,QAAQ,CAAC,YAAY,EAAE,QAAQ,CAAC,eAAe,CAAC,cAAc,CAAC,CAAC,CAAA;IAChE,QAAQ,CAAC,QAAQ,EAAE,SAAS,QAAQ,CAAC,cAAc,CAAC,EAAE,CAAA;IACtD,qFAAqF;IACrF,YAAY,CAAC,OAAO,EAAE,UAAU,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IAChD,kFAAkF;IAClF,SAAS,IAAI,QAAQ,CAAC,oBAAoB,CAAC,CAAA;IAC3C,yEAAyE;IACzE,iBAAiB,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,QAAQ,CAAC;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,OAAO,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC,CAAA;IAC3G,wEAAwE;IACxE,iBAAiB,CAAC,KAAK,CAAC,EAAE,QAAQ,CAAC;QAAE,QAAQ,IAAI,OAAO,CAAC,IAAI,CAAC,CAAA;KAAE,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IACjF,qFAAqF;IACrF,UAAU,IAAI,QAAQ,CAAC,2BAA2B,CAAC,CAAA;IACnD,QAAQ,IAAI,OAAO,CAAC,IAAI,CAAC,CAAA;CAC1B;AAED;;;;;;GAMG;AACH,wBAAsB,mCAAmC,CACvD,KAAK,EAAE,6BAA6B,GACnC,OAAO,CAAC,6BAA6B,CAAC,CAgLxC"}

package/lib/benchmark/strategy-owner.js

@@ -0,0 +1,282 @@
+var __rewriteRelativeImportExtension = (this && this.__rewriteRelativeImportExtension) || function (path, preserveJsx) {
+    if (typeof path === "string" && /^\.\.?\//.test(path)) {
+        return path.replace(/\.(tsx)$|((?:\.d)?)((?:\.[^./]+?)?)\.([cm]?)ts$/i, function (m, tsx, d, ext, cm) {
+            return tsx ? preserveJsx ? ".jsx" : ".js" : d && (!ext || !cm) ? m : (d + ext + "." + cm.toLowerCase() + "js");
+        });
+    }
+    return path;
+};
+/**
+ * Ephemeral, production-native owner runtime for one strategy benchmark cell.
+ *
+ * This is deliberately a Host helper rather than a benchmark fixture.  It
+ * composes the public DSH services used by Delivery and keeps every durable
+ * artifact below the caller-provided benchmark state root.  It never opens a
+ * real channel adapter and does not discover a user's DSH profile.
+ */
+import { createHash } from 'node:crypto';
+import { lstat, mkdir, realpath } from 'node:fs/promises';
+import { isAbsolute, join, resolve } from 'node:path';
+import { Context } from '@deepseek-ai/cordis';
+import AgentLoop from '@deepseek-ai/dsh-agent-loop';
+import { AgentRegistry } from '@deepseek-ai/dsh-agent';
+import { LlmAdapter, LlmRuntime } from '@deepseek-ai/dsh-llm';
+import { JsonlSessionPersistence } from '@deepseek-ai/dsh-session-persistence-jsonl';
+import { SessionStore } from '@deepseek-ai/dsh-session';
+import { SessionProjectionRegistry } from '@deepseek-ai/dsh-session-projection';
+import SystemPrompt from '@deepseek-ai/dsh-system-prompt';
+import ToolRuntime from '@deepseek-ai/dsh-tools';
+import { AssistantPolicyService } from '@dsh-enhanced/assistant-policy';
+import { benchmarkSnapshot } from './schema.js';
+const MAX_TEXT_BYTES = 64 * 1024;
+const safeId = (value, label) => {
+    if (typeof value !== 'string' || !/^[A-Za-z0-9][A-Za-z0-9._-]{0,127}$/u.test(value))
+        throw new Error(`invalid ${label}`);
+    return value;
+};
+const safeRoute = (value, label) => {
+    if (typeof value !== 'string' || !/^[A-Za-z0-9][A-Za-z0-9._:/-]{0,511}$/u.test(value))
+        throw new Error(`invalid ${label}`);
+    return value;
+};
+const positive = (value, label, max) => {
+    if (!Number.isSafeInteger(value))
+        throw new Error(`invalid ${label}`);
+    const number = value;
+    if (number < 1 || number > max)
+        throw new Error(`invalid ${label}`);
+    return number;
+};
+const inside = (parent, child) => child === parent || child.startsWith(`${parent}/`);
+async function privateCanonicalDirectory(path, label) {
+    const metadata = await lstat(path);
+    if (!metadata.isDirectory() || metadata.isSymbolicLink() || (metadata.mode & 0o077) !== 0) {
+        throw new Error(`${label} must be a private non-symlink directory`);
+    }
+    if (process.getuid !== undefined && metadata.uid !== process.getuid())
+        throw new Error(`${label} must belong to the current OS user`);
+    const canonical = await realpath(path);
+    if (canonical !== path)
+        throw new Error(`${label} must be canonical`);
+    return canonical;
+}
+/**
+ * Compose the real DSH/Delivery stack needed by a native strategy cell.
+ *
+ * Goals, Verifier, the in-process Subagent provider, the outer meter, and the
+ * model adapter deliberately remain caller-owned: they differ by cell or arm
+ * and must be installed before `sendPublicInbound()`.
+ */
+export async function createBenchmarkStrategyOwnerRuntime(input) {
+    const ctx = input.ctx;
+    if (['llm', 'sessions', 'agents', 'tools', 'assistantDelivery'].some(name => ctx.get(name) !== undefined)) {
+        throw new Error('benchmark owner requires a fresh dedicated Context');
+    }
+    const workspace = resolve(input.workspace);
+    const stateRoot = resolve(input.stateRoot);
+    if (!isAbsolute(input.workspace) || !isAbsolute(input.stateRoot) || inside(workspace, stateRoot) || inside(stateRoot, workspace)) {
+        throw new Error('benchmark workspace and state root must be distinct absolute directories');
+    }
+    const cellId = safeId(input.cellId, 'benchmark cell id');
+    const provider = safeRoute(input.provider, 'provider');
+    const model = safeRoute(input.model, 'model');
+    const preset = safeId(input.agentPreset ?? 'benchmark', 'agent preset');
+    const maxOutputTokens = positive(input.maxOutputTokens, 'max output tokens', 1_000_000);
+    if (typeof input.persona !== 'string' || Buffer.byteLength(input.persona, 'utf8') > MAX_TEXT_BYTES)
+        throw new Error('invalid benchmark persona');
+    const persona = input.persona;
+    const policyRules = benchmarkSnapshot(input.policyRules);
+    const allowedToolNames = Object.freeze([...(input.allowedToolNames ?? ['goal_create', 'goal_strategy'])].map(value => safeId(value, 'allowed tool name')));
+    if (new Set(allowedToolNames).size !== allowedToolNames.length)
+        throw new Error('duplicate allowed tool name');
+    const goalContinuationTimeoutMs = input.goalContinuationTimeoutMs === undefined ? 0 : positive(input.goalContinuationTimeoutMs, 'goal continuation timeout', 300_000);
+    const fingerprint = createHash('sha256').update(`${cellId}\0${workspace}\0${stateRoot}`).digest('hex').slice(0, 24);
+    const channel = `benchmark-${fingerprint.slice(0, 12)}`;
+    const account = `cell-${fingerprint.slice(12)}`;
+    const principal = Object.freeze({ channel, account, tenant: 'isolated', user: `owner-${fingerprint}` });
+    const conversation = Object.freeze({ channel, account, tenant: 'isolated', kind: 'dm', chat: `private-${fingerprint}` });
+    const deliveryPackage = '@dsh-enhanced/assistant-delivery';
+    const deliveryModule = await import(__rewriteRelativeImportExtension(deliveryPackage));
+    if (typeof deliveryModule.AssistantDeliveryService !== 'function' || typeof deliveryModule.externalPrincipalId !== 'function')
+        throw new Error('benchmark Delivery peer lacks required public APIs');
+    const principalId = deliveryModule.externalPrincipalId(principal);
+    const Delivery = deliveryModule.AssistantDeliveryService;
+    let delivery;
+    const authorityId = `benchmark-owner-${fingerprint}`;
+    const bootstrapId = `benchmark-bootstrap-${fingerprint}`;
+    const runtimeRoot = join(stateRoot, `owner-${fingerprint}`);
+    const sent = [];
+    let paired = false;
+    let inboundSent = false;
+    let closed = false;
+    let shutdownFlight;
+    let eventNumber = 0;
+    let removeAdapter;
+    let removeChannel;
+    await privateCanonicalDirectory(workspace, 'benchmark workspace');
+    await privateCanonicalDirectory(stateRoot, 'benchmark state root');
+    await mkdir(runtimeRoot, { recursive: false, mode: 0o700 });
+    await privateCanonicalDirectory(runtimeRoot, 'benchmark runtime root');
+    try {
+        await ctx.plugin(LlmRuntime);
+        await ctx.plugin(SessionStore);
+        new SessionProjectionRegistry(ctx);
+        await ctx.plugin(SystemPrompt, { includeHarnessIdentity: false, includeRuntimeContext: true, persona: '' });
+        ctx.systemPrompt.section({ name: 'benchmark-strategy-persona', order: 0, text: persona, complete: true });
+        await ctx.plugin(ToolRuntime, { mode: 'native' });
+        await ctx.plugin(AgentRegistry);
+        await ctx.plugin(JsonlSessionPersistence, { root: join(runtimeRoot, 'sessions'), compression: 'none', writeBatchMaxDelayMs: 1 });
+        // Delivery first persists an unstarted Session, then disposes and resumes
+        // it for the inbound turn. JSONL flush alone leaves an empty log lazy;
+        // explicitly materialize its real header without fabricating an event.
+        ctx.on('session/flush', session => ctx.sessionPersistence.ensureMaterialized(session));
+        await ctx.plugin(AssistantPolicyService, { databasePath: join(runtimeRoot, 'policy.sqlite'), toolDefaultEffect: 'deny', rules: [
+                { id: 'benchmark-pair-issue', effect: 'allow', subject: { kind: 'external', id: `local:${bootstrapId}` }, actions: ['pair.issue'], resource: { kind: 'message', id: 'pairing' }, context: { initiators: ['foreground'] } },
+                { id: 'benchmark-owner-ingest', effect: 'allow', subject: { kind: 'external', id: principalId }, actions: ['pair.confirm', 'ingest'], resource: { kind: 'message', id: '*' }, context: { initiators: ['external'] } },
+                { id: 'benchmark-owner-reply', effect: 'allow', subject: { kind: 'agent', id: preset, workspace, principal: principalId }, actions: ['reply'], resource: { kind: 'message', id: '*' }, context: { initiators: ['external'] } },
+                ...policyRules,
+            ] });
+        ctx.on('agent/created', ({ agent }) => {
+            // Delivery creates all benchmark foreground agents after the arm has
+            // mounted its public tools.  Restrict only global tools; arm-scoped tools
+            // still need an explicit Policy allow and are therefore not a bypass.
+            agent.ctx.tools.restrict({ allow: allowedToolNames });
+        });
+        await ctx.plugin(Delivery, {
+            databasePath: join(runtimeRoot, 'delivery.sqlite'), spoolPath: join(runtimeRoot, 'spool'), schedulerEnabled: false,
+            defaultWorkspace: workspace, defaultAgentPreset: preset, policyRef: 'benchmark-owner', agentProvider: provider, agentModel: model,
+            agentMaxOutputTokens: maxOutputTokens, agentMaxAutoContinuationTurns: 0, agentGoalContinuationTimeoutMs: goalContinuationTimeoutMs,
+            ownerRoutes: [{ id: authorityId, conversation, principal, workspace, agentPreset: preset, policyRef: 'benchmark-owner', minimumGeneration: 1 }],
+        });
+        delivery = ctx.get('assistantDelivery');
+        if (!delivery || typeof delivery.validateOwnerRoute !== 'function')
+            throw new Error('benchmark Delivery service unavailable');
+        await ctx.plugin(AgentLoop, { agents: [] });
+    }
+    catch (error) {
+        try {
+            await ctx.fiber.dispose();
+        }
+        catch (cleanup) {
+            throw new AggregateError([error, cleanup], 'benchmark owner runtime setup and disposal failed');
+        }
+        throw error;
+    }
+    const adapter = {
+        channel, account,
+        capabilities: Object.freeze({ reconcileUnknownSend: false, receipts: [], formats: ['plain'] }),
+        async start() { },
+        async send(intent) {
+            // The only channel is a local capture sink.  No network, provider
+            // notification, browser, or user profile is reachable from this adapter.
+            sent.push(structuredClone(intent));
+            return { outcome: 'accepted', providerMessageId: `benchmark-${createHash('sha256').update(intent.idempotencyKey).digest('hex').slice(0, 32)}` };
+        },
+    };
+    const assertOpen = () => { if (closed)
+        throw new Error('benchmark owner runtime is shut down'); };
+    const snapshotOutbound = () => Object.freeze(sent.map(intent => Object.freeze(structuredClone(intent))));
+    const ownerScope = () => {
+        assertOpen();
+        const route = delivery.validateOwnerRoute({ authorityId, principalId, workspace, agentPreset: preset });
+        return Object.freeze({ workspace, preset, principalId, owner: Object.freeze({ principalRecordId: route.principalRecordId, principalVersion: route.principalVersion }), route });
+    };
+    const waitForQuiescence = async (goals) => {
+        assertOpen();
+        // A second Delivery tick flushes a reply queued by the completed foreground
+        // turn.  Goals is caller supplied because this helper never installs it.
+        await delivery.tick();
+        await delivery.whenIdle();
+        await goals?.whenIdle();
+        await delivery.tick();
+        await delivery.whenIdle();
+        await goals?.whenIdle();
+        const health = delivery.health();
+        if (health.pendingInbox !== 0 || health.pendingOutbox !== 0 || health.deadLetterInbox !== 0
+            || health.deadLetterOutbox !== 0 || health.unknownOutbox !== 0 || health.pendingPresentations !== 0 || health.deadPresentations !== 0) {
+            throw new Error('benchmark Delivery work is pending or unresolved');
+        }
+    };
+    try {
+        removeChannel = await delivery.registerAdapter(adapter);
+    }
+    catch (error) {
+        try {
+            await ctx.fiber.dispose();
+        }
+        catch (cleanup) {
+            throw new AggregateError([error, cleanup], 'benchmark owner adapter setup and disposal failed');
+        }
+        throw error;
+    }
+    return Object.freeze({
+        ctx, workspace, stateRoot, runtimeRoot, principal, principalId, ownerRouteId: authorityId, conversation,
+        get outbound() { return snapshotOutbound(); },
+        async installModel(modelAdapter) {
+            assertOpen();
+            if (removeAdapter !== undefined)
+                throw new Error('benchmark model adapter is already installed');
+            if (!(modelAdapter instanceof LlmAdapter))
+                throw new Error('benchmark model adapter must be an LlmAdapter');
+            removeAdapter = ctx.llm.registerAdapter([provider], modelAdapter);
+            try {
+                await ctx.llm.resolveModelInfo(provider, model);
+            }
+            catch (error) {
+                removeAdapter();
+                removeAdapter = undefined;
+                throw error;
+            }
+        },
+        pairOwner() {
+            assertOpen();
+            if (paired)
+                throw new Error('benchmark owner is already paired');
+            const pairing = delivery.issuePairing(bootstrapId, principal);
+            const confirmed = delivery.confirmPairing({ challengeId: pairing.challenge.id, principal, code: pairing.code });
+            if (confirmed.role !== 'owner' || confirmed.status !== 'active' || confirmed.id === '') {
+                throw new Error('benchmark owner pairing was not confirmed');
+            }
+            const owner = Object.freeze({ principalRecordId: confirmed.id, principalVersion: confirmed.version });
+            paired = true;
+            return owner;
+        },
+        async sendPublicInbound(text) {
+            assertOpen();
+            if (!paired || inboundSent)
+                throw new Error('benchmark owner inbound must follow one pairing and precede shutdown');
+            if (removeAdapter === undefined)
+                throw new Error('benchmark model adapter is not installed');
+            if (typeof text !== 'string' || text.trim() === '' || Buffer.byteLength(text, 'utf8') > MAX_TEXT_BYTES)
+                throw new Error('invalid benchmark inbound text');
+            inboundSent = true;
+            const accepted = await delivery.acceptInbound({ channel, account, eventId: `benchmark-inbound-${fingerprint}-${++eventNumber}`,
+                occurredAt: Date.now(), principal, conversation, kind: 'text', text });
+            return Object.freeze(accepted);
+        },
+        waitForQuiescence,
+        ownerScope,
+        shutdown() {
+            if (shutdownFlight !== undefined)
+                return shutdownFlight;
+            closed = true;
+            shutdownFlight = (async () => {
+                for (const agent of ctx.agents.list())
+                    agent.cancel({ kind: 'hook', reason: 'benchmark-owner-shutdown' });
+                const errors = [];
+                for (const cleanup of [() => delivery.whenIdle(), () => removeChannel?.(), () => removeAdapter?.(), () => ctx.fiber.dispose()]) {
+                    try {
+                        await cleanup();
+                    }
+                    catch (error) {
+                        errors.push(error);
+                    }
+                }
+                if (errors.length > 0)
+                    throw new AggregateError(errors, 'benchmark owner shutdown failed');
+            })();
+            return shutdownFlight;
+        },
+    });
+}
+//# sourceMappingURL=strategy-owner.js.map