@danceiny/gotry 0.0.1-rc.18 → 0.0.1-rc.20

package/dist/src/artifact-gate.js

@@ -1,4 +1,4 @@
-import { flightClaimVerdict, itineraryInvariants, latestFactsForRouteDate } from './bookable-facts.js';
+import { flightClaimVerdict, hotelClaimVerdict, itineraryInvariants, latestFactsForRouteDate } from './bookable-facts.js';
 export const ARTIFACT_GATE_SCHEMA = 'gotry_artifact_gate.v1';
 const FLIGHT_NO = /(?<![A-Za-z0-9])([A-Z0-9]{2}\d{3,4})(?![\d])/g;
 const CARRIER_DIRECT = /(?<![A-Za-z0-9])([A-Z0-9]{2})(?!\d)(?:\s|[一-龥]){0,6}直飞/g;
@@ -26,6 +26,12 @@ function hasDirectAssertion(line) {
     const before = line.slice(Math.max(0, idx - 3), idx);
     return !(before.endsWith('无') || before.endsWith('不') || before.endsWith('没有') || before.endsWith('未见'));
 }
+function occ1(line, map) {
+    for (const city of Object.keys(map.city_alias)){
+        if (line.includes(city)) return city;
+    }
+    return undefined;
+}
 function routeCtxOf(text, map, defaultYear) {
     const occ = [];
     for (const [city, iata] of Object.entries(map.city_alias)){
@@ -72,13 +78,21 @@ export function extractClaims(markdown, map, opts) {
         flights: [],
         policies: [],
         airports: [],
-        direct_lines: []
+        direct_lines: [],
+        hotels: [],
+        anchors: new Map()
     };
     const lines = markdown.split('\n');
     let section = {};
+    const ANCHOR = /fact:([0-9a-f]{16})/;
     for(let i = 0; i < lines.length; i++){
         const line = lines[i];
         const lineNo = i + 1;
+        const anchorMatch = line.match(ANCHOR);
+        if (anchorMatch) {
+            claims.anchors.set(lineNo, anchorMatch[1].toLowerCase());
+            continue;
+        }
         if (/^#{1,4}\s/.test(line)) {
             const ctx = routeCtxOf(line, map, opts?.trip_year);
             section = {
@@ -145,6 +159,20 @@ export function extractClaims(markdown, map, opts) {
                 has_as_of: AS_OF_WORD.test(line)
             });
         }
+        const HOTEL_ASSERT = /酒店|住宿|客栈|民宿/;
+        const HOTEL_BOOKABLE = /有房|可订|在架|已核验|已验证|可住|✓|✅/;
+        if (HOTEL_ASSERT.test(line) && HOTEL_BOOKABLE.test(line)) {
+            const stays = [
+                ...line.matchAll(/(?<!\d)(\d{4}-\d{2}-\d{2})(?!\d)/g)
+            ].map((m)=>m[1]);
+            claims.hotels.push({
+                line: lineNo,
+                text: line.trim().slice(0, 160),
+                destination: occ1(line, map),
+                check_in: stays[0],
+                check_out: stays[1]
+            });
+        }
         for (const [city, carriers] of Object.entries(map.carrier_airport)){
             const airports = [
                 ...new Set(Object.values(carriers))
@@ -176,6 +204,66 @@ export function gateArtifact(markdown, facts, map, opts) {
     const claims = extractClaims(markdown, map, opts);
     const violations = [];
     let traceable = 0;
+    for (const [lineNo, factId] of claims.anchors){
+        const f = facts.find((x)=>x.fact_id === factId);
+        if (!f) {
+            violations.push({
+                kind: 'fact_anchor_unknown',
+                line: lineNo,
+                detail: `渲染锚点 fact:${factId} 不在事实注册表——锚点被手改或伪造,产物不可信`
+            });
+            continue;
+        }
+        if (f.kind !== 'policy' && f.bookability === 'unavailable_exact_date') {
+            violations.push({
+                kind: 'not_in_source',
+                line: lineNo,
+                detail: `锚点事实为 exact-date 负事实(${f.fetched_at ?? ''})——负事实对应的可住/可订断言不得出现`
+            });
+            continue;
+        }
+        traceable++;
+    }
+    for (const c of claims.hotels){
+        if (claims.anchors.has(c.line)) continue;
+        if (!c.destination) {
+            violations.push({
+                kind: 'unverifiable_hotel_claim',
+                line: c.line,
+                detail: `酒店可住断言缺目的地上下文,无法回溯 exact-date 事实——fail closed 按未核验处理`
+            });
+            continue;
+        }
+        const r = hotelClaimVerdict(facts, {
+            destination: c.destination,
+            check_in: c.check_in,
+            check_out: c.check_out
+        });
+        if (r.verdict === 'traceable') {
+            traceable++;
+            continue;
+        }
+        if (r.verdict === 'unverified') {
+            violations.push({
+                kind: 'unverifiable_hotel_claim',
+                line: c.line,
+                detail: r.reason
+            });
+        } else {
+            violations.push({
+                kind: 'not_in_source',
+                line: c.line,
+                detail: r.reason
+            });
+        }
+        if (CHECK_MARK.test(c.text)) {
+            violations.push({
+                kind: 'unconditional_check',
+                line: c.line,
+                detail: '对未核验酒店使用无条件 ✓/✅——只有 bookable_exact_date 才允许确定性标记'
+            });
+        }
+    }
     for (const c of claims.flights){
         if (c.carrier_only) {
             const hasCarrier = facts.some((f)=>f.kind === 'flight' && f.bookability === 'bookable_exact_date' && f.nonstop !== false && f.flight_no.startsWith(c.carrier_only) && (!c.origin || f.route.origin === c.origin || f.route.origin_airport === c.origin) && (!c.destination || f.route.destination === c.destination || f.route.dest_airport === c.destination) && (!c.date || f.date === c.date));
@@ -277,7 +365,7 @@ export function gateArtifact(markdown, facts, map, opts) {
         const nos = [
             ...line.matchAll(FLIGHT_NO)
         ].map((m)=>m[1].toUpperCase());
-        const matched = facts.filter((f)=>f.kind !== 'policy' && nos.includes(f.flight_no.toUpperCase()));
+        const matched = facts.filter((f)=>(f.kind === 'flight' || f.kind === 'train') && nos.includes(f.flight_no.toUpperCase()));
         const allProtected = nos.length > 0 && matched.length === nos.length && matched.every((f)=>f.protected_connection === true);
         if (!allProtected) {
             violations.push({

package/dist/src/benchmark-agent-conformance.js

@@ -12,6 +12,13 @@ export const BENCHMARK_BRIDGE_RETRY_CALL_NOT_ALLOWED = 'BENCHMARK_BRIDGE_RETRY_C
 export const BENCHMARK_CONFORMANCE_STATE_UNAVAILABLE = 'BENCHMARK_CONFORMANCE_STATE_UNAVAILABLE';
 const IDENTIFIER = /^[A-Za-z][A-Za-z0-9_.-]*$/;
 const MAX_TERMINAL_BYTES = 1024 * 1024;
+const BENCHMARK_TOOL_RESULT_SCHEMA_VERSION = 'gotry_benchmark_tool_result_v1';
+const BENCHMARK_DOMAIN_RECOVERIES = new Set([
+    'none',
+    'retry_same',
+    'revise_arguments',
+    'choose_alternative'
+]);
 function plainObject(value) {
     return typeof value === 'object' && value !== null && !Array.isArray(value);
 }
@@ -29,10 +36,13 @@ function invalidTerminal() {
         error: 'invalid_terminal_output'
     };
 }
+function stripReasoningBlocks(raw) {
+    return raw.replace(/<think>[\s\S]*?<\/think>\s*/gi, '');
+}
 export function parseBenchmarkTerminal(raw, config) {
     if (!validateTerminalOutputConfig(config)) return invalidTerminal();
     if (Buffer.byteLength(raw, 'utf8') > config.max_bytes) return invalidTerminal();
-    const trimmed = raw.trim();
+    const trimmed = stripReasoningBlocks(raw).trim();
     const opening = `<${config.tag}>`;
     const closing = `</${config.tag}>`;
     if (!trimmed.startsWith(opening) || !trimmed.endsWith(closing)) return invalidTerminal();
@@ -87,13 +97,23 @@ function bridgeResultStatus(data, callId) {
     try {
         const parsed = JSON.parse(text);
         if (!plainObject(parsed)) return undefined;
-        if (parsed.ok === true) return {
-            ok: true
+        const keys = Object.keys(parsed).sort().join(',');
+        if (parsed.ok === true && keys === 'ok,result' && (plainObject(parsed.result) || Array.isArray(parsed.result))) return {
+            kind: 'result'
         };
+        if (parsed.ok === true && keys === 'ok,outcome' && plainObject(parsed.outcome)) {
+            const outcomeKeys = Object.keys(parsed.outcome).sort().join(',');
+            if (outcomeKeys === 'code,recovery,schema_version,status' && parsed.outcome.schema_version === BENCHMARK_TOOL_RESULT_SCHEMA_VERSION && (parsed.outcome.status === 'miss' || parsed.outcome.status === 'error') && typeof parsed.outcome.code === 'string' && IDENTIFIER.test(parsed.outcome.code) && typeof parsed.outcome.recovery === 'string' && BENCHMARK_DOMAIN_RECOVERIES.has(parsed.outcome.recovery)) return {
+                kind: 'domain'
+            };
+            return undefined;
+        }
         if (parsed.ok === false && typeof parsed.error === 'string') {
+            const exactFailure = parsed.error === 'runner_failed' ? keys === 'error,ok' || keys === 'error,exit_code,ok,signal' && (parsed.exit_code === null || typeof parsed.exit_code === 'number' && Number.isInteger(parsed.exit_code)) && (parsed.signal === null || typeof parsed.signal === 'string') : parsed.error === 'invalid_arguments' ? keys === 'error,ok,reason' && typeof parsed.reason === 'string' : keys === 'error,ok';
+            if (!exactFailure) return undefined;
             const code = parsed.error === 'timed_out' ? BENCHMARK_BRIDGE_TIMED_OUT : parsed.error === 'runner_failed' ? BENCHMARK_BRIDGE_RUNNER_FAILED : parsed.error === 'spawn_failed' ? BENCHMARK_BRIDGE_SPAWN_FAILED : parsed.error === 'output_truncated' ? BENCHMARK_BRIDGE_OUTPUT_TRUNCATED : BENCHMARK_BRIDGE_CALL_FAILED;
             return {
-                ok: false,
+                kind: 'failure',
                 code
             };
         }
@@ -150,8 +170,8 @@ export function createBenchmarkAgentConformance(projection) {
                 }
                 if (typeof event.data.callId !== 'string') return;
                 const args = parseToolArguments(event.data.arguments);
-                const query = args && plainObject(args.query) ? args.query : undefined;
-                if (query?.action === 'call' && typeof query.tool === 'string' && projection.allowedTools.includes(query.tool)) {
+                const flat = args && plainObject(args) ? args : undefined;
+                if (flat?.action === 'call' && typeof flat.tool === 'string' && projection.allowedTools.includes(flat.tool)) {
                     state.validCallIds.add(event.data.callId);
                 }
                 return;
@@ -161,8 +181,11 @@ export function createBenchmarkAgentConformance(projection) {
                 const callId = event.data.message.source.callId;
                 if (typeof callId !== 'string' || !state.validCallIds.has(callId)) return;
                 const resultStatus = bridgeResultStatus(event.data, callId);
-                if (resultStatus?.ok === true) {
+                if (typeof event.data.step === 'number') state.lastBridgeResponseStep = event.data.step;
+                if (resultStatus?.kind === 'result') {
                     state.successfulResultStep = typeof event.data.step === 'number' ? event.data.step : state.successfulResultStep;
+                } else if (resultStatus?.kind === 'domain') {
+                    state.domainOutcomeStep = typeof event.data.step === 'number' ? event.data.step : state.domainOutcomeStep;
                 } else {
                     state.bridgeCallFailed = true;
                     state.bridgeFailureCode = resultStatus?.code ?? BENCHMARK_BRIDGE_CALL_FAILED;
@@ -191,13 +214,13 @@ export function createBenchmarkAgentConformance(projection) {
                     code: BENCHMARK_BRIDGE_RETRY_CALL_NOT_ALLOWED
                 };
             }
-            if (state.successfulResultStep === undefined) {
-                if (state.bridgeCallFailed) {
-                    return {
-                        kind: 'reject',
-                        code: state.bridgeFailureCode ?? BENCHMARK_BRIDGE_CALL_FAILED
-                    };
-                }
+            if (state.bridgeCallFailed && state.successfulResultStep === undefined) {
+                return {
+                    kind: 'reject',
+                    code: state.bridgeFailureCode ?? BENCHMARK_BRIDGE_CALL_FAILED
+                };
+            }
+            if (state.successfulResultStep === undefined && state.domainOutcomeStep === undefined) {
                 if (state.retryCount >= MAX_CONFORMANCE_RETRIES) {
                     return {
                         kind: 'reject',
@@ -212,7 +235,7 @@ export function createBenchmarkAgentConformance(projection) {
                 };
             }
             const terminal = state.lastAssistant;
-            const terminalIsValid = terminal !== undefined && !terminal.interrupted && terminal.step > state.successfulResultStep && parseBenchmarkTerminal(terminal.text, projection.terminal).ok;
+            const terminalIsValid = terminal !== undefined && !terminal.interrupted && terminal.step > state.lastBridgeResponseStep && parseBenchmarkTerminal(terminal.text, projection.terminal).ok;
             if (terminalIsValid) return {
                 kind: 'accept'
             };
@@ -261,7 +284,7 @@ function systemSection(projection) {
         text: [
             'Benchmark execution contract:',
             `- Translate every task instruction to use a CLI, shell, Python, or agent_env.cli into the native tool ${projection.toolName}; do not merely describe the intended command.`,
-            `- Call it with exactly {"query":{"action":"call","tool":"<one of: ${allowed}>","arguments":{...}}}.`,
+            `- Call it with exactly {"action":"call","tool":"<one of: ${allowed}>","arguments":{...}}.`,
             '- action:"tools" is discovery only and does not satisfy the required environment call.',
             `- After a successful tool result, reply only <${projection.terminal.tag}>{...one JSON object...}</${projection.terminal.tag}> with no prose or code fence.`,
             '- If a terminal-format correction arrives, reuse the existing result and do not call the tool again.'