@namzu/sdk 42.0.2 → 44.0.0

package/dist/eval/judge.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"judge.d.ts","sourceRoot":"","sources":["../../src/eval/judge.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,gCAAgC,CAAA;AACjE,OAAO,KAAK,EAA4B,MAAM,EAAE,MAAM,YAAY,CAAA;AAElE;;;;;;;;;;;;;;GAcG;AACH,MAAM,WAAW,iBAAiB;IACjC,qEAAqE;IACrE,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,QAAQ,EAAE,WAAW,CAAA;IACrB,KAAK,EAAE,MAAM,CAAA;IACb;;;;;;;OAOG;IACH,MAAM,EAAE,MAAM,CAAA;IACd;;;;;;;;;;;OAWG;IACH,KAAK,CAAC,EAAE,MAAM,CAAA;IACd;;;;;;OAMG;IACH,iBAAiB,CAAC,EAAE,OAAO,CAAA;IAC3B;;;;;;OAMG;IACH,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB;;;;OAIG;IACH,mBAAmB,CAAC,EAAE,MAAM,CAAA;CAC5B;AAwID,wBAAgB,WAAW,CAAC,MAAM,EAAE,iBAAiB,GAAG,MAAM,CA6C7D"}
+{"version":3,"file":"judge.d.ts","sourceRoot":"","sources":["../../src/eval/judge.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,gCAAgC,CAAA;AACjE,OAAO,KAAK,EAA6B,MAAM,EAAE,MAAM,YAAY,CAAA;AAEnE;;;;;;;;;;;;;;GAcG;AACH,MAAM,WAAW,iBAAiB;IACjC,qEAAqE;IACrE,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,QAAQ,EAAE,WAAW,CAAA;IACrB,KAAK,EAAE,MAAM,CAAA;IACb;;;;;;;OAOG;IACH,MAAM,EAAE,MAAM,CAAA;IACd;;;;;;;;;;;OAWG;IACH,KAAK,CAAC,EAAE,MAAM,CAAA;IACd;;;;;;OAMG;IACH,iBAAiB,CAAC,EAAE,OAAO,CAAA;IAC3B;;;;;;OAMG;IACH,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB;;;;OAIG;IACH,mBAAmB,CAAC,EAAE,MAAM,CAAA;CAC5B;AAwID,wBAAgB,WAAW,CAAC,MAAM,EAAE,iBAAiB,GAAG,MAAM,CA6C7D"}

package/dist/eval/judge.js

@@ -70,9 +70,9 @@ function parseVerdict(reply, scale) {
         reason: typeof reason === 'string' && reason.length > 0 ? reason : '(judge gave no reason)',
     };
 }
-function buildPrompt(config, run, evalCase, scale) {
+function buildPrompt(config, turn, evalCase, scale) {
     const limit = config.maxOutputChars ?? DEFAULT_MAX_OUTPUT_CHARS;
-    const answer = run.output ?? '';
+    const answer = turn.output ?? '';
     const truncated = answer.length > limit;
     const sections = [
         'Grade the ANSWER against the RUBRIC.',
@@ -87,7 +87,7 @@ function buildPrompt(config, run, evalCase, scale) {
             : JSON.stringify(evalCase.expected)}`);
     }
     if (config.includeTrajectory === true) {
-        sections.push('', `TOOLS CALLED, in order:\n${run.toolCalls.length > 0 ? run.toolCalls.join(' -> ') : '(none)'}`);
+        sections.push('', `TOOLS CALLED, in order:\n${turn.toolCalls.length > 0 ? turn.toolCalls.join(' -> ') : '(none)'}`);
     }
     sections.push('', `ANSWER:\n${truncated ? answer.slice(0, limit) : answer}`);
     if (truncated) {
@@ -111,11 +111,11 @@ export function judgeScorer(config) {
     }
     return {
         name: config.name ?? 'judge',
-        async score(run, evalCase, signal) {
+        async score(turn, evalCase, signal) {
             const response = await collectChatCompletion(provider.chatStream({
                 model: config.model,
-                messages: [{ role: 'user', content: buildPrompt(config, run, evalCase, scale) }],
-                // The same run must grade the same way twice, or a
+                messages: [{ role: 'user', content: buildPrompt(config, turn, evalCase, scale) }],
+                // The same turn must grade the same way twice, or a
                 // regression cannot be told from sampling noise.
                 temperature: 0,
                 maxTokens: 512,

package/dist/eval/judge.js.map

@@ -1 +1 @@
-{"version":3,"file":"judge.js","sourceRoot":"","sources":["../../src/eval/judge.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,qBAAqB,EAAE,MAAM,wCAAwC,CAAA;AAC9E,OAAO,EAAE,0BAA0B,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAA;AAsE/F,MAAM,aAAa,GAAG,CAAC,CAAA;AACvB,MAAM,wBAAwB,GAAG,MAAM,CAAA;AAOvC;;;;;;;;GAQG;AACH,SAAS,YAAY,CAAC,KAAa,EAAE,KAAa;IACjD,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAA;IAChC,IAAI,KAAK,KAAK,CAAC,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,kCAAkC,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAA;IAE1F,IAAI,KAAK,GAAG,CAAC,CAAA;IACb,IAAI,QAAQ,GAAG,KAAK,CAAA;IACpB,IAAI,OAAO,GAAG,KAAK,CAAA;IACnB,IAAI,GAAG,GAAG,CAAC,CAAC,CAAA;IACZ,KAAK,IAAI,CAAC,GAAG,KAAK,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAC3C,MAAM,EAAE,GAAG,KAAK,CAAC,CAAC,CAAC,CAAA;QACnB,IAAI,OAAO,EAAE,CAAC;YACb,OAAO,GAAG,KAAK,CAAA;YACf,SAAQ;QACT,CAAC;QACD,IAAI,EAAE,KAAK,IAAI,EAAE,CAAC;YACjB,OAAO,GAAG,IAAI,CAAA;YACd,SAAQ;QACT,CAAC;QACD,IAAI,EAAE,KAAK,GAAG,EAAE,CAAC;YAChB,QAAQ,GAAG,CAAC,QAAQ,CAAA;YACpB,SAAQ;QACT,CAAC;QACD,IAAI,QAAQ;YAAE,SAAQ;QACtB,IAAI,EAAE,KAAK,GAAG;YAAE,KAAK,EAAE,CAAA;aAClB,IAAI,EAAE,KAAK,GAAG,EAAE,CAAC;YACrB,KAAK,EAAE,CAAA;YACP,IAAI,KAAK,KAAK,CAAC,EAAE,CAAC;gBACjB,GAAG,GAAG,CAAC,CAAA;gBACP,MAAK;YACN,CAAC;QACF,CAAC;IACF,CAAC;IACD,IAAI,GAAG,KAAK,CAAC,CAAC;QACb,MAAM,IAAI,KAAK,CAAC,+CAA+C,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAA;IAEtF,IAAI,MAAe,CAAA;IACnB,IAAI,CAAC;QACJ,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,CAAA;IACjD,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACd,MAAM,IAAI,KAAK,CACd,mCAAmC,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CACrF,CAAA;IACF,CAAC;IAED,MAAM,KAAK,GAAI,MAA8B,CAAC,KAAK,CAAA;IACnD,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;QAC1D,MAAM,IAAI,KAAK,CAAC,oCAAoC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAA;IAC5F,CAAC;IACD,IAAI,KAAK,GAAG,CAAC,IAAI,KAAK,GAAG,KAAK,EAAE,CAAC;QAChC,qEAAqE;QACrE,qEAAqE;QACrE,2BAA2B;QAC3B,MAAM,IAAI,KAAK,CAAC,kBAAkB,KAAK,oBAAoB,KAAK,qBAAqB,CAAC,CAAA;IACvF,CAAC;IAED,MAAM,MAAM,GAAI,MAA+B,CAAC,MAAM,CAAA;IACtD,OAAO;QACN,KAAK;QACL,MAAM,EAAE,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,wBAAwB;KAC3F,CAAA;AACF,CAAC;AAED,SAAS,WAAW,CACnB,MAAyB,EACzB,GAAY,EACZ,QAAkB,EAClB,KAAa;IAEb,MAAM,KAAK,GAAG,MAAM,CAAC,cAAc,IAAI,wBAAwB,CAAA;IAC/D,MAAM,MAAM,GAAG,GAAG,CAAC,MAAM,IAAI,EAAE,CAAA;IAC/B,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM,GAAG,KAAK,CAAA;IAEvC,MAAM,QAAQ,GAAa;QAC1B,sCAAsC;QACtC,EAAE;QACF,YAAY,MAAM,CAAC,MAAM,EAAE;QAC3B,EAAE;QACF,UAAU,OAAO,QAAQ,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE;KAChG,CAAA;IAED,IAAI,QAAQ,CAAC,QAAQ,KAAK,SAAS,EAAE,CAAC;QACrC,QAAQ,CAAC,IAAI,CACZ,EAAE,EACF,mDACC,OAAO,QAAQ,CAAC,QAAQ,KAAK,QAAQ;YACpC,CAAC,CAAC,QAAQ,CAAC,QAAQ;YACnB,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,QAAQ,CACpC,EAAE,CACF,CAAA;IACF,CAAC;IAED,IAAI,MAAM,CAAC,iBAAiB,KAAK,IAAI,EAAE,CAAC;QACvC,QAAQ,CAAC,IAAI,CACZ,EAAE,EACF,4BAA4B,GAAG,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,SAAS,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE,CAC9F,CAAA;IACF,CAAC;IAED,QAAQ,CAAC,IAAI,CAAC,EAAE,EAAE,YAAY,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAA;IAC5E,IAAI,SAAS,EAAE,CAAC;QACf,+DAA+D;QAC/D,qBAAqB;QACrB,QAAQ,CAAC,IAAI,CACZ,EAAE,EACF,gCAAgC,KAAK,wFAAwF,CAC7H,CAAA;IACF,CAAC;IAED,QAAQ,CAAC,IAAI,CACZ,EAAE,EACF,2DAA2D,KAAK,yCAAyC,EACzG,yCAAyC,KAAK,wDAAwD,CACtG,CAAA;IAED,OAAO,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;AAC3B,CAAC;AAED,MAAM,UAAU,WAAW,CAAC,MAAyB;IACpD,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,IAAI,aAAa,CAAA;IAC3C,MAAM,QAAQ,GAAG,qBAAqB,CAAC,MAAM,CAAC,QAAQ,EAAE;QACvD,aAAa,EAAE,0BAA0B,CAAC,MAAM,CAAC,mBAAmB,CAAC;KACrE,CAAC,CAAA;IACF,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC,EAAE,CAAC;QAC3C,MAAM,IAAI,KAAK,CAAC,sDAAsD,KAAK,EAAE,CAAC,CAAA;IAC/E,CAAC;IACD,IAAI,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvC,MAAM,IAAI,KAAK,CACd,qHAAqH,CACrH,CAAA;IACF,CAAC;IAED,OAAO;QACN,IAAI,EAAE,MAAM,CAAC,IAAI,IAAI,OAAO;QAC5B,KAAK,CAAC,KAAK,CAAC,GAAY,EAAE,QAAkB,EAAE,MAAoB;YACjE,MAAM,QAAQ,GAAG,MAAM,qBAAqB,CAC3C,QAAQ,CAAC,UAAU,CAAC;gBACnB,KAAK,EAAE,MAAM,CAAC,KAAK;gBACnB,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,CAAC,MAAM,EAAE,GAAG,EAAE,QAAQ,EAAE,KAAK,CAAC,EAAE,CAAC;gBAChF,mDAAmD;gBACnD,iDAAiD;gBACjD,WAAW,EAAE,CAAC;gBACd,SAAS,EAAE,GAAG;gBACd,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC7B,CAAC,CACF,CAAA;YAED,MAAM,OAAO,GAAG,YAAY,CAAC,QAAQ,CAAC,OAAO,CAAC,OAAO,IAAI,EAAE,EAAE,KAAK,CAAC,CAAA;YACnE,OAAO;gBACN,KAAK,EAAE,OAAO,CAAC,KAAK,GAAG,KAAK;gBAC5B,MAAM,EAAE,OAAO,CAAC,MAAM;gBACtB,OAAO,EAAE;oBACR,KAAK,EAAE,OAAO,CAAC,KAAK;oBACpB,KAAK;oBACL,sDAAsD;oBACtD,0DAA0D;oBAC1D,uDAAuD;oBACvD,YAAY;oBACZ,WAAW,EAAE,QAAQ,CAAC,KAAK,CAAC,WAAW;iBACvC;aACD,CAAA;QACF,CAAC;KACD,CAAA;AACF,CAAC"}
+{"version":3,"file":"judge.js","sourceRoot":"","sources":["../../src/eval/judge.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,qBAAqB,EAAE,MAAM,wCAAwC,CAAA;AAC9E,OAAO,EAAE,0BAA0B,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAA;AAsE/F,MAAM,aAAa,GAAG,CAAC,CAAA;AACvB,MAAM,wBAAwB,GAAG,MAAM,CAAA;AAOvC;;;;;;;;GAQG;AACH,SAAS,YAAY,CAAC,KAAa,EAAE,KAAa;IACjD,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAA;IAChC,IAAI,KAAK,KAAK,CAAC,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,kCAAkC,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAA;IAE1F,IAAI,KAAK,GAAG,CAAC,CAAA;IACb,IAAI,QAAQ,GAAG,KAAK,CAAA;IACpB,IAAI,OAAO,GAAG,KAAK,CAAA;IACnB,IAAI,GAAG,GAAG,CAAC,CAAC,CAAA;IACZ,KAAK,IAAI,CAAC,GAAG,KAAK,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAC3C,MAAM,EAAE,GAAG,KAAK,CAAC,CAAC,CAAC,CAAA;QACnB,IAAI,OAAO,EAAE,CAAC;YACb,OAAO,GAAG,KAAK,CAAA;YACf,SAAQ;QACT,CAAC;QACD,IAAI,EAAE,KAAK,IAAI,EAAE,CAAC;YACjB,OAAO,GAAG,IAAI,CAAA;YACd,SAAQ;QACT,CAAC;QACD,IAAI,EAAE,KAAK,GAAG,EAAE,CAAC;YAChB,QAAQ,GAAG,CAAC,QAAQ,CAAA;YACpB,SAAQ;QACT,CAAC;QACD,IAAI,QAAQ;YAAE,SAAQ;QACtB,IAAI,EAAE,KAAK,GAAG;YAAE,KAAK,EAAE,CAAA;aAClB,IAAI,EAAE,KAAK,GAAG,EAAE,CAAC;YACrB,KAAK,EAAE,CAAA;YACP,IAAI,KAAK,KAAK,CAAC,EAAE,CAAC;gBACjB,GAAG,GAAG,CAAC,CAAA;gBACP,MAAK;YACN,CAAC;QACF,CAAC;IACF,CAAC;IACD,IAAI,GAAG,KAAK,CAAC,CAAC;QACb,MAAM,IAAI,KAAK,CAAC,+CAA+C,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAA;IAEtF,IAAI,MAAe,CAAA;IACnB,IAAI,CAAC;QACJ,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,CAAA;IACjD,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACd,MAAM,IAAI,KAAK,CACd,mCAAmC,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CACrF,CAAA;IACF,CAAC;IAED,MAAM,KAAK,GAAI,MAA8B,CAAC,KAAK,CAAA;IACnD,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;QAC1D,MAAM,IAAI,KAAK,CAAC,oCAAoC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAA;IAC5F,CAAC;IACD,IAAI,KAAK,GAAG,CAAC,IAAI,KAAK,GAAG,KAAK,EAAE,CAAC;QAChC,qEAAqE;QACrE,qEAAqE;QACrE,2BAA2B;QAC3B,MAAM,IAAI,KAAK,CAAC,kBAAkB,KAAK,oBAAoB,KAAK,qBAAqB,CAAC,CAAA;IACvF,CAAC;IAED,MAAM,MAAM,GAAI,MAA+B,CAAC,MAAM,CAAA;IACtD,OAAO;QACN,KAAK;QACL,MAAM,EAAE,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,wBAAwB;KAC3F,CAAA;AACF,CAAC;AAED,SAAS,WAAW,CACnB,MAAyB,EACzB,IAAc,EACd,QAAkB,EAClB,KAAa;IAEb,MAAM,KAAK,GAAG,MAAM,CAAC,cAAc,IAAI,wBAAwB,CAAA;IAC/D,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,IAAI,EAAE,CAAA;IAChC,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM,GAAG,KAAK,CAAA;IAEvC,MAAM,QAAQ,GAAa;QAC1B,sCAAsC;QACtC,EAAE;QACF,YAAY,MAAM,CAAC,MAAM,EAAE;QAC3B,EAAE;QACF,UAAU,OAAO,QAAQ,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE;KAChG,CAAA;IAED,IAAI,QAAQ,CAAC,QAAQ,KAAK,SAAS,EAAE,CAAC;QACrC,QAAQ,CAAC,IAAI,CACZ,EAAE,EACF,mDACC,OAAO,QAAQ,CAAC,QAAQ,KAAK,QAAQ;YACpC,CAAC,CAAC,QAAQ,CAAC,QAAQ;YACnB,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,QAAQ,CACpC,EAAE,CACF,CAAA;IACF,CAAC;IAED,IAAI,MAAM,CAAC,iBAAiB,KAAK,IAAI,EAAE,CAAC;QACvC,QAAQ,CAAC,IAAI,CACZ,EAAE,EACF,4BAA4B,IAAI,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE,CAChG,CAAA;IACF,CAAC;IAED,QAAQ,CAAC,IAAI,CAAC,EAAE,EAAE,YAAY,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAA;IAC5E,IAAI,SAAS,EAAE,CAAC;QACf,+DAA+D;QAC/D,qBAAqB;QACrB,QAAQ,CAAC,IAAI,CACZ,EAAE,EACF,gCAAgC,KAAK,wFAAwF,CAC7H,CAAA;IACF,CAAC;IAED,QAAQ,CAAC,IAAI,CACZ,EAAE,EACF,2DAA2D,KAAK,yCAAyC,EACzG,yCAAyC,KAAK,wDAAwD,CACtG,CAAA;IAED,OAAO,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;AAC3B,CAAC;AAED,MAAM,UAAU,WAAW,CAAC,MAAyB;IACpD,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,IAAI,aAAa,CAAA;IAC3C,MAAM,QAAQ,GAAG,qBAAqB,CAAC,MAAM,CAAC,QAAQ,EAAE;QACvD,aAAa,EAAE,0BAA0B,CAAC,MAAM,CAAC,mBAAmB,CAAC;KACrE,CAAC,CAAA;IACF,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC,EAAE,CAAC;QAC3C,MAAM,IAAI,KAAK,CAAC,sDAAsD,KAAK,EAAE,CAAC,CAAA;IAC/E,CAAC;IACD,IAAI,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvC,MAAM,IAAI,KAAK,CACd,qHAAqH,CACrH,CAAA;IACF,CAAC;IAED,OAAO;QACN,IAAI,EAAE,MAAM,CAAC,IAAI,IAAI,OAAO;QAC5B,KAAK,CAAC,KAAK,CAAC,IAAc,EAAE,QAAkB,EAAE,MAAoB;YACnE,MAAM,QAAQ,GAAG,MAAM,qBAAqB,CAC3C,QAAQ,CAAC,UAAU,CAAC;gBACnB,KAAK,EAAE,MAAM,CAAC,KAAK;gBACnB,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,CAAC,MAAM,EAAE,IAAI,EAAE,QAAQ,EAAE,KAAK,CAAC,EAAE,CAAC;gBACjF,oDAAoD;gBACpD,iDAAiD;gBACjD,WAAW,EAAE,CAAC;gBACd,SAAS,EAAE,GAAG;gBACd,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC7B,CAAC,CACF,CAAA;YAED,MAAM,OAAO,GAAG,YAAY,CAAC,QAAQ,CAAC,OAAO,CAAC,OAAO,IAAI,EAAE,EAAE,KAAK,CAAC,CAAA;YACnE,OAAO;gBACN,KAAK,EAAE,OAAO,CAAC,KAAK,GAAG,KAAK;gBAC5B,MAAM,EAAE,OAAO,CAAC,MAAM;gBACtB,OAAO,EAAE;oBACR,KAAK,EAAE,OAAO,CAAC,KAAK;oBACpB,KAAK;oBACL,sDAAsD;oBACtD,0DAA0D;oBAC1D,uDAAuD;oBACvD,YAAY;oBACZ,WAAW,EAAE,QAAQ,CAAC,KAAK,CAAC,WAAW;iBACvC;aACD,CAAA;QACF,CAAC;KACD,CAAA;AACF,CAAC"}

package/dist/eval/scorers.d.ts

@@ -1,6 +1,6 @@
-import type { EvalCase, EvalRun, Score, Scorer } from './types.js';
+import type { EvalCase, EvalTurn, Score, Scorer } from './types.js';
 /**
- * How closely the run's tool sequence matched the expected one, as F1 over
+ * How closely the turn's tool sequence matched the expected one, as F1 over
  * the longest common subsequence.
  *
  * Trajectory, not final answer. Namzu's most load-bearing behavior is
@@ -11,15 +11,15 @@ import type { EvalCase, EvalRun, Score, Scorer } from './types.js';
  * tool calls where it took one. Final-answer scoring cannot see that;
  * this can.
  *
- * Extra calls cut precision, missing calls cut recall, so a run that does
- * the right thing wastefully and a run that skips a step score
+ * Extra calls cut precision, missing calls cut recall, so a turn that does
+ * the right thing wastefully and a turn that skips a step score
  * differently — which is the distinction a final-answer score collapses.
  */
 export declare function trajectoryScorer(): Scorer;
-/** The run settled cleanly rather than erroring or being cut off. */
+/** The turn settled cleanly rather than erroring or being cut off. */
 export declare function completionScorer(acceptable?: readonly string[]): Scorer;
 /**
- * The run stayed within a step budget.
+ * The turn stayed within a step budget.
  *
  * A regression that makes the agent take four turns where it took one is
  * invisible to correctness scoring and very visible on the bill.
@@ -28,7 +28,7 @@ export declare function stepBudgetScorer(maxSteps: number): Scorer;
 /** The final text contains every required substring. */
 export declare function containsScorer(...required: string[]): Scorer;
 /**
- * Judge the run with a caller-supplied predicate.
+ * Judge the turn with a caller-supplied predicate.
  *
  * The escape hatch for anything the built-in scorers do not cover. For
  * a model-graded judge, reach for `judgeScorer` first: it handles the
@@ -36,5 +36,5 @@ export declare function containsScorer(...required: string[]): Scorer;
  * an ordinal scale rather than a float, disclosed truncation, and a
  * failed call that reports as unjudged rather than as a zero.
  */
-export declare function customScorer(name: string, fn: (run: EvalRun, evalCase: EvalCase) => Score | Promise<Score>): Scorer;
+export declare function customScorer(name: string, fn: (turn: EvalTurn, evalCase: EvalCase) => Score | Promise<Score>): Scorer;
 //# sourceMappingURL=scorers.d.ts.map

package/dist/eval/scorers.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"scorers.d.ts","sourceRoot":"","sources":["../../src/eval/scorers.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,YAAY,CAAA;AAyBlE;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,gBAAgB,IAAI,MAAM,CAwCzC;AAED,qEAAqE;AACrE,wBAAgB,gBAAgB,CAC/B,UAAU,GAAE,SAAS,MAAM,EAAmC,GAC5D,MAAM,CAqBR;AAED;;;;;GAKG;AACH,wBAAgB,gBAAgB,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAezD;AAED,wDAAwD;AACxD,wBAAgB,cAAc,CAAC,GAAG,QAAQ,EAAE,MAAM,EAAE,GAAG,MAAM,CAmB5D;AAED;;;;;;;;GAQG;AACH,wBAAgB,YAAY,CAC3B,IAAI,EAAE,MAAM,EACZ,EAAE,EAAE,CAAC,GAAG,EAAE,OAAO,EAAE,QAAQ,EAAE,QAAQ,KAAK,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,GAC9D,MAAM,CAER"}
+{"version":3,"file":"scorers.d.ts","sourceRoot":"","sources":["../../src/eval/scorers.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,YAAY,CAAA;AAyBnE;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,gBAAgB,IAAI,MAAM,CAwCzC;AAED,sEAAsE;AACtE,wBAAgB,gBAAgB,CAC/B,UAAU,GAAE,SAAS,MAAM,EAAmC,GAC5D,MAAM,CAqBR;AAED;;;;;GAKG;AACH,wBAAgB,gBAAgB,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAezD;AAED,wDAAwD;AACxD,wBAAgB,cAAc,CAAC,GAAG,QAAQ,EAAE,MAAM,EAAE,GAAG,MAAM,CAmB5D;AAED;;;;;;;;GAQG;AACH,wBAAgB,YAAY,CAC3B,IAAI,EAAE,MAAM,EACZ,EAAE,EAAE,CAAC,IAAI,EAAE,QAAQ,EAAE,QAAQ,EAAE,QAAQ,KAAK,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,GAChE,MAAM,CAER"}

package/dist/eval/scorers.js

@@ -2,7 +2,7 @@
  * Longest common subsequence length between two tool sequences.
  *
  * Subsequence, not set intersection: order carries meaning in a
- * trajectory. Reading a file before editing it is not the same run as
+ * trajectory. Reading a file before editing it is not the same turn as
  * editing then reading, and a set-based score cannot tell them apart.
  */
 function lcsLength(a, b) {
@@ -20,7 +20,7 @@ function lcsLength(a, b) {
     return table[a.length]?.[b.length] ?? 0;
 }
 /**
- * How closely the run's tool sequence matched the expected one, as F1 over
+ * How closely the turn's tool sequence matched the expected one, as F1 over
  * the longest common subsequence.
  *
  * Trajectory, not final answer. Namzu's most load-bearing behavior is
@@ -31,16 +31,16 @@ function lcsLength(a, b) {
  * tool calls where it took one. Final-answer scoring cannot see that;
  * this can.
  *
- * Extra calls cut precision, missing calls cut recall, so a run that does
- * the right thing wastefully and a run that skips a step score
+ * Extra calls cut precision, missing calls cut recall, so a turn that does
+ * the right thing wastefully and a turn that skips a step score
  * differently — which is the distinction a final-answer score collapses.
  */
 export function trajectoryScorer() {
     return {
         name: 'trajectory',
-        score(run, evalCase) {
+        score(turn, evalCase) {
             const expected = evalCase.expectedTools ?? [];
-            const actual = run.toolCalls;
+            const actual = turn.toolCalls;
             if (expected.length === 0 && actual.length === 0) {
                 return { score: 1, reason: 'no tools expected and none called' };
             }
@@ -72,31 +72,31 @@ export function trajectoryScorer() {
         },
     };
 }
-/** The run settled cleanly rather than erroring or being cut off. */
+/** The turn settled cleanly rather than erroring or being cut off. */
 export function completionScorer(acceptable = ['end_turn', 'stop_condition']) {
     return {
         name: 'completion',
-        // Binary by construction: the run either settled cleanly or it did
+        // Binary by construction: the turn either settled cleanly or it did
         // not, so there is no score between 0 and 1 for a mean to soften.
         // Averaged in, a hard 0 here is carried by three good fuzzy scores.
         severity: 'gate',
-        score(run) {
-            if (run.error) {
-                return { score: 0, reason: `run failed: ${run.error}` };
+        score(turn) {
+            if (turn.error) {
+                return { score: 0, reason: `turn failed: ${turn.error}` };
             }
-            const ok = run.stopReason !== undefined && acceptable.includes(run.stopReason);
+            const ok = turn.stopReason !== undefined && acceptable.includes(turn.stopReason);
             return {
                 score: ok ? 1 : 0,
                 reason: ok
-                    ? `settled as ${run.stopReason}`
-                    : `settled as ${run.stopReason ?? 'unknown'}, expected one of ${acceptable.join(', ')}`,
-                details: { stopReason: run.stopReason },
+                    ? `settled as ${turn.stopReason}`
+                    : `settled as ${turn.stopReason ?? 'unknown'}, expected one of ${acceptable.join(', ')}`,
+                details: { stopReason: turn.stopReason },
             };
         },
     };
 }
 /**
- * The run stayed within a step budget.
+ * The turn stayed within a step budget.
  *
  * A regression that makes the agent take four turns where it took one is
  * invisible to correctness scoring and very visible on the bill.
@@ -104,8 +104,8 @@ export function completionScorer(acceptable = ['end_turn', 'stop_condition']) {
 export function stepBudgetScorer(maxSteps) {
     return {
         name: 'step-budget',
-        score(run) {
-            const used = run.steps.length;
+        score(turn) {
+            const used = turn.steps.length;
             return {
                 score: used <= maxSteps ? 1 : Math.max(0, maxSteps / used),
                 reason: used <= maxSteps
@@ -123,8 +123,8 @@ export function containsScorer(...required) {
         // A required fragment is required. Partial credit here reads as
         // "most of the answer was there", which is not what was asked.
         severity: 'gate',
-        score(run) {
-            const text = run.output ?? '';
+        score(turn) {
+            const text = turn.output ?? '';
             const missing = required.filter((r) => !text.includes(r));
             return {
                 score: required.length === 0 ? 1 : (required.length - missing.length) / required.length,
@@ -137,7 +137,7 @@ export function containsScorer(...required) {
     };
 }
 /**
- * Judge the run with a caller-supplied predicate.
+ * Judge the turn with a caller-supplied predicate.
  *
  * The escape hatch for anything the built-in scorers do not cover. For
  * a model-graded judge, reach for `judgeScorer` first: it handles the

package/dist/eval/scorers.js.map

@@ -1 +1 @@
-{"version":3,"file":"scorers.js","sourceRoot":"","sources":["../../src/eval/scorers.ts"],"names":[],"mappings":"AAEA;;;;;;GAMG;AACH,SAAS,SAAS,CAAC,CAAoB,EAAE,CAAoB;IAC5D,MAAM,KAAK,GAAe,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,EAAE,GAAG,EAAE,CACnE,IAAI,KAAK,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAC/B,CAAA;IACD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACpC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACpC,MAAM,GAAG,GAAG,KAAK,CAAC,CAAC,CAAC,CAAA;YACpB,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,CAAA;YACzB,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI;gBAAE,SAAQ;YAC3B,GAAG,CAAC,CAAC,CAAC;gBACL,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,CAAA;QAC1F,CAAC;IACF,CAAC;IACD,OAAO,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAA;AACxC,CAAC;AAED;;;;;;;;;;;;;;;GAeG;AACH,MAAM,UAAU,gBAAgB;IAC/B,OAAO;QACN,IAAI,EAAE,YAAY;QAClB,KAAK,CAAC,GAAY,EAAE,QAAkB;YACrC,MAAM,QAAQ,GAAG,QAAQ,CAAC,aAAa,IAAI,EAAE,CAAA;YAC7C,MAAM,MAAM,GAAG,GAAG,CAAC,SAAS,CAAA;YAE5B,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBAClD,OAAO,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,mCAAmC,EAAE,CAAA;YACjE,CAAC;YACD,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBAC3B,OAAO;oBACN,KAAK,EAAE,CAAC;oBACR,MAAM,EAAE,+BAA+B,MAAM,CAAC,MAAM,KAAK,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE;oBAC7E,OAAO,EAAE,EAAE,MAAM,EAAE;iBACnB,CAAA;YACF,CAAC;YACD,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBACzB,OAAO;oBACN,KAAK,EAAE,CAAC;oBACR,MAAM,EAAE,YAAY,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,4BAA4B;oBACpE,OAAO,EAAE,EAAE,QAAQ,EAAE;iBACrB,CAAA;YACF,CAAC;YAED,MAAM,OAAO,GAAG,SAAS,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAA;YAC3C,MAAM,SAAS,GAAG,OAAO,GAAG,MAAM,CAAC,MAAM,CAAA;YACzC,MAAM,MAAM,GAAG,OAAO,GAAG,QAAQ,CAAC,MAAM,CAAA;YACxC,MAAM,EAAE,GAAG,SAAS,GAAG,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,SAAS,GAAG,MAAM,CAAC,GAAG,CAAC,SAAS,GAAG,MAAM,CAAC,CAAA;YAEzF,OAAO;gBACN,KAAK,EAAE,EAAE;gBACT,MAAM,EACL,EAAE,KAAK,CAAC;oBACP,CAAC,CAAC,qBAAqB,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE;oBAC3C,CAAC,CAAC,WAAW,OAAO,IAAI,QAAQ,CAAC,MAAM,uBAAuB,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE;gBACjH,OAAO,EAAE,EAAE,SAAS,EAAE,MAAM,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,EAAE;aACzD,CAAA;QACF,CAAC;KACD,CAAA;AACF,CAAC;AAED,qEAAqE;AACrE,MAAM,UAAU,gBAAgB,CAC/B,aAAgC,CAAC,UAAU,EAAE,gBAAgB,CAAC;IAE9D,OAAO;QACN,IAAI,EAAE,YAAY;QAClB,mEAAmE;QACnE,kEAAkE;QAClE,oEAAoE;QACpE,QAAQ,EAAE,MAAM;QAChB,KAAK,CAAC,GAAY;YACjB,IAAI,GAAG,CAAC,KAAK,EAAE,CAAC;gBACf,OAAO,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,eAAe,GAAG,CAAC,KAAK,EAAE,EAAE,CAAA;YACxD,CAAC;YACD,MAAM,EAAE,GAAG,GAAG,CAAC,UAAU,KAAK,SAAS,IAAI,UAAU,CAAC,QAAQ,CAAC,GAAG,CAAC,UAAU,CAAC,CAAA;YAC9E,OAAO;gBACN,KAAK,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;gBACjB,MAAM,EAAE,EAAE;oBACT,CAAC,CAAC,cAAc,GAAG,CAAC,UAAU,EAAE;oBAChC,CAAC,CAAC,cAAc,GAAG,CAAC,UAAU,IAAI,SAAS,qBAAqB,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;gBACxF,OAAO,EAAE,EAAE,UAAU,EAAE,GAAG,CAAC,UAAU,EAAE;aACvC,CAAA;QACF,CAAC;KACD,CAAA;AACF,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,gBAAgB,CAAC,QAAgB;IAChD,OAAO;QACN,IAAI,EAAE,aAAa;QACnB,KAAK,CAAC,GAAY;YACjB,MAAM,IAAI,GAAG,GAAG,CAAC,KAAK,CAAC,MAAM,CAAA;YAC7B,OAAO;gBACN,KAAK,EAAE,IAAI,IAAI,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;gBAC1D,MAAM,EACL,IAAI,IAAI,QAAQ;oBACf,CAAC,CAAC,GAAG,IAAI,IAAI,QAAQ,QAAQ;oBAC7B,CAAC,CAAC,gBAAgB,IAAI,oBAAoB,QAAQ,YAAY;gBAChE,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;aAC3B,CAAA;QACF,CAAC;KACD,CAAA;AACF,CAAC;AAED,wDAAwD;AACxD,MAAM,UAAU,cAAc,CAAC,GAAG,QAAkB;IACnD,OAAO;QACN,IAAI,EAAE,UAAU;QAChB,gEAAgE;QAChE,+DAA+D;QAC/D,QAAQ,EAAE,MAAM;QAChB,KAAK,CAAC,GAAY;YACjB,MAAM,IAAI,GAAG,GAAG,CAAC,MAAM,IAAI,EAAE,CAAA;YAC7B,MAAM,OAAO,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAA;YACzD,OAAO;gBACN,KAAK,EAAE,QAAQ,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,GAAG,QAAQ,CAAC,MAAM;gBACvF,MAAM,EACL,OAAO,CAAC,MAAM,KAAK,CAAC;oBACnB,CAAC,CAAC,aAAa,QAAQ,CAAC,MAAM,qBAAqB;oBACnD,CAAC,CAAC,YAAY,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;gBACpC,OAAO,EAAE,EAAE,OAAO,EAAE;aACpB,CAAA;QACF,CAAC;KACD,CAAA;AACF,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,YAAY,CAC3B,IAAY,EACZ,EAAgE;IAEhE,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,EAAE,EAAE,CAAA;AAC3B,CAAC"}
+{"version":3,"file":"scorers.js","sourceRoot":"","sources":["../../src/eval/scorers.ts"],"names":[],"mappings":"AAEA;;;;;;GAMG;AACH,SAAS,SAAS,CAAC,CAAoB,EAAE,CAAoB;IAC5D,MAAM,KAAK,GAAe,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,EAAE,GAAG,EAAE,CACnE,IAAI,KAAK,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAC/B,CAAA;IACD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACpC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACpC,MAAM,GAAG,GAAG,KAAK,CAAC,CAAC,CAAC,CAAA;YACpB,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,CAAA;YACzB,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI;gBAAE,SAAQ;YAC3B,GAAG,CAAC,CAAC,CAAC;gBACL,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,CAAA;QAC1F,CAAC;IACF,CAAC;IACD,OAAO,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAA;AACxC,CAAC;AAED;;;;;;;;;;;;;;;GAeG;AACH,MAAM,UAAU,gBAAgB;IAC/B,OAAO;QACN,IAAI,EAAE,YAAY;QAClB,KAAK,CAAC,IAAc,EAAE,QAAkB;YACvC,MAAM,QAAQ,GAAG,QAAQ,CAAC,aAAa,IAAI,EAAE,CAAA;YAC7C,MAAM,MAAM,GAAG,IAAI,CAAC,SAAS,CAAA;YAE7B,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBAClD,OAAO,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,mCAAmC,EAAE,CAAA;YACjE,CAAC;YACD,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBAC3B,OAAO;oBACN,KAAK,EAAE,CAAC;oBACR,MAAM,EAAE,+BAA+B,MAAM,CAAC,MAAM,KAAK,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE;oBAC7E,OAAO,EAAE,EAAE,MAAM,EAAE;iBACnB,CAAA;YACF,CAAC;YACD,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBACzB,OAAO;oBACN,KAAK,EAAE,CAAC;oBACR,MAAM,EAAE,YAAY,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,4BAA4B;oBACpE,OAAO,EAAE,EAAE,QAAQ,EAAE;iBACrB,CAAA;YACF,CAAC;YAED,MAAM,OAAO,GAAG,SAAS,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAA;YAC3C,MAAM,SAAS,GAAG,OAAO,GAAG,MAAM,CAAC,MAAM,CAAA;YACzC,MAAM,MAAM,GAAG,OAAO,GAAG,QAAQ,CAAC,MAAM,CAAA;YACxC,MAAM,EAAE,GAAG,SAAS,GAAG,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,SAAS,GAAG,MAAM,CAAC,GAAG,CAAC,SAAS,GAAG,MAAM,CAAC,CAAA;YAEzF,OAAO;gBACN,KAAK,EAAE,EAAE;gBACT,MAAM,EACL,EAAE,KAAK,CAAC;oBACP,CAAC,CAAC,qBAAqB,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE;oBAC3C,CAAC,CAAC,WAAW,OAAO,IAAI,QAAQ,CAAC,MAAM,uBAAuB,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE;gBACjH,OAAO,EAAE,EAAE,SAAS,EAAE,MAAM,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,EAAE;aACzD,CAAA;QACF,CAAC;KACD,CAAA;AACF,CAAC;AAED,sEAAsE;AACtE,MAAM,UAAU,gBAAgB,CAC/B,aAAgC,CAAC,UAAU,EAAE,gBAAgB,CAAC;IAE9D,OAAO;QACN,IAAI,EAAE,YAAY;QAClB,oEAAoE;QACpE,kEAAkE;QAClE,oEAAoE;QACpE,QAAQ,EAAE,MAAM;QAChB,KAAK,CAAC,IAAc;YACnB,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;gBAChB,OAAO,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,gBAAgB,IAAI,CAAC,KAAK,EAAE,EAAE,CAAA;YAC1D,CAAC;YACD,MAAM,EAAE,GAAG,IAAI,CAAC,UAAU,KAAK,SAAS,IAAI,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,UAAU,CAAC,CAAA;YAChF,OAAO;gBACN,KAAK,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;gBACjB,MAAM,EAAE,EAAE;oBACT,CAAC,CAAC,cAAc,IAAI,CAAC,UAAU,EAAE;oBACjC,CAAC,CAAC,cAAc,IAAI,CAAC,UAAU,IAAI,SAAS,qBAAqB,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;gBACzF,OAAO,EAAE,EAAE,UAAU,EAAE,IAAI,CAAC,UAAU,EAAE;aACxC,CAAA;QACF,CAAC;KACD,CAAA;AACF,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,gBAAgB,CAAC,QAAgB;IAChD,OAAO;QACN,IAAI,EAAE,aAAa;QACnB,KAAK,CAAC,IAAc;YACnB,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAA;YAC9B,OAAO;gBACN,KAAK,EAAE,IAAI,IAAI,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;gBAC1D,MAAM,EACL,IAAI,IAAI,QAAQ;oBACf,CAAC,CAAC,GAAG,IAAI,IAAI,QAAQ,QAAQ;oBAC7B,CAAC,CAAC,gBAAgB,IAAI,oBAAoB,QAAQ,YAAY;gBAChE,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;aAC3B,CAAA;QACF,CAAC;KACD,CAAA;AACF,CAAC;AAED,wDAAwD;AACxD,MAAM,UAAU,cAAc,CAAC,GAAG,QAAkB;IACnD,OAAO;QACN,IAAI,EAAE,UAAU;QAChB,gEAAgE;QAChE,+DAA+D;QAC/D,QAAQ,EAAE,MAAM;QAChB,KAAK,CAAC,IAAc;YACnB,MAAM,IAAI,GAAG,IAAI,CAAC,MAAM,IAAI,EAAE,CAAA;YAC9B,MAAM,OAAO,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAA;YACzD,OAAO;gBACN,KAAK,EAAE,QAAQ,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,GAAG,QAAQ,CAAC,MAAM;gBACvF,MAAM,EACL,OAAO,CAAC,MAAM,KAAK,CAAC;oBACnB,CAAC,CAAC,aAAa,QAAQ,CAAC,MAAM,qBAAqB;oBACnD,CAAC,CAAC,YAAY,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;gBACpC,OAAO,EAAE,EAAE,OAAO,EAAE;aACpB,CAAA;QACF,CAAC;KACD,CAAA;AACF,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,YAAY,CAC3B,IAAY,EACZ,EAAkE;IAElE,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,EAAE,EAAE,CAAA;AAC3B,CAAC"}

package/dist/eval/types.d.ts

@@ -1,4 +1,4 @@
-import type { StepResult } from '../types/run/step.js';
+import type { StepResult } from '../types/session/step.js';
 import type { ScoreUncertainty } from './uncertainty.js';
 /**
  * One case in a dataset: an input, and what a good run looks like.
@@ -12,7 +12,7 @@ export interface EvalCase<TInput = unknown> {
     /** Free-form expectation, consumed by whichever scorers you attach. */
     expected?: unknown;
     /**
-     * Tool names the run is expected to call, in order. Consumed by
+     * Tool names the turn is expected to call, in order. Consumed by
      * `trajectoryScorer`.
      */
     expectedTools?: readonly string[];
@@ -20,8 +20,8 @@ export interface EvalCase<TInput = unknown> {
     scorers?: readonly Scorer[];
 }
 /** What actually happened, handed to every scorer. */
-export interface EvalRun {
-    /** Final assistant text, when the run produced one. */
+export interface EvalTurn {
+    /** Final assistant text, when the turn produced one. */
     output: string | null;
     /** Schema-validated final output, when `structuredOutput` was requested. */
     structuredOutput?: unknown;
@@ -32,13 +32,13 @@ export interface EvalRun {
     totalTokens: number;
     totalCostUsd: number;
     durationMs: number;
-    /** Set when the run threw rather than settling. */
+    /** Set when the turn threw rather than settling. */
     error?: string;
 }
 /**
- * A judgement about one run.
+ * A judgement about one turn.
  *
- * `reason` is REQUIRED, not optional. A bare number tells you a run got
+ * `reason` is REQUIRED, not optional. A bare number tells you a turn got
  * worse without telling you how, which is exactly when you need to know —
  * so every scorer has to be able to explain itself, and one that cannot
  * is a scorer that should not exist.
@@ -54,7 +54,7 @@ export interface Score {
      * unreachable, its verdict did not parse.
      *
      * A failed measurement is not a measurement of zero. Scoring it zero
-     * says "the run was bad" when the truth is "we do not know", and the
+     * says "the turn was bad" when the truth is "we do not know", and the
      * two demand opposite responses: one is a regression to chase, the
      * other is a broken harness to fix. This mattered little while every
      * scorer was a pure function — a throw there is a bug — and matters a
@@ -99,7 +99,7 @@ export interface Scorer {
      * scorer may ignore it; a scorer that owns I/O should carry it to that
      * transport so timed-out work is stopped rather than merely abandoned.
      */
-    score(run: EvalRun, evalCase: EvalCase, signal?: AbortSignal): Score | Promise<Score>;
+    score(turn: EvalTurn, evalCase: EvalCase, signal?: AbortSignal): Score | Promise<Score>;
     /** See {@link ScorerSeverity}. Default `soft`. */
     severity?: ScorerSeverity;
     /**
@@ -121,7 +121,7 @@ export interface Scorer {
 export type CaseStatus = 'passed' | 'failed' | 'inconclusive';
 export interface CaseResult {
     case: string;
-    run: EvalRun;
+    turn: EvalTurn;
     scores: Record<string, Score>;
     /** Mean of this case's AVAILABLE scores. Zero when none were. */
     mean: number;
@@ -146,8 +146,8 @@ export interface ExperimentReport {
      * How much of {@link mean} is signal.
      *
      * A mean on its own has been read as a result, and at the n a
-     * hand-built suite has it usually is not one: two runs three points
-     * apart are normally the same run twice.
+     * hand-built suite has it usually is not one: two experiments three points
+     * apart are normally the same experiment twice.
      *
      * Optional, and deliberately so after trying it the other way. A suite
      * file is loaded at runtime and may be plain JavaScript, so a required

package/dist/eval/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/eval/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAA;AACtD,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAA;AAExD;;;;;GAKG;AACH,MAAM,WAAW,QAAQ,CAAC,MAAM,GAAG,OAAO;IACzC,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,MAAM,CAAA;IACb,uEAAuE;IACvE,QAAQ,CAAC,EAAE,OAAO,CAAA;IAClB;;;OAGG;IACH,aAAa,CAAC,EAAE,SAAS,MAAM,EAAE,CAAA;IACjC,sEAAsE;IACtE,OAAO,CAAC,EAAE,SAAS,MAAM,EAAE,CAAA;CAC3B;AAED,sDAAsD;AACtD,MAAM,WAAW,OAAO;IACvB,uDAAuD;IACvD,MAAM,EAAE,MAAM,GAAG,IAAI,CAAA;IACrB,4EAA4E;IAC5E,gBAAgB,CAAC,EAAE,OAAO,CAAA;IAC1B,KAAK,EAAE,SAAS,UAAU,EAAE,CAAA;IAC5B,wDAAwD;IACxD,SAAS,EAAE,SAAS,MAAM,EAAE,CAAA;IAC5B,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,WAAW,EAAE,MAAM,CAAA;IACnB,YAAY,EAAE,MAAM,CAAA;IACpB,UAAU,EAAE,MAAM,CAAA;IAClB,mDAAmD;IACnD,KAAK,CAAC,EAAE,MAAM,CAAA;CACd;AAED;;;;;;;GAOG;AACH,MAAM,WAAW,KAAK;IACrB,iEAAiE;IACjE,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,EAAE,MAAM,CAAA;IACd,2EAA2E;IAC3E,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IACjC;;;;;;;;;;;;;;;;OAgBG;IACH,WAAW,CAAC,EAAE,OAAO,CAAA;CACrB;AAED;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,MAAM,MAAM,cAAc,GAAG,MAAM,GAAG,MAAM,CAAA;AAE5C,MAAM,WAAW,MAAM;IACtB,IAAI,EAAE,MAAM,CAAA;IACZ;;;;;OAKG;IACH,KAAK,CAAC,GAAG,EAAE,OAAO,EAAE,QAAQ,EAAE,QAAQ,EAAE,MAAM,CAAC,EAAE,WAAW,GAAG,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,CAAA;IACrF,kDAAkD;IAClD,QAAQ,CAAC,EAAE,cAAc,CAAA;IACzB;;;;;;;OAOG;IACH,SAAS,CAAC,EAAE,MAAM,CAAA;CAClB;AAED;;;;;GAKG;AACH,MAAM,MAAM,UAAU,GAAG,QAAQ,GAAG,QAAQ,GAAG,cAAc,CAAA;AAE7D,MAAM,WAAW,UAAU;IAC1B,IAAI,EAAE,MAAM,CAAA;IACZ,GAAG,EAAE,OAAO,CAAA;IACZ,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC,CAAA;IAC7B,iEAAiE;IACjE,IAAI,EAAE,MAAM,CAAA;IACZ,MAAM,EAAE,UAAU,CAAA;IAClB,wCAAwC;IACxC,MAAM,EAAE,OAAO,CAAA;IACf;;;;;;OAMG;IACH,WAAW,CAAC,EAAE,SAAS,MAAM,EAAE,CAAA;CAC/B;AAED,MAAM,WAAW,gBAAgB;IAChC,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,SAAS,UAAU,EAAE,CAAA;IAC5B,qDAAqD;IACrD,IAAI,EAAE,MAAM,CAAA;IACZ;;;;;;;;;;;;;;OAcG;IACH,WAAW,CAAC,EAAE,gBAAgB,CAAA;IAC9B,MAAM,EAAE,MAAM,CAAA;IACd,MAAM,EAAE,MAAM,CAAA;IACd,uDAAuD;IACvD,YAAY,EAAE,MAAM,CAAA;IACpB;;;;OAIG;IACH,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAChC,UAAU,EAAE,MAAM,CAAA;CAClB"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/eval/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,0BAA0B,CAAA;AAC1D,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAA;AAExD;;;;;GAKG;AACH,MAAM,WAAW,QAAQ,CAAC,MAAM,GAAG,OAAO;IACzC,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,MAAM,CAAA;IACb,uEAAuE;IACvE,QAAQ,CAAC,EAAE,OAAO,CAAA;IAClB;;;OAGG;IACH,aAAa,CAAC,EAAE,SAAS,MAAM,EAAE,CAAA;IACjC,sEAAsE;IACtE,OAAO,CAAC,EAAE,SAAS,MAAM,EAAE,CAAA;CAC3B;AAED,sDAAsD;AACtD,MAAM,WAAW,QAAQ;IACxB,wDAAwD;IACxD,MAAM,EAAE,MAAM,GAAG,IAAI,CAAA;IACrB,4EAA4E;IAC5E,gBAAgB,CAAC,EAAE,OAAO,CAAA;IAC1B,KAAK,EAAE,SAAS,UAAU,EAAE,CAAA;IAC5B,wDAAwD;IACxD,SAAS,EAAE,SAAS,MAAM,EAAE,CAAA;IAC5B,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,WAAW,EAAE,MAAM,CAAA;IACnB,YAAY,EAAE,MAAM,CAAA;IACpB,UAAU,EAAE,MAAM,CAAA;IAClB,oDAAoD;IACpD,KAAK,CAAC,EAAE,MAAM,CAAA;CACd;AAED;;;;;;;GAOG;AACH,MAAM,WAAW,KAAK;IACrB,iEAAiE;IACjE,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,EAAE,MAAM,CAAA;IACd,2EAA2E;IAC3E,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IACjC;;;;;;;;;;;;;;;;OAgBG;IACH,WAAW,CAAC,EAAE,OAAO,CAAA;CACrB;AAED;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,MAAM,MAAM,cAAc,GAAG,MAAM,GAAG,MAAM,CAAA;AAE5C,MAAM,WAAW,MAAM;IACtB,IAAI,EAAE,MAAM,CAAA;IACZ;;;;;OAKG;IACH,KAAK,CAAC,IAAI,EAAE,QAAQ,EAAE,QAAQ,EAAE,QAAQ,EAAE,MAAM,CAAC,EAAE,WAAW,GAAG,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,CAAA;IACvF,kDAAkD;IAClD,QAAQ,CAAC,EAAE,cAAc,CAAA;IACzB;;;;;;;OAOG;IACH,SAAS,CAAC,EAAE,MAAM,CAAA;CAClB;AAED;;;;;GAKG;AACH,MAAM,MAAM,UAAU,GAAG,QAAQ,GAAG,QAAQ,GAAG,cAAc,CAAA;AAE7D,MAAM,WAAW,UAAU;IAC1B,IAAI,EAAE,MAAM,CAAA;IACZ,IAAI,EAAE,QAAQ,CAAA;IACd,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC,CAAA;IAC7B,iEAAiE;IACjE,IAAI,EAAE,MAAM,CAAA;IACZ,MAAM,EAAE,UAAU,CAAA;IAClB,wCAAwC;IACxC,MAAM,EAAE,OAAO,CAAA;IACf;;;;;;OAMG;IACH,WAAW,CAAC,EAAE,SAAS,MAAM,EAAE,CAAA;CAC/B;AAED,MAAM,WAAW,gBAAgB;IAChC,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,SAAS,UAAU,EAAE,CAAA;IAC5B,qDAAqD;IACrD,IAAI,EAAE,MAAM,CAAA;IACZ;;;;;;;;;;;;;;OAcG;IACH,WAAW,CAAC,EAAE,gBAAgB,CAAA;IAC9B,MAAM,EAAE,MAAM,CAAA;IACd,MAAM,EAAE,MAAM,CAAA;IACd,uDAAuD;IACvD,YAAY,EAAE,MAAM,CAAA;IACpB;;;;OAIG;IACH,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAChC,UAAU,EAAE,MAAM,CAAA;CAClB"}

package/dist/eval/uncertainty.d.ts

@@ -1,7 +1,7 @@
 /**
  * How much of a score is signal.
  *
- * A suite reported a mean and nothing else, so two runs differing by three
+ * A suite reported a mean and nothing else, so two experiments differing by three
  * points read as a difference. At the n a hand-built suite has, that is
  * usually noise, and there was no number on the page that would have said
  * so.
@@ -57,7 +57,7 @@ export declare function uncertaintyOf(scores: readonly number[]): ScoreUncertain
  *
  * Names the interval rather than only the mean, because the mean alone is
  * the thing that has been over-read. An interval spanning most of the
- * scale says the suite cannot currently tell two runs apart, and that is
+ * scale says the suite cannot currently tell two experiments apart, and that is
  * the most useful sentence such a suite can produce.
  */
 export declare function describeUncertainty(mean: number, u: ScoreUncertainty): string;

package/dist/eval/uncertainty.js

@@ -1,7 +1,7 @@
 /**
  * How much of a score is signal.
  *
- * A suite reported a mean and nothing else, so two runs differing by three
+ * A suite reported a mean and nothing else, so two experiments differing by three
  * points read as a difference. At the n a hand-built suite has, that is
  * usually noise, and there was no number on the page that would have said
  * so.
@@ -78,7 +78,7 @@ export function uncertaintyOf(scores) {
  *
  * Names the interval rather than only the mean, because the mean alone is
  * the thing that has been over-read. An interval spanning most of the
- * scale says the suite cannot currently tell two runs apart, and that is
+ * scale says the suite cannot currently tell two experiments apart, and that is
  * the most useful sentence such a suite can produce.
  */
 export function describeUncertainty(mean, u) {

package/dist/execution/code-runtime/types.d.ts

@@ -18,8 +18,8 @@
  *    backend that cannot withhold those is not a backend for this.
  *  - **Everything it can do, it does by asking.** The only channel out is a
  *    call back into the host, which is what makes NZ-EXEC-06's dispatch
- *    through the run's own `ToolRegistry` possible — the program cannot
- *    reach a tool the run did not grant, because it cannot reach anything.
+ *    through the turn's own `ToolRegistry` possible — the program cannot
+ *    reach a tool the turn did not grant, because it cannot reach anything.
  *  - **Bounded.** Wall clock and output, both enforced by the backend
  *    rather than requested of the program.
  *
@@ -53,8 +53,8 @@ export interface HostCallContext {
  * What the host will answer.
  *
  * A function rather than a table, so the host decides per call. NZ-EXEC-06
- * makes this the run's `ToolRegistry` and its permission gate, which is
- * exactly the point: the program's reach is the run's reach, resolved at
+ * makes this the turn's `ToolRegistry` and its permission gate, which is
+ * exactly the point: the program's reach is the turn's reach, resolved at
  * the moment of the call rather than frozen when the program started.
  */
 export type HostCallHandler = (request: HostCallRequest, context: HostCallContext) => Promise<HostCallResult>;

package/dist/execution/code-runtime/types.js

@@ -18,8 +18,8 @@
  *    backend that cannot withhold those is not a backend for this.
  *  - **Everything it can do, it does by asking.** The only channel out is a
  *    call back into the host, which is what makes NZ-EXEC-06's dispatch
- *    through the run's own `ToolRegistry` possible — the program cannot
- *    reach a tool the run did not grant, because it cannot reach anything.
+ *    through the turn's own `ToolRegistry` possible — the program cannot
+ *    reach a tool the turn did not grant, because it cannot reach anything.
  *  - **Bounded.** Wall clock and output, both enforced by the backend
  *    rather than requested of the program.
  *

package/dist/execution/code-runtime/worker.d.ts

@@ -16,7 +16,7 @@ export interface WorkerCodeRuntimeOptions {
  * Runs untrusted JavaScript in a QuickJS interpreter inside a worker thread.
  * QuickJS owns the guest globals, functions, and promises. Node capabilities
  * never enter that realm; constructors and dynamic imports cannot recover them.
- * A fresh interpreter and bounded imported WASM memory are created per run.
+ * A fresh interpreter and bounded imported WASM memory are created per turn.
  * The worker makes wall-clock cancellation independent of interpreter progress.
  */
 export declare class WorkerCodeRuntime implements CodeRuntime {

package/dist/execution/code-runtime/worker.js

@@ -36,7 +36,7 @@ function decodeValue(json, maxBytes) {
  * Runs untrusted JavaScript in a QuickJS interpreter inside a worker thread.
  * QuickJS owns the guest globals, functions, and promises. Node capabilities
  * never enter that realm; constructors and dynamic imports cannot recover them.
- * A fresh interpreter and bounded imported WASM memory are created per run.
+ * A fresh interpreter and bounded imported WASM memory are created per turn.
  * The worker makes wall-clock cancellation independent of interpreter progress.
  */
 export class WorkerCodeRuntime {

package/dist/invariants/index.d.ts

@@ -3,7 +3,7 @@
  *
  * Three states, not two, for the same reason {@link InvariantOutcome} shares
  * its shape with NZ-BOOT-02's `CapabilityProbe`: a check that could not run
- * — no live run to point it at, a dependency not wired yet — is not the same
+ * — no live turn to point it at, a dependency not wired yet — is not the same
  * fact as a check that ran and found nothing wrong, and collapsing the two
  * turns "I do not know" into the strongest possible wrong answer. See
  * "an optional dependency may not degrade a check".

package/dist/manager/agent/child-session.d.ts

@@ -0,0 +1,49 @@
+import type { SessionLog, SessionRecordDraft } from '../../store/session-log/index.js';
+import type { SessionId } from '../../types/ids/index.js';
+import { type ChildSessionMeta } from '../../types/session/records.js';
+/**
+ * A child session, as its parent sees it (spec §4.2, "Children").
+ *
+ * A delegated agent runs as a child session: its own log at
+ * `<parent-session-dir>/subagents/<child-id>.jsonl`, a convenience document
+ * `<child-id>.meta.json` beside it, and, in the parent's log, a
+ * `child_session_spawned` record when it starts and a `child_session_ended`
+ * record when it settles. This module holds the pieces of that which do not
+ * depend on who holds the parent's lease: the meta document, the ended
+ * record derived from the child's own log, and the process-local lookup of
+ * a live child's log.
+ */
+/** `child_session_ended`, as the parent's writer appends it. */
+export type ChildSessionEndedDraft = Extract<SessionRecordDraft, {
+    type: 'child_session_ended';
+}>;
+/**
+ * The parent's `child_session_ended` record for a child, read from the
+ * child's own terminal record so the two cannot disagree: the status, stop
+ * reason, answer message, usage and cost are copied from the settlement of
+ * the child's last `turn_completed` or `turn_failed`. `null` while the child
+ * has no settled turn.
+ *
+ * The draft carries no `turnId`. The parent's writer adds the spawning turn's
+ * id when that turn is still open, and leaves it out otherwise (the child
+ * belongs to the turn its `child_session_spawned` names).
+ */
+export declare function childSessionEnded(childLog: SessionLog): Promise<ChildSessionEndedDraft | null>;
+/** A `<child-id>.meta.json` that belongs to another child or another parent. */
+export declare class ChildSessionMetaOwnerError extends Error {
+    readonly name = "ChildSessionMetaOwnerError";
+}
+/**
+ * Write `<child-id>.meta.json`, atomically and with an owner check: an
+ * existing document must name the same child, parent and parent turn, or the
+ * write is refused rather than repointing it. The document is a convenience
+ * only; the child's log wins on any disagreement.
+ */
+export declare function writeChildSessionMeta(path: string, meta: ChildSessionMeta): Promise<void>;
+/** The meta document at `path`, or `null` when there is none. Throws on one that does not parse. */
+export declare function readChildSessionMeta(path: string): Promise<ChildSessionMeta | null>;
+/** Register a live child's log; the returned function removes it. */
+export declare function registerChildSessionLog(log: SessionLog): () => void;
+/** The log of a child session running in this process, if there is one. */
+export declare function childSessionLog(childSessionId: SessionId): SessionLog | undefined;
+//# sourceMappingURL=child-session.d.ts.map

package/dist/manager/agent/child-session.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"child-session.d.ts","sourceRoot":"","sources":["../../../src/manager/agent/child-session.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,UAAU,EAAE,kBAAkB,EAAE,MAAM,kCAAkC,CAAA;AACtF,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,0BAA0B,CAAA;AACzD,OAAO,EAAE,KAAK,gBAAgB,EAA0B,MAAM,gCAAgC,CAAA;AAG9F;;;;;;;;;;;GAWG;AAEH,gEAAgE;AAChE,MAAM,MAAM,sBAAsB,GAAG,OAAO,CAAC,kBAAkB,EAAE;IAAE,IAAI,EAAE,qBAAqB,CAAA;CAAE,CAAC,CAAA;AAEjG;;;;;;;;;;GAUG;AACH,wBAAsB,iBAAiB,CACtC,QAAQ,EAAE,UAAU,GAClB,OAAO,CAAC,sBAAsB,GAAG,IAAI,CAAC,CAoBxC;AAED,gFAAgF;AAChF,qBAAa,0BAA2B,SAAQ,KAAK;IACpD,SAAkB,IAAI,gCAA+B;CACrD;AAED;;;;;GAKG;AACH,wBAAsB,qBAAqB,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,gBAAgB,GAAG,OAAO,CAAC,IAAI,CAAC,CAe/F;AAED,oGAAoG;AACpG,wBAAsB,oBAAoB,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,gBAAgB,GAAG,IAAI,CAAC,CASzF;AAYD,qEAAqE;AACrE,wBAAgB,uBAAuB,CAAC,GAAG,EAAE,UAAU,GAAG,MAAM,IAAI,CAKnE;AAED,2EAA2E;AAC3E,wBAAgB,eAAe,CAAC,cAAc,EAAE,SAAS,GAAG,UAAU,GAAG,SAAS,CAEjF"}

package/dist/manager/agent/child-session.js

@@ -0,0 +1,94 @@
+import { mkdir, readFile } from 'node:fs/promises';
+import { dirname } from 'node:path';
+import { ChildSessionMetaSchema } from '../../types/session/records.js';
+import { atomicWriteFile } from '../../utils/atomic-write.js';
+/**
+ * The parent's `child_session_ended` record for a child, read from the
+ * child's own terminal record so the two cannot disagree: the status, stop
+ * reason, answer message, usage and cost are copied from the settlement of
+ * the child's last `turn_completed` or `turn_failed`. `null` while the child
+ * has no settled turn.
+ *
+ * The draft carries no `turnId`. The parent's writer adds the spawning turn's
+ * id when that turn is still open, and leaves it out otherwise (the child
+ * belongs to the turn its `child_session_spawned` names).
+ */
+export async function childSessionEnded(childLog) {
+    let ended = null;
+    for await (const { record } of childLog.read({ mode: 'tolerant' })) {
+        if (record.type !== 'turn_completed' && record.type !== 'turn_failed')
+            continue;
+        const { settlement } = record;
+        ended = {
+            type: 'child_session_ended',
+            childSessionId: childLog.sessionId,
+            status: settlement.status,
+            ...(record.type === 'turn_completed' && record.stopReason !== undefined
+                ? { stopReason: record.stopReason }
+                : {}),
+            ...(settlement.resultMessageId !== undefined
+                ? { resultMessageId: settlement.resultMessageId }
+                : {}),
+            usage: settlement.usage,
+            cost: settlement.cost,
+        };
+    }
+    return ended;
+}
+/** A `<child-id>.meta.json` that belongs to another child or another parent. */
+export class ChildSessionMetaOwnerError extends Error {
+    name = 'ChildSessionMetaOwnerError';
+}
+/**
+ * Write `<child-id>.meta.json`, atomically and with an owner check: an
+ * existing document must name the same child, parent and parent turn, or the
+ * write is refused rather than repointing it. The document is a convenience
+ * only; the child's log wins on any disagreement.
+ */
+export async function writeChildSessionMeta(path, meta) {
+    const next = ChildSessionMetaSchema.parse(meta);
+    const current = await readChildSessionMeta(path);
+    if (current !== null &&
+        (current.sessionId !== next.sessionId ||
+            current.parentSessionId !== next.parentSessionId ||
+            current.parentTurnId !== next.parentTurnId)) {
+        throw new ChildSessionMetaOwnerError(`${path} describes child session ${current.sessionId} of ${current.parentSessionId}; refusing to overwrite it with ${next.sessionId} of ${next.parentSessionId}.`);
+    }
+    await mkdir(dirname(path), { recursive: true, mode: 0o700 });
+    await atomicWriteFile(path, `${JSON.stringify(next)}\n`, { mode: 0o600 });
+}
+/** The meta document at `path`, or `null` when there is none. Throws on one that does not parse. */
+export async function readChildSessionMeta(path) {
+    let raw;
+    try {
+        raw = await readFile(path, 'utf8');
+    }
+    catch (error) {
+        if (error.code === 'ENOENT')
+            return null;
+        throw error;
+    }
+    return ChildSessionMetaSchema.parse(JSON.parse(raw));
+}
+/**
+ * The logs of the child sessions running in this process, by child id.
+ *
+ * A disk child's log can be reopened from its path; an in-memory one exists
+ * only as the object its manager built. The parent's writer reads the
+ * child's terminal record from here to append `child_session_ended`. Each
+ * entry lives as long as its child's spawn record in the `AgentManager`.
+ */
+const liveChildLogs = new Map();
+/** Register a live child's log; the returned function removes it. */
+export function registerChildSessionLog(log) {
+    liveChildLogs.set(log.sessionId, log);
+    return () => {
+        if (liveChildLogs.get(log.sessionId) === log)
+            liveChildLogs.delete(log.sessionId);
+    };
+}
+/** The log of a child session running in this process, if there is one. */
+export function childSessionLog(childSessionId) {
+    return liveChildLogs.get(childSessionId);
+}
+//# sourceMappingURL=child-session.js.map

package/dist/manager/agent/child-session.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"child-session.js","sourceRoot":"","sources":["../../../src/manager/agent/child-session.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAA;AAClD,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAA;AAGnC,OAAO,EAAyB,sBAAsB,EAAE,MAAM,gCAAgC,CAAA;AAC9F,OAAO,EAAE,eAAe,EAAE,MAAM,6BAA6B,CAAA;AAkB7D;;;;;;;;;;GAUG;AACH,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACtC,QAAoB;IAEpB,IAAI,KAAK,GAAkC,IAAI,CAAA;IAC/C,IAAI,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,IAAI,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC,EAAE,CAAC;QACpE,IAAI,MAAM,CAAC,IAAI,KAAK,gBAAgB,IAAI,MAAM,CAAC,IAAI,KAAK,aAAa;YAAE,SAAQ;QAC/E,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,CAAA;QAC7B,KAAK,GAAG;YACP,IAAI,EAAE,qBAAqB;YAC3B,cAAc,EAAE,QAAQ,CAAC,SAAS;YAClC,MAAM,EAAE,UAAU,CAAC,MAAM;YACzB,GAAG,CAAC,MAAM,CAAC,IAAI,KAAK,gBAAgB,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS;gBACtE,CAAC,CAAC,EAAE,UAAU,EAAE,MAAM,CAAC,UAAU,EAAE;gBACnC,CAAC,CAAC,EAAE,CAAC;YACN,GAAG,CAAC,UAAU,CAAC,eAAe,KAAK,SAAS;gBAC3C,CAAC,CAAC,EAAE,eAAe,EAAE,UAAU,CAAC,eAAe,EAAE;gBACjD,CAAC,CAAC,EAAE,CAAC;YACN,KAAK,EAAE,UAAU,CAAC,KAAK;YACvB,IAAI,EAAE,UAAU,CAAC,IAAI;SACrB,CAAA;IACF,CAAC;IACD,OAAO,KAAK,CAAA;AACb,CAAC;AAED,gFAAgF;AAChF,MAAM,OAAO,0BAA2B,SAAQ,KAAK;IAClC,IAAI,GAAG,4BAA4B,CAAA;CACrD;AAED;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,qBAAqB,CAAC,IAAY,EAAE,IAAsB;IAC/E,MAAM,IAAI,GAAG,sBAAsB,CAAC,KAAK,CAAC,IAAI,CAAC,CAAA;IAC/C,MAAM,OAAO,GAAG,MAAM,oBAAoB,CAAC,IAAI,CAAC,CAAA;IAChD,IACC,OAAO,KAAK,IAAI;QAChB,CAAC,OAAO,CAAC,SAAS,KAAK,IAAI,CAAC,SAAS;YACpC,OAAO,CAAC,eAAe,KAAK,IAAI,CAAC,eAAe;YAChD,OAAO,CAAC,YAAY,KAAK,IAAI,CAAC,YAAY,CAAC,EAC3C,CAAC;QACF,MAAM,IAAI,0BAA0B,CACnC,GAAG,IAAI,4BAA4B,OAAO,CAAC,SAAS,OAAO,OAAO,CAAC,eAAe,mCAAmC,IAAI,CAAC,SAAS,OAAO,IAAI,CAAC,eAAe,GAAG,CACjK,CAAA;IACF,CAAC;IACD,MAAM,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAA;IAC5D,MAAM,eAAe,CAAC,IAAI,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,IAAI,EAAE,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAA;AAC1E,CAAC;AAED,oGAAoG;AACpG,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,IAAY;IACtD,IAAI,GAAW,CAAA;IACf,IAAI,CAAC;QACJ,GAAG,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAA;IACnC,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QAChB,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO,IAAI,CAAA;QACnE,MAAM,KAAK,CAAA;IACZ,CAAC;IACD,OAAO,sBAAsB,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAA;AACrD,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,aAAa,GAAG,IAAI,GAAG,EAAyB,CAAA;AAEtD,qEAAqE;AACrE,MAAM,UAAU,uBAAuB,CAAC,GAAe;IACtD,aAAa,CAAC,GAAG,CAAC,GAAG,CAAC,SAAS,EAAE,GAAG,CAAC,CAAA;IACrC,OAAO,GAAG,EAAE;QACX,IAAI,aAAa,CAAC,GAAG,CAAC,GAAG,CAAC,SAAS,CAAC,KAAK,GAAG;YAAE,aAAa,CAAC,MAAM,CAAC,GAAG,CAAC,SAAS,CAAC,CAAA;IAClF,CAAC,CAAA;AACF,CAAC;AAED,2EAA2E;AAC3E,MAAM,UAAU,eAAe,CAAC,cAAyB;IACxD,OAAO,aAAa,CAAC,GAAG,CAAC,cAAc,CAAC,CAAA;AACzC,CAAC"}