stable-harness 0.0.23 → 0.0.24

package/docs/guides/quality-gates.md

@@ -84,9 +84,12 @@ When `synthesis.enabled` is true, the runtime may recover from a rejected final
 answer by building an `evidence_only` report from successful observed tool or
 delegated-task outputs. This is a control-plane synthesis path, not a domain
 writer: it does not call more tools, does not add pretrained facts, and runs the
-same execution review again before delivery. Control outputs such as invalid
-input, approval blocks, or repeated-call limits are preserved as blockers or
-evidence gaps rather than treated as factual evidence.
+same execution review again before delivery. The built-in presentation layer
+groups observed facts into generic user-facing sections such as confirmed
+context, observed data points, time-bound evidence, evidence limits, and compact
+sources used. It does not contain domain-specific report templates. Control
+outputs such as invalid input, approval blocks, or repeated-call limits are
+preserved as blockers or evidence gaps rather than treated as factual evidence.
 
 BetterCall remains focused on individual tool-call reliability. Quality gates
 consume tool and repair diagnostics as evidence, but they live in Stable Harness

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "stable-harness",
-  "version": "0.0.23",
+  "version": "0.0.24",
   "type": "module",
   "description": "Stable application runtime and operator control plane for agent workspaces.",
   "license": "Apache-2.0",

package/packages/core/dist/quality/synthesis.js

@@ -1 +1 @@
-import{controlBlockers as e,controlGaps as n,successfulEvidenceItems as t}from"./event-evidence.js";export function synthesizeEvidenceOnlyReport(s,r,i){if(!i.enabled||!i.synthesis.enabled||"evidence_only"!==i.synthesis.mode)return;if("pass"===r.verdict||!function hasRecoverableSynthesisIssue(e){return e.issues.some(e=>"control_blocker"!==e.code)}(r))return;const o=t(s.events).slice(-i.synthesis.maxEvidenceItems),c=e(s.events),u=n(s.events);if(0===o.length&&0===c.length&&0===u.length)return;const a=function detectSynthesisLanguage(e){const n=e.workspace.runtime.responseLanguage;if(function isRecord(e){return"object"==typeof e&&null!==e&&!Array.isArray(e)}(n)){const e=readString(n.language)??readString(n.locale)??readString(n.target);if(e&&/^(?:zh|zh-|chinese|中文|汉语|漢語)/iu.test(e))return"zh";if(e&&/^(?:en|en-|english)$/iu.test(e))return"en"}return/\p{Script=Han}/u.test(e.request.input)?"zh":"en"}(s);return"zh"===a?function buildChineseReport(e,n,t){return["# 有依据的报告","","Stable Harness 已拒绝上一版最终回答，因为它未被已观察到的运行时证据完全支持。以下报告只根据已完成的工具或委托任务证据生成。","","## 已完成的证据来源",...sourceLines(e,"zh"),"","## 有证据支持的事实",...evidenceLines(e,"zh"),"","## 证据缺口与阻塞",...gapLines(n,t,"zh")].join("\n")}(o,c,u):function buildEnglishReport(e,n,t){return["# Grounded report","","Stable Harness rejected the previous final answer because it was not fully supported by observed runtime evidence. This report is synthesized only from completed tool or delegated-task evidence.","","## Completed evidence sources",...sourceLines(e,"en"),"","## Evidence-backed facts",...evidenceLines(e,"en"),"","## Evidence gaps and blockers",...gapLines(n,t,"en")].join("\n")}(o,c,u)}function sourceLines(e,n){return 0===e.length?["zh"===n?"- 未观察到成功的工具或委托任务证据。":"- No successful tool or delegated-task evidence was observed."]:[...e.reduce((e,n)=>e.set(n.source,(e.get(n.source)??0)+1),new Map)].map(([e,n])=>`- ${e}${n>1?` (${n})`:""}`)}function evidenceLines(e,n){return 0===e.length?["zh"===n?"- 证据缺口：没有可用于生成事实性结论的成功证据。":"- Evidence gap: no successful tool or delegated-task evidence was available."]:e.map(e=>`- ${e.source}: ${function formatEvidence(e){const n=e.replace(/\s+/gu," ").trim();return n.length>1200?`${n.slice(0,1197)}...`:n}(e.output)}`)}function gapLines(e,n,t){return 0===e.length&&0===n.length?["zh"===t?"- 未观察到未解决的运行时证据缺口或阻塞。":"- No unresolved runtime evidence gaps or blockers were observed."]:"zh"===t?[...e.map(e=>`- 阻塞：${e}`),...n.map(e=>`- 证据缺口：${e}`)]:[...e.map(e=>`- Blocked: ${e}`),...n.map(e=>`- Evidence gap: ${e}`)]}function readString(e){return"string"==typeof e&&e.trim()?e.trim():void 0}
+import{controlBlockers as e,controlGaps as t,successfulEvidenceItems as n}from"./event-evidence.js";export function synthesizeEvidenceOnlyReport(s,r,c){if(!c.enabled||!c.synthesis.enabled||"evidence_only"!==c.synthesis.mode)return;if("pass"===r.verdict||!function hasRecoverableSynthesisIssue(e){return e.issues.some(e=>"control_blocker"!==e.code)}(r))return;const i=n(s.events).slice(-c.synthesis.maxEvidenceItems),o=e(s.events),u=t(s.events);if(0===i.length&&0===o.length&&0===u.length)return;const a=function detectSynthesisLanguage(e){const t=e.workspace.runtime.responseLanguage;if(isRecord(t)){const e=readString(t.language)??readString(t.locale)??readString(t.target);if(e&&/^(?:zh|zh-|chinese|中文|汉语|漢語)/iu.test(e))return"zh";if(e&&/^(?:en|en-|english)$/iu.test(e))return"en"}return/\p{Script=Han}/u.test(e.request.input)?"zh":"en"}(s);return"zh"===a?function buildChineseReport(e,t,n){const s=evidenceFacts(e);return["# 有依据的报告","","本报告只使用本次运行中已完成的工具或委托任务证据；未被证据支持的最终回答内容已被丢弃。","","## 证据摘要",...summaryLines(e,t,n,"zh"),"",...structuredFactSections(s,"zh"),"","## 证据缺口与阻塞",...gapLines(t,n,"zh"),"","## 使用的证据来源",...sourceLines(sourceSummary(e),"zh")].join("\n")}(i,o,u):function buildEnglishReport(e,t,n){const s=evidenceFacts(e);return["# Grounded report","","This report uses only completed tool or delegated-task evidence observed in this run. Unsupported final-answer claims were discarded.","","## Evidence summary",...summaryLines(e,t,n,"en"),"",...structuredFactSections(s,"en"),"","## Evidence gaps and blockers",...gapLines(t,n,"en"),"","## Sources used",...sourceLines(sourceSummary(e),"en")].join("\n")}(i,o,u)}function sourceSummary(e){const t=new Map;for(const n of e){const e=t.get(n.source)??{source:n.source,count:0,facts:[]};e.count+=1,e.facts.push(...factLines(n.output)),t.set(n.source,e)}return[...t.values()].map(t=>({...t,facts:t.facts.length>0?t.facts.slice(0,8):[fallbackEvidenceText(evidenceTextForSource(e,t.source))]}))}function evidenceFacts(e){return e.flatMap(e=>{const t=factLines(e.output);return(t.length>0?t:[fallbackEvidenceText(e.output)]).map(t=>({source:e.source,text:t,kind:classifyFact(t)}))}).slice(0,40)}function summaryLines(e,t,n,s){if(0===e.length)return["zh"===s?"- 未观察到成功的工具或委托任务证据。":"- No successful tool or delegated-task evidence was observed."];const r=new Set(e.map(e=>e.source)).size,c=0===t.length&&0===n.length?"zh"===s?"未观察到未解决的运行时证据缺口或阻塞。":"No unresolved runtime evidence gaps or blockers were observed.":"zh"===s?`仍有 ${t.length+n.length} 个运行时证据缺口或阻塞。`:`${t.length+n.length} runtime evidence gaps or blockers remain.`;return["zh"===s?`- 已使用 ${e.length} 条完成证据，来自 ${r} 个来源。`:`- Used ${e.length} completed evidence item(s) from ${r} source(s).`,`- ${c}`]}function structuredFactSections(e,t){return 0===e.length?["zh"===t?"- 证据缺口：没有可用于生成事实性结论的成功证据。":"- Evidence gap: no successful tool or delegated-task evidence was available."]:[{kind:"context",title:"zh"===t?"## 已确认背景":"## Confirmed context"},{kind:"data",title:"zh"===t?"## 观察到的数据":"## Observed data points"},{kind:"timeBound",title:"zh"===t?"## 近期或时间相关证据":"## Recent or time-bound evidence"},{kind:"limit",title:"zh"===t?"## 证据限制":"## Evidence limits"},{kind:"other",title:"zh"===t?"## 其他观察":"## Additional observations"}].flatMap(n=>{const s=e.filter(e=>e.kind===n.kind).slice(0,8);return s.length>0?[n.title,...s.map(e=>function factLine(e,t){const n=humanSourceLabel(e.source);return"zh"===t?`- ${e.text}（来源：${n}）`:`- ${e.text} (source: ${n})`}(e,t)),""]:[]}).filter((e,t,n)=>""!==e||t<n.length-1)}function sourceLines(e,t){return 0===e.length?["zh"===t?"- 未使用成功证据来源。":"- No successful evidence source was used."]:e.map(e=>`- ${humanSourceLabel(e.source)}${e.count>1?` (${e.count})`:""}: ${e.source}`)}function gapLines(e,t,n){return 0===e.length&&0===t.length?["zh"===n?"- 未观察到未解决的运行时证据缺口或阻塞。":"- No unresolved runtime evidence gaps or blockers were observed."]:"zh"===n?[...e.map(e=>`- 阻塞：${e}`),...t.map(e=>`- 证据缺口：${e}`)]:[...e.map(e=>`- Blocked: ${e}`),...t.map(e=>`- Evidence gap: ${e}`)]}function formatEvidence(e){const t=e.replace(/\s+/gu," ").trim();return t.length>1200?`${t.slice(0,1197)}...`:t}function factLines(e){const t=function factLinesFromJson(e){try{const t=JSON.parse(e);return isRecord(t)?Object.entries(t).filter(([e])=>!/^(?:status|controlStatus)$/iu.test(e)).slice(0,8).map(([e,t])=>`${e}: ${formatEvidence(function stringifyJsonValue(e){return"string"==typeof e?e:JSON.stringify(e)}(t))}`):[]}catch{return[]}}(e);return t.length>0?t:function splitPlainTextFacts(e){const t=function stripControlPreamble(e){return e.replace(/^(?:Status:\s*(?:completed|success|ok|recorded)\b\.?\s*)+/iu,"").replace(/^(?:Evidence tool:\s*[A-Za-z0-9_.-]+\b\.?\s*)+/iu,"").trim()}(e);return t.split(/\r?\n|;\s*/u).map(e=>e.trim()).map(stripControlFragments).filter(e=>e&&!function isControlFact(e){return/^Status:\s*(?:completed|success|ok|recorded)$/iu.test(e)||/^Evidence tool:\s*[A-Za-z0-9_.-]+$/iu.test(e)}(e))}(e).slice(0,8).map(formatEvidence)}function stripControlFragments(e){return e.replace(/\bStatus:\s*(?:completed|success|ok|recorded)\b\.?\s*/giu,"").replace(/\bEvidence tool:\s*[A-Za-z0-9_.-]+\b\.?\s*/giu,"").trim()}function evidenceTextForSource(e,t){return e.find(e=>e.source===t)?.output??""}function fallbackEvidenceText(e){return formatEvidence(e)}function classifyFact(e){return/(?:gap|blocked|missing|unavailable|unsupported|缺口|阻塞|缺失|不支持|无法|未提供)/iu.test(e)?"limit":/(?:news|headline|recent|latest|filing|event|公告|新闻|近期|最新|披露|\b20\d{2}[-/年])/iu.test(e)?"timeBound":/(?:[$€£¥%]|\b\d[\d,]*(?:\.\d+)?\b)/u.test(e)?"data":/(?:name|company|symbol|resolved|overview|query|名称|公司|代码|概览|识别)/iu.test(e)?"context":"other"}function humanSourceLabel(e){const t=e.replace(/([a-z])([A-Z])/gu,"$1 $2").split(/[_:.\-\s]+/u).map(e=>e.trim()).filter(Boolean);return 0===t.length?e:t.map(e=>e.charAt(0).toUpperCase()+e.slice(1)).join(" ")}function isRecord(e){return"object"==typeof e&&null!==e&&!Array.isArray(e)}function readString(e){return"string"==typeof e&&e.trim()?e.trim():void 0}