dsh-industry-research 0.1.3 → 0.2.0

package/CHANGELOG.md

@@ -5,6 +5,30 @@ All notable changes to this project are documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.2.0] - 2026-08-23
+
+### Added
+
+- Deterministic `chain.json` → SVG network renderer (`src/chain-svg.ts`): tier-column layout, directed arrow edges, and dual-rule bottleneck detection (funnel: in-degree ≥ 2 and out-degree ≤ 1; hub: in-degree ≥ 2 and out-degree ≥ 2) with highlight colors. `industry_map` gains a `renderSvg` flag that writes `chain.svg` beside `chain.json` and returns the bottleneck summary; all user text is XML-escaped and the output is byte-stable for the same input.
+- Status assertions with dates: chain nodes and company cards may carry a `status` (`public`/`private`/`acquired`/`IPO`) that must carry a non-future `statusAsOf`; missing or future dates fail validation loud.
+- Price discipline and ticker validation: company cards accept sourced `metrics` (every value requires `source` + `asOf`) and a `ticker` checked against built-in formats (A-share 6 digits, US 1–5 letters, HK 1–5 digits), with the `scan.strictTicker` config exemption.
+- Research depth routing (`src/depth.ts`): `industry_map` / `industry_track` / `company_scan` accept a `depth` argument (quick/standard/comprehensive, default standard) that deterministically scales web collection — quick uses minimal source counts, comprehensive uses maximum — with loud failure out of range.
+- Per-company scan failure isolation: `company_scan` accepts a `companies` batch argument; one company's failure (bad file, path escape, validation) no longer aborts the batch and is reported in a `failures` list with its reason while the rest produce normally.
+- Artifact version ledger (`src/versions.ts` + `versions.jsonl` at the industry root): `industry_map` / `company_scan` / `industry_report` append a root-relative path + SHA-256 + injected timestamp + change type on every artifact write; artifact reads verify against the ledger and fail loud on a hash mismatch (unrecorded artifacts are skipped).
+- Built-in industry taxonomy anchor table (`src/taxonomy.ts`, 国民经济行业分类 大类 code-name-keywords); chain nodes may declare a `taxonomyCode` that must hit the table, with loud failure for unknown codes.
+- Evidence category labels: timeline entries may carry an `evidenceCategory` (`confirmed-catalyst` / `market-narrative` / `forum-buzz` / `technical-confirmation` / `macro-amplifier` / `background-noise`) validated on write; the auto-draft groups timeline entries by category.
+- Delivery contract validation (`validateDeliveryContract`): `industry_report` runs a deterministic pre-production check (complete blocks, status assertions with dates, value assertions with source + asOf, no placeholder residue) and fails loud instead of emitting a half-assembled report.
+- Research-state memory (`src/research-state.ts` + `research-state.json` per industry): `industry_map` / `industry_track` persist the last run's depth, latest artifact hashes, source URLs, evidence-category counts, gaps, and an injected timestamp, and return a deterministic "vs last run" delta (new/removed sources, unchanged evidence); a corrupt state fails loud and the write is registered in `versions.jsonl`.
+- Red-team adversarial review (`src/adversarial.ts`): `industry_report` runs the deterministic `adversarialCheck` machine-check baseline (dead evidence links, unsourced/undated numbers, statuses without dates) and, when `ctx.get('jobs')` is mounted, spawns a devil's-advocate review job that writes `red-review-note.md`; without `jobs` it records `review: skipped(jobs unavailable)`.
+- Per-company parallel scan: `company_scan` batch mode accepts `parallel` (default false); with `parallel: true` and a mounted `ctx.jobs` it fans each company out into an independent job with failure isolation and reports the execution `mode`, otherwise it falls back to the sequential path.
+
+### Deviations
+
+- **chain.json Slot client visualization UI** — needs a new client face plus packaging changes; the deterministic `chain.svg` artifact already covers the core value.
+- **Industry-expert subagent team (model orchestration)** — covered by failure isolation, depth routing, and adversarial review; a full model team exceeds this repo's determinism-first principle.
+- **Exa external retrieval service** — third-party API dependency, out of scope.
+- **Skill self-evolution (Hermes shadow agent)** — exceeds the DSH skill-provider mechanism.
+
 ## [0.1.3] - 2026-08-23
 
 ### Added

package/README.es.md

@@ -90,24 +90,25 @@ Todos los ajustes son campos `Config` de Schemastery; los valores inválidos fal
 | `track.maxFetchesPerCall` | `10` | Presupuesto de capturas por llamada `industry_track`. |
 | `scan.maxFileBytes` | `1048576` | Tope de lectura por archivo de datos de empresa. |
 | `scan.maxFigureCandidates` | `100` | Presupuesto de líneas candidatas por escaneo. |
+| `scan.strictTicker` | `true` | Los tickers de tarjetas deben coincidir con un formato integrado (A-share 6 dígitos, EE. UU. 1–5 letras, HK 1–5 dígitos); `false` exime la verificación. |
 
 ## Tools & surfaces
 
-### `industry_map({ industry, seed?, seedFiles?, web?, chain? })`
+### `industry_map({ industry, seed?, seedFiles?, web?, chain?, renderSvg?, depth? })`
 
-Con `chain`: valida (aristas colgadas, valores sin fuente, tiers ilegales, ids duplicados — fallo ruidoso con la lista completa) y persiste `chain.json`, luego lista las lagunas explícitas. Sin `chain`: devuelve el mapa actual, las fuentes registradas y un resumen opcional de estructura vía `ctx.web` para iterar. Emite `industry-research/map`.
+Con `chain`: valida (aristas colgadas, valores sin fuente, tiers ilegales, ids duplicados, status/statusAsOf, `taxonomyCode` desconocido — fallo ruidoso con la lista completa) y persiste `chain.json`, luego lista las lagunas explícitas y los nodos cuello de botella. Sin `chain`: devuelve el mapa actual, las fuentes registradas y un resumen opcional de estructura vía `ctx.web` para iterar. Con `renderSvg: true` también escribe un `chain.svg` determinista. `depth` escala la asistencia web. Emite `industry-research/map`.
 
-### `industry_track({ industry, topics?, since? })`
+### `industry_track({ industry, topics?, since?, depth?, evidenceCategory? })`
 
-Busca cada tema a través de `ctx.web`, filtra por las listas y `since`, captura instantáneas (SHA-256) dentro del presupuesto y fusiona en `timeline.jsonl` (deduplicación por URL normalizada, con tope). Las fuentes cuya instantánea falló se conservan como entradas solo-cita con la razón en `note`. Falla en voz alta nombrando la capacidad faltante cuando `ctx.web` no está montada o `offline: true`. Emite `industry-research/track`.
+Busca cada tema a través de `ctx.web`, filtra por las listas y `since`, captura instantáneas (SHA-256) dentro del presupuesto y fusiona en `timeline.jsonl` (deduplicación por URL normalizada, con tope). `depth` escala el presupuesto; `evidenceCategory` etiqueta cada entrada y se valida contra la enumeración de seis categorías. Las fuentes cuya instantánea falló se conservan como entradas solo-cita con la razón en `note`. Falla en voz alta nombrando la capacidad faltante cuando `ctx.web` no está montada o `offline: true`. Emite `industry-research/track`.
 
-### `company_scan({ name, dataFiles?, web? })`
+### `company_scan({ name | companies, dataFiles?, web?, status?, statusAsOf?, ticker?, metrics?, depth?, parallel? })`
 
-Lee los archivos de datos del workspace (`.md/.txt/.csv/.tsv/.json`; v1 no analiza PDF), los hashea, extrae esquemas Markdown y líneas candidatas de cifras, adjunta citas `ctx.web` opcionales y persiste la tarjeta. Los archivos rechazados regresan con razones; todo lo que la tarjeta no puede establecer es una laguna explícita.
+Lee los archivos de datos del workspace (`.md/.txt/.csv/.tsv/.json`; v1 no analiza PDF), los hashea, extrae esquemas Markdown y líneas candidatas de cifras, adjunta citas `ctx.web` opcionales y persiste la tarjeta. Un `status` requiere un `statusAsOf` no futuro; un `ticker` debe coincidir con un formato integrado salvo `scan.strictTicker: false`; cada valor de `metrics` debe llevar `source` + `asOf`. `companies` (lote) aísla una empresa fallida sin abortar el lote; `parallel: true` distribuye cada empresa en un job independiente cuando `ctx.jobs` está montado (si no, vuelve a secuencial, reflejado en `mode`). Los archivos rechazados regresan con razones; todo lo que la tarjeta no puede establecer es una laguna explícita.
 
 ### `industry_report({ industry, sections?, companies?, draft? })`
 
-Reúne la evidencia (`E-chain`, `E-timeline`, `E-company-<slug>`) y valida tu `draft` (secciones + claims; cada `evidenceIds` debe referenciar evidencia registrada) o construye el borrador mecánico (métricas con fuente y entradas recientes se convierten en claims). Ruta del motor: directorio sellado + `sealHash` + veredictos por claim. Ruta integrada: Markdown versionado + manifiesto, claims honestamente marcados `unverified`. Emite `industry-research/report`.
+Reúne la evidencia (`E-chain`, `E-timeline`, `E-company-<slug>`), verifica los artefactos leídos contra `versions.jsonl` (el hash que no coincide falla en voz alta) y valida tu `draft` (secciones + claims; cada `evidenceIds` debe referenciar evidencia registrada) o construye el borrador mecánico (métricas con fuente y entradas recientes se convierten en claims, agrupadas por `evidenceCategory`). Un contrato de entrega determinista corre antes de producir y falla en voz alta ante bloques faltantes, marcadores de posición o aserciones sin fuente/fecha. Una comprobación adversaria determinista siempre corre, y un job de revisión rojo (`red-review-note.md`) se lanza cuando `ctx.jobs` está montado (si no, `review: skipped(jobs unavailable)`). Ruta del motor: directorio sellado + `sealHash` + veredictos por claim. Ruta integrada: Markdown versionado + manifiesto, claims honestamente marcados `unverified`. Emite `industry-research/report`.
 
 ## Skills
 
@@ -119,7 +120,11 @@ Ambas se cargan bajo demanda con la herramienta estándar `skill` (`加载 indus
 ## Data layout
 
 ```
+<workspace>/<industryRoot>/versions.jsonl             registro de versiones (SHA-256 + timestamp + cambio)
+<workspace>/<industryRoot>/<industria>/research-state.json  memoria de estado de investigación
+<workspace>/<industryRoot>/<industria>/red-review-note.md    industry_report (revisión roja, jobs)
 <workspace>/<industryRoot>/<industria>/chain.json      industry_map
+<workspace>/<industryRoot>/<industria>/chain.svg       industry_map (renderSvg: true)
 <workspace>/<industryRoot>/<industria>/timeline.jsonl  industry_track
 <workspace>/<industryRoot>/<industria>/sources.json    registro de fuentes citables (S1, S2, …)
 <workspace>/<industryRoot>/<industria>/notes/          notas semilla

package/README.hi.md

@@ -90,24 +90,25 @@ dsh plugin --profile demo remove dsh-industry-research    # अनइंस्
 | `track.maxFetchesPerCall` | `10` | प्रति `industry_track` कॉल स्नैपशॉट-फ़ेच बजट। |
 | `scan.maxFileBytes` | `1048576` | कंपनी डेटा फ़ाइलों की प्रति-फ़ाइल रीड सीमा। |
 | `scan.maxFigureCandidates` | `100` | प्रति कंपनी स्कैन आंकड़ा-उम्मीदवार बजट। |
+| `scan.strictTicker` | `true` | कार्ड टिकर को बिल्ट-इन फ़ॉर्मेट से मेल खाना चाहिए (A-share 6 अंक, US 1–5 अक्षर, HK 1–5 अंक); `false` फ़ॉर्मेट जाँच को छूट देता है। |
 
 ## Tools & surfaces
 
-### `industry_map({ industry, seed?, seedFiles?, web?, chain? })`
+### `industry_map({ industry, seed?, seedFiles?, web?, chain?, renderSvg?, depth? })`
 
-`chain` के साथ: वैलिडेट करता है (लटकती एज, बिना-स्रोत मान, अवैध tier, डुप्लिकेट id — पूरी समस्या सूची के साथ स्पष्ट विफलता) और `chain.json` सहेजता है, फिर स्पष्ट अंतराल स्लॉट सूचीबद्ध करता है। `chain` के बिना: वर्तमान मानचित्र, दर्ज स्रोत और वैकल्पिक `ctx.web` श्रृंखला-संरचना सारांश लौटाता है ताकि मॉडल पुनरावृत्ति कर सके। `industry-research/map` उत्सर्जित करता है।
+`chain` के साथ: वैलिडेट करता है (लटकती एज, बिना-स्रोत मान, अवैध tier, डुप्लिकेट id, status/statusAsOf, अज्ञात `taxonomyCode` — पूरी समस्या सूची के साथ स्पष्ट विफलता) और `chain.json` सहेजता है, फिर स्पष्ट अंतराल स्लॉट व बॉटलनेक नोड सूचीबद्ध करता है। `chain` के बिना: वर्तमान मानचित्र, दर्ज स्रोत और वैकल्पिक `ctx.web` श्रृंखला-संरचना सारांश लौटाता है ताकि मॉडल पुनरावृत्ति कर सके। `renderSvg: true` पर एक नियतात्मक `chain.svg` भी लिखता है। `depth` web सहायता को स्केल करता है। `industry-research/map` उत्सर्जित करता है।
 
-### `industry_track({ industry, topics?, since? })`
+### `industry_track({ industry, topics?, since?, depth?, evidenceCategory? })`
 
-`ctx.web` से प्रत्येक विषय खोजता है, सूचियों व `since` से फ़िल्टर करता है, कॉल बजट के भीतर स्नैपशॉट (SHA-256) फ़ेच करता है, और `timeline.jsonl` में मर्ज करता है (नॉर्मलाइज़्ड URL से डीडुप, सीमा सहित)। जिन स्रोतों का स्नैपशॉट विफल हुआ वे कारण `note` में लिखकर केवल-उद्धरण एंट्री के रूप में रहते हैं। `ctx.web` अनमाउंटेड या `offline: true` होने पर गुम क्षमता का नाम बताते हुए स्पष्ट विफल। `industry-research/track` उत्सर्जित करता है।
+`ctx.web` से प्रत्येक विषय खोजता है, सूचियों व `since` से फ़िल्टर करता है, कॉल बजट के भीतर स्नैपशॉट (SHA-256) फ़ेच करता है, और `timeline.jsonl` में मर्ज करता है (नॉर्मलाइज़्ड URL से डीडुप, सीमा सहित)। `depth` बजट स्केल करता है; `evidenceCategory` हर एंट्री को लेबल करता है और छह-श्रेणी एनम से वैलिडेट होता है। जिन स्रोतों का स्नैपशॉट विफल हुआ वे कारण `note` में लिखकर केवल-उद्धरण एंट्री के रूप में रहते हैं। `ctx.web` अनमाउंटेड या `offline: true` होने पर गुम क्षमता का नाम बताते हुए स्पष्ट विफल। `industry-research/track` उत्सर्जित करता है।
 
-### `company_scan({ name, dataFiles?, web? })`
+### `company_scan({ name | companies, dataFiles?, web?, status?, statusAsOf?, ticker?, metrics?, depth?, parallel? })`
 
-workspace डेटा फ़ाइलें (`.md/.txt/.csv/.tsv/.json`; v1 PDF पार्स नहीं करता) पढ़ता, हैश करता, Markdown रूपरेखा व आंकड़ा-उम्मीदवार पंक्तियाँ निकालता है, वैकल्पिक `ctx.web` उद्धरण जोड़ता है और कार्ड सहेजता है। अस्वीकृत फ़ाइलें कारणों सहित लौटती हैं; कार्ड जो स्थापित नहीं कर सकता वह स्पष्ट अंतराल है।
+workspace डेटा फ़ाइलें (`.md/.txt/.csv/.tsv/.json`; v1 PDF पार्स नहीं करता) पढ़ता, हैश करता, Markdown रूपरेखा व आंकड़ा-उम्मीदवार पंक्तियाँ निकालता है, वैकल्पिक `ctx.web` उद्धरण जोड़ता है और कार्ड सहेजता है। `status` को एक गैर-भविष्य `statusAsOf` चाहिए; `ticker` को `scan.strictTicker: false` के अलावा बिल्ट-इन फ़ॉर्मेट से मेल खाना चाहिए; हर `metrics` मान को `source` + `asOf` चाहिए। `companies` (बैच) एक विफल कंपनी को बैच रोके बिना अलग करता है; `parallel: true` और `ctx.jobs` माउंट होने पर हर कंपनी को स्वतंत्र job में फ़ैन-आउट करता है (वरना क्रमिक पथ पर लौटता है, `mode` में दर्शित)। अस्वीकृत फ़ाइलें कारणों सहित लौटती हैं; कार्ड जो स्थापित नहीं कर सकता वह स्पष्ट अंतराल है।
 
 ### `industry_report({ industry, sections?, companies?, draft? })`
 
-एविडेंस (`E-chain`, `E-timeline`, `E-company-<slug>`) जुटाता है और या तो आपका `draft` वैलिडेट करता है (sections + claims; हर claim के `evidenceIds` दर्ज एविडेंस को संदर्भित करने चाहिए) या यांत्रिक ऑटो-ड्राफ़्ट बनाता है (स्रोतित मेट्रिक्स और हालिया टाइमलाइन एंट्री claims बनती हैं)। इंजन पथ: सील्ड डायरेक्टरी + `sealHash` + प्रति-claim निष्कर्ष। फ़ॉलबैक पथ: संस्करणित Markdown + मैनिफ़ेस्ट, claims ईमानदारी से `unverified` अंकित। `industry-research/report` उत्सर्जित करता है।
+एविडेंस (`E-chain`, `E-timeline`, `E-company-<slug>`) जुटाता है, पढ़े गए आर्टिफ़ैक्ट `versions.jsonl` से वेरिफ़ाई करता है (बेमेल हैश स्पष्ट विफल) और या तो आपका `draft` वैलिडेट करता है (sections + claims; हर claim के `evidenceIds` दर्ज एविडेंस को संदर्भित करने चाहिए) या यांत्रिक ऑटो-ड्राफ़्ट बनाता है (स्रोतित मेट्रिक्स और हालिया टाइमलाइन एंट्री claims बनती हैं, `evidenceCategory` से समूहित)। उत्पादन से पहले एक नियतात्मक डिलीवरी कॉन्ट्रैक्ट चलता है और गुम ब्लॉक, प्लेसहोल्डर या बिना-स्रोत/बिना-तिथि कथन पर स्पष्ट विफल होता है। एक नियतात्मक विरोधी मशीन-जाँच हमेशा चलती है, और `ctx.jobs` माउंट होने पर एक रेड-टीम समीक्षा job (`red-review-note.md`) शुरू होता है (वरना `review: skipped(jobs unavailable)`)। इंजन पथ: सील्ड डायरेक्टरी + `sealHash` + प्रति-claim निष्कर्ष। फ़ॉलबैक पथ: संस्करणित Markdown + मैनिफ़ेस्ट, claims ईमानदारी से `unverified` अंकित। `industry-research/report` उत्सर्जित करता है।
 
 ## Skills
 
@@ -119,7 +120,11 @@ workspace डेटा फ़ाइलें (`.md/.txt/.csv/.tsv/.json`; v1 PDF
 ## Data layout
 
 ```
+<workspace>/<industryRoot>/versions.jsonl             संस्करण लेजर (SHA-256 + टाइमस्टैम्प + बदलाव)
+<workspace>/<industryRoot>/<उद्योग>/research-state.json  अनुसंधान-स्थिति स्मृति
+<workspace>/<industryRoot>/<उद्योग>/red-review-note.md    industry_report (रेड-टीम समीक्षा, jobs)
 <workspace>/<industryRoot>/<उद्योग>/chain.json      industry_map
+<workspace>/<industryRoot>/<उद्योग>/chain.svg       industry_map (renderSvg: true)
 <workspace>/<industryRoot>/<उद्योग>/timeline.jsonl  industry_track
 <workspace>/<industryRoot>/<उद्योग>/sources.json    उद्धरण योग्य स्रोत रजिस्ट्री (S1, S2, …)
 <workspace>/<industryRoot>/<उद्योग>/notes/          सीड नोट्स

package/README.md

@@ -32,10 +32,11 @@
 
 ## What you get
 
-- **`industry_map`** — build or update an industry chain map (`chain.json`): upstream/midstream/downstream nodes, edges, and metric slots. A metric with a `value` must carry a `sourceRef`; a value-less slot is an explicit gap. Seed notes/files and an optional `ctx.web` digest are registered as citable sources (`S1`, `S2`, …).
-- **`industry_track`** — public-source policy/news tracking over the official `ctx.web` seam: per-topic search, host allow/block lists, `since` filtering, bounded snapshot fetches with SHA-256 provenance hashes, and append + dedupe into `timeline.jsonl` with a retention cap. Fails loud when the web capability is unreachable.
-- **`company_scan`** — company cards (`card.json` + `card.md`) from your workspace data files (SHA-256-hashed, with Markdown outlines and figure-candidate lines so every number cites a file and a line), plus an optional `ctx.web` citation complement. Text formats only in v1 (no PDF).
-- **`industry_report`** — one auditable report from the map + timeline + cards. With a mounted `ctx.researchReport` engine the evidence/sections/claims are sealed by its `assemble` and per-claim verdicts come back; without one, the builtin fallback writes versioned `reports/<YYYYMMDD-HHmmss>/report.md` + `manifest.json` with a SHA-256 source-traceability table and honestly labels `engine: 'builtin-fallback'`.
+- **`industry_map`** — build or update an industry chain map (`chain.json`): upstream/midstream/downstream nodes, edges, and metric slots. A metric with a `value` must carry a `sourceRef`; a value-less slot is an explicit gap. Node `status` assertions carry a `statusAsOf`, and an optional `taxonomyCode` must hit the built-in 国民经济行业分类 大类 table. With `renderSvg: true` it also writes a deterministic, tier-layered `chain.svg` network diagram with funnel/hub bottleneck highlighting. A `depth` argument scales the web assist (quick/standard/comprehensive). Seed notes/files and an optional `ctx.web` digest are registered as citable sources (`S1`, `S2`, …).
+- **`industry_track`** — public-source policy/news tracking over the official `ctx.web` seam: per-topic search, host allow/block lists, `since` filtering, bounded snapshot fetches with SHA-256 provenance hashes, and append + dedupe into `timeline.jsonl` with a retention cap. Entries may carry a validated `evidenceCategory`; a `depth` argument scales source counts. Fails loud when the web capability is unreachable.
+- **`company_scan`** — company cards (`card.json` + `card.md`) from your workspace data files (SHA-256-hashed, with Markdown outlines and figure-candidate lines so every number cites a file and a line), plus an optional `ctx.web` citation complement. Cards carry a listing `status` (with `statusAsOf`), a format-checked `ticker`, and sourced price/value `metrics` (each with `source` + `asOf`). Batch mode (`companies`) isolates one bad company without aborting the batch; `depth` scales the web complement. Text formats only in v1 (no PDF).
+- **`industry_report`** — one auditable report from the map + timeline + cards. Before producing, a deterministic delivery contract checks blocks, status/price assertions, and placeholder residue (fail loud, no half-assembled report). Timeline entries are grouped by `evidenceCategory`. With a mounted `ctx.researchReport` engine the evidence/sections/claims are sealed by its `assemble` and per-claim verdicts come back; without one, the builtin fallback writes versioned `reports/<YYYYMMDD-HHmmss>/report.md` + `manifest.json` with a SHA-256 source-traceability table and honestly labels `engine: 'builtin-fallback'`.
+- **Artifact version ledger** — `industry_map` / `company_scan` / `industry_report` append to `versions.jsonl` (root-relative path + SHA-256 + timestamp + change type) on every artifact write; reads verify the hash and fail loud on mismatch.
 - **Two methodology skills** — `industry-research-method` (chain decomposition, supply/demand framework, sourcing discipline, gap-declaration discipline) and `company-research-method` (company framework, public-source list, compliance wording).
 - **Typed Cordis events** — `industry-research/map`, `industry-research/track`, `industry-research/report` emitted after each committed artifact.
 
@@ -91,24 +92,25 @@ All tunables are Schemastery `Config` fields; invalid values fail the profile lo
 | `track.maxFetchesPerCall` | `10` | Snapshot-fetch budget per `industry_track` call. |
 | `scan.maxFileBytes` | `1048576` | Per-file read cap for company data files. |
 | `scan.maxFigureCandidates` | `100` | Figure-candidate budget per company scan. |
+| `scan.strictTicker` | `true` | Company-card tickers must match a built-in format (A-share 6 digits, US 1–5 letters, HK 1–5 digits); `false` exempts the format check. |
 
 ## Tools & surfaces
 
-### `industry_map({ industry, seed?, seedFiles?, web?, chain? })`
+### `industry_map({ industry, seed?, seedFiles?, web?, chain?, renderSvg?, depth? })`
 
-Called with `chain`: validates (dangling edges, unsourced values, illegal tiers, duplicate ids — loud failure with the full issue list) and persists `chain.json`, then lists the explicit gap slots. Called without `chain`: returns the current map, the registered sources, and an optional `ctx.web` chain-structure digest so the model can iterate. Emits `industry-research/map`.
+Called with `chain`: validates (dangling edges, unsourced values, illegal tiers, duplicate ids, status/statusAsOf, unknown `taxonomyCode` — loud failure with the full issue list) and persists `chain.json`, then lists the explicit gap slots and bottleneck nodes. Called without `chain`: returns the current map, the registered sources, and an optional `ctx.web` chain-structure digest so the model can iterate. With `renderSvg: true` it also writes a deterministic `chain.svg`. `depth` (quick/standard/comprehensive) scales the web assist. Returns a `delta` summary vs the previous `research-state.json`. Emits `industry-research/map`.
 
-### `industry_track({ industry, topics?, since? })`
+### `industry_track({ industry, topics?, since?, depth?, evidenceCategory? })`
 
-Searches each topic through `ctx.web`, filters by the allow/block lists and `since`, fetches page snapshots (SHA-256) within the call budget, and merges into `timeline.jsonl` (dedupe by normalized URL, capped). Sources whose snapshot failed are kept as citation-only entries with the reason in `note`. Fails loud naming the missing capability when `ctx.web` is unmounted or `offline: true`. Emits `industry-research/track`.
+Searches each topic through `ctx.web`, filters by the allow/block lists and `since`, fetches page snapshots (SHA-256) within the call budget, and merges into `timeline.jsonl` (dedupe by normalized URL, capped). `depth` scales the source/fetch budget; `evidenceCategory` labels every appended entry and is validated against the six-category enum. Sources whose snapshot failed are kept as citation-only entries with the reason in `note`. Returns a `delta` summary vs the previous `research-state.json`. Fails loud naming the missing capability when `ctx.web` is unmounted or `offline: true`. Emits `industry-research/track`.
 
-### `company_scan({ name, dataFiles?, web? })`
+### `company_scan({ name | companies, dataFiles?, web?, status?, statusAsOf?, ticker?, metrics?, depth?, parallel? })`
 
-Reads the workspace data files (`.md/.txt/.csv/.tsv/.json`; v1 does not parse PDF), hashes them, extracts Markdown outlines and figure-candidate lines, optionally attaches `ctx.web` citations, and persists the card. Rejected files come back with reasons; everything the card cannot establish is an explicit gap.
+Reads the workspace data files (`.md/.txt/.csv/.tsv/.json`; v1 does not parse PDF), hashes them, extracts Markdown outlines and figure-candidate lines, optionally attaches `ctx.web` citations, and persists the card. A `status` requires a non-future `statusAsOf`; a `ticker` must match a built-in format unless `scan.strictTicker: false`; every `metrics` value must carry `source` + `asOf`. `companies` (batch) isolates one bad company without aborting the batch; `parallel: true` fans each company into an independent job when `ctx.jobs` is mounted (otherwise it falls back to sequential, reported in the result `mode`). Rejected files come back with reasons; everything the card cannot establish is an explicit gap.
 
 ### `industry_report({ industry, sections?, companies?, draft? })`
 
-Assembles evidence (`E-chain`, `E-timeline`, `E-company-<slug>`) and either validates your `draft` (sections + claims; every claim's `evidenceIds` must reference registered evidence) or builds the mechanical auto-draft (sourced metrics and recent timeline entries become claims). Engine path: sealed directory + `sealHash` + per-claim verdicts. Fallback path: versioned Markdown + manifest, claims honestly marked `unverified`. Emits `industry-research/report`.
+Assembles evidence (`E-chain`, `E-timeline`, `E-company-<slug>`), verifies the read artifacts against `versions.jsonl` (hash mismatch fails loud), and either validates your `draft` (sections + claims; every claim's `evidenceIds` must reference registered evidence) or builds the mechanical auto-draft (sourced metrics and recent timeline entries become claims, grouped by `evidenceCategory`). A deterministic delivery contract runs before production and fails loud on any missing block, placeholder, or unsourced/undated assertion. A deterministic adversarial machine-check always runs, and a red-team review job (`red-review-note.md`) is spawned when `ctx.jobs` is mounted (else `review: skipped(jobs unavailable)`). Engine path: sealed directory + `sealHash` + per-claim verdicts. Fallback path: versioned Markdown + manifest, claims honestly marked `unverified`. Emits `industry-research/report`.
 
 ## Skills
 
@@ -120,10 +122,14 @@ Both load on demand through the standard `skill` tool (`加载 industry-research
 ## Data layout
 
 ```
+<workspace>/<industryRoot>/versions.jsonl              artifact version ledger (SHA-256 + timestamp + change)
+<workspace>/<industryRoot>/<industry>/research-state.json  research-state memory
 <workspace>/<industryRoot>/<industry>/chain.json      industry_map
+<workspace>/<industryRoot>/<industry>/chain.svg       industry_map (renderSvg: true)
 <workspace>/<industryRoot>/<industry>/timeline.jsonl  industry_track
 <workspace>/<industryRoot>/<industry>/sources.json    citable source registry (S1, S2, …)
 <workspace>/<industryRoot>/<industry>/notes/          seed notes
+<workspace>/<industryRoot>/<industry>/red-review-note.md  industry_report (red-team review, jobs)
 <workspace>/<industryRoot>/<industry>/reports/<ts>/   industry_report (report.md + manifest.json)
 <workspace>/<industryRoot>/companies/<name>/card.*    company_scan
 ```

package/README.pt.md

@@ -90,24 +90,25 @@ Todos os ajustes são campos `Config` de Schemastery; valores inválidos falham
 | `track.maxFetchesPerCall` | `10` | Orçamento de capturas por chamada `industry_track`. |
 | `scan.maxFileBytes` | `1048576` | Teto de leitura por arquivo de dados de empresa. |
 | `scan.maxFigureCandidates` | `100` | Orçamento de linhas candidatas por varredura. |
+| `scan.strictTicker` | `true` | Tickers de cartões devem corresponder a um formato integrado (A-share 6 dígitos, EUA 1–5 letras, HK 1–5 dígitos); `false` isenta a verificação. |
 
 ## Tools & surfaces
 
-### `industry_map({ industry, seed?, seedFiles?, web?, chain? })`
+### `industry_map({ industry, seed?, seedFiles?, web?, chain?, renderSvg?, depth? })`
 
-Com `chain`: valida (arestas pendentes, valores sem fonte, tiers ilegais, ids duplicados — falha ruidosa com a lista completa) e persiste `chain.json`, depois lista as lacunas explícitas. Sem `chain`: devolve o mapa atual, as fontes registradas e um resumo opcional de estrutura via `ctx.web` para iterar. Emite `industry-research/map`.
+Com `chain`: valida (arestas pendentes, valores sem fonte, tiers ilegais, ids duplicados, status/statusAsOf, `taxonomyCode` desconhecido — falha ruidosa com a lista completa) e persiste `chain.json`, depois lista as lacunas explícitas e os nós gargalo. Sem `chain`: devolve o mapa atual, as fontes registradas e um resumo opcional de estrutura via `ctx.web` para iterar. Com `renderSvg: true` também grava um `chain.svg` determinístico. `depth` escala a assistência web. Emite `industry-research/map`.
 
-### `industry_track({ industry, topics?, since? })`
+### `industry_track({ industry, topics?, since?, depth?, evidenceCategory? })`
 
-Busca cada tópico via `ctx.web`, filtra pelas listas e `since`, captura snapshots (SHA-256) dentro do orçamento e funde em `timeline.jsonl` (deduplicação por URL normalizada, com teto). Fontes cujo snapshot falhou são mantidas como entradas só-citação com a razão em `note`. Falha em voz alta nomeando a capacidade ausente quando `ctx.web` não está montada ou `offline: true`. Emite `industry-research/track`.
+Busca cada tópico via `ctx.web`, filtra pelas listas e `since`, captura snapshots (SHA-256) dentro do orçamento e funde em `timeline.jsonl` (deduplicação por URL normalizada, com teto). `depth` escala o orçamento; `evidenceCategory` rotula cada entrada e é validado contra a enumeração de seis categorias. Fontes cujo snapshot falhou são mantidas como entradas só-citação com a razão em `note`. Falha em voz alta nomeando a capacidade ausente quando `ctx.web` não está montada ou `offline: true`. Emite `industry-research/track`.
 
-### `company_scan({ name, dataFiles?, web? })`
+### `company_scan({ name | companies, dataFiles?, web?, status?, statusAsOf?, ticker?, metrics?, depth?, parallel? })`
 
-Lê os arquivos de dados do workspace (`.md/.txt/.csv/.tsv/.json`; v1 não analisa PDF), calcula hashes, extrai esboços Markdown e linhas candidatas de números, anexa citações `ctx.web` opcionais e persiste o cartão. Arquivos rejeitados retornam com razões; tudo o que o cartão não consegue estabelecer é uma lacuna explícita.
+Lê os arquivos de dados do workspace (`.md/.txt/.csv/.tsv/.json`; v1 não analisa PDF), calcula hashes, extrai esboços Markdown e linhas candidatas de números, anexa citações `ctx.web` opcionais e persiste o cartão. Um `status` exige um `statusAsOf` não futuro; um `ticker` deve corresponder a um formato integrado salvo `scan.strictTicker: false`; cada valor de `metrics` deve levar `source` + `asOf`. `companies` (lote) isola uma empresa falha sem abortar o lote; `parallel: true` distribui cada empresa em um job independente quando `ctx.jobs` está montado (senão volta a sequencial, refletido em `mode`). Arquivos rejeitados retornam com razões; tudo o que o cartão não consegue estabelecer é uma lacuna explícita.
 
 ### `industry_report({ industry, sections?, companies?, draft? })`
 
-Reúne a evidência (`E-chain`, `E-timeline`, `E-company-<slug>`) e valida o seu `draft` (seções + claims; cada `evidenceIds` deve referenciar evidência registrada) ou constrói o rascunho mecânico (métricas com fonte e entradas recentes viram claims). Rota do motor: diretório selado + `sealHash` + veredictos por claim. Rota integrada: Markdown versionado + manifesto, claims honestamente marcados `unverified`. Emite `industry-research/report`.
+Reúne a evidência (`E-chain`, `E-timeline`, `E-company-<slug>`), verifica os artefatos lidos contra `versions.jsonl` (hash divergente falha em voz alta) e valida o seu `draft` (seções + claims; cada `evidenceIds` deve referenciar evidência registrada) ou constrói o rascunho mecânico (métricas com fonte e entradas recentes viram claims, agrupadas por `evidenceCategory`). Um contrato de entrega determinístico roda antes de produzir e falha em voz alta ante blocos faltantes, marcadores de posição ou asserções sem fonte/data. Uma verificação adversária determinística sempre roda, e um job de revisão vermelho (`red-review-note.md`) é lançado quando `ctx.jobs` está montado (senão `review: skipped(jobs unavailable)`). Rota do motor: diretório selado + `sealHash` + veredictos por claim. Rota integrada: Markdown versionado + manifesto, claims honestamente marcados `unverified`. Emite `industry-research/report`.
 
 ## Skills
 
@@ -119,7 +120,11 @@ Ambas carregam sob demanda pela ferramenta padrão `skill` (`加载 industry-res
 ## Data layout
 
 ```
+<workspace>/<industryRoot>/versions.jsonl             registro de versões (SHA-256 + timestamp + mudança)
+<workspace>/<industryRoot>/<indústria>/research-state.json  memória de estado de pesquisa
+<workspace>/<industryRoot>/<indústria>/red-review-note.md    industry_report (revisão vermelha, jobs)
 <workspace>/<industryRoot>/<indústria>/chain.json      industry_map
+<workspace>/<industryRoot>/<indústria>/chain.svg       industry_map (renderSvg: true)
 <workspace>/<industryRoot>/<indústria>/timeline.jsonl  industry_track
 <workspace>/<industryRoot>/<indústria>/sources.json    registro de fontes citáveis (S1, S2, …)
 <workspace>/<industryRoot>/<indústria>/notes/          notas-semente

package/README.zh.md

@@ -90,24 +90,25 @@ dsh plugin --profile demo remove dsh-industry-research    # 卸载
 | `track.maxFetchesPerCall` | `10` | 每次 `industry_track` 调用的快照抓取预算。 |
 | `scan.maxFileBytes` | `1048576` | 公司数据文件的单文件读取上限（字节）。 |
 | `scan.maxFigureCandidates` | `100` | 每次公司扫描的数字候选行预算。 |
+| `scan.strictTicker` | `true` | 公司卡 ticker 必须匹配内置格式（A 股 6 位数字、美股 1–5 字母、港股 1–5 数字）；`false` 豁免格式校验。 |
 
 ## Tools & surfaces
 
-### `industry_map({ industry, seed?, seedFiles?, web?, chain? })`
+### `industry_map({ industry, seed?, seedFiles?, web?, chain?, renderSvg?, depth? })`
 
-带 `chain` 调用：校验（悬空边、无来源数值、非法 tier、重复 id——带完整问题清单响亮失败）并落盘 `chain.json`，随后列出显式缺口槽位。不带 `chain` 调用：返回当前图、已登记来源与可选 `ctx.web` 产业链结构摘要，供模型迭代。发出 `industry-research/map` 事件。
+带 `chain` 调用：校验（悬空边、无来源数值、非法 tier、重复 id、status/statusAsOf、未知 `taxonomyCode`——带完整问题清单响亮失败）并落盘 `chain.json`，随后列出显式缺口槽位与瓶颈节点。不带 `chain` 调用：返回当前图、已登记来源与可选 `ctx.web` 产业链结构摘要，供模型迭代。`renderSvg: true` 时另写确定性的 `chain.svg`。`depth`（quick/standard/comprehensive）调节 web 辅助检索规模。返回与上次 `research-state.json` 相比的 `delta` 摘要。发出 `industry-research/map` 事件。
 
-### `industry_track({ industry, topics?, since? })`
+### `industry_track({ industry, topics?, since?, depth?, evidenceCategory? })`
 
-经 `ctx.web` 逐主题检索，按白/黑名单与 `since` 过滤，在调用预算内抓取页面快照（SHA-256），合并进 `timeline.jsonl`（按规范化 URL 去重，保留上限）。快照失败的来源以纯引用条目保留并在 `note` 中说明原因。`ctx.web` 未挂载或 `offline: true` 时响亮失败并指明缺失能力。发出 `industry-research/track` 事件。
+经 `ctx.web` 逐主题检索，按白/黑名单与 `since` 过滤，在调用预算内抓取页面快照（SHA-256），合并进 `timeline.jsonl`（按规范化 URL 去重，保留上限）。`depth` 调节来源/抓取预算；`evidenceCategory` 为本批条目打分类标签并按六类枚举校验。快照失败的来源以纯引用条目保留并在 `note` 中说明原因。返回与上次 `research-state.json` 相比的 `delta` 摘要。`ctx.web` 未挂载或 `offline: true` 时响亮失败并指明缺失能力。发出 `industry-research/track` 事件。
 
-### `company_scan({ name, dataFiles?, web? })`
+### `company_scan({ name | companies, dataFiles?, web?, status?, statusAsOf?, ticker?, metrics?, depth?, parallel? })`
 
-读取工作区数据文件（`.md/.txt/.csv/.tsv/.json`；v1 不解析 PDF），计算哈希，提取 Markdown 大纲与数字候选行，可选附加 `ctx.web` 引用，落盘速览卡。未采纳文件带原因返回；卡片无法确立的内容一律写入显式缺口。
+读取工作区数据文件（`.md/.txt/.csv/.tsv/.json`；v1 不解析 PDF），计算哈希，提取 Markdown 大纲与数字候选行，可选附加 `ctx.web` 引用，落盘速览卡。`status` 必须带非未来 `statusAsOf`；`ticker` 必须匹配内置格式（除非 `scan.strictTicker: false`）；每条 `metrics` 数值必须带 `source` + `asOf`。`companies`（批量）单公司失败不中断整批；`parallel: true` 且挂载 `ctx.jobs` 时按公司 fan-out 为独立 job（否则回退顺序路径并在结果 `mode` 注明）。未采纳文件带原因返回；卡片无法确立的内容一律写入显式缺口。
 
 ### `industry_report({ industry, sections?, companies?, draft? })`
 
-组装证据（`E-chain`、`E-timeline`、`E-company-<slug>`），或校验你提供的 `draft`（章节 + claims；每条 claim 的 `evidenceIds` 必须引用已登记证据），或机械自动生成草稿（有来源的指标与近期时间线条目生成 claims）。引擎路径：封存目录 + `sealHash` + 逐 claim 结论。降级路径：版本化 Markdown + manifest，claims 如实标记 `unverified`。发出 `industry-research/report` 事件。
+组装证据（`E-chain`、`E-timeline`、`E-company-<slug>`），读取时对照 `versions.jsonl` 校验哈希（不一致响亮失败），或校验你提供的 `draft`（章节 + claims；每条 claim 的 `evidenceIds` 必须引用已登记证据），或机械自动生成草稿（有来源的指标与近期时间线条目生成 claims，按 `evidenceCategory` 分组）。产出前跑确定性交付契约校验（缺区块/占位符/无来源或无日期断言即响亮失败）。确定性对抗机器检查始终执行；挂载 `ctx.jobs` 时派生红方审阅 job 写回 `red-review-note.md`（否则 `review: skipped(jobs unavailable)`）。引擎路径：封存目录 + `sealHash` + 逐 claim 结论。降级路径：版本化 Markdown + manifest，claims 如实标记 `unverified`。发出 `industry-research/report` 事件。
 
 ## Skills
 
@@ -119,10 +120,14 @@ dsh plugin --profile demo remove dsh-industry-research    # 卸载
 ## Data layout
 
 ```
+<工作区>/<industryRoot>/versions.jsonl        工件版本清单（SHA-256 + 时间戳 + 变更类型）
+<工作区>/<industryRoot>/<行业>/research-state.json  研究状态记忆
 <工作区>/<industryRoot>/<行业>/chain.json      industry_map
+<工作区>/<industryRoot>/<行业>/chain.svg       industry_map（renderSvg: true）
 <工作区>/<industryRoot>/<行业>/timeline.jsonl  industry_track
 <工作区>/<industryRoot>/<行业>/sources.json    可引用来源登记（S1、S2……）
 <工作区>/<industryRoot>/<行业>/notes/          种子笔记
+<工作区>/<industryRoot>/<行业>/red-review-note.md  industry_report（红方审阅，需 jobs）
 <工作区>/<industryRoot>/<行业>/reports/<ts>/   industry_report（report.md + manifest.json）
 <工作区>/<industryRoot>/companies/<公司>/card.*  company_scan
 ```

package/cordis.patch.yml

@@ -42,3 +42,7 @@
           maxFileBytes: 1048576
           # Maximum figure-candidate lines surfaced per company scan.
           maxFigureCandidates: 100
+          # When true, company-card tickers must match a built-in format
+          # (A-share 6 digits, US 1-5 letters, HK 1-5 digits); false exempts
+          # the format check.
+          strictTicker: true