alphacouncil-agent 0.9.1 → 0.9.4

package/docs/INSTALL.md

@@ -10,14 +10,24 @@ plain MCP server in the Claude desktop app.
 ---
 
 
-## npm
+## GitHub v0.9.4 prerelease
+
+This acceptance release is GitHub-only:
+
+```bash
+npm install -g github:Zhao73/alphacouncil-agent#v0.9.4
+```
+
+## npm channel
 
 ```bash
 npm install -g alphacouncil-agent@next
 ```
 
-`0.9.1` is intentionally on `next`; unqualified `alphacouncil-agent` follows the stable
-`latest` tag and does not install this non-GA preview.
+The GitHub-only 0.9.4 acceptance does not run `npm publish` or change npm dist-tags.
+`alphacouncil-agent@next` installs whichever preview npm currently serves; verify it with
+`npm view alphacouncil-agent dist-tags --json` rather than assuming it is 0.9.4. The
+unqualified package follows the stable `latest` tag.
 
 Then point any MCP host at the `alphacouncil-agent` binary. For Claude Code:
 
@@ -57,7 +67,7 @@ persona set loads, and what the data directory holds.
 One command, `/alpha`. Modes are arguments.
 
 ```text
-/alpha <ticker>              full council — displays all methods and confirms this run's selection
+/alpha <ticker>              full council — displays all methods; headless run is bounded <=30m
 /alpha <ticker> quick        quick_v1 — 4 analysts incl. news + 1-4 methods + one parallel debate round (<=10m)
 /alpha <ticker> screen       mechanical filings screen only        (no model spend)
 /alpha <ticker> options      IV term structure, skew, positioning  (no model spend)
@@ -69,8 +79,35 @@ One command, `/alpha`. Modes are arguments.
 The four marked *no model spend* call keyless data tools and spawn no subagents. **Start
 there** — they show real data at no cost, so you can see the shape of the thing before
 committing a fan-out. Full and quick both require a fresh method selection receipt. Model
-spend is not a fixed seat count: selected method seats may execute deterministically or be
-recorded `out_of_scope`, while evidence and debate workers are model calls.
+spend is not a fixed seat count. In plugin-managed headless mode every selected v3 seat runs
+its deterministic policy (which may return `out_of_scope`) and then one isolated voice
+worker explains that frozen result; evidence and debate seats are also model workers. The
+voice is a recorded provisional method result, not the named person's words.
+
+### What full headless means
+
+Full remains `full_v2`. When launched through plugin-managed headless
+`analyze_symbol(council_mode="full")`, it has a hard maximum of 1800000 ms from durable
+queueing through terminal artifact persistence:
+
+- all eight mandatory evidence workers start in one parallel wave;
+- after the fail-closed evidence barrier, each selected physical v3 method freezes its
+  deterministic stance and gets one isolated voice worker that cannot alter it;
+- Bull and Bear start in parallel within each of three rounds, with a barrier between rounds
+  and exact Round-2-question to Round-3-answer binding;
+- the PM and deterministic report persistence use the same global clock.
+
+Callers and `ALPHACOUNCIL_FULL_TOTAL_MS` may lower the deadline, never raise it. At expiry
+the server saves a terminal `incomplete` run naming failed/timed-out/skipped seats. The
+30-minute bound guarantees terminal persistence, not complete results when search, model
+transport or data providers are unavailable. `plan_visible_run` has no such enforceable
+deadline because the external host owns and schedules its subagents.
+
+A full `user_response.md` lists all eight analyst statuses and summaries, every selected
+stable master ID with its stance and isolated-worker output/status, and a system-owned price
+snapshot with currency/time/source or an explicit quote-data gap. System-owned labels are
+localized for Chinese (`zh-CN`), English, Japanese and Korean, and worker prompts carry the
+run language.
 
 ### What `quick` means
 

package/docs/evaluation/persona-v3-release-evidence.md

@@ -1,6 +1,6 @@
 # PersonaPack v3 GA release evidence
 
-Status: package/plugin version `0.9.0-solo-test.3`, channel `solo_test`; verifier contracts implemented,
+Status: package/plugin version `0.9.2`, channel `solo_test`; verifier contracts implemented,
 but no production pass, approval artifact or signature claim is included in this repository.
 
 ## Solo-test boundary
@@ -119,5 +119,5 @@ node scripts/check-persona-v3-ga.mjs \
 ```
 
 `--release-manifest` remains only as a migration diagnostic and always blocks GA. The
-current worktree is version 0.9.0-solo-test.3 and has no approved physical production release
+current worktree is version 0.9.2 and has no approved physical production release
 or signed GA evidence set, so the default production-GA command must fail.

package/docs/plans/0.9.0-personapack-v3.md

@@ -1,6 +1,6 @@
 # AlphaCouncil 0.9.0 - PersonaPack v3 execution tracker
 
-Status: `0.9.0-solo-test.3` **solo-test prerelease**; not formal production GA. Bare
+Snapshot status on 2026-07-27: `0.9.0-solo-test.3` **solo-test prerelease**; not formal production GA. Bare
 `0.9.0` remains the future GA target, and the production release gates at the end of this
 document remain fail-closed.
 
@@ -8,7 +8,7 @@ Revision: **1** (2026-07-27). Revision 1 adds corpus-baseline and human-adjudica
 a real-grounding refusal contract for Taleb, and a preregistered historical/shadow protocol
 for any future `N_eff` estimate. It does not claim that those gates have already passed.
 
-## Current solo-test snapshot
+## 2026-07-27 solo-test snapshot
 
 - Package/plugin/build-profile version: `0.9.0-solo-test.3`; channel: `solo_test`.
 - Physical v3 packs: **26/26**; explicit solo-test loader/compiler: **26/26**.
@@ -31,7 +31,7 @@ research, compute, reject and remember, not by imitating a famous person's tone.
 and executed through the same catalog/runtime shape. It does not prove attributed methods,
 operational admission or production eligibility.
 
-The release does not describe 26 opinions as 26 independent samples. It records shared-model,
+The planned release does not describe 26 opinions as 26 independent samples. It records shared-model,
 shared-evidence and shared-source correlation, preserves minority reports, and reports
 effective diversity only after measured ablations.
 

package/docs/releases/v0.9.2.md

@@ -0,0 +1,62 @@
+# AlphaCouncil Agent v0.9.2 — Quick Deadline Hotfix (Non-GA)
+
+`v0.9.2` supersedes `v0.9.1` on npm's `next` dist-tag. It preserves the same bounded
+quick-council feature and assurance scope, while fixing a clean-CI failure in the timeout
+implementation. It is a GitHub prerelease and is not a PersonaPack production-GA release.
+
+## Why 0.9.2 exists
+
+After `0.9.1` was accepted by npm, the first clean GitHub Actions matrix showed the quick
+grounding timeout being cancelled with `Promise resolution is still pending but the event
+loop has already resolved` on Linux, macOS and Windows. The deadline timers had been
+`unref()`'d. In a short-lived process with no other referenced handles, Node could exit
+before the timer fired, so the claimed hard deadline was not self-sustaining.
+
+npm versions are immutable. The published `0.9.1` bytes were therefore not replaced or
+hidden; `0.9.2` carries the correction and `0.9.1` is marked as superseded.
+
+## Fix
+
+- Grounding and linked-operation deadline timers remain referenced until they abort or are
+  cleared.
+- The existing abort propagation remains intact: a timeout cancels the underlying quote,
+  options, macro, filing, screen and non-US market requests.
+- The global quick deadline still includes grounding, retries, queueing, evidence, method
+  seats, Bull/Bear, PM synthesis and deterministic finalization.
+- An isolated child-process regression proves the linked deadline fires when it is the only
+  remaining event-loop handle. The Windows quick-analysis fixture now launches its fake
+  worker through an explicit Node `.cmd` wrapper, matching the production invocation shape.
+
+## Unchanged quick contract
+
+- Complete 26-seat catalog display before research; quick accepts one to four explicit
+  stable IDs and rejects `all`.
+- Four parallel evidence seats: `market_data`, `earnings_deep_dive`,
+  `valuation_long_short`, `news_industry_management`.
+- One wave of selected method seats, one parallel Bull/Bear statement round and one short
+  PM synthesis.
+- Hard end-to-end ceiling of 600,000ms, which may only be lowered.
+- Recent-news display includes only dated sources in the 120 days up to `as_of`.
+- Explicit `degraded` terminal state and system-owned failure ledger.
+- `quick_v1` is not equivalent to `full_v2`; it does not run three-round exact Q&A or the
+  adversarial source/numeric/counterevidence verifier stage.
+- Method-seat output is a provisional recorded lens result, not a quotation from the named
+  person.
+
+## Assurance boundary
+
+- Build channel: `solo_test`.
+- `production_eligible=false`.
+- `method_model_eligible=false`.
+- Physical provisional packs: 26.
+- Provisional `operator_lens`: 26.
+- Operational seats: 0.
+- Validated `method_model` seats: 0.
+- Live external host E2E evidence: 0/4.
+
+Changing `pack_version` to `0.9.2` mechanically changes all 26 pack hashes and the ten
+simulation binding artifacts. That administrative hash churn does not change method logic,
+evidence approval, experiment status or maturity.
+
+The formal GA gate remains fail-closed. This hotfix does not promote method maturity or
+claim production readiness.

package/docs/releases/v0.9.3.md

@@ -0,0 +1,86 @@
+# AlphaCouncil Agent v0.9.3 — Bounded Full Council and Complete Handoff (Non-GA)
+
+`v0.9.3` is a non-GA runtime and reporting correction released as a GitHub prerelease.
+This acceptance release does not publish npm or mutate dist-tags. It preserves `full_v2`
+and `quick_v1`; it does not promote any PersonaPack
+seat, relax the selection gate or claim production GA.
+
+## Why 0.9.3 exists
+
+Earlier full headless runs could take substantially longer than the useful user window:
+the eight evidence seats were concurrency-limited, Bull/Bear sides were awaited serially,
+and full had no global queue-to-persistence deadline. The concise handoff could also sample
+analyst groups and omit selected Master Bench results even when their artifacts existed.
+Some v3 seats completed only through the deterministic policy, so users reasonably expected
+an individual method explanation but saw no dedicated model-worker output.
+
+Those are product defects, not investment conclusions. `v0.9.3` changes scheduling,
+visibility and terminalization while preserving the evidence and maturity boundaries.
+
+## Full headless runtime contract
+
+Plugin-managed headless `analyze_symbol(council_mode="full")` now has a hard maximum of
+1,800,000 ms from durable queueing through terminal artifact persistence:
+
+- all eight mandatory evidence workers start in one parallel wave;
+- the evidence barrier remains fail-closed after one bounded parse-only repair;
+- each selected physical v3 method first executes its deterministic policy and freezes its
+  stance, then launches one isolated voice worker for that stable ID;
+- the voice worker may explain the frozen result in the run language but cannot change the
+  stance, invent missing typed facts or speak as the real named person;
+- Bull and Bear launch in parallel within each of the three rounds, with a barrier between
+  rounds and exact Round-2-question to Round-3-answer binding;
+- the PM starts only after both Round-3 sides pass, and deterministic final assembly shares
+  the same global deadline.
+
+A caller or environment can lower the limit, never raise it. At deadline expiry the server
+stops opening downstream work and persists a terminal fail-closed `incomplete` run naming
+every timed-out, failed and skipped role. Thirty minutes therefore guarantees a durable,
+auditable terminal state; it does **not** guarantee that all seats complete when search,
+model transport or external data providers are unavailable or slow.
+
+The deadline belongs to plugin-managed headless `analyze_symbol`. A visible-host run created
+through `plan_visible_run` is scheduled by the external host; the plugin cannot force-stop
+those host subagents and makes no 30-minute promise for that path.
+
+## Complete user handoff
+
+The full `user_response.md` no longer hides the council behind a three-group synopsis. It
+must visibly include:
+
+- a system-owned price snapshot with price, currency, quote timestamp, exchange/feed and
+  source when available, or an explicit unavailable-data gap;
+- every selected stable master ID, its frozen stance and its isolated voice-worker
+  explanation/status;
+- all eight mandatory analyst task IDs, statuses and summaries;
+- key earnings, forward expectations, company/industry news, valuation, position guidance,
+  invalidation conditions, terminal/report-quality state, elapsed/deadline state and paths.
+
+Master output remains a **recorded provisional method-seat result**. It is not a quotation,
+endorsement, current statement or simulated personal advice from the named person.
+
+System-owned catalog/report/handoff labels and failure text support `zh-CN`, English,
+Japanese and Korean. Worker prompts receive the run language. Stable IDs and JSON field
+names remain English for artifact compatibility.
+
+## Quick contract
+
+`quick_v1` remains explicit, plugin-managed and bounded to 600,000 ms. Its fixed four-seat
+evidence wave, 1-4 method limit, one parallel Bull/Bear statement and short PM are unchanged.
+It still rejects `all`, does not run three-round exact Q&A or adversarial verifiers, and
+cannot be represented as `full_v2`.
+
+## Assurance boundary
+
+- Build channel: `solo_test`.
+- `production_eligible=false`.
+- `method_model_eligible=false`.
+- Physical provisional packs: 26.
+- Provisional `operator_lens`: 26.
+- Operational seats: 0.
+- Validated `method_model` seats: 0.
+- Human source/formula approvals and approval signatures: 0.
+
+One worker per selected method makes each recorded result visible; it does not create 26
+independent base models, prove behavioral differentiation or promote a seat's maturity.
+Formal production GA remains fail-closed.

package/docs/releases/v0.9.4.md

@@ -0,0 +1,79 @@
+# AlphaCouncil Agent v0.9.4 — Locale-Consistent Council Output (Non-GA)
+
+`v0.9.4` is a non-GA GitHub prerelease that corrects the locale contract exposed by the
+post-release QQQ Stage 0 acceptance. It does not publish to npm or change npm dist-tags.
+
+## Why 0.9.4 exists
+
+The `v0.9.3` runtime correctly bounded a plugin-managed full council to a thirty-minute
+terminal state and made every selected method-seat statement visible. Its selection layer
+still had a serious observability defect: a `zh-CN` catalog returned English `method` copy,
+Japanese and Korean silently fell back to English, `auto` did not infer the prompt language,
+and confirmation text was always English. A structurally complete English PM report could
+also pass `report_quality` for a Japanese or Korean run.
+
+Those are product defects, not investment findings.
+
+## Selector and receipt corrections
+
+- `zh-CN`, English, Japanese and Korean resolve to one canonical locale before catalog,
+  intent and receipt hashes are created.
+- `auto` infers Chinese, Japanese and Korean scripts from the unchanged run prompt.
+- Unsupported explicit selector locales fail with `UNSUPPORTED_SELECTION_LANGUAGE`; the
+  runtime no longer labels an English fallback as customer-language output.
+- Every one of the 26 physical solo-test manifests carries four-locale title, identity,
+  method and best-for copy. All 26 Chinese method descriptions are now Chinese rather than
+  copied English, and reader-facing `best_for` no longer exposes machine snake-case domain
+  IDs in Chinese, Japanese or Korean.
+- Human-facing maturity has `maturity_label`; the stable machine enum remains unchanged.
+- Selection and confirmation text is localized, while stable IDs, receipt tokens, hashes,
+  JSON keys and maturity enums remain machine-stable.
+- Receipt consumption revalidates the complete selection-record binding and recomputes
+  `selection_hash`; a tampered hash or crossed record fails before a run is created or the
+  one-time receipt is consumed.
+
+## Worker and report language integrity
+
+- Plugin-managed Evidence, dedicated method voice, Bull/Bear and PM packets are checked for
+  the requested reader language. A wrong-language response receives one bounded parse-only,
+  no-search repair opportunity and otherwise fails closed. Visible-host record tools reject
+  wrong-language packets before writing or marking a seat complete so the host can retry it.
+- The language audit treats Han-only fragments as shared script instead of claiming that
+  every Kanji sentence is Chinese. Kana and Hangul take precedence, and English must contain
+  dominant English prose rather than a few company names or finance acronyms.
+- Quick PM now uses the same parse/language/repair path as Bull and Bear. A real debate
+  timeout or transport failure produces a localized failure packet, never a fictitious
+  `DRY_RUN`, and reader-language mismatch remains distinct from malformed JSON.
+- Japanese and Korean timeout, malformed-output, deterministic method-decline and PM
+  fallback text is localized. Failure diagnostics remain separate from investment evidence.
+- `report_quality.json` records `requested_locale`, `observed_locale`, `language_status`,
+  target-script counts and mismatched sections. A complete English report cannot pass a
+  Japanese or Korean run.
+- The system-owned Master Bench heading no longer appends English to non-English headings,
+  and delayed-price labeling no longer leaks provider English notes into localized handoff.
+- Full visible-host recording requires all six Bull/Bear round artifacts, ordered round
+  barriers and the exact Round-2-question to Round-3-answer binding before the PM can finish.
+  Same-content retries are idempotent; conflicting round replays are rejected.
+
+## Package acceptance
+
+The release package check creates a physical npm tarball with lifecycle scripts disabled,
+installs it offline into a temporary prefix, starts that installed MCP server, and checks all
+four selector locales, exactly 31 MCP tools, exactly 26 catalog seats, one stable-ID
+selection, one receipt consumption and replay rejection. Full source, generated-pack and
+installed-tarball checks remain mandatory.
+
+## Runtime boundary
+
+- `full_v2` remains bounded queue-to-terminal at 1,800,000 ms for plugin-managed headless
+  execution. A deadline failure is persisted as incomplete; it is not a completeness promise.
+- Each selected physical v3 method still receives an isolated explanation worker only after
+  its deterministic stance is frozen. These are provisional project-derived method seats,
+  not the named people's statements, endorsements or current views.
+- Build channel: `solo_test`.
+- `production_eligible=false`.
+- `method_model_eligible=false`.
+- Physical provisional packs: 26.
+- Operational seats: 0.
+- Validated `method_model` seats: 0.
+- Formal GA remains fail-closed.

package/docs/report-contract.md

@@ -22,6 +22,16 @@ writes the mode-appropriate versions of:
 The report and handoff must call a named master result a recorded method-seat or lens result.
 It is not a quote from, endorsement by, or current statement of the named person.
 
+System-owned selector, report and handoff labels are localized for `zh-CN`, `en`, `ja` and
+`ko`. Each worker receives the canonical run language, while stable IDs and JSON field names
+remain English. Reader-facing evidence, method, debate and PM fields are checked against the
+requested locale. Plugin-managed workers receive one bounded no-search language/JSON repair,
+then fail closed; visible-host record tools reject the wrong-language packet before writing
+it so the host may retry. `report_quality.json` records requested/observed locale, keeps
+Han-only fragments explicitly inconclusive, and cannot pass a Japanese or Korean run whose
+report body is English. Unsupported explicit selector locales are rejected instead of being
+silently mislabeled as localized English.
+
 ## full_v2 Contract
 
 Full remains the default. Its `final_report.md` visibly covers:
@@ -48,11 +58,46 @@ Full remains the default. Its `final_report.md` visibly covers:
 - confidence
 - source table
 
+Its `user_response.md` must also visibly carry:
+
+- a system-owned price snapshot with price, currency, quote timestamp, exchange/feed and
+  source when available, or an explicit statement that the quote is unavailable;
+- every selected stable master ID, its frozen deterministic stance, its isolated voice-worker
+  explanation/status and a clear `not a quote` attribution boundary;
+- all eight mandatory analyst task IDs, statuses and summaries, including failures or gaps;
+- terminal status, contract, report quality, elapsed time, deadline state and artifact paths.
+
 All mandatory full evidence roles must be completed. If one still fails after the one bounded
 parse-only repair, full fails closed at the evidence barrier: no master, bull/bear or PM model
 call is started. The run is persisted as `incomplete` with the failed evidence and skipped
 downstream roles named. A partial PM opinion never converts that run to complete.
 
+## Full Runtime Budget
+
+Plugin-managed headless `analyze_symbol(council_mode="full")` has a non-overridable maximum
+of 1800000 ms from durable queueing through terminal artifact persistence. A caller or
+environment may lower it, never raise it. The execution topology is:
+
+1. the eight mandatory evidence workers start in one parallel wave;
+2. after the evidence barrier, every selected physical v3 method runs its deterministic
+   policy and freezes a stance, then receives one isolated voice worker that can explain but
+   cannot change that stance;
+3. Bull and Bear start together within each of three rounds, with a barrier before the next
+   round; the PM starts only after both Round-3 outputs pass exact Q&A validation;
+4. deterministic assembly and persistence consume the same global clock.
+
+At deadline expiry the run stops opening downstream work and persists fail-closed as
+`incomplete`, naming timed-out, failed and skipped roles. This is a terminal-persistence
+guarantee, not a promise of full-seat success when search, model transport or data sources
+are unavailable. The deadline does not apply to `plan_visible_run`: an external host owns
+those subagents and the plugin cannot force-stop them.
+
+Visible full runs use the same six role-by-round audit records and exact Q&A gate. Each
+Bull/Bear call declares round 1, 2 or 3; the server rejects out-of-order calls, altered replay
+content and a PM submitted before both Round-3 records pass. Identical role/round retries are
+idempotent. This preserves workflow completeness but does not give host-owned subagents the
+plugin-managed 30-minute deadline.
+
 ## quick_v1 Contract
 
 Quick is explicit and plugin-managed through headless `analyze_symbol`. It is not available
@@ -154,7 +199,9 @@ Every handoff includes status, report contract, report quality, rating, winner,
 one judgment paragraph, valuation/position, material gaps and file locations.
 
 A full handoff additionally carries the key earnings result, forward setup, news/voice
-signals and top invalidation conditions.
+signals and top invalidation conditions. It lists every selected method seat and all eight
+mandatory analysts rather than sampling a subset, and includes the system-owned price
+snapshot or an explicit quote-data gap.
 
 A quick handoff instead names:
 
@@ -175,10 +222,12 @@ Both handoffs list `final_report.md`, `artifact_index.md`, `all_agents.md`, and
 | "The user only asked if they can enter, so four bullets are enough." | Keep chat concise, but write the mode-appropriate report and artifact index. |
 | "Quick passed report quality, so it passed full." | `quick_v1` can only pass `quick_v1`; retain `full_council_equivalent=false`. |
 | "Ten minutes expired, so silently omit a seat." | Record the exact degraded/incomplete ledger; never manufacture evidence or extend the ceiling. |
+| "Thirty minutes expired, so finish full with the seats that returned." | Persist full as `incomplete`, name every missing/skipped role and keep the saved partial evidence; never synthesize a complete verdict. |
 | "One evidence task failed, but the PM has an opinion." | Apply the mode-specific evidence gate; full fails before downstream calls, quick may only use its explicit degraded rules. |
 | "The source table mentions the news, so the news section can be skipped." | News findings need their own visible section; quick recent news must also pass its date window. |
 | "The final report exists, so chat can hide file locations." | The handoff lists the saved report, index, trace and quality file. |
 | "The master said this." | Call it a recorded method-seat result, never a quote from the named person. |
+| "The full report has a bench table, so the handoff can hide the individual seats." | Full handoff lists every selected stable ID, stance and isolated-worker result/status. |
 
 ## How the Quality Gate Checks Reports
 

package/knowledge/ai-assisted-solo/experiments/runs/a.json

@@ -18,12 +18,12 @@
   "formal_h_satisfied": false,
   "human_reference": false,
   "input_binding": {
-    "input_hash": "sha256:633cc82231d542dbc55c41a65fb234104735180d4bea554c42b7ef2db8c7db44",
+    "input_hash": "sha256:3e3ba26efc5e2b26c1f7549c845af116b7a07b9b8814cb16bd0bbf633289d3eb",
     "path": "simulation-input.json"
   },
   "network_allowed": false,
   "production_effect": "none",
-  "result_hash": "sha256:25e121198a6cbb96554df3c005349e98e7d47a5f948bb0b3cf437d1375541f1b",
+  "result_hash": "sha256:00da168490d6f96bd98e05a69427686ed2b19b9a6e1c23033e004de3798df762",
   "results": {
     "blocked_fail_closed_count": 0,
     "completion_status": "completed_machine_simulation",
@@ -34,7 +34,7 @@
         "fact_pack_hash": "sha256:0adab0b69b1e8ba756deaeac3c457e044dd5a7195b1847893db0953772220e41",
         "frozen_decision_hash": "sha256:ddfd120be4926ef3b400a07f8af44c277d170f4c8f48364579b5b21ae48fa661",
         "native_state": "provisional_own_at_price",
-        "pack_hash": "sha256:9d8260b5256a67c74be7d2d25545c13efd2fcae8cd95cf247c8f48cc36fc8ee5",
+        "pack_hash": "sha256:acd69a6e493f5b1faa050d3266d12ed6d97a7e698daeba073727c085ff5c63dd",
         "persona_id": "master_buffett",
         "policy_execution_hash": "sha256:5e853405b4dd96e4760f842c0726e26bf7f87e9e3cc79a6b90e3ceb58a2211a5",
         "score_ratio": 1,
@@ -54,7 +54,7 @@
     "verifier_checks": []
   },
   "reviewer_kind": "ai",
-  "run_artifact_hash": "sha256:dfd795cbdab47b35a82f5c97d132c15c12ca83cc1d2a63fbf5be58823059c3ea",
+  "run_artifact_hash": "sha256:01f7b32aea7e26f216d62f3d1123770e39defa026bb83f65dbb4475d572a7280",
   "run_id": "A",
   "schema_version": 1,
   "simulation_scope": "single deterministic seat surrogate; formal single-model answer is not simulated"

package/knowledge/ai-assisted-solo/experiments/runs/b.json

@@ -16,7 +16,7 @@
   "formal_h_satisfied": false,
   "human_reference": false,
   "input_binding": {
-    "input_hash": "sha256:633cc82231d542dbc55c41a65fb234104735180d4bea554c42b7ef2db8c7db44",
+    "input_hash": "sha256:3e3ba26efc5e2b26c1f7549c845af116b7a07b9b8814cb16bd0bbf633289d3eb",
     "path": "simulation-input.json"
   },
   "network_allowed": false,
@@ -37,7 +37,7 @@
     "verifier_checks": []
   },
   "reviewer_kind": "ai",
-  "run_artifact_hash": "sha256:43f0e68bacb64753bff985b7a1c6e726490fec70c4250a376a6cf9e0fea48372",
+  "run_artifact_hash": "sha256:8e089c92d91ec2e790a3d253a539376c706c3d7a99a71dad249d41f213ac6de0",
   "run_id": "B",
   "schema_version": 1,
   "simulation_scope": "frozen-input control digest only; eight LLM analyst outputs are not simulated"

package/knowledge/ai-assisted-solo/experiments/runs/c.json

@@ -16,7 +16,7 @@
   "formal_h_satisfied": false,
   "human_reference": false,
   "input_binding": {
-    "input_hash": "sha256:633cc82231d542dbc55c41a65fb234104735180d4bea554c42b7ef2db8c7db44",
+    "input_hash": "sha256:3e3ba26efc5e2b26c1f7549c845af116b7a07b9b8814cb16bd0bbf633289d3eb",
     "path": "simulation-input.json"
   },
   "network_allowed": false,
@@ -37,7 +37,7 @@
     "verifier_checks": []
   },
   "reviewer_kind": "ai",
-  "run_artifact_hash": "sha256:2db7b9fad7e7a7f1056640431d6c6af56e7182683e5fdb51b4040b88a9a48676",
+  "run_artifact_hash": "sha256:cd7b259120c835df126e2e445e131a5d595e4efbd059c27d20a9f596390744d3",
   "run_id": "C",
   "schema_version": 1,
   "simulation_scope": "prompt-catalog binding only; prompt-only LLM outputs are not simulated"

package/knowledge/ai-assisted-solo/experiments/runs/d13.json

@@ -30,12 +30,12 @@
   "formal_h_satisfied": false,
   "human_reference": false,
   "input_binding": {
-    "input_hash": "sha256:633cc82231d542dbc55c41a65fb234104735180d4bea554c42b7ef2db8c7db44",
+    "input_hash": "sha256:3e3ba26efc5e2b26c1f7549c845af116b7a07b9b8814cb16bd0bbf633289d3eb",
     "path": "simulation-input.json"
   },
   "network_allowed": false,
   "production_effect": "none",
-  "result_hash": "sha256:03e60cc07be3ea141b5cec29e1ecc09413181be84abf2c90483ec41a7d7b7c15",
+  "result_hash": "sha256:37d5a3ce7d7beaddebfb7dfa41404027478b16179735a077a071a789e886df90",
   "results": {
     "blocked_fail_closed_count": 0,
     "completion_status": "completed_machine_simulation",
@@ -46,7 +46,7 @@
         "fact_pack_hash": "sha256:702d4dbd2513e429997e9e77b69f8f5dd924bf4c01d8c7accf11645f2a39664c",
         "frozen_decision_hash": "sha256:1d568408ad261a4899b0faca1b8506d2c088408587bf370cb163e15b4f975e3b",
         "native_state": "provisional_undervalued",
-        "pack_hash": "sha256:e2860158ecf8ae9ae91f0276f3585579f3a90c166062cc0d88500eec518447f9",
+        "pack_hash": "sha256:31a3466e5ad820f398e87bf81c7e04d94d66442029993d519333d817c01f9100",
         "persona_id": "master_damodaran",
         "policy_execution_hash": "sha256:be74915d6f24c7a7223bf449b805bbb419bdf724f482ccf7df48efb87afc3e41",
         "score_ratio": 1,
@@ -60,7 +60,7 @@
         "fact_pack_hash": "sha256:76cca34c2e39ea5898330406a49612004eb5ac6f5a9fc3bac57c06f72bce028f",
         "frozen_decision_hash": "sha256:84dd5be279cf1ee37990d2cdc58a16ee00d99995df6dc48ea86525626764f97f",
         "native_state": "provisional_margin_of_safety",
-        "pack_hash": "sha256:c401a14385bb67d5f4135718972ba10c670e549e4fa3c3836f3f11c493577436",
+        "pack_hash": "sha256:4f8bb4a076bccf522315beac6725dcbb34c8a96cc5911e6a0ac4ae915a9e893b",
         "persona_id": "master_graham",
         "policy_execution_hash": "sha256:5cd51dcd115c45d1b2c7afdba56d2db078869825eb0958afade69beed7421b43",
         "score_ratio": 1,
@@ -74,7 +74,7 @@
         "fact_pack_hash": "sha256:301b59fe11ccf10706684f270e8b7185260545dc68b13c39438f39b53e93efab",
         "frozen_decision_hash": "sha256:a2d28669d1d782d84f8af0f45ecc18609b23d32cb0ff29fc4777bc3585a6e349",
         "native_state": "provisional_engagement_candidate",
-        "pack_hash": "sha256:6cc2aa3e5cf2698de3e42321531ae77f11e0b32cc40c5f303cfae3f5b638560f",
+        "pack_hash": "sha256:34150f8e64223a6ca752cd8e96f8da7fdbd5556d50db8abc8232cf8c7e73943d",
         "persona_id": "master_ackman",
         "policy_execution_hash": "sha256:f6f80cef0c69e2f755e2f1c20dc5886cba2a5e6e0dd675fb1b6eb56330ada427",
         "score_ratio": 1,
@@ -88,7 +88,7 @@
         "fact_pack_hash": "sha256:46c6f2d5014a20e4f96254cb363b71eedf66beb33204a12a4579e08116afb249",
         "frozen_decision_hash": "sha256:6a8785ec84078c6f8daad9c6a18a71779196795ccdb6d3a6ac34b58da7c90b07",
         "native_state": "provisional_asymmetric_innovation",
-        "pack_hash": "sha256:2d5e3101a82b4097b8a6b75085f4ff13fb0fbaa347f262cd8d3bed94808ce672",
+        "pack_hash": "sha256:a002878b7c572ff70c231c339beccb9ab690652759261d94fe113096a319f5d7",
         "persona_id": "master_cathie_wood",
         "policy_execution_hash": "sha256:66b39b370506643e72ede933211accd910945072cdeaf4f9785d2316324a4cae",
         "score_ratio": 1,
@@ -102,7 +102,7 @@
         "fact_pack_hash": "sha256:c249a152df048b50d725d1454faf6081b1bf2ef7b02dd8352275bee559f3fa13",
         "frozen_decision_hash": "sha256:c184bca80fca0d0ab573789c97539dfdd0d1cb63e567829c0d500b15c660b796",
         "native_state": "provisional_monitor",
-        "pack_hash": "sha256:8259649c35488d58f4912c29f4f73285995c598700c460692b07e2ee8a479a97",
+        "pack_hash": "sha256:144ee322b64d24a46232588b9fd8bed5c0140e49ba7db891a3dfe43c28d6c07c",
         "persona_id": "master_munger",
         "policy_execution_hash": "sha256:1bfbc42a862cab4e876e0f98991a8cce7b1bd0dc08a239ce0897d97fc02411b2",
         "score_ratio": 1,
@@ -116,7 +116,7 @@
         "fact_pack_hash": "sha256:a5b193f40fec598fb159fbe5e37dd9782e2660cee697c20e9a59b7f7f7bfa959",
         "frozen_decision_hash": "sha256:40be9e02a61ec476335d0a5c0f36dee44de7e03cb6f3d34bb0ac658b792b06af",
         "native_state": "provisional_structural_mispricing",
-        "pack_hash": "sha256:76eb044893c547e58ecf6e4c8731c91f58568640232c0c1c4a7e4b06580d44c2",
+        "pack_hash": "sha256:5d0d0a8123487f1f8c129624524f2fbc2f532b99874ba8f9aee4dc14ac3ceb2f",
         "persona_id": "master_burry",
         "policy_execution_hash": "sha256:3dc3d19e4a61dd494d3072695f5c829e0e60fd7891720cd9dcf6769a531f6995",
         "score_ratio": 1,
@@ -130,7 +130,7 @@
         "fact_pack_hash": "sha256:5b1b600295eeee302313ddf47dc1eb639a469e2c9b5c4aa33a63eba9ec1e6ce5",
         "frozen_decision_hash": "sha256:f4c16ec558bb51672b53b675ec8a8a6b50453537a32897602c4bae5a2ef2be40",
         "native_state": "provisional_asymmetric_candidate",
-        "pack_hash": "sha256:408a7b0965e2fb4fb984ac9490d5959c225fbf9c0b384fb2d719b0fbad22a38a",
+        "pack_hash": "sha256:8bacb4b8e388f709d478299bd2d81bb8f79571d96b9c1881e24f6d9b8567f85c",
         "persona_id": "master_pabrai",
         "policy_execution_hash": "sha256:38edf349b7d5b42136634cca7b5544932692fee42ff26ebfe03c46e339c2d66e",
         "score_ratio": 1,
@@ -144,7 +144,7 @@
         "fact_pack_hash": "sha256:ce643fb45ddef2af2f9d06c2f01853fa0f8c9ce4c8599abcb1d912f7fef3d95d",
         "frozen_decision_hash": "sha256:c0bd26b8ee5b9fa7ac7f35038917b74d7c9173f127e62fab2b4b2bd17e441617",
         "native_state": "provisional_convex_opportunity",
-        "pack_hash": "sha256:84486c7f0877153383e577714243572fdbe88d8e022f68fbbe6a583082457802",
+        "pack_hash": "sha256:e04b2f08dbda8b786afa09544ca9c9b71e6a06d0488511799f0eb0ac43cef25b",
         "persona_id": "master_taleb",
         "policy_execution_hash": "sha256:685910a55811876df60c9890f083d93b1a95f11ac6799d06ecb97f11be106491",
         "score_ratio": 1,
@@ -158,7 +158,7 @@
         "fact_pack_hash": "sha256:36e4568b935d73e4bf27285ac7c2e26f4a3c4cb63ae84569accfef36af1633ee",
         "frozen_decision_hash": "sha256:11a2d8045d0f225a14985f2b48b55f203a84edaf773928d27535f36def9a4537",
         "native_state": "provisional_category_opportunity",
-        "pack_hash": "sha256:81bda7495aac1aeef35e99c74171ff77c1d0eef57225f89ac84a4985b3f885c1",
+        "pack_hash": "sha256:dc628a6b2d69be1cce786e38dd50342193216d10dcd7839cf8cf0dc5d8941820",
         "persona_id": "master_lynch",
         "policy_execution_hash": "sha256:25a035ba9912aa3e362927fa8eb14e376657e5e758a79d366ed4f4776cbe99a2",
         "score_ratio": 1,
@@ -172,7 +172,7 @@
         "fact_pack_hash": "sha256:91c21c62a5508f53773b13714956008301a96f35455421f0a029aca9f6d96846",
         "frozen_decision_hash": "sha256:7e2d1fda2363ad67e46aea321bfca662d5273788a0b0c05c4e31238c4d69c117",
         "native_state": "provisional_long_duration_quality",
-        "pack_hash": "sha256:b6ba9d263137936227500c17af79d04daf481f38b934022ae046b2344ee8dda2",
+        "pack_hash": "sha256:042a406c1b77414b7a77fc5d8a06b575efbeabf0a51f97f429f33884f190f822",
         "persona_id": "master_fisher",
         "policy_execution_hash": "sha256:9e620b5bb630cd42e5b482c6a01a18b210d615669a5100e6856cfcb3b095bc90",
         "score_ratio": 1,
@@ -186,7 +186,7 @@
         "fact_pack_hash": "sha256:3c00b9082899e1bc040b4e6e967e094a1c24d53f603b0996d508bda4d3278fb8",
         "frozen_decision_hash": "sha256:b0130956191645565781d0c23c7e42e772cff80efeb00d4b3ff169c24bb1e308",
         "native_state": "provisional_concentrated_growth_candidate",
-        "pack_hash": "sha256:adcc2e4633f7ab3e7a9e4c018cb97a41ba69cb30c2abf09bc06c2c78dc83d26c",
+        "pack_hash": "sha256:b7a062137a3537cea59864fe82165db67bc0d45a35d005577cdf7e04fa3f07e7",
         "persona_id": "master_jhunjhunwala",
         "policy_execution_hash": "sha256:b8f06017f86e3c59ed3fec0311772bcbf13230e08275c5686280f8a360380bc6",
         "score_ratio": 1,
@@ -200,7 +200,7 @@
         "fact_pack_hash": "sha256:6110836b4444c3976467c4ab707a8f6e38c6cc19a1c5dc41a53884a03066fb6d",
         "frozen_decision_hash": "sha256:1953083b763be405572008a7e8b639f745f60fadb8cdf29b68c7951e84858335",
         "native_state": "provisional_asymmetric_setup",
-        "pack_hash": "sha256:14c122f22b9f6d0c4ac09bebd9783ad663e6c7b966a65d47e3c1915ea81598cf",
+        "pack_hash": "sha256:aca213dba751fa47c4ecf572690981190ca78bd28a83d1fd8d5c0f3648253050",
         "persona_id": "master_druckenmiller",
         "policy_execution_hash": "sha256:05524438b4cb011ed69794efc854510b34cbe2df73a05340f5844a6d500dd774",
         "score_ratio": 1,
@@ -214,7 +214,7 @@
         "fact_pack_hash": "sha256:0adab0b69b1e8ba756deaeac3c457e044dd5a7195b1847893db0953772220e41",
         "frozen_decision_hash": "sha256:ddfd120be4926ef3b400a07f8af44c277d170f4c8f48364579b5b21ae48fa661",
         "native_state": "provisional_own_at_price",
-        "pack_hash": "sha256:9d8260b5256a67c74be7d2d25545c13efd2fcae8cd95cf247c8f48cc36fc8ee5",
+        "pack_hash": "sha256:acd69a6e493f5b1faa050d3266d12ed6d97a7e698daeba073727c085ff5c63dd",
         "persona_id": "master_buffett",
         "policy_execution_hash": "sha256:5e853405b4dd96e4760f842c0726e26bf7f87e9e3cc79a6b90e3ceb58a2211a5",
         "score_ratio": 1,
@@ -234,7 +234,7 @@
     "verifier_checks": []
   },
   "reviewer_kind": "ai",
-  "run_artifact_hash": "sha256:b1a29e61f61e7059e515b5933cc3c52b920cb68b4ee4912a30801f45d59952ca",
+  "run_artifact_hash": "sha256:d5e59ee75e51b7c4297b19bd790a9f32cc61eb34f43ca89d1c9788dc99e959bc",
   "run_id": "D13",
   "schema_version": 1,
   "simulation_scope": "13 physical provisional deterministic seats on frozen synthetic typed facts"