v1.9.1 — Evidence validation and terminal conclusions

AlphaCouncil 1.9.1

Fixes a false financial-evidence rejection and makes the terminal's research result clear.

• Normalize unavailable coverage gaps and their required open-question copies with identical reader escaping. Q&A, HTML characters and whitespace no longer create a false mismatch. Actual source/retrieval requirements and evidence/deadline gates are preserved.
• Count successful evidence/method/debate/report outputs independently of saved files. Timeouts and skipped stages stay visible; background-process completion is no longer displayed as research completion.
• Render the entire saved conclusion below progress and reveal it once when a live run ends. Scroll by arrows, mouse wheel or full-page Page Up/Down; complete reports remain accessible from the report list.
• Hide HTML disclosure tags and machine sentinels in terminal reading while retaining all recorded statements and source links.
• Add bounded primary/final failure diagnostics without raw model traces, with result labels in all 12 interface languages.

The existing incomplete run is preserved, not silently retried or rewritten as a completed investment analysis. PersonaPacks remain provisional version 0.9.4. No new paid stock-research acceptance run was performed.

Installation

Download and extract the complete archive for your system. Run ./alphacouncil on macOS/Linux or .\alphacouncil.cmd on Windows. Keep the launcher, app/ and runtime/ together. Node and the official Codex runtime are bundled. A service login/API key is required for new research; archives remain unsigned.

Verification

• 73/73 terminal tests and 1,685/1,685 full source checks passed locally; 14/14 company-dossier tests include the escaping regression and strict missing-retrieval rejection.
• Native CI passed tests, builds and relocated-archive startup on macOS arm64, Linux x64 and Windows x64.
• The downloaded CI macOS archive and locally installed client passed a 131×38 / 80×24 PTY walkthrough: inline conclusion paging, complete-report search/back and primary/final failure diagnostics. Original research artifact hashes remained unchanged; zero model-generation calls.
• Initial local validation caught a changelog format error (fixed) and an existing five-second timeout fixture (isolated and full reruns passed without weakening assertions).

All 10 core/Work matrix jobs passed on the reviewed source; the merged main tree is identical.

The tagged release workflow passed source, Work gateway, packaged-host and npm publication checks.

v1.9.0

AlphaCouncil Terminal 1.9.0

This release fixes terminal navigation and replaces the manual model-ID prompt with a browsable model directory.

• Fixed toolbar with Back, History and New research; inline fields with clickable Save/Cancel, cursor movement and Ctrl+U clearing.
• Delayed mouse sequences and mouse motion cannot become Back or unexpected page actions. Rapid clicks cannot pass through into a newly opened page. Small terminals cannot activate hidden Start buttons.
• Searchable model catalogs, retry and explicit manual entry. Codex reads its official app-server model directory; GLM documentation candidates are labeled separately from account-access checks.
• Named DeepSeek, Kimi, GLM, OpenRouter and OpenCode Zen/Go connections alongside Codex, Claude API, OpenAI API and custom endpoints. Known gateway models select the documented Chat Completions, Responses or Messages protocol.
• Searchable history and complete recorded reports; returning restores list position. Run views show the saved instrument, progress, actual worker failures and recent events.
• All interface additions cover 12 languages. The repository logo remains on the startup home only; no additional research-question field is required.

Native terminal downloads

Download and extract the complete archive for your system, then run ./alphacouncil on macOS/Linux or .\alphacouncil.cmd on Windows. Keep the launcher, app/ and runtime/ together. The archives bundle Node and the official Codex runtime; no separate installation of either is needed. SHA-256 and build-information files accompany each archive.

These are unsigned archives. A selected service account or API key is still required for real research. Kimi Code and GLM Coding Plan are information entries, not unrestricted research API subscriptions. OpenCode Go includes its coding-use notice; plan eligibility for this app's research use has not been confirmed. Codex authorization remains in the official runtime.

Verification

• 70 terminal regression tests passed locally; 1,684 full source checks passed on the local rerun. The complete Node/platform matrix passed on the tested commit: source CI.
• Native terminal tests, builds and relocated-archive startup passed on macOS arm64, Linux x64 and Windows x64: native CI.
• An independent macOS PTY walkthrough exercised the installed archive at 80×24 and 129×40, including actual Codex model-directory reading, model selection, inline editing/cancellation, synthetic history, report search and return navigation. It made no paid model-generation requests.
• npm latest is published as 1.9.0 through the tagged OIDC workflow: publication workflow.
• An initial local five-second fixture and one CI 50-millisecond assertion exposed existing test timing instability; neither core code nor timeout assertions were weakened. Final CI and publication state are recorded by the linked workflow runs.

Model listing and protocol fixtures are distinct from paid account access and live stock-research acceptance. No provider's paid account was used for live research acceptance. The existing evidence, method-selection receipts and rating contracts remain unchanged; all 26 PersonaPacks remain provisional version 0.9.4.

v1.8.1

This patch fixes terminal typing and simplifies stock research setup.

• Prevent full-screen redraws from scrolling at the bottom-right corner. The real cursor follows the input text, and long Unicode input keeps its editable tail visible.
• Enter a stock ticker without a separate research-question field.
• Show the repository's alpha/price-trend logo only on the startup language page, using ordinary terminal character cells.

Download and extract the complete archive for your machine. Keep its launcher, runtime/ and app/ together; no separate Node/npm/Codex installation is needed.

Platform	Archive	Start in the extracted folder
macOS Apple Silicon	alphacouncil-terminal-1.8.1-darwin-arm64.tar.gz	./alphacouncil
Linux x64	alphacouncil-terminal-1.8.1-linux-x64.tar.gz	./alphacouncil
Windows x64	alphacouncil-terminal-1.8.1-win32-x64.tar.gz	.\alphacouncil.cmd

Validation: 1,684 source tests and 40 terminal tests pass. The VT regression reproduces scrolling with the previous renderer and records zero scrolls with the fix. The macOS PTY flow verifies home-only branding, ticker-only setup, per-character cursor position, and resizing between 80×24 and 120×36. The compatibility matrix, Work tests, fuzzing and three-platform native builds also pass.

A separate duplicate Windows push run hit an unchanged method-worker state assertion; the complete same-tree PR Windows check passed. The exact cause of the duplicate-run failure remains undiagnosed.

Archives remain unsigned. The macOS Terminal GUI and native IME composition could not be controlled by the automation tool, so PTY/VT checks are not described as native GUI acceptance. No live stock research or model-account acceptance was rerun for this UI patch.

Terminal usage and authentication · Changes

v1.8.0

AlphaCouncil now has a standalone, mouse-enabled terminal client. Start with language, ticker and research question, then choose a model connection, full/quick settings and method seats. Completed evidence, method statements, individual debate rounds, questions/answers, reports and sources open as saved, scrollable content.

Download the archive for your machine, extract the complete archive and keep its launcher, runtime/ and app/ together:

Platform	Archive	Start inside the extracted folder
macOS Apple Silicon	alphacouncil-terminal-1.8.0-darwin-arm64.tar.gz	./alphacouncil
Linux x64	alphacouncil-terminal-1.8.0-linux-x64.tar.gz	./alphacouncil
Windows x64	alphacouncil-terminal-1.8.0-win32-x64.tar.gz	.\alphacouncil.cmd

Node and the official Codex runtime are included. No separate Node/npm/Codex installation is needed for these archives. Original npm and host/plugin installations retain the existing MCP command. Each archive has a SHA-256 file and build metadata.

• 12 interface and research languages: Chinese, English, Japanese, Korean, Spanish, French, German, Brazilian Portuguese, Italian, Russian, Vietnamese and Indonesian. Source quotations and machine IDs remain unchanged.
• Official Codex / ChatGPT sign-in; Anthropic, OpenAI and compatible API keys. API model/tool access is checked; Codex checks official login state. Missing API web search is disclosed before research starts. API usage has explicit request/output-token limits; these are not dollar limits.
• Independent runners continue when the interface closes. History and attach reopen the same run. Explicit stopping preserves the resulting state and artifacts.
• System credential storage where available, or session-only keys. The 26 provisional method packs and existing evidence, rating, language and one-use selection gates remain in force.

Validation: 1,684 local source tests, 36 terminal tests, the source/Work compatibility matrix, development fuzzing, and native archive tests on all three platforms. Native startup checks use an empty PATH and repeat after relocation to a directory with spaces. A real macOS PTY flow, actual Keychain save/read/delete, and one small real Codex JSON-marker call also passed.

The archives are unsigned. Full live stock research and real API-account acceptance for Anthropic/OpenAI/compatible providers were not performed for this terminal release; a passing interface/build test does not establish those results. See terminal setup and authentication and changes.

v1.7.0

v1.7.0 improves research setup, calculation accuracy and report reading while retaining all 26 provisional method seats and the existing PM rating thresholds.

• Start from a concise, named configuration with compatible remembered choices and a fresh confirmation. Keep the full method catalog available.
• Check the complete method input requirements. Retain frozen headless abstention explanations without redundant model calls.
• Correct the historical annual free-cash-flow yield calculation and replace the Simons deployable-signal label with an unusual-session-move observation.
• Lead reports with the conclusion, preserve expandable evidence and the full method ledger, and add run search and progress to the local viewer.
• Request up to ten years of daily price history, disclose actual coverage and observed drawdown, and cache successful requests for one hour without changing observation timestamps.
• Update existing transitive dependencies and expose the existing installation diagnostic as npm run doctor.

No additional data client, API key or runtime dependency is introduced by these changes. Moomoo and TradingView integrations are not included; they remain follow-up work requiring verified access, coverage and permitted use.

Validation: 1,655 source tests and 5 Work gateway tests passed locally. All 21 branch/PR CI checks and the release workflow passed. The downloaded npm package contains 1,202 files and matches the candidate package checksum. GUI checks used clearly labelled synthetic data. A real AAPL history request returned 2,514 sessions and a verified cache hit. These results do not establish investment performance, human method approval or live multi-host council completion.

See the detailed release notes and upgrade PR.

v1.6.0

What's Changed

• feat: calibrate one-year council ratings (v1.6.0) by @Zhao73 in #18

Full Changelog: v1.5.0...v1.6.0

v1.5.0

What's Changed

• Release 1.2.3 reliable full-council runtime by @Zhao73 in #11
• Release 1.4.0 — proportional gates, and a withheld vote that still speaks by @Zhao73 in #13
• fix: harden dossier grounding and verification by @Zhao73 in #12
• v1.5 candidate: evidence, runtime, and release-truth gates by @Zhao73 in #15
• release: finalize v1.5.0 by @Zhao73 in #17

Full Changelog: v1.2.2...v1.5.0

AlphaCouncil Agent v1.2.2

What's Changed

• fix: harden full-council source acquisition by @Zhao73 in #10

Full Changelog: v1.2.1...v1.2.2

AlphaCouncil Agent v1.2.1

Adaptive company source acquisition

• Discovers issuer-owned IR, product, blog and filing sources from current regulator filings instead of relying on an NVDA-only source list.
• Builds a bounded starter evidence pack for US and non-US companies and shares the complete pack with every analysis and method seat.
• Requires every operating-company coverage item to resolve as a cited reported actual, reproducible proxy, bounded scenario model, or audited unavailable result after the frozen source ladder is exhausted.
• Adds like-for-like local observation history for rolling metrics and 90-day revision support without fabricating unavailable history.
• Adds the get_company_sources MCP tool and expands the installed surface to 34 tools.

Verification

• Local portable suite: 1182/1182 passed.
• GitHub CI: both push and pull-request matrices passed on Ubuntu Node 18/20/22, macOS Node 20, and Windows Node 20; transport-runtime passed.
• Live source smoke covered NVDA, GOOGL and Toyota 7203.T.

AlphaCouncil Agent 1.2.0

What's Changed

• release: AlphaCouncil 1.2.0 verified full-council breadth by @Zhao73 in #8

Full Changelog: v1.1.9...v1.2.0