GuanDan Lab v0.3.0-beta.23

• Follow the actual Chinese or English coaching text in ElevenLabs and browser speech.
• Give Wang, Gu and Lin distinct server-side voices with a backward-compatible fallback.
• Default interactive synthesis to eleven_flash_v2_5 with explicit language codes.
• Extend deployment checks and documentation for per-character voice configuration.

GuanDan Lab v0.3.0-beta.22

• Rename the direct live-lobby entry to “在线匹配” / “Match online”.
• Keep online matchmaking directly available without a foundation-course gate.
• Add mobile E2E coverage for the public label.

GuanDan Lab v0.3.0-beta.9

Privacy-safe AI match reviews

• Optional compatible-model post-match strategy classification from public events and non-hidden local metrics
• Strict controlled style/advice codes: provider free text is never displayed
• Local scores, evidence, ethics reminder, and coaching copy remain authoritative
• Explicit loading, fallback, privacy, local/cloud save, and retry states
• Browser and provider timeouts plus SSRF, origin, rate, size, and secret-redaction guards
• Scroll-safe result dialog on small and landscape screens

Verification

119 unit/contract tests, coverage gates, production build, 25 browser flows, CodeQL, dependency review, GitGuardian, npm audit, and Compose validation all pass.

Full changelog: https://github.com/Mereithhh/guandan-lab/blob/main/CHANGELOG.md

GuanDan Lab v0.3.0-beta.8 · Mini Endgame Bridge

5-minute bridge from rules to a full deal\n\n- Five deterministic 5–7 card Chen-table endgames\n- Immediate rules and table-manner feedback in Chinese and English\n- Public-information-only fixtures; every candidate action is checked by the production rules engine\n- Two partnership-yield cases without pretending to know hidden cards\n- Versioned on-device progress, first-choice scoring, retry and restart\n\nVerification: 108 unit/contract tests, coverage thresholds, production build, 24 passing desktop/mobile E2E checks (6 intentional project skips), dependency audit with 0 vulnerabilities, CodeQL/dependency review/GitGuardian green, and two independent no-P0/P1 reviews.\n\nTry it: https://guandan-bootcamp.miromind-0889.chatgpt.site

GuanDan Lab v0.3.0-beta.7

What changed

• Mobile same-rank stacks now retain a 44px direct tap step while keeping their visual grouping and individual selection.
• Added a browser regression that measures the tap step and clicks the first two overlapped cards.
• Updated official checkout/setup-node actions across CI and security workflows.
• Kept unsupported TypeScript, ESLint and Node type major migrations manual while preserving Dependabot security updates.

Verification

Release commit: 4e76d3b

• 103 unit tests passed
• 22 browser tests passed; 6 project-specific skips were intentional
• Mobile 390×844 audit verified no page overflow, complete action buttons, full hand reachability and direct selection
• Typecheck, lint, production build, npm production audit, dependency review, CodeQL and GitGuardian passed
• Independent user and adversarial reviews found no P0/P1; the only wording P2 was corrected before release

The public demo intentionally remains in local Agent / device voice mode. Paid providers, cloud persistence, OAuth and live matching are opt-in self-hosted Beta capabilities.

GuanDan Lab v0.3.0-beta.6

What changed

• AI opponents now default to a readable 2.2-second pace, with a persisted 3.5-second explanation mode.
• The in-table history shows every action in the current deal; finished deals remain available in full replay.
• Same-rank cards use clearer stacks and count badges on desktop and mobile, while individual cards remain selectable.
• Added a complete English README and deployment guide, CodeQL/dependency review, a scoped security review record, and an original international launch visual.

Verification

Release commit: 3a7ed09

• 103 unit tests passed
• 21 browser tests passed; 5 project-specific skips were intentional
• 1200×900 overflow regression verified first and last cards are directly clickable
• Typecheck, lint, production build, dependency review, CodeQL and GitGuardian passed
• Independent user and adversarial reviews found no remaining P0/P1

The public demo intentionally uses the deterministic local Agent and device voice. Paid models, ElevenLabs, persistence, Google OAuth and live matching remain opt-in self-hosted Beta capabilities.

GuanDan Lab v0.3.0-beta.5

What changed

• Added a typed Chinese/English onboarding experience across the home page, mastery course, full rulebook, footer, support flow and OAuth return notices.
• Added a keyboard-accessible, locally persisted language switch without duplicating game logic.
• Added 32 competition-rule conformance checks: 30 table-driven fixtures, one response-right scenario and one provenance check.
• Added route-level compatible Agent and ElevenLabs contracts for auth headers, timeouts, redirects, malformed upstreams, audio validation and secret redaction.
• Hardened card-ID validation and browser-speech fallback behavior.

Verification

At commit f3e028dc963230b54864ff349b447009b8f84851, 103 tests passed with 93.47% line and 84.19% branch coverage. TypeScript, ESLint, the production build, desktop/mobile E2E, Docker build and the Compose capability smoke test passed in PR #13. Provider-route contracts cover authentication headers, unsafe base URLs, redirects, timeouts, malformed upstreams, audio validation and secret redaction.

GuanDan Lab v0.3.0-beta.4

What changed

• AI turns now start in a slower 1.4-second comfort mode, with a 2.2-second explanation mode available.
• Play history stays visible and expandable during a deal.
• Your hand groups and overlaps cards of the same rank for faster scanning.
• Training entrances now lead to distinct course, memory, and full-deal experiences.
• The footer truthfully discloses the active Agent and voice provider modes.
• Docker Compose now preserves .env.local capability settings and is smoke-tested in CI.
• README and launch materials now use real product screenshots and a verified walkthrough GIF.

Verification

62 unit tests passed with 93.44% line coverage; production build and desktop/mobile E2E passed; CI and security checks passed; independent adversarial review found no remaining P0/P1.

GuanDan Lab v0.3.0-beta.3

新增双副牌“已见牌减法”训练：

• 训练王、A、2 与完整 2～A 级牌范围
• 错题自动用新牌序重练同类目标
• 计数与九宫格分别记录准确率，并安全校验/限制本地历史
• 保留位置九宫格，覆盖桌面与移动端端到端测试
• 独立对抗审查无剩余 P0/P1

完整变更见 PR #10。

v0.3.0-beta.21

Full self-hosted public demo

• Makes https://guandan.mereith.com the sole public Demo URL.
• Opens the four-player live lobby directly to anonymous guests.
• Preserves the always-available complete local AI match.
• Improves mobile online-hand scrolling and keeps touch actions reachable above bottom navigation.
• Confirms compatible-model Agent and ElevenLabs modes with local fallbacks.

Verification

• Production four-guest matchmaking, private hand projections, legal action and cancellation passed
• Production compatible Agent returned a locally validated move
• Production ElevenLabs returned valid MP3 audio
• 176 unit tests passed
• 36 end-to-end tests passed, 10 skipped
• lint, typecheck, build, CodeQL, dependency review and secret scanning passed