Lab digest — 2026-W24 (W22–W24 catch-up)
Window: 2026-05-25 (Mon) → 2026-06-12, UTC — three weeks, catch-up digest (W22 and W23 were never written; this entry honestly covers the gap rather than backfilling fake weeklies)
🏆 Headline achievements
- OpenAlice Embed moved from skeleton to near-complete product.
- Tenant dashboard: embed.blal.pro.
- One-line widget: embed.js.
- agent-lite runtime landed.
- RAG knowledge base shipped with a 6h re-index loop.
- EU-first TTS voice chain is in place, with Mistral-native default.
- Live operator takeover shipped through inbox, SSE, and operator API.
- Proactive triggers v1 landed: time, scroll, exit, and URL rules.
- Trigger arming happens pre-consent.
- HMAC signature guard is enforced.
- Pricing v2 shipped: Starter €49, Pro €249, Business €799.
- Pricing now includes compare table and FAQ.
- Brand wave collapsed the mono-brand sakura palette across products.
- Agent persona renamed Aria → Lily.
- Landing visuals moved to living surfaces.
- WebGL ASCII-emergence portrait landed.
- Mandelbrot ASCII hero field landed as the lab logo motif.
- Leads inbox gained the paper-toss animation.
- Multi-viewport audit + fix wave is in flight.
- Atlas leapt from catalogue to code map.
- U7 indexed ~50K symbols across ~60 repos.
- U7.1 added 8,317 Rust impl-block methods.
- U7.2 shipped a real AST call graph.
- Call graph: 289K edges across 64 repos.
- Internal call resolution is ~49% fully resolved.
- Previous state was 15–19% candidate-grade.
- atlas-first agent hook landed.
- Work-claims collision primitive landed.
- Org task board shipped: board_tasks, /board page, atlas.board.* MCP tools.
- Board was seeded with live priorities.
- /graph fixed monorepo workspace crates drawn as external satellites.
- alice-core side refactors continued.
- lab, ranker, and inspector unified into one cost-ladder research rig.
- Net removal: −9.3K LOC.
- 6 biggest god-files split into modules.
- Byte-identical golden gates protected the splits.
- +34 tests closed test-debt.
- Conservative tool-output compactor landed as opt-in.
- Ops shipped enough rails to keep the pace manageable.
- bench.blal.pro UI-bench platform is live.
- Codex OAuth re-auth runbook exists.
- Agent memory reorganised into per-agent folders.
📊 By the numbers
- Total repos: 68.
- Total features: 2,694.
- Last scan: 2026-06-12T03:14:36.563338Z.
- Probed repos: 27.
- Reachable repos: 18.
- Unreachable repos: 9.
- 24h uptime: 66.66666666666667%.
- Feature status split:
- unknown: 1,674.
- beta: 311.
- ga: 226.
- alpha: 218.
- stable: 198.
- experimental: 66.
- HELD (branch u-slice4-pii-toggle, pending merge to mvp): 1.
- Selected high-volume category counters:
- openalice/critic: 176.
- agi/safety: 132.
- openalice/agi_core: 121.
- agi/skills: 110.
- openalice/agi_safety: 99.
- openalice/cot: 55.
- embed-app/page: 48.
- atlas/api: 42.
- atlas/web: 39.
- openalice/sessions: 38.
- app/account: 37.
- openalice/lane: 33.
- ui/primitive: 30.
- atlas/ingestor: 28.
- app/lib: 23.
- embed-app/route: 22.
- openalice/prompt: 22.
- openalice/stages: 22.
- openalice/tools: 22.
- landing/embed: 21.
- embed-app/lib: 21.
🔭 Next
- Finish the in-flight multi-viewport audit and fix wave.
- Keep OpenAlice Embed focused on the shipped surface: dashboard, embed.js, agent-lite runtime, RAG, voice, takeover, proactive triggers, HMAC guard, and pricing v2.
- Use Atlas first for repo shape, runtime deps, features, health, graph, and board tasks.
- Use the call graph and work-claims primitive before touching sibling repos.
- Keep alice-core refactors gated by byte-identical goldens and targeted tests.
- Use bench.blal.pro for UI checks.
- Keep Codex OAuth re-auth and per-agent memory layout as operational basics.