Milestones
List view
Fable-5 a11y audit (2026-07-05, static): partially keyboard-operable, marginally SR-usable. Explorer browse table rows are mouse-only (can't open a law by keyboard); the streaming chat reply has no live region (SR silence); 5 overlays declare aria-modal but never trap focus; non-existent Tailwind classes silently no-op breaking dark contrast + splash progress bar. See memory/fable_a11y_audit_2026-07-05.md
No due date•0/5 issues closedFable-5 adversarial review (2026-07-05, static): mostly sound on classic sinks, but the API has zero auth/CSRF while exposing side-effecting endpoints — the sharpest spawns an MCP server's command as a subprocess (CSRF->RCE via the multipart bundle endpoint; unauth RCE if networked). Audit-log hash chain is keyless + never verified. See memory/fable_redteam_audit_2026-07-05.md
No due date•0/5 issues closedFable-5 frontend perf audit (2026-07-05, static, post #712/#713): well-tuned at the frame level but unprotected at the input level — search boxes fire un-debounced/un-abortable backend searches per keystroke, and LawDetail mounts one react-markdown parse per clause with no windowing. See memory/fable_frontend_perf_audit_2026-07-05.md
No due date•0/5 issues closedFable-5 backend perf audit (2026-07-05, static): cold start is a 1-3min single-threaded double-pass re-parse of the 12k-law corpus; the cold graph pass pins every Law in an unbounded cache (~3x text) → GB resident; the chat agentic loop runs blocking work on the event loop. See memory/fable_backend_perf_audit_2026-07-05.md
No due date•0/5 issues closedFable-5 editor UX audit (2026-07-05): core typing loop is solid but the seams between citations/templates/AI/comments leak — normal actions silently lose or misplace work (autosave cancelled on nav, AI inserts against wrong selection, comments invisible after creation, 1-click template delete). See memory/fable_editor_ux_audit_2026-07-05.md
No due date•0/5 issues closedFable-5 micro-copy audit (2026-07-05): copy split between a careful i18n layer and ~15 components that bypass it (EN/ES mix in editor + wizard); no glossary (every surface invents its term); some labels misstate legal facts (RD-ley mislabeled, mock graph stats, v1.0-v1.3). See memory/fable_microcopy_audit_2026-07-05.md
No due date•0/5 issues closedFable-5 trust audit (2026-07-05): ZERO AI/legal disclaimers in the frontend; inferred graph edges, LLM answers, AI drafts, placeholder-embedder scores and derived 'vigente' status all rendered as fact; several strings fabricate facts. See memory/fable_trust_audit_2026-07-05.md
No due date•0/5 issues closedFable-5 IA audit (2026-07-04): /search route fully orphaned (no entry points), Settings/Editor/Communities unreachable on mobile, editor multi-doc unreachable, palette knows only 5 commands, help advertises 4 nonexistent shortcuts. See memory/fable_ia_audit_2026-07-04.md
No due date•0/5 issues closedFable-5 data-fidelity audit (2026-07-03): parsed law content vs BOE. 75% of laws swallow disposiciones into the last article; 452 expired/annulled shown as vigente; phantom articles, dropped ranges, preámbulo/anexos never served. See memory/fable_data_fidelity_audit_2026-07-03.md
No due date•0/5 issues closedFable-5 search-UX audit (2026-07-03): matcher can't match what a lawyer types (multi-word + unaccented → 0 results), results strand the user at the top of the law, browse shows 20 of 12k with no pagination. See memory/fable_search_ux_audit_2026-07-03.md
No due date•0/5 issues closedFable-5 resilience audit (2026-07-03): never wedge at splash, every failure explains itself, no silent data loss. See memory/fable_resilience_audit_2026-07-03.md
No due date•0/5 issues closedFixes from the Fable-5 legal-chat audit. Not trustworthy yet: citations evaporate on persistence, no system prompt, zero AI disclaimer, and the agentic loop breaks cloud RAG (works only on Ollama). Bones are good.
No due date•0/4 issues closedFixes from the Fable-5 app-wide design audit. Not slop at system level, but craft decays surface-level: dead motion layer (missing Tailwind plugins), Editor prose doesn't render, ~15 dead controls, 301 ad-hoc font sizes, 3 competing accent hues.
No due date•1/4 issues closedFixes from the Fable-5 law-reading audit. Not shippable as the core surface: the text renders as a blob the markdown destroys, apartados renumber (legal-accuracy bug), empty article headings, disposiciones misfiled, hierarchy never rendered, 147ch line.
No due date•0/4 issues closedFixes from the Fable-5 onboarding + model-setup audit. A non-technical lawyer cannot self-serve setup today: 2 competing wizards, the wizard's model choice is dead code, Ollama install dead-end, env-var cloud copy, no key validation.
No due date•0/4 issues closedFixes from the Fable-5 audit of the graph viewer (epic #782). Monochrome hairball, dead controls, fabricated copy, no hover, unshipped global view.
No due date•0/4 issues closedFixes from the Fable-5 audit of graph cross-reference inference (epic #775). The graph mis-targets many cited laws + inverts consolidation edges — not trustworthy for 'leyes relacionadas' until Sprint 1 lands.
No due date•0/4 issues closedMake laws findable: subjects taxonomy, faceted search, by-number/short-name, browse by community, related laws (epic #671).
No due date•8/8 issues closedWYSIWYG legal-document editor: TipTap base, typed citations to the corpus, template uploader/filler, AI-assisted drafting on the RAG stack, inline comments/notes. Umbrella issue #108.
No due date•5/8 issues closedExtensive automated tests targeting features that silently broke despite green CI (fixture-vs-reality, content assertions, frontend-backend integration, e2e, perf budgets).
No due date•0/1 issues closedComponent-by-component visual/UX overhaul toward a clean, cohesive, acrylic design language. Kill the AI-slop look; add product vision.
No due date•15/18 issues closedDistribuir LexFlow como aplicación de escritorio standalone: backend FastAPI y frontend React empaquetados en un único binario por sistema operativo, sin Docker ni Python requeridos para el usuario final. Hecho hasta ahora: Dockerfile y docker-compose con bind-mount read-only del submódulo legalize-es (PR #52, ajuste en PR #112) y pipeline de releases con release-please (PR #62, fix de credenciales en PR #112, secret PAT añadido tras la primera incidencia en main). Queda: PyInstaller spec para el backend (#38), montaje del build de React detrás de FastAPI con catch-all SPA (#66), instaladores nativos (.exe, .dmg, .AppImage), code signing en Windows con Azure Trusted Signing y en macOS con Apple Developer (#127), wrapper Tauri 2 con PyInstaller como sidecar (#125) y auto-update con el plugin de Tauri (#128). Fuera de alcance hasta después del primer release público: distribución por Microsoft Store o App Store.
No due date•6/13 issues closed