Skip to content

TO-DO.md

Latest commit

 

History

History
218 lines (194 loc) · 13.8 KB

TO-DO.md

File metadata and controls

218 lines (194 loc) · 13.8 KB

Repo TO-DO - active backlog

Design ethos: modular, intelligent, functional, logged, tested, and documented. Real methods only; never mocks or fakes. Every release ships with green tests and accurate docs.

This file tracks live work after the v3.4.0 release (latest published release: v3.4.0, tagged 2026-06-12). Historical release detail belongs in CHANGELOG.md; generated counts and project rosters belong in docs/_generated/COUNTS.md and docs/_generated/active_projects.md.

Live state snapshot

Refreshed on 2026-06-13 on branch codex/template-exemplar-forkability after the forkability and verifier-first roadmap pass. Re-run the commands in the Source column before copying any number into prose; live counts belong in docs/_generated/COUNTS.md, not hard-coded here.

Gate or surface Current value Source
Package version 3.4.0 pyproject.toml#[project].version
Latest published release v3.4.0 (tagged + GitHub release published 2026-06-12; CHANGELOG [3.4.0] body) gh release list, git tag
Public source scope infrastructure plus nine public exemplar src/ trees uv run python -m infrastructure.project.public_scope source-paths
Public exemplars template_active_inference, template_autoresearch_project, template_autoscientists, template_code_project, template_newspaper, template_prose_project, template_sia, template_template, template_textbook docs/_generated/active_projects.md
Canonical generated facts Importable infrastructure packages, infrastructure Python-module count, project-scope + publishing test collections, and per-exemplar coverage — all live-derived; do not hard-code here docs/_generated/COUNTS.md (regenerate with uv run python scripts/generate_counts.py --write)
Open GitHub PRs 0 open /opt/homebrew/bin/gh-axi pr list
Docs lint status links-only, consistency-only, and doc-pairs all clean (re-verified 2026-06-13) uv run python scripts/lint_docs.py --links-only --quiet --json, --consistency-only, --doc-pairs-only
Mermaid lint status clean with chunked batch rendering under the default total budget uv run python scripts/lint_docs.py --mermaid-only --quiet --json
Release verification baseline drift --strict, COUNTS/skills/exports gates, tracked-project/generated-artifact guards, docs-lint (links/consistency), reporting+evidence+repro suites (877), and the LLM suite (1244, live Ollama) all green at the v3.4.0 commit v3.4.0 release + local command history

Recently shipped

Keep this section short. Details live in release notes or archived audits.

  • AutoResearch reviewer-boundary closeout (2026-06-13, [Unreleased]): closed AR-REPORT-ERGONOMICS-1, AR-BENCHMARK-ERGONOMICS-1, and AR-SOURCE-LEDGER-2. The loop now emits output/data/autoresearch_evidence_overview.json, output/reports/autoresearch_evidence_overview.md, and output/data/benchmark_boundary.json; the source-ledger contract validates ledger citekeys against BibTeX and manuscript citations, rejects invalid tiers/future dates, and prints offline age/tier summaries. Verified with uv run pytest projects/templates/template_autoresearch_project/tests/ -q (224 passed).

  • Template forkability + verifier roadmap pass (2026-06-13, [Unreleased]): closed TODO-REBASE-2 (live PR count is 0; active backlog restored instead of _No open items_), REGRESSION-PIN-1 (the first real template_code_project regression pins now collect and pass with a mutation negative control), and AI-VIZ-SPLIT-1 (src/visualizations/figures.py split into focused semantic and simulation modules while preserving the figure facade). AI-SEMANTIC-FIXPOINT-1 is now closed: the shared fixed-point orchestrator drives manuscript variables, sheaf semantic outputs, and contract settlement; the selected fixed-point cluster passed 21 tests on 2026-06-13.

  • Generated-report web design (2026-06-12, [Unreleased]): modernized the base HTML report/dashboard template (infrastructure/reporting/html_templates.py) with CSS design tokens, dark mode (prefers-color-scheme), WCAG-AA status contrast, fluid type, and a mobile breakpoint — template contract + deterministic output preserved (7 template + 875 reporting tests green).

  • Scoped-improvement sweep (2026-06-12, [Unreleased]): closed all three freshly-scoped items. WEBDESIGN-EXTEND-1 — extracted html_templates.shared_css() as the single design-token source; the pipeline report, interactive dashboard, and web renderer all now anchor to --brand-1 + a prefers-color-scheme block (108 tests). LINKCHECK-PERF-1 — link_audit_core prunes excluded/gitignored dirs before descending + single-pass file reads (~15–28× faster; 303 link tests + a timed regression). TEXLIVE-2026-BEAMER-1 — latex_utils.compile_latex tolerates the benign \reserved@a kernel warning when a valid PDF results (the beamer test now passes on TeX Live 2026; genuine failures still raise).

  • Backlog closeout + comprehensive review (2026-06-12): closed REPRO-VERIFY-1 (the repro bundle now fails closed on declared-but-absent outputs — a declared output absent at build rebases onto the project tree and verify returns ok=False; pinned by tests/infra_tests/publishing/test_repro_bundle_absent_output.py) and EVIDENCE-CLAIM-1 (the claim-ledger candidate list gained an output/data/*claims*.json glob so the template_autoresearch_project exemplar's autoresearch_claims.json is ingested as claim nodes + supports edges; pinned by test_ingest_claims_autoresearch_ledger_yields_nodes_and_supports and test_build_evidence_graph_ingests_real_autoresearch_ledger). Shipped alongside a large multi-pass review (dead-code removal, run.config matrix runner, SOURCE_DATE_EPOCH determinism, the canonical_facts.md→COUNTS.md generator, exemplar-support-tier tagging, methods-plan gate, validation↔ autoresearch decoupling, and an Ollama test-model fix). Full detail in CHANGELOG.md.

  • Backlog sweep + RedTeam hardening (2026-06-08): closed 6 backlog items — PIPELINE-INCR-FLAG-1 (--incremental CLI flag on both entrypoints), DASHBOARD-CLI-1 (release-readiness dashboard discoverable + subprocess test), REPRO-MULTI-1 (--all-public multi-exemplar repro bundles), PROSE-GATE-WIRE-1 (opt-in report-only prose gate in Stage 6, disabled = byte-identical), SIA-HARNESS-2 (fixture/live separation tests + docs), and LOG-SEP-CENTRAL-1 (33 banner literals routed through width constants + lint). Verified TODO-REBASE-1, ARCH-CONFTEST-1, and AI-GATE-CACHE-1 already satisfied. A verifier-first RedTeam pass then fixed a real SIA fail-closed hole (max_generations<1 returned a vacuous empty run) and closed five green-wash test gaps. Commits 33e5ca71, c8381d9a. One out-of-scope finding logged as REPRO-VERIFY-1. RELEASE-TAG-1 deferred (needs a clean tree + sign-off).

  • v3.3.1 release train (2026-06-07): completed Pandoc DOCX output (embed figures + resolve cross-references in infrastructure/rendering/pipeline.py), ran a deep per-exemplar quality pass across the eight tracked public templates, completed and cross-linked sidecar publication metadata for all nine public exemplars, and reconciled the generated project-scope collection count to 216. Full detail in CHANGELOG.md.

  • v3.3.0 release train (2026-06-07): closed the three Medium backlog items (READFILE-SAFE-1, CI-MATRIX-DYNAMIC-1, LOG-CLEAN-1) and all five Major items (EVIDENCE-GRAPH-1, REPRO-BUNDLE-1, DASHBOARD-1, PLUGIN-STAGES-1, INCREMENTAL-PIPELINE-1) — the two pipeline-core features are opt-in/default-off. The shipped artifacts were re-verified on disk and under test on 2026-06-08 (169 dedicated tests pass; no stubs). Full detail in CHANGELOG.md. The genuine residual next-increments of these features are now tracked as new backlog items below (EVIDENCE-CLAIM-1, PIPELINE-INCR-FLAG-1, DASHBOARD-CLI-1, REPRO-MULTI-1, PROSE-GATE-WIRE-1).

  • Reference-existence + AI-writing infrastructure (2026-06-06): new infrastructure/reference/verification/ deterministic anti-hallucination gate (Crossref→OpenAlex/arXiv resolver, SQLite cache, offline-first) and infrastructure/validation/content/ai_writing.py AI-writing fingerprint detector (validation.cli prose-quality); wired into the docs/prompts workflows. Clean-room distillation of academic-research-skills ideas (CC-BY-NC-4.0); no code vendored.

  • Infra test parallelization (2026-06-06): pytest-xdist + -n auto on the CI test-infra job cut each leg from ~892s to ~585s; suite verified parallel-safe; default -v dropped from addopts.

  • Docs accuracy sweep (2026-06-06): audited docs/ + every infrastructure/*/{SKILL,README,AGENTS}.md; corrected examples that cited methods/params/CLI flags/test paths that did not exist (every fixed command re-verified to resolve).

  • v3.2.0 release train (2026-06-04): DOCX/EPUB renderers + format toggles (FMT-BUNDLE-1), Active Inference validation spine v2 (AI-SPINE-V2), infrastructure coverage gaps rebaselined (COVERAGE-REBASE-1), GitHub supply-chain hardening (GH-PIN-1, GH-ACTIONLINT-1, GH-AUTOMERGE-1), and the XML parser policy (DEP-DEFUSEDXML-1). See CHANGELOG.md.

  • v3.1.0 release train: SIA joined the public exemplar scope; Active Inference gained the first validation-spine tracks; docs signposting moved to projects/templates/...; public coverage orchestration was hardened. See CHANGELOG.md.

  • Thermo-nuclear infra/docs audit (2026-06-08): Waves A–E closed — doc contract, READFILE-SAFE-1 CLI, AGENTS v3.3 completeness, evidence-graph SUPPORTS/registry bridge, autoresearch validation split. Approve — see docs/audit/archived/thermo-nuclear-code-quality-2026-06-08.md.

  • Thermo-nuclear remediation waves: 2026-05-29 and 2026-05-30 blockers and branch deltas are closed. See docs/audit/archived/thermo-nuclear-code-quality-2026-05-29.md and docs/audit/archived/thermo-nuclear-code-quality-2026-05-30.md.

Active backlog

AI-GATE-PERF-2 — Reduce active-inference gate runtime after correctness

  • Problem: template_active_inference gate tests still have one heavy explicit roadmap/sheaf write path; clean-state fixed-point and gate setup now avoid redundant regeneration, but the full project suite still needs a durations-driven pass.
  • Why it matters: slow gates discourage local verification and make future semantic changes harder to review.
  • Smallest next step: complete the remaining project-local MEDIUM-TEST-PERF-1 split with cheap source-only negative controls plus one end-to-end refresh characterization.
  • Acceptance check: focused mutation tests still catch the known failure classes, --durations=20 shows fewer redundant artifact refreshes, and the manuscript gate still passes.
  • Out of scope: weakening coverage, skipping pymdp-backed evidence, or dropping the end-to-end refresh characterization.

Known divergences from CHANGELOG.md

As of 2026-06-12 there are no known divergences: pyproject.toml, CHANGELOG.md, and the published tag all agree at 3.4.0 (the prior "v3.3.1 bumped but not released" gap was resolved by cutting v3.4.0, which folds the dated [3.3.1] entry and all post-3.3.1 work into one published release). The docs-lint (links/consistency/doc-pairs), drift, and canonical-facts gates are clean.

If a new drift appears between CHANGELOG.md, TO-DO.md, generated facts, or .github/workflows/ci.yml, fix forward and record the current source of truth here instead of rewriting shipped changelog entries.

Conventions

  • Backlog IDs are stable. Use them in commit messages when closing related work; do not silently reuse retired IDs for new work.
  • ID scheme: <AREA>-<SLUG>-<N>, where AREA names the surface — GH (GitHub workflows/CI), ARCH (architecture/test rules), LOG (logging), AI (Active Inference exemplar), SIA (SIA exemplar), DEP (dependencies), FMT (rendering formats), PIPELINE/DASHBOARD/REPRO/PROSE/EVIDENCE (shipped-capability follow-ups), COVERAGE/READFILE/CI-MATRIX (named gates/patterns), and TODO (backlog hygiene itself). N is a monotonic counter within that area+slug.
  • Every active item must include a problem, why it matters, smallest next step, acceptance check, and out-of-scope boundary.
  • Completion requires evidence from a command, file diff, or generated artifact; do not check off items from intention. Before retiring an item, confirm its artifact exists on disk (and under test) this session — never from a commit message alone.
  • Re-baseline measured facts instead of copying old numbers from this file.
  • Keep private or rotating project names out of public docs; link to docs/_generated/active_projects.md for public scope.
  • Preserve the no-mocks testing policy, project coverage floors, and generated artifact guard when closing backlog items.

See also