Research Wiki · branch wiki-setup

From Zero to a Queryable Wiki

≈190
interlinked pages built in one night — 0 lint errors, every commit.

22 → 23 June 2026 · 21 commits · 13:43 → 00:56

The arc, in one breath


Stood it up → grew an equations layer → made it queryable.


A pile of PDFs became a knowledge base you query instead of re-reading.

By the numbers

49
unique source papers
109
topic pages
9
result (theorem) pages
92
stubs filled tonight
35
agents in one workflow
21
commits
3
bib entries fixed (real DOIs)
0
lint errors — all night

The night, in six moves

  • 13:43Stand up the wiki — 61 candidate md → 49 unique papers; 11 dup-groups collapsed.
  • 13:43Equations layer · Cycle 1 — notation + terminology registries, result pages, syntheses.
  • 16:33Equations · Cycle 2 — 17 proofs consolidated & wired; 3 new result pages.
  • 17:20Make it queryable — regime/layer axes + 35-agent backfill + 4 Dataview boards.
  • 19:21Reorg — retire synthesis type, dashboards/ folder, sims-wiki window.
  • 00:54Provenance + bib + fill — word-ledger, real DOIs, all 92 stubs filled.

Move 1 — Stood up the wiki

61 candidate markdowns →
49 unique papers after content-dedup.

  • 11 cross-naming dup-groups collapsed
  • 1 non-paper quarantined
  • 49 source + 102 topic stubs, all review: draft
DEDUP REALITY

Snapshots held only ~600 MB of provable duplication, not the ~3 GB estimated — most of library/ is genuine content.

4 foundational PDFs exist only in snapshots → acquisition queue.

Move 2 — The equations layer

Canonical registries promoted from the private master:
notation.md (inertia M not H; quaternion ε/η) + terminology.md.

6 result pages · 15 topic Key-Equations · proofs in generated/math/.

RESOLVED, NOT A CONFLICT

hat vs breve — the 7-DOF "disagreement" was a communication gap.

hat = full 12×12 Γ⁻ᵀMΓ⁻¹
breve = reduced 9×9 after the CoM (mE) decouples

Both write-ups agree. Logged centrally.

Move 3 — Made it queryable

Two typed axes on every page (not tags): regime ∈ {free_flying, free_floating, fixed_base, general} · layer ∈ {nominal, risk_aware}.

4 Dataview boards

equations checklist · clusters · open-questions · verify Kanban

The payoff query

WHERE regime = "free_floating" → an 18-paper re-derivation watch-list for our actuated base.

Distribution — regime: general 98 · free_flying 42 · free_floating 26 · fixed_base 7.

Move 4 — Reorg, provenance & bib

Reorg — retired the synthesis type (suits humanities, not engineering); dashboards/ folder; no-underscore naming rule for curated files; a local sims-wiki window to check math vs simulation.

Provenance ledger — words_with_provenance.md:
oracle → test-oracle · fiber → geometric mechanics · ghost → self-motion manifold · rule → not-a-term.

Bib — chiaverini1994review + zar2005spearman added (real DOIs) → undefined-cite errors gone.

Three findings worth keeping

STABILITY

ω_b ripple was numerics

A forward-Euler period-2 instability — blows up when dt·μ_max > 2 (dt*≈0.0156 s). Exonerates the γ & singularity layer.
GEOMETRY

Elbow singularity

det J ∝ sin θ₃ · sin θ₅ — independent of link mass & length.
REGIME

zhang2020 is free-flying

Its force includes the base drive F_B as an active input — not free-floating. Load-bearing fix.

The method that paid off

MAP — REDUCE — WRITE

Agents classify. One script writes.

  • 35 agents read pages → emit JSON (read-only, in parallel)
  • 1 deterministic apply_classification.py is the sole writer — idempotent, key-preserving
  • Zero parallel-write races
  • A bad call is a one-line JSON diff, never a corrupted page
  • Everything stays review: draft for your verify pass

Reusable pattern — the seed for a future generated/insight-guru/SKILL.md.

Parked for you · human-in-the-loop

  • Verify the 92 filled topic pages — spot-check vs sources, then flip draft → verified.
  • The big reorg is uncommitted (228 deletions) — decide commit/cleanup as its own pass.
  • Notation escalation — generalized_inertia_matrix.md keeps H symbols vs the M lock.
  • `chiaverini1994review` has a bib entry but no source yet → conversion pile.
  • Phase-D proofs now unblocked — error floor + singularity handling are next.
end of report

The wiki now answers questions
the PDFs used to hold hostage.


Full raw notes archived in _done/Jun23_26.md · live tasks in wiki/ONGOING.md.

Self-contained Marp deck — no external theme needed. Render: marp --html generated/summaries/wiki_development.md -o wiki_development.html (or open in the Marp for VS Code preview.) Palette lifted from the workspace VS Code theme: green #42b883 / #65c89b, purple #945bc4, ink #15202b.