docs(research): adopt Layer-2 thread-state contract + vault research wiring — Phase 30 prep

- .planning/research/THREAD.md: hard core, H-RS-substrate (pivoting), kill criteria
  incl. the 2026-09-30 pivot gate, claim table, EXP-RS-11 active
- .planning/research/CONVENTIONS.md: append-only convention lock (C-1..C-7 seeded
  from Phase 29 practice + EXP-RS-11 pre-registration)
- CLAUDE.md: research-thread wiring section (thread state, experiment registry,
  baseline rule, /consult //commission //scribe-debrief, cartographer channel,
  notebook workspace)
- .gitignore: .cartographer-notes.md (cartographer run 2026-07-02)

Pivot to Path C approved by human 2026-07-02; see .cartographer-notes.md and the
vault hypothesis-ledger.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: JasperSeehofer

.gitignore

@@ -5,3 +5,4 @@ data-*/
 
 # Gardener sweep notes (local audit trail; not tracked)
 .gardener-notes.md
+.cartographer-notes.md

.planning/research/CONVENTIONS.md

@@ -0,0 +1,16 @@
+# Conventions ledger — Dynamical LBD thread
+
+*Append-only convention lock (Layer-2 pack anatomy, vault:
+`wiki/analyses/research-routine-packs-spec.md`). Every experiment/artifact declares which
+conventions it uses; changing a CRITICAL row requires re-verifying all downstream consumers.
+Never edit rows — append a superseding row and mark the old one.*
+
+| # | convention | value | set by | criticality |
+|---|---|---|---|---|
+| C-1 | Pre-2015 cutoff | `--published-before 2014-12-31` (inclusive, lexicographic on ISO dates) | Phase 29 | CRITICAL |
+| C-2 | Corpus | `data-kuramoto/` — 421 papers, 35 Louvain communities (bidirectional S2 crawl from 10 Feynman-pair seeds) | Phase 29 | CRITICAL |
+| C-3 | Benchmark gate | shared 10-pair Feynman-reduction set; evaluable pair = maps to non-Other communities; gate `n_eval ≥ 3` AND `BENCH_P10 > 0.15` (EXP-RS-06 baseline; ≥ 0.30 = success) | Phase 29 roadmap gate | CRITICAL |
+| C-4 | Node vectors | c-TF-IDF, top-50 terms per node (`--tfidf-top-n 50`), from `export-louvain-graph` | Phase 29 | high |
+| C-5 | Export schema | `{louvain_params, corpus_fingerprint, nodes, communities, edges}`; "Other" community excluded; uniform edge weight 1.0 | Phase 28/29 | high |
+| C-6 | EXP-RS-11 τ sweep | cosine threshold τ ∈ {0.2, 0.3, 0.4, 0.5}; headline prediction at τ=0.3; connectivity precheck `n_cc/N ≤ 0.05` required before `compute_K_stable` | EXP-RS-11 pre-registration | CRITICAL |
+| C-7 | Notebook workspace | LBD notebooks + exports live in `../professional-vault/prototypes/` (`kuramoto_lbd_v0*.ipynb`, `data/`); Rust-side prototypes in `prototypes/` here | Phase 29 practice | high |

.planning/research/THREAD.md

@@ -0,0 +1,51 @@
+# Thread state — Dynamical LBD (Gen-4)
+
+*Layer-2 thread-state contract (vault: `wiki/analyses/research-routine-packs-spec.md`). Read by
+the vault's `/cartographer` bridge in place of the retired `.gpd/` state. Keep current: same-day
+update after every experiment run (vault: `wiki/meta/research-operating-manual.md`, "per
+numerical experiment / run"). Vault mirror of the hypothesis rows: `wiki/meta/hypothesis-ledger.md`.*
+
+## Hard core
+
+The knowledge graph is a *dynamical system*; bridges *emerge* from generation, not static
+scoring (Gen-4 LBD — vault: `wiki/concepts/dynamical-lbd.md`, the three acceptance criteria).
+
+## Live hypotheses (mirror of vault hypothesis-ledger)
+
+| id | statement | discriminating experiment | status |
+|---|---|---|---|
+| H-RS-substrate | Cellular **sheaves** over the Louvain community graph detect multi-causal bridges better than RAFs or Kuramoto | 4-tier benchmark incl. multi-causal joint-removal ablation on the shared 10-pair Feynman set | live (**pivoting** — substrate rebuild via EXP-RS-11) |
+
+## Kill criteria
+
+- **Method-level:** sheaf near-section frustration does not beat the brute-force baseline
+  (vault: `wiki/concepts/brute-force-lbd-baseline.md`) on held-out bridges.
+- **Pivot gate (time-bound, set 2026-07-02, human-approved):** if the Path C TF-IDF
+  semantic-edge substrate (EXP-RS-11) yields **<3 evaluable Feynman pairs** or
+  **`BENCH_P10 ≤ 0.15`** on the shared 10-pair set by **2026-09-30**, kill the
+  dynamical-substrate line and revert to the brute-force baseline.
+
+## Current claims
+
+| claim | status | evidence |
+|---|---|---|
+| Dynamical-LBD on the pre-2015 cond-mat *citation* graph is empirically infeasible (~41 components / 153 nodes → `K_stable` bisection diverges) | verified (Phase 29 FAIL, 2026-05-05) | `.planning/phases/29-kuramoto-corpus-build/29-VERIFICATION.md` |
+| Sheaf near-section frustration ranks bridges on this corpus | HOLD — untestable on VOID corpus (T2 precision@10 = 0.000) | `prototypes/SHEAF_V01_RESULTS.md` |
+| TF-IDF cosine edges (τ=0.3) make the same corpus connected enough for spectral/dynamical LBD (`n_cc/N ≤ 0.05`, largest CC ≥ 80%) | predicted — untested (EXP-RS-11, pre-registered) | vault: `wiki/meta/agentic-experiments-research.md` § EXP-RS-11 |
+
+## Active experiment
+
+**EXP-RS-11** (→ Phase 30) — TF-IDF cosine semantic-edge substrate. Full pre-registered
+hypothesis / predictions / setup / metrics: vault `wiki/meta/agentic-experiments-research.md`
+§ EXP-RS-11. Decision provenance: `.cartographer-notes.md` (pivot approved 2026-07-02, human;
+recorded in vault hypothesis-ledger, research-portfolio, decision-timeline).
+
+## Claim history
+
+`CLAIMS.jsonl` (commission-compatible) is created by the first `/commission --research` run on
+this thread; until then the table above is the claim record. (Exact format is an open Layer-2
+spec question — session feedback welcome.)
+
+## Last verification
+
+2026-05-05 — Phase 29 verification (FAIL verdict, honest negative; gate not reached).

CLAUDE.md

@@ -168,6 +168,30 @@ cargo run --bin resyn -- export-louvain-graph \
 # 4. Run kuramoto_lbd_v03.ipynb in professional-vault/prototypes/
 ```
 
+## Research thread wiring (professional-vault)
+
+Besides the product track, this repo hosts the **Dynamical LBD (Gen-4)** research thread,
+managed by the vault at `~/Repositories/professional-vault`. For research phases (exploratory
+LBD experiments, EXP-RS-*), follow the vault's research structure on top of the GSD scaffolding:
+
+- **Thread state** (read by the vault's `/cartographer`): `.planning/research/THREAD.md` (hard
+  core, live hypotheses, kill criteria, claims) + `.planning/research/CONVENTIONS.md`
+  (append-only convention lock — honor it; append, never edit). Update THREAD.md same-day after
+  every experiment run.
+- **Experiment registry**: EXP-RS-* rows in the vault's `wiki/meta/agentic-experiments-research.md`.
+  Predictions are pre-registered there **before** running — never adjusted post-hoc.
+- **Baseline rule**: no dynamical method is interesting until it beats the brute-force baseline
+  (vault: `wiki/concepts/brute-force-lbd-baseline.md`) on the shared 10-pair benchmark.
+- **Skills**: `/consult research "<question>"` — read-only vault lookups; `/commission --research`
+  — independent falsification of a claim/verdict before accepting it; `/scribe-debrief` at
+  session end (auto-files lessons to the vault).
+- **Cartographer channel**: `.cartographer-notes.md` (gitignored) — proposals from vault research
+  reviews land there; go/kill/pivot decisions are the human's, recorded in the vault
+  hypothesis-ledger.
+- **Notebook workspace**: Python LBD notebooks + exports live in
+  `../professional-vault/prototypes/` (see the v03 workflow above); Rust-side prototypes in
+  `prototypes/` here.
+
 ## Important Notes
 
 - arXiv rate limiting: `ArxivHTMLDownloader` enforces configurable delays (default 3s) between requests using `tokio::time::sleep`. Violating this causes request blocks.