Separation placement review — 2026-10-06

Arc 7x5n (GOAL 1 vuip). Measured on main 8b6039d by two read-only review agents and the coordinating session (https://claude.ai/code/session_012qoycyCSGidZqW245vXhze). This page records what must move, or be cut, before litlfred/cat-harness(-tools) and litlfred/folio-assistant-{core,sci} are seeded from their staging directories.

The owner’s placement rule, verbatim (2026-10-06):

folio-asst-core focused on content authroiung, review pubilcation. cat-harness on SDLC, tooling, decision making, general methdologies, basic common infra/state mgmt

The owner on pace, verbatim: “not a race to separate. do it properly.”

Where each layer stands

bun run cat seed:ready --layer <L> (no --rehearse), 06:30Z:

layer	verdict	rule	layer PRs	moving PRs	upward declared paths	standalone
cat-harness	not yet	Rule_NextLayer	12	5 (#2229, #2197, #2192, #2189, #2080)	0 of 134	not measured (needs --rehearse); 375 failing on 10-06
cat-harness-tools	not yet	Rule_NextLayer	5	0	0	not measured
folio-assistant-core	not yet	Rule_NextLayer	6	1 (#2080)	0	not measured
folio-assistant-sci	not yet	Rule_LayerMoves	2	2 (#2192, #2080)	0 of 1	not measured

All four target repositories are empty: git ls-remote returns no refs, 2026-10-06.

Graph dependencies

check	result
check:import-direction	0 wrong-direction imports in every instance. 76 import(expr) calls in 40 files cannot be determined
check:partition	1,259 modules, 0 unassigned, 0 wrong-direction edges in cat-harness code
check:tools-closure	bootstrap-tools imports only itself and its declared packages
cat-harness-tools imports	all stay within {bootstrap-tools, cat-harness, cat-harness-tools}; needs matches
check:reference-direction:strict	4,552 wrong-direction mentions (prose, names, paths) in 524 files; 162 files name several instances above them; 8 stale PENDING entries

Imports are clean. Upward paths and mentions are not. A mention that is a link or a path breaks when the layer stands alone.

Hard-coded upward paths that no import gate sees (prerequisite)

• cat-harness/scripts/check-term-mapping.ts:99 walks folio-assistant-core/glossary.
• cat-harness/scripts/external-schemas.ts:336 reads folio-assistant-core/schemas.
• cat-harness/skills/kg/graph-management/kg-detangle.ts:92 names folio-assistant-core/schemas.
• cat-harness/tools/index.ts:2512: Tool glossary-build runs folio-assistant-core/scripts/build-glossary.ts. Line 1544 names folio-assistant-core/ns.jsonld.
• cat-harness/schemas/translation-tools.ts:221 points at a core BPMN.
• cat-harness/scripts/gates.ts:199,237,248 runs fhir-harness/scripts/*. task-io.ts and regen-after-merge.ts name smart-* tasks.
• cat-harness-tools/scripts/merge-train.ts:250-256 runs smart-base scripts.
• qa-refresh.ts:162 hardcodes ["cat-harness","smart-base","who-iris"]. check-secret-leaks.ts:203 ROOTS includes who-iris.
• cat-harness/schemas/jsonld.ts defines core’s vocabulary under folio-assistant-core: (38 lines), including the sci terms leanRef and sorryFree.

Code still in cat-harness (D2: the KG repository holds no code)

• 1,475 tracked .ts files: 641 tests and 834 non-test. The 10-01 figure was 1,522.
• Non-test by directory: content 317 (189 content/docs block manifests, about 128 content/pipeline), scripts 277, schemas 172, src 41, skills 11, translations 7, tools 6, types 2, adapters 1.
• 220 KG-node .ts → code edges (y9r6). 189 of them are content/docs manifests importing schemas/builders.ts (172) and schemas/webpage.ts (17).
• Moving cat-harness’s code wholesale into cat-harness-tools (S5 txue) creates no upward import edges. Upward risk comes only from moving modules above tools.

Move candidates

Skills, processes, docs: new, not covered by any ruling

item	now → proposed	confidence
skills/sdlc/sdlc-core/diff.md	cat-harness → core (content review)	high
docs/guides/reseeding-the-lean-cache.md, docs/sage-mcp.md	cat-harness → sci	high
docs/concepts/architecture/folio-board-requirements.md	cat-harness → core (boards already ruled to core)	high
docs/guides/new-content-type.md	cat-harness → core	med
docs/qou-migration-checklist.md	cat-harness → sci or the qou folio	med
docs/process/publication-workflow.md (+5 locales)	split: content processes to core; the index stays	med
skills/ui/ui-core/html-rendering-qc.md, docs-generation.md	split: generic half stays; content/paper half to core/sci	med
folio-assistant-core/skills/library/cataloguing/glossary-build.md	split: LaTeX half to sci; SKOS half and name stay	med
skills/sdlc/sdlc-core/staging-review.md	keep or split, never whole: 2 cat-harness BPMN name it	low
docs/proposals/translation-block-audit.md	cat-harness → core	low
19 generated docs/processes/* pages for core/sci processes	retarget gen-processes-viz.ts; do not move by hand	high

Code and schemas

module group	now → proposed	status	inbound / upward if moved
Lean, TeX, proof, witness code and Tool nodes	CH/CHT → sci	ruled (Q4, partition)	78 / 5
maths checkers (q-usage, triviality, vacuity, conjectural-*)	CH → sci	high	(counted above)
_folio-chapter-profiles.qou.ts	CH → the qou folio, as data	high	2 / 1
authoring/publication Tool nodes (glossary-build, ingest-*, upload-url, l1-complete-check, folio-block-*, …)	CH/tools → core/tools	high	n/a
folio MCP surfaces (folio-init, readme-*, preview, translation, …)	CHT → core	ruled in part (MCP placement; mer2)	2 / 1
review store and block CRUD (feedback, qa-agent-write, blocks/*, manifest-entries)	CH/CHT → core	ruled (C1)	12 / 2
folio/glossary graph typologies, glossary-ledger	CH → core	ruled (q2wn)	30 / 22
content-object model (types, builders, webpage, block-kinds, …)	stays (owner, 2026-10-06)	ruled: keep	310 / 219
bibliography (citations, bib-qa, validate-references*, …)	CH → core	med-high	27 / 6
library / L1 intake	CH → core	partition-triaged	48 / 2
content validation and publication pipeline	CH → core	med-high	36 / 2
authoring/review scripts (narratives, summaries, todos, gen-docs-pages, …)	CH → core	medium	45 / 5
downward: KG materialisation and subscription (kg-materialize, materialization, kg-subscribe, kg-instantiate)	core → CHT	medium	15 / 0
schemas/jsonld.ts vocabulary	split: core vocab to core, Lean terms to sci	medium	not measured

Ruled moves still pending (not relitigated here)

• Q1: five watchers and narrative-asserts-code to sci.
• Q2: content-graph, uses-editorial-review and corpus-grep split.
• Q3: Milnor to sci; markdown-render-check split.
• PR2: getting-started, repo-conversion, board-windows, board-diagram-interchange and evidence-appraisal to core, plus the editorial set: 14 skills to core folio-editorial, 3 to core one-voice, 6 to sci paper-editorial. The per-item list was never written into the tree. A proposed assignment is in the review report.
• PR9: the content-type doc pages.

Open owner decisions

1. Ruling 1A against the 2026-10-06 rule: RULED 2026-10-06. The owner said “do content split propoerly across repos”, then “that needs to be done before F” (GOAL 5 resumes after the split). Read as a SPLIT: the method write-ups (the “how”) stay in cat-harness as general methodology, and the operational content-authoring, review and publication skills, processes and code move to core, each split cleanly with no upward reference left behind. The question as it was put: 1A (2026-09-30) and the completed PR3 keep a generic content group in cat-harness: content-lifecycle (8 skills), the voice-review skills (3), technical-documentation, review-task, review-narrative and voice-review.bpmn. Today’s rule says content authoring and review belong in core. Which wins decides the largest single skills move.
2. The content-object model: RULED 2026-10-06, KEEP in cat-harness. The owner chose “Keep in cat-harness” from three options (split with seams staying; keep; move all to core). It is common infrastructure every layer builds on. Nothing moves, and cat-harness’s content/docs manifests keep importing builders/webpage. Under D2 those builders still go to cat-harness-tools with the rest of the code (S5), which is downward and allowed. The question as it was put: The content-object model (schemas/types, builders, webpage, block-kinds, …): core, or kept below as the seam every layer builds on? Moving it puts 219 importers below it, 189 of them cat-harness’s own content/docs manifests.

Prerequisites for a clean split, in order

1. In-flight PRs land; their sessions are wrapping up.
2. The owner’s decisions above.
3. Hard-coded upward paths are cut: replaced by declaration lookups or by registration on load.
4. S5 txue: cat-harness code moves wholesale to cat-harness-tools, including the y9r6 cut of the 220 KG-node → code edges.
5. Ruled and approved moves land, one concern group per PR, with a rewording pass wherever a staying file names a moved one.
6. Upward mentions that are links or paths are fixed (check:reference-direction).
7. The standalone failures reach 0 (ho66; Session A).
8. bun run cat seed:ready --layer <L> --rehearse answers settled; then the seed PRs go to each empty repository, on a branch.