feat(cockpit): /helix2 v4-bake route, vessel CAP per-bake, and the measured v4 addressing plan

[AdaWorldAPI]

Summary

Three related things, none of which touch the v3 bake or anything downstream of it:

#	change	files
1	/helix2 — a v4-bake sibling of /helix	cockpit/src/BodyHelix2.tsx, cockpit/src/main.tsx
2	vessel CAP is per-bake — v3 keeps 2.0, v4 takes 1.2	crates/osint-bake/tools/fill_body_soa.py
3	v4 plan §10 + §11 — the addressing and vessel measurements	claude-notes/plans/2026-09-06-fma-bake-v4-plan.md

No bake was produced or modified. No producer was run. There is still no v4 artifact; this ships the surface that will show one and the knob it will use.

---

1. /helix2

The fourth application of the repo's own #64 pattern, not a new idea — /helix is "Standalone so it can never break /body (#64)", GeoHelix is "a verbatim copy of BodyHelix, #64-style", GenomeHelix is "standalone so it can never break /cpic".

BodyHelix2.tsx is a verbatim fork whose only behavioural difference is the manifest key:

route	manifest key
/helix	helix_latest
/helix2	helix_v4_latest

Same loader, same BSO2 ver-6 decoder, same Signed360 shading — the renderer is held constant, so any visible difference between the routes is the bake, never the viewer. That is the whole point of having both.

Two safety boundaries, deliberately distinguished:

• /helix is protected by the fork itself.
• medcare-rs is protected by not re-baking over what it pins. It never loads q2 routes; it consumes the release artifact body.soa.gz, SHA256-pinned in its scripts/fetch-frontend-assets.sh where "a URL without a matching pin is a hard error". A new manifest key plus a new artifact leaves that untouched — replacing helix_latest's asset is what would break it, no matter how many routes exist.

fetchSoa's CANONICAL-ONLY rule is kept deliberately: falling back to the v3 artifact would make /helix2 a copy of /helix that looks like a successful v4 render.

Two defects fixed after review (codex P2 ×2, both confirmed against source, both resolved):

• Server LOD is now off on this route. /api/body/lod cascades over the v3 bake's compile-time bounds (include_bytes!(".../assets/body.blocks")) and returns actions indexed by v3 concept rows. BodyHelix already disables it for geo scenes on exactly that argument; a v4 bake changes concept rows by construction, so folding v3 actions into d.vrow would hide visible structures silently, while looking like a successful v4 render. The toggle renders disabled with the reason rather than vanishing.
• /helix2 added to the scene selector. The <select> falls back to /helix for any path absent from sceneOptions, so it claimed the v3 body was selected while rendering v4, with no way back.

2. Vessel CAP: 2.0 → 1.2, for v4 only

At CAP = 2.0 a ring may be twice the vessel's own calibre, which is the mechanism behind the over-thick arteries on the live /helix render. v4 uses 1.2 — bend allowance from +100 % to +20 % (a 0.004-calibre vessel's ceiling falls 0.0080 → 0.0048).

Implemented as CAP = float(os.environ.get("BODY_FILL_CAP", "2.0")). The default is the v3 value, so an unset variable reproduces the v3 artifact byte-identically. Nothing invokes the script from a driver, so the v4 procedure sets it explicitly (recorded in plan §11.1a):

BODY_FILL_CAP=1.2 python3 crates/osint-bake/tools/fill_body_soa.py <soa_dir>

Two limits stated rather than discovered later: it bounds the bend balloon only — min(RMAX, caliber * CAP) cannot make a vessel thinner than its measured calibre — and it narrows the damage without fixing the reconstructed centerline that causes it.

3. Plan §10 + §11 — measured, to be researched

Nothing is ruled; §6 stays unresearched. Headline figures, all from BodyParts3D 3.0:

• is_a max depth is exactly 24. Overflow at today's 6 levels: 95.93 %; at 12: 48.60 %; at 24: 0.00 %.
• is_a is a tree (104,696/104,698 single-parent), so parent = mask −1 is valid there; part_of is a DAG (536 of 1,522 multi-parent), so it is not. Sizing part_of at 12 makes the shortest-vs-deepest question moot.
• The "uncle" edge reference works only on a part_of-ordered address (0 unreachable, max climb 5, one-byte encodable); on an is_a-ordered one it fails (23 unreachable). The ordering relation is the real decision, not the slot count.
• The many2many hubs already exist as FMA region/Set of nodes, and Set of is a trigger, not a container: only 30 of 3,409 have any members, because the extent is the join { p ∈ parts(O) : p is_a T }.
• Connectivity is recoverable, correcting an earlier claim in the same section: 74 % of labels are <Predicate> of <Object> and the object resolves by name — Tributary of 90 %, Branch of 46 % — yielding ~490 vessel edges with no new source.

Open items recorded: the 203/281/301 skeleton-baseline ambiguity, 23 empty-FMAID rows that collide at address 0, the post-mint one-byte reachability falsifier, and that 24 has zero headroom until P-5 runs against 4.0.

Verification

npx tsc --noEmit -p . → exit 0; npm run build (tsc && vite build) → green.

Stated plainly: this fork has no frontend CI. ts-test-suite.yml and hub-client-e2e.yml carry if: github.repository == 'quarto-dev/q2' and can never run in AdaWorldAPI/q2, so their permanent skipped status is structural, not a failure — and the local build plus the reviews are the only gates on the .tsx. The Rust test-suite.yml does run here and is the gate on everything else. No browser render was performed, so /helix2 is typechecked and built, not visually confirmed.

Bake artifacts verified byte-identical (9/9 checksums) before every commit.

🤖 Generated with Claude Code

https://claude.ai/code/session_012wrzeZAdwGYTCKoxamwQht

Summary by CodeRabbit

• New Features

  • Added a new /helix2 viewer for exploring the latest V4 visualization.
  • Supports anatomy, map, terrain, and Garmin scenes with orbit controls, layer filtering, x-ray mode, concept search, and focused navigation.
  • Added terrain decoding, relief and sky rendering, color data, and improved error reporting.
  • Added V4 vessel rendering with configurable caliber limits for more controlled visual proportions.

• Documentation

  • Documented updated zipper addressing, vessel modeling, connectivity considerations, and open bake decisions.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Essentials

Run ID: 3eeee160-4392-4b21-90d5-ad222e2e4444

📥 Commits

Reviewing files that changed from the base of the PR and between 7c07315 and 04e899f.

📒 Files selected for processing (4)

• claude-notes/plans/2026-09-06-fma-bake-v4-plan.md
• cockpit/src/BodyHelix2.tsx
• cockpit/src/main.tsx
• crates/osint-bake/tools/fill_body_soa.py

---

📝 Walkthrough

Walkthrough

Adds measured V4 anatomy research, configurable vessel filling, and a standalone /helix2 viewer. The viewer decodes BSO2 artifacts, renders multiple scene types, loads canonical V4 manifests, and exposes search, layer, focus, and x-ray controls.

Changes

V4 anatomy research

Layer / File(s)	Summary
Measured zipper addressing claude-notes/plans/2026-09-06-fma-bake-v4-plan.md	Defines 24 is_a slots, additional part_of edge handling, taxonomy measurements, derived Set of joins, and open validation decisions.
Vessel way proposal claude-notes/plans/2026-09-06-fma-bake-v4-plan.md, crates/osint-bake/tools/fill_body_soa.py	Documents ordered vessel ways, BODY_FILL_CAP, recovered connectivity, type relationships, and vessel-model validation conditions.

BodyHelix2 V4 viewer

Layer / File(s)	Summary
BSO2 artifact decoding cockpit/src/BodyHelix2.tsx	Adds version 8 radix-grid decoding, legacy BSO2 decoding, normal reconstruction, colors, concept metadata, and outlier filtering.
Scene rendering and interaction cockpit/src/BodyHelix2.tsx	Adds Three.js scene setup, shaders, terrain handling, scene filtering, camera interaction, x-ray mode, focus movement, and disabled server LOD.
Artifact loading and route wiring cockpit/src/BodyHelix2.tsx, cockpit/src/main.tsx	Loads canonical and Garmin artifacts, connects viewer state and controls, and exposes /helix2 with no V3 fallback.

Estimated code review effort: 4 (Complex) | ~60 minutes

Sequence Diagram(s)

sequenceDiagram
  participant Viewer as BodyHelix2
  participant Manifest as CanonicalManifest
  participant Decoder as BSO2Decoder
  participant Renderer as ThreeScene
  Viewer->>Manifest: request helix_v4_latest
  Manifest-->>Viewer: return artifact payload
  Viewer->>Decoder: decode BSO2 data
  Decoder->>Renderer: provide geometry and metadata
  Viewer->>Renderer: apply layers, focus, and scene controls
  Renderer-->>Viewer: render scene state

Loading

Suggested reviewers: claude

Poem

A rabbit maps the bake,
Vessels follow ordered ways,
Helix lights the scene,
Zippers hold the anatomy,
Clear paths guide each review.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

---

_{Comment @coderabbitai help to get the list of available commands.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_2763bc16-d55b-461f-b72b-381a2685edb5)

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 57dfc0c2a5

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".