osint/torso: helix-anchor codec (x265-for-gaussians), measured — 2.8× + free crisp colour

[AdaWorldAPI]

What

The Helix-anchor codec — "x265 for gaussians" — as a measurement tool that proves the design (anchor + motion + residual + helix scan) on the real torso splat before it's wired into the render/animation path.

Stacked on #53 (claude/torso-anisotropic-map) → auto-retargets as the stack merges.

tools/spl_codec.py maps the full x265 pipeline onto signals already in SPL2 + torso.nodes.json (zero new data):

x265	gaussian codec
scan order	helix = 3D Morton of position = identity/GUID order (locality-preserving)
I-frame anchor	FMA node (SoA centroid + per-node colour) — random-access per structure
motion vector	gaussian offset from its node anchor
residual	helix-ordered zig-zag delta of (motion, normal)
—	colour = anchor-predicted → 0 per-gaussian bytes (a node palette)

Measured (231,515 gaussians)

SPL2 raw : 4.86 MB  (21.0 B/gaussian)
SPL3     : 1.73 MB  ( 7.47 B/gaussian)  →  2.8× smaller   (zlib entropy stand-in)
colour   : EXACT, 887 B for ALL colour  (178-entry node palette)
position : round-trip RMSE 0.00001  (16-bit quant, effectively lossless)
node_row : RLE 35 KB / 231K gaussians  (structures contiguous in helix order)
streams  : motion 1.02 MB · normal 671 KB · rows 35 KB · palette 887 B

What it proves

• "Crisp colours without overhead" is provably exact. Colour is fully predicted by the node anchor (per-structure flat), so it costs 0 per-gaussian bytes and cannot bleed across a structure boundary — crisp by construction. All colour for the whole torso = 887 bytes.
• The anchor = node mapping gives random access (decode one structure from its anchor via the O(1) switch) and stops residual drift (each structure resets, like an I-frame).
• Geometry is near-lossless at 16-bit; node_row RLE confirms structures are contiguous along the helix.
• The normal stream (671 KB) is the clear optimization target — high-frequency, so octahedral encoding + a real range/CABAC coder (zlib is only the stand-in) is the next win.

Not in this PR (next increments, scoped in the plan)

• octahedral normals + range coder (shrink the 671 KB normal stream);
• decode SPL3 at cockpit load + anisotropic / edge-aware reconstruction (node_row-bounded + normal-oriented → crisp colours in the render);
• animation: deform the node anchors → motion-skinned gaussians follow (Motion-Blender GS — the FMA partonomy is the rig).

This PR is the measurement that de-risks all three.

🤖 Generated with Claude Code

https://claude.ai/code/session_01TzqvDqbFRzyx17EkLKBoZF

---

Generated by Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 3af94e06-4a92-4a1b-a989-f16d9e4632a4

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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