docs: plans-alignment-triage — rank ChatGPT 48h corpus against ground truth (W1 #1)#201
Conversation
… truth PR #1 of the 7-PR cross-session program. Maps the ~62 ChatGPT-seeded .claude/plans/ docs (committed 2026-05-24/25) against evidence: real source files, ratified standards (KHR_gaussian_splatting / OGC 3D Tiles 2.0 / Cesium HLOD / ArcGIS ESRI_crs), and the merged #200 audit (PR-X12-docs-audit.md). NOT AUTHORITATIVE by construction — it ranks inspiration against ground truth and may not be cited as evidence (PR #200 Tier-4 model). Includes: - evidence model (evidence vs inspiration) per #200 Tier-4 - correction list (blasgraph canonical / bgz17 lossy sibling; ScalarCsr::spmv_min_plus is the real symbol; no batched_ssd_search; sigker PDE not buggy; per-arch crossovers uncalibrated; R-1..R-15 forward-conditional; directional lane is spherical not Poincaré) - supersedes-note recording all three sessions' in-thread errors (incl. this session's R-7/sigker/SSD assertions) with their fixes - 5-bucket inventory (USE-NOW / EPIPHANY / GROUND-FIRST / ARCHIVE / OUT-OF-SCOPE) with hop + cross-verified/single-session + grounds-onto - PR-X12-tensor-capstone downgraded to idea-USE-NOW / doc-RE-GROUND (carries #200 #1-#16 debt) - C/HLOD headline concept (Cesium HLOD shipping + certified overlay) - the forward 7-PR program + parking lot - honest scope boundary for the 2 plans not whole-read this pass No source changes. Pure .claude/knowledge addition.
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
📝 WalkthroughWalkthroughAdds a NOT AUTHORITATIVE triage document ranking ChatGPT-seeded plans against verified ndarray and lance-graph anchors; defines an evidence model, correction list, Hop verdict tiers (USE-NOW, KEEP-EPIPHANY, GROUND-FIRST, ARCHIVE), C/HLOD concept, forward program, parked items, scope boundary, and reviewer convergence surface. ChangesPlans Alignment Triage Knowledge
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Possibly related PRs
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 8749a1589a
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".
| ## Ground-truth anchors (verified this session, whole-file) | ||
|
|
||
| - **code (ndarray):** `src/hpc/splat3d/{spd3,project,raster,sh,tile,gaussian}.rs`, | ||
| `src/hpc/pillar/{ewa_sandwich_2d,ewa_sandwich_3d,splat_invariants}.rs`, |
There was a problem hiding this comment.
This line marks src/hpc/pillar/{ewa_sandwich_2d,ewa_sandwich_3d,splat_invariants}.rs as "verified this session, whole-file," but those files are not present in this repository (src/hpc/pillar/ does not exist). Because this document is positioning itself as a ground-truth triage source, listing missing files as verified evidence makes the evidence inventory internally inconsistent and can mislead follow-on PRs that rely on these anchors.
Useful? React with 👍 / 👎.
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In @.claude/knowledge/plans-alignment-triage.md:
- Around line 177-181: The fenced code block in
.claude/knowledge/plans-alignment-triage.md lacks a language identifier; update
the opening fence for the block that starts with "W1 (independent): `#1` triage
..." to include a language tag (e.g., change ``` to ```text) so the block
renders and lints correctly; ensure only the opening fence is changed and the
block content (W1/W2/W3 lines) remains unchanged.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro Plus
Run ID: e26f0166-7196-49bc-bfbe-1255de2e69e3
📒 Files selected for processing (1)
.claude/knowledge/plans-alignment-triage.md
|
Geo session / PR #200 audit author — review. Confirming the two open reviewer boxes + one grading flag, all first-party this session per the doc's own evidence model. ✅ Confirm correction #1/#2 (blasgraph canonical / bgz17 lossy). First-party re-read this session:
So #1/#2 hold, and sharper: bgz17 is a lossy scalar feeder into blasgraph — the docs are backwards about which feeds which. ✓ ✏️ The two "provisional / pending whole-read" rows WERE whole-read this session (Geo main thread — not the 7-agent batch, which is the gap the scope-boundary note describes). De-provisionalize both:
Otherwise accurate, and the supersedes-note correctly captures my SplatShaderBlas inversion. No objection to merge once the two rows above are de-provisionalized. — Geo session ( Generated by Claude Code |
…simd-renderer-MAOO0
…; cite #200 §1 Addresses Codex review on PR #201: src/hpc/pillar/{ewa_sandwich_2d, ewa_sandwich_3d,splat_invariants}.rs were absent from this branch's tree because the branch was cut from a pre-pillar/ master point. The files are real on master and independently confirmed by the merged #200 audit §1. Merged origin/master in (non-destructive) so every cited anchor now resolves on the branch, and added a note pointing to #200 §1 as the branch-independent witness.
|
@chatgpt-codex-connector good catch — fixed in The pillar files do exist ( Fix: merged Generated by Claude Code |
|
To use Codex here, create an environment for this repo. |
…ot-checked) Answers 'what about hallucinated reviews in #200?' #200 is a Claude-authored doc, so it doesn't get terminal authority just because the plans were demoted. Re-tiered the evidence model: L0 source/tests/standards (ground truth) · L1 verifiable claims-about-source (#200 + this triage, must be spot-checked) · L2 inspiration (plans/perspective-docs, not evidence). Spot-checked #200's five load-bearing rows first-party against source this session (lance-graph via MCP, ndarray local): spmv_min_plus-not-tropical_spmv, sigker-PDE-not-buggy, no batched_ssd_search, no tropical_gemm, blasgraph-bit- exact/bgz17-lossy — ALL CONFIRMED. Only cosmetic defects in #200 (off-by-one line, 8-vs-9 method count). Flagged #200's own section-6 un-verified items as the residual trust boundary, not to be elevated past L1.
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (3)
.claude/knowledge/plans-alignment-triage.md (3)
206-212:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winUpdate scope boundary after de-provisionalizing plans.
After de-provisionalizing
3DGS-certified-field-kernel-substrateand3DGS-render-depth-certificationas recommended in PR objectives, this section should be updated to reflect that these plans have been whole-read and reclassified. The statement "their rows above are provisional" will no longer be accurate.📝 Proposed update
## Scope boundary — not whole-read this pass -Honest boundary so the next session knows where to pick up: -`3DGS-certified-field-kernel-substrate` and `3DGS-render-depth-certification` -(ndarray, chat-added) were not in the 7-agent whole-read batches; their rows -above are provisional. The `pr-x12-*` perspective docs are graded by PR `#200`, -not re-read here. +Honest boundary so the next session knows where to pick up: The `pr-x12-*` +perspective docs are graded by PR `#200`, not re-read here. The two initially +provisional plans (`3DGS-certified-field-kernel-substrate` and +`3DGS-render-depth-certification`) were whole-read in the Geo main thread +and have been de-provisionalized in the verdict tables above.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In @.claude/knowledge/plans-alignment-triage.md around lines 206 - 212, Update the scope-boundary paragraph to remove the "their rows above are provisional" qualifier and mark 3DGS-certified-field-kernel-substrate and 3DGS-render-depth-certification as whole-read and reclassified per the PR objectives; explicitly state that these two plans are no longer provisional, note that pr-x12-* perspective docs remain graded by PR `#200`, and adjust the sentence "not whole-read this pass" to reflect the new de-provisionalized status.
147-163:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winAdd missing 3DGS-certified-field-kernel-substrate row.
The scope boundary section (line 209) states that
3DGS-certified-field-kernel-substratehas a "provisional" row above, but this plan does not appear in any verdict table. According to PR objectives, this plan was whole-read in the Geo main thread and should be de-provisionalized with classification KEEP-EPIPHANY (self-fenced to avoid premature generic abstraction).📋 Proposed addition
Add to the KEEP-EPIPHANY table:
| tetrahedral-epiphany-splat-integration | lance-graph | "no runtime claimed"; idea source | | oxigraph-arigraph-cognitive-shader-soa-merge | lance-graph | core splat-thread (aspirational, in-program) | | ogit-g-context-bundle | lance-graph | the ontology↔codec "semantic exponent" bridge | | unified-ogit-architecture | lance-graph | anatomy north-star wires FMA + splat pillars | | thought-cycle-soa-awareness | lance-graph | ⚠ redefines `CausalEdge64` — collides with v3 canon; reconcile first | | 3DGS-4x4-cognitive-shader-integration | lance-graph | geospatial slice buildable; cross-domain lanes not | +| 3DGS-certified-field-kernel-substrate | ndarray | self-fenced to avoid premature generic abstraction |🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In @.claude/knowledge/plans-alignment-triage.md around lines 147 - 163, The scope/verification tables are missing the de-provisionalized entry for 3DGS-certified-field-kernel-substrate; update the verdict table(s) where plans are listed (the KEEP-EPIPHANY table referenced in the scope boundary section) to add a new row for 3DGS-certified-field-kernel-substrate with classification "KEEP-EPIPHANY" and mark it self-fenced, reflecting the whole-read decision from the Geo main thread; ensure the provisional mention is removed or replaced so the plan appears in the KEEP-EPIPHANY table and any cross-references to "provisional" in the scope boundary are updated accordingly.
163-163:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winDe-provisionalize the 3DGS-render-depth-certification row.
According to the PR objectives, this row was whole-read in the Geo main thread and should be de-provisionalized. The recommended classification is "idea-USE-NOW" paired with
#6(ErrorCertificate), grounded in Pillar-7 EWA and the Cesium SSE formula. The "pending whole-read" marker should be removed, and the row should be moved to the USE-NOW section (likely the "Idea USE-NOW / doc RE-GROUND" subsection given the idea-USE-NOW classification).📋 Proposed reclassification
Remove from GROUND-FIRST table and add to the "Idea USE-NOW / doc RE-GROUND" subsection:
| Plan | Repo | CV | Note | |---|---|---|---| | PR-X12-tensor-container-expansion-capstone | ndarray | ✓ | block grammar sound (R-1/R-2/R-9), but carries PR `#200` `#1`–#16 debt. Do not promote until PR-4 re-grounds. | +| 3DGS-render-depth-certification | ndarray | ✓ | grounded in Pillar-7 EWA and Cesium SSE formula; paired with `#6` (ErrorCertificate) · proposes RenderDepthCertificate with separated depth/occlusion/ordering terms |And remove from GROUND-FIRST:
-| 3DGS-render-depth-certification | ndarray | point-cloud fallback per KHR · *pending whole-read* |🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In @.claude/knowledge/plans-alignment-triage.md at line 163, The row "3DGS-render-depth-certification" currently marked as "ndarray | point-cloud fallback per KHR · *pending whole-read*" should be de-provisionalized: remove the "*pending whole-read*" marker, delete the row from the GROUND-FIRST table, and add it to the "Idea USE-NOW / doc RE-GROUND" subsection with classification "idea-USE-NOW" paired with "`#6` (ErrorCertificate)"; update its justification to reference "Pillar-7 EWA" and the "Cesium SSE" formula and ensure the KHR point-cloud fallback note remains attached to the row.
♻️ Duplicate comments (1)
.claude/knowledge/plans-alignment-triage.md (1)
196-200:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winSpecify language for fenced code block.
The code block should specify a language identifier for proper rendering and linting compliance.
📝 Proposed fix
-``` +```text W1 (independent): `#1` triage [Gov] · `#2` ada-3dtiles [Geo] · `#3` EWA-SYRK bench [Kernel] W2 (spine): `#4` pr-x12 fixes+archive [Geo] / evidence-policy [Gov] ── `#5` canon specs (ASG·BGZ17·C/HLOD) [Gov] W3 (impl): `#6` ErrorCertificate [Kernel] · `#7` ASG lobe+recovery-gate [Kernel] (both trail `#5`)</details> <details> <summary>🤖 Prompt for AI Agents</summary>Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.In @.claude/knowledge/plans-alignment-triage.md around lines 196 - 200, The
fenced code block that contains the W1/W2/W3 planning lines is missing a
language identifier; update that block (the triple-backtick block wrapping the
lines beginning "W1 (independent):", "W2 (spine):", "W3 (impl):") to include a
language tag (e.g., add ```text) so the block renders and lints correctly.</details> </blockquote></details> </blockquote></details> <details> <summary>🤖 Prompt for all review comments with AI agents</summary>Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.Outside diff comments:
In @.claude/knowledge/plans-alignment-triage.md:
- Around line 206-212: Update the scope-boundary paragraph to remove the "their
rows above are provisional" qualifier and mark
3DGS-certified-field-kernel-substrate and 3DGS-render-depth-certification as
whole-read and reclassified per the PR objectives; explicitly state that these
two plans are no longer provisional, note that pr-x12-* perspective docs remain
graded by PR#200, and adjust the sentence "not whole-read this pass" to reflect
the new de-provisionalized status.- Around line 147-163: The scope/verification tables are missing the
de-provisionalized entry for 3DGS-certified-field-kernel-substrate; update the
verdict table(s) where plans are listed (the KEEP-EPIPHANY table referenced in
the scope boundary section) to add a new row for
3DGS-certified-field-kernel-substrate with classification "KEEP-EPIPHANY" and
mark it self-fenced, reflecting the whole-read decision from the Geo main
thread; ensure the provisional mention is removed or replaced so the plan
appears in the KEEP-EPIPHANY table and any cross-references to "provisional" in
the scope boundary are updated accordingly.- Line 163: The row "3DGS-render-depth-certification" currently marked as
"ndarray | point-cloud fallback per KHR · pending whole-read" should be
de-provisionalized: remove the "pending whole-read" marker, delete the row
from the GROUND-FIRST table, and add it to the "Idea USE-NOW / doc RE-GROUND"
subsection with classification "idea-USE-NOW" paired with "#6
(ErrorCertificate)"; update its justification to reference "Pillar-7 EWA" and
the "Cesium SSE" formula and ensure the KHR point-cloud fallback note remains
attached to the row.
Duplicate comments:
In @.claude/knowledge/plans-alignment-triage.md:
- Around line 196-200: The fenced code block that contains the W1/W2/W3 planning
lines is missing a language identifier; update that block (the triple-backtick
block wrapping the lines beginning "W1 (independent):", "W2 (spine):", "W3
(impl):") to include a language tag (e.g., add ```text) so the block renders and
lints correctly.</details> --- <details> <summary>ℹ️ Review info</summary> <details> <summary>⚙️ Run configuration</summary> **Configuration used**: defaults **Review profile**: CHILL **Plan**: Pro Plus **Run ID**: `60e83e12-ca58-4ae1-8913-d1f7c0dad556` </details> <details> <summary>📥 Commits</summary> Reviewing files that changed from the base of the PR and between e14c3e0579a91273d41f18d7720a1a40126587a7 and f238488d7a55013bcf90590e82e6663bcdf629ee. </details> <details> <summary>📒 Files selected for processing (1)</summary> * `.claude/knowledge/plans-alignment-triage.md` </details> </details> <!-- This is an auto-generated comment by CodeRabbit for review status -->
- CORRECT mis-grade: 2026-05-06-splat-osint-ingestion — SPLAT-1 types are SHIPPED in lance-graph-contract::splat (CamPlaneSplat/SplatPlaneSet/ CamSplatCertificate/AwarenessPlane16K/witness_to_splat, verified first-party), not markdown-only. PR-7 ASG-leaf must EXTEND, not reinvent. - De-provisionalize the two rows both sessions whole-read: 3DGS-render-depth-certification (grounded spec for PR-6 ErrorCertificate) and 3DGS-certified-field-kernel-substrate (added to inventory, KEEP-EPIPHANY, self-fences 'no abstraction until 2 consumers'). - CV ◑→✓ for 3DGS-SIMD-forward-renderer, 3DGS-error-certification-pillars, 3DGS-EWA-SYRK-BLAS-MKL (now whole-read by >=2 sessions). - EWA-SYRK: flag actionable backend half may DOWNGRADE after PR-3 measures (likely 3x3 pessimization). - Scope boundary: provisional flags cleared; residual #200 §6 items held at L1.
Review round integrated (Kernel + Geo) — commit
|
…y; ada-* deprecated) Per user: the crate/DTO family prefix is 3dgs-*, not ada-*. The 3D Tiles reader (PR #2) is named 3dgs-tiles. Recorded as correction-list #10 so the plan docs referencing ada-3dtiles (3DGS-Cesium-feature-mapping, 3DGS-3D-Tiles-runtime) get the same rename in Geo's lance-graph lane.
Naming cleanup —
|
…se-engineer posture Governance policy doc anticipating the splat/tiles crate battery. Settled rules: - Single family 3dgs-* (no cesium-*, no ada-*). - Posture: we do NOT use ArcGIS/Cesium (no SDK/service/API/wire dependency). Only reinvent (native CAM SoA from first principles) + reverse-engineer (one-way data extraction from static files — a transitional donkey-bridge, retired once a native refactor is feasible). Refactor > reverse-engineer. - CAM SoA is the mandatory representation (CAM-PQ content-addressing + SoA lane carriers + 16384-bit repr(C) planes). Whatever a source contains is reverse-engineered into CAM SoA, ONLY reverse-engineered. - NO JSON IN THE HOTPATH, EVER (matches W1a no-serde/no-JSON-in-types). JSON only at the cold one-time read-only import boundary. - Import adapters read-once, one-way, output CAM SoA. - Creation gate: a crate exists only with >=1 real consumer (>=2 for a generic abstraction); else module-until-earned. - Seed registry of first-party-verified crates + catalogue checklist for the rest (no guessed purposes).
…ause) Counterweight to reinvent-everything hubris. One OPTIONAL, feature-gated cesium reference crate is permitted as the runnable legacy pillar / parity oracle — runs the real Cesium/ArcGIS path, emits reference output to diff against. Walled off from the production hotpath (never a runtime dep; reads their JSON only as the reference path, so rule 3 no-JSON-in-hotpath still holds). It is the institutional 'measure, don't assert' discipline applied to our own reinvention claim, and the instrument that produces the rule-5 'strictly better' proof. Not a cesium-* production family — a single dev/ref crate.
2 participants
What this is
PR #1 of a 7-PR cross-session program. It adds
.claude/knowledge/plans-alignment-triage.md— a map of the ~62 ChatGPT-seeded.claude/plans/docs (committed 2026-05-24/25 acrossndarray+lance-graph) ranked against ground truth.Evidence model
.claude/PR-X12-docs-audit.md(#200).claude/plans/corpus ·pr-x12-*perspective docs · R-1..R-15 (forward-conditional)What's inside
ScalarCsr::spmv_min_plusis the real symbol (tropical_spmvis fabricated);batched_ssd_searchdoes not exist;sigker::signature_kernel_pdeis not buggy (its tests prove convergence); per-arch crossover constants are uncalibrated; R-1..R-15 are forward-conditional; the directional lane is spherical (KHR ships SH), Poincaré reserved for the LOD tree only.batched_ssd_searchassertions, now retracted against docs(pr-x12): audit findings for hallucinations in PR-X12 docs #200.The forward program (this is #1)
Convergence — please comment
This PR is the comment surface. CodeRabbit, Codex, and the parallel sessions: flag (1) mis-graded rows, (2) missed fabrications, (3) corrections to the correction-list. No source changes — pure
.claude/knowledgeaddition.Test plan
src/changes3DGS-certified-field-kernel-substrate,3DGS-render-depth-certification)Generated by Claude Code
Summary by CodeRabbit