docs(plans)+test: rebaseline #497 OCR plans to #498 + gating probes (5-specialist framing)

[AdaWorldAPI]

Follow-up to the merged #497 (OCR-transcode plans) + #498 (helix Signed360 + GUID keystone). Five specialists (cascade-architect / family-codec-smith / palette-engineer / dto-soa-savant / truth-architect) framed the merged #497 plans against the post-#498 substrate; this PR applies the corrections + specs the gating probes. Consolidated framing: EPIPHANIES.md E-OCR-PLAN-DRIFT-1.

Two showstoppers the framing caught

1. "reversible without a hash" is false in code — no residue→rank inverse exists; deepnsm/vocabulary.rs is a stored string-table keyed by rank, every decode takes a known rank as input. Reframed: node = identity → content-store lookup; codebook = repair signal (I-VSA-IDENTITIES), not reversible text.
2. "Morton-tile stacked-pyramid perturbation-shader cascade" does not exist (0 hits; Morton is explicitly rejected for Hilbert). Purged from 3 deliverables → real primitives (framebuffer::build_mipmap_pyramid / splat3d/depth_cascade / CAKES ladder).

Plan corrections (all 7 #497 docs)

• HelixResidue 48 B → 6 B everywhere (it's a stored Signed360 place index, not a 48-byte field); budgets/carve rebaselined (Full 154→112, [32,186)→[32,144)); headers post-#496 → post-#498.
• §0 anti-invention tripwires: dropped ValueSchema::Ocr (ride Full/Compressed or mint a class); de-overloaded Meta (confidence→Energy, provenance→Plasticity, OOV→content-store); flagged TurbovecResidue as the edge codec (rank-only) — glyph→word uses DeepNSM CamCodes.
• D-OCR-50 marked PARTIALLY SHIPPED (LayoutBlock::to_node_row landed in feat(contract): GUID decode→read-mode keystone + helix Signed360 right-size + OCR→NodeRow transcode #498) — re-cast as "extend", not "build".
• HHTL = a coherent layout-address trie, but classid-marked as layout-addressed (forgoing the similarity-basin reading).
• master critical path 42 → 53 → 42 → {50,51} → 53 — resolves the open CodeRabbit Major on docs(plan): Tesseract → tesseract-rs 1:1 transcode (LSTM hosted via embedanything) — v2 #497.
• bit-repro caveat: DeepNSM repair is f32 → carve out of frozen-mode + pin floor_version.

Probes (gate the unmeasured claims)

New ocr-probes-v1.md specs 4 gating probes — OCR-RT (residue→rank round-trip), OCR-DET (repair determinism), OCR-POST (GGUF posterior parity), OCR-SCHEMA (ValueSchema fit) — plus 3 cascade perf probes. The big claims (int8-exact LSTM, bit-reproducible diff, ~200k-LOC 1:1 layout) are CONJECTURE until these run.

• OCR-SCHEMA shipped as a contract test (ocr::tests::ocr_schema_fit_rides_existing_preset_no_new_variant) — proves OCR rides Compressed/Full, no new enum variant.

Board: EPIPHANIES.md E-OCR-PLAN-DRIFT-1, AGENT_LOG.md entry. contract lib green; clippy/fmt clean.

https://claude.ai/code/session_01D2WSmezQBNC3bUdHuGfGmo

---

Generated by Claude Code

Summary by CodeRabbit

• Bug Fixes

  • Rebaseline and correct OCR transcode planning; resolved identified architectural inconsistencies and migration issues.

• Documentation

  • Updated integration plans to reflect post-release architecture updates.
  • Added new OCR gating probes specification document for staged validation framework.
  • Updated agent logs and epiphany records documenting resolved issues and corrected plans.

• Tests

  • Added validation test confirming OCR functionality integrates with existing schema presets without introducing new variants.

[coderabbitai]

📝 Walkthrough

Walkthrough

Rebaselines seven OCR transcode plan documents from post-#497 to post-#498 architecture, removing two showstoppers (false reversibility rationale and nonexistent Morton-tile cascade). Adds ocr-probes-v1.md with four gating probes and three cascade-performance probes. Ships the OCR-SCHEMA deliverable as a new contract test in crates/lance-graph-contract/src/ocr.rs.

Changes

OCR Transcode Rebaseline and OCR-SCHEMA Shipment

Layer / File(s)	Summary
OCR-SCHEMA contract test crates/lance-graph-contract/src/ocr.rs	Adds ocr_schema_fit_rides_existing_preset_no_new_variant asserting HelixResidue, TurbovecResidue, EntityType, and Fingerprint are already present in ValueSchema::Compressed, that ValueSchema::Full covers all OCR-touched tenants, and that both schemas are layout-preserving—proving no new ValueSchema::Ocr variant is needed.
New gating probe specification .claude/plans/ocr-probes-v1.md	Introduces the staged probe queue: four primary gating probes (OCR-RT, OCR-DET, OCR-POST, OCR-SCHEMA) with explicit claims, pass/fail criteria, and costs, plus three secondary cascade-performance probes and a DAG honesty section establishing OCR-RT as the highest-leverage gate before funding the full layout transcode.
ocr-canonical-soa-integration-v1.md rewrite .claude/plans/ocr-canonical-soa-integration-v1.md	Updates front-matter to post-#498 (ENVELOPE_LAYOUT_VERSION = 2, Signed360 48B→6B); revises HHTL/OCR class addressing to bootstrap on classid = 0x0000_0000; rewrites ValueSchema preset section to forbid a new ValueSchema::Ocr variant and correct tenant semantics; replaces Hamming with L1 CAM-PQ cascade repair; tightens bit-reproducibility harness; rewrites D-OCR-50/51 deliverables.
Supporting plan and board updates .claude/plans/soa-centroid-attention-field-synthesis-v1.md, .claude/plans/tesseract-rs-ast-dll-codegen-v1.md, .claude/plans/tesseract-rs-transcode-master-v1.md, .claude/board/AGENT_LOG.md, .claude/board/EPIPHANIES.md	Removes Morton-tile framing from the centroid attention field plan and adds CONJECTURE gate; advances post references to #498 and adds CONJECTURE qualifications on determinism in codegen and master transcode plans; records the rebaseline outcome, showstoppers found, and next steps in agent log and epiphanies board.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• AdaWorldAPI/lance-graph#496: Introduced ValueTenant, VALUE_TENANTS, and ValueSchema::{Compressed, Full, ...} presets that the new contract test directly asserts against.
• AdaWorldAPI/lance-graph#497: The plan corrections in this PR (disallowing ValueSchema::Ocr variant, removing false reversibility rationale) directly close showstoppers identified from the approach described in #497.
• AdaWorldAPI/lance-graph#498: Shipped LayoutBlock::to_node_row, ENVELOPE_LAYOUT_VERSION = 2, and the OCR→NodeRow transcoding that all updated plan docs now reference as the post-feat(contract): GUID decode→read-mode keystone + helix Signed360 right-size + OCR→NodeRow transcode #498 architecture baseline.

Poem

🐇 Hop hop, the plans are set straight,
No Morton tiles, no phantom cascade gate,
Six bytes of Signed360 lead the way,
OCR-SCHEMA shipped green today!
The probes stand guard before the big transcode—
This rabbit won't fund work on a broken road. 🗺️

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the main change: rebaseling OCR plans (#497) to post-#498 architecture with gating probes and a 5-specialist review framing.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

---

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 and usage tips.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 24d3fd843a

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".

[coderabbitai]

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/plans/ocr-probes-v1.md:
- Around line 30-33: The 99% pass threshold in the OCR-RT probe definition
allows lossy residue→rank mappings to incorrectly pass the reversibility gate.
Replace the threshold-based pass criterion (≥ 99%) with an exact requirement
that fails on any miss, ensuring only perfect round-trips satisfy this gate. If
tolerance is desired, separate it into a distinct quality probe rather than
mixing it into the reversibility assertion, so the corrected plans' claim about
text-as-identity and codebook-as-repair-signal can be properly validated.

🪄 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: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 2324b59e-e828-4fee-8cd8-4c7e02a16568

📥 Commits

Reviewing files that changed from the base of the PR and between 4acea26 and 24d3fd8.

📒 Files selected for processing (8)

• .claude/board/AGENT_LOG.md
• .claude/board/EPIPHANIES.md
• .claude/plans/ocr-canonical-soa-integration-v1.md
• .claude/plans/ocr-probes-v1.md
• .claude/plans/soa-centroid-attention-field-synthesis-v1.md
• .claude/plans/tesseract-rs-ast-dll-codegen-v1.md
• .claude/plans/tesseract-rs-transcode-master-v1.md
• crates/lance-graph-contract/src/ocr.rs