docs(plan): Git porcelain for epistemic algebra, causal Pickaxe and Lance 13 history

[AdaWorldAPI]

Why this PR

The proposed Thought Git API in #1599 and candidate-resolution semantics in #1604 still need a precise boundary between cheap speculative revision in RAM, Lance dataset/version branching and persistence, and causal Pickaxe over semantic provenance and temporal epistemic horizons. This captures the operator's proposed Git × Lance 13/LanceDB 0.40 × stable rows × temporal.rs connection without turning each of millions of folds into a storage transaction.

Git porcelain as the public API (2026-10-18 extension)

Decision: expand #1605, not open #1606. This is the same architecture seen from the human/agent interface: status, branch, add -p, diff --semantic, merge, rebase, blame, bisect, pickaxe, squash, checkout and revert are mapped to existing D-TGA thought operations, RelAlg normalization, Datalog dependency/retraction, R2IL/mask-risc execution, NAL evidence revision, temporal horizons and the single Lance writer.

Invariants: Git is porcelain, not a VM; commands selecting identical typed intents must yield identical masked results. Staging proposals does not certify their truth; presence XOR alone is not a semantic value diff; Git ancestry cannot certify causality; historical queries enforce pinned read horizons and authorization. No second evaluator or new CE64 encoding.

New negative controls: CPG-18..25 cover porcelain-to-IR equivalence, read-only status, unchanged-presence/changed-value diff, add -p not proving claims, source-aware merge, nonmonotonic bisect/lineage refusal, historical authorization and CLI/AST parity. All are symbolic preregistrations, not executed tests.

Source-grounded findings

• Lance 13.0.0 Dataset exposes versions, create_branch, checkout_branch, checkout_version, tags, delta, read_version_transaction, restore and stable-row-ID migration. Pinned API
• LanceDB 0.40.0 targets Lance 13.0.0 and documents Rust Table version/checkout/tag/branch primitives. Rust Table API. Integration with this repository's default workspace dependencies is not yet verified; Cargo manifests still carry older lockstep surfaces while Lance13 probes are isolated.
• temporal.rs already separates owner-local trajectories from reader horizon, with Contemporary, Anachronistic, Spoiler, Unknowable. QueryReference::default() uses unbounded u64::MAX despite Strict; causal/as-of Pickaxe MUST pin a concrete head. The data-dependency DependsClosure is currently a seam whose default NoDeps cannot certify cause.
• NodeGuid ≠ stable Lance _rowid ≠ physical _rowaddr. Storage branch/version numbers require a dataset+branch+version composite ref, and the row mapping must be checked across compaction and updates.
• The 64k dirty-row benchmark used Lance 11, with ~1.7 ns for a small fold and ~1.7 ms minimum nonempty commit; this is not a Lance13 measurement. Source: .claude/board/entries/2026-09-24-cost-scales-with-dirty-rows-replay-budget-before-materialization.md.

Files

• .claude/plans/causal-pickaxe-resident-git-lance13-temporal-v1.md: precise design, identity and temporal boundaries, mapping of Git verbs to existing in-memory mask/D-TGA operations and native Lance durable refs, causal Pickaxe grades, dependency gates, and benchmark arms.
• docs/research/causal-pickaxe-version-temporal-fixtures.json: 26 pre-registered symbolic falsifiers, not yet executed. Includes stable-row movement, branch version collisions, read-only checkout, strict-horizon leakage, HLC/owner-local deinterlacing, source-root independence, alternative sufficient proofs/overdetermination, premise deletion requiring DRed, replay retention after squash, NoDeps failure, authorization and idempotent writer recovery.

What it deliberately does NOT do

No code or dependency bump, new VM, per-fold persistence, second WAL, duplicate thought-Git lifecycle, CE64 bit repurposing, implicit native Git merge implementation, or claim of physical causality merely from commit ancestry. Pickaxe is a candidate RelAlg/Datalog query over existing address/delta masks, with temporal filtering and independently justified derivation / counterfactual stages.

Landing order

1. Isolated pinned Lance13 + LanceDB0.40 API and row/branch probe.
2. D-TGA resident history and one-writer artifact gate, with codebook-sealed NodeGuid mapping.
3. Strict pinned temporal reads + Datalog source dependency provenance.
4. Counterfactual replay and negative controls.
5. Only then evaluate whether native Lance durable experimental branches beat plain squash references.

Review question: can we use native Lance branches/tags for durable checkpoints while keeping the thought Git DAG resident and the cause of a conclusion a provenance/ablation query, never a mere commit-parent edge?

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 7c13519c-4934-46b5-8c9b-354f806b69e3

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Warning

Billing warning: we have not been able to collect payment for this subscription for more than 72 hours. Please update the payment method or pay any pending invoices in Billing to avoid service interruption.

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.}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-18T14:32:22.314882Z	42c65b7	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 42c65b7c6a

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".

[chatgpt-codex-connector]

Update the prerequisite for the migrated workspace

At this commit, Cargo.toml:275-277 and Cargo.toml:348-360 already define Lance 13 and LanceDB 0.40 as the default workspace dependencies, so the claim that the root documents a 7/0.30 lockstep and that the whole-workspace migration has not happened is stale. Following this rollout instruction would keep validation in an isolated probe instead of testing the integration that now exists; rewrite the preflight around the current workspace pins while separately identifying any excluded consumers that still require older versions.

Useful? React with 👍 / 👎.

[AdaWorldAPI]

Fixed in #1618 (f95c72fe): CPG-0 and the prerequisite now name the workspace pins Lance 13.* and LanceDB 0.40.*, and reuse the H14c-5 row-identity measurement instead of a separate probe.

---

Generated by Claude Code