[AdaWorldAPI]

This adds the pre-registration for thought git (D-TGB-0) and the slot pool (D-TGB-1) from thought-git-basics-v1.md (merged in #1629). Both live in a new lab probe, crates/thought-git-probe. It is its own workspace, excluded from the main one, and depends only on lance-graph-contract.

D-TGB-0: fixtures, from-scratch oracle, signal checks

• Fixtures (src/fixtures.rs):
  • one write intention per Rubicon outcome;
  • a stack of unremarkable thoughts;
  • a contradiction against a given;
  • a divergent pair;
  • a regression scanpath;
  • a cold/hot surprise pair;
  • an unwritten lane.
• Oracle (src/oracle.rs): computes the next version from scratch.
  • Only a Commit thought lands.
  • A given is never overwritten. An agreeing write is not stored; a different value is recorded as a contradiction.
  • If two thoughts in one version write different values to the same cell, nothing is stored and a divergence is recorded.
  • An unwritten cell reads as absent.
• PREREGISTRATION.md: the falsifier, the disable and the fixture for D-TGB-1..10.

D-TGB-1: slot pool (src/slots.rs)

Operator decisions, 2026-10-20: "The easiest way in the beginning is Fifo vs pruning the least used lanes", then "Squashing if the least used are in a dependency chain".

branch takes a free slot. When all 63 are taken, the policy picks one candidate:

• Eviction::Fifo picks the oldest;
• Eviction::LeastUsed picks the one used least, with ties going to the oldest.

What happens to the candidate depends on its place in a dependency chain:

• if it has a live parent, it is squashed into that parent;
• else, if it has a live child, its oldest child is squashed into it;
• only if it has neither is it evicted.

The parent the new thought branches from is never the candidate. free (used for squash, discard and land) returns a slot, and any children it had move to its parent. A handle carries its generation, so a stale handle is refused.

Tests (25)

• 9 oracle tests, 10 signal tests and 6 slot tests.
• Signal tests run on the real contract code. Each signal has a can-fire case and a can-stay-silent case: regressions, the cold/hot xor, the kanban routes, the MUL gate, and Gadamer.
• Every rule has a disable, run after commit. Each turned the named test red and was then restored.

disable	red test(s)
land every outcome	only_commit_lands
store agreeing writes	agreement_with_a_given_is_not_stored
overwrite the given	a_different_value_for_a_given_is_refused_and_recorded
last writer wins	divergence and squash tests
read missing as 0	absence test (plus two others)
ignore the slot generation	stale-handle assertions in tests/slots.rs
LeastUsed keyed on age only	least_used_evicts_the_idlest_root
evict instead of squashing a chain	a_least_used_thought_in_a_chain_is_squashed_into_its_parent, a_candidate_with_a_child_absorbs_it
let the new thought's parent be the candidate	branching_from_the_candidate_keeps_it

cargo test, cargo clippy --all-targets -- -D warnings and cargo fmt are clean, run from crates/thought-git-probe.

Plan notes added in this PR

• Polyfill, never legacy.
• Direct crate calls for internal timing.
• The write path is Rust, not an API.
• Folding datafrog through Quack (D-TGB-10).
• What a landing writes into CE64 bits 59..63 (pearl, MUL, Gadamer).
• The kanban planning vocabulary (D-TGB-9).
• The MORK API surface (D-MPS-9), with the MORK licence state recorded as OPEN.
• crepe (Soufflé-style) and ascent (datafrog+, lattices, BYODS, rayon) as candidate Datalog engines. Their role is still OPEN.

Not in this PR

The falsifiers for CE64 bits 59..63 on edges (pearl::hydrate, pearl::revise) need lance-graph-planner. They move to D-TGB-3.

Board

• STATUS_BOARD: D-TGB-0 and D-TGB-1 are In PR.
• Plan items are updated.
• New entry: 2026-10-20-d-tgb-0-preregistration.md.
• AGENT_LOG: today's Haiku runs.
• The root Cargo.toml adds the probe to exclude.

🤖 Generated with Claude Code

https://claude.ai/code/session_01VncKoyG6JkpsoqhqTGTv3f

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• 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 2026-10-20T17:30:14.991010Z	17537a3	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.

[coderabbitai]

Add Carrot credits or activate Agent usage billing to use Autofix

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 17537a323a

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]

Keep the branch parent out of chain squashing

When the pool is full and the policy selects a root whose oldest child is the from slot, this branch removes that very parent: for example, under FIFO with thought 1 parented to thought 0, branch(..., Some(thought_1_slot)) skips thought 1 as the candidate, selects thought 0, and then chooses thought 1 here as from and releases it. The call succeeds but unexpectedly makes the supplied parent handle stale, contrary to the registered invariant that a new thought's parent is never removed; child selection must also exclude keep, or a different candidate must be chosen.

Useful? React with 👍 / 👎.

[AdaWorldAPI]

Confirmed: reproduced with FIFO, thought 1 parented to thought 0, then branching from thought 1. The fix is 8b6b585, carried by #1631 because this PR was already merged.

When the candidate root's oldest child is the new thought's parent, the root is now squashed into that child. The child then takes the root's place in the tree, so its handle stays live.

Test: branching_from_the_candidates_child_keeps_the_child. It is red without the fix, and red again if the child does not take over the root's parent.

---

Generated by Claude Code