Repository navigation
planner: derivation ledger and load-bearing ablation (D-EPI-COT C0–C2) - #1840
Merged
Merged
Conversation
…n (D-EPI-COT C1/C2) BeliefArena keeps one premise pair per statement and an empty stamp on derived entries, so it cannot say which observations a conclusion needs. DerivationLedger records every observation with its source and every derivation with its law, premises and chain, and evaluates any statement under an ablation: proofs over pooled premise evidence, NARS revision across disjoint roots only, chains that still derive it, and a single/joint ablation classification (critical, synergistic, redundant). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMPWTwovjYJEUMuGPb4
The 34 contract recipe kernels change scalars only; the five planner tactics are the term-logic base. Records what the certification folds guarantee and what C1/C2 (nars::support) add, with the falsifiers and the disables that turn them red. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMPWTwovjYJEUMuGPb4
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars.
 |
…t evidence selection, depth bound
- derive checks the step's terms against its law's figure (the figures the
planner tactics use) and refuses otherwise; admit records a tactic
Candidate directly, so the tactics stay the proposing authority.
- sensitivity reports Δfrequency, Δconfidence, chains and roots lost: a
conclusion can survive an ablation while its evidence collapses. load is
unchanged.
- pooling selects the root-disjoint set of proofs with the most total
evidence weight; a greedy pick by single-proof confidence could suppress
two independent proofs that outweigh one stronger shared-root proof.
- MAX_DEPTH refuses a path deeper than 64; MAX_STEPS bounded work only.
Truncated becomes Refusal { Steps, Depth, Pool }.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XAqKMPWTwovjYJEUMuGPb4
…selection, bounds Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMPWTwovjYJEUMuGPb4
…ith the bits 50..52 conflicts Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMPWTwovjYJEUMuGPb4
AdaWorldAPI
marked this pull request as ready for review
October 28, 2026 19:08
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
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[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: f8c582ed0d
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".
- grounded_since counts a derivation only if a proof uses it, not when its conclusion already has other support (codex P2). - Pooled::chains is computed per chain with every other chain's derivations removed, instead of from the pooled proof's combined provenance, which mixed chains wherever a premise had proofs from two (codex P2). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMPWTwovjYJEUMuGPb4
This was referenced Oct 28, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
D-EPI-COT C0–C2: a substrate audit, plus a derivation ledger that says which sources carry a conclusion and what happens to its truth when they are removed.
C0 — audit (
docs/research/epistemic-cot/c0-substrate-audit.md)What exists today, and what each piece can and cannot answer:
BeliefArena;CertificationModeland its statistical gaps;EpistemicState5.C1/C2 —
nars::support::DerivationLedgerThe ledger sits beside
BeliefArena, which is unchanged. It is append-only and holds observations(stmt, truth, source)and admitted derivations(chain, law, premises, conclusion).NotAFigure). The figures are the ones the planner tactics use (RCR abduction, TR analogy, CAS induction/deduction).admittakes a tacticCandidatedirectly.Ablation(sources and/or derivations removed):load()asks whether the conclusion still exists:FirstCritical,SecondCritical,Synergistic,Redundant,NotLoadBearingorAbsent;sensitivity()asks how strong it still is: Δf, Δc, chains lost and roots lost. A conclusion can survive while its evidence collapses.Refusal::{Steps, Depth, Pool}. The evidence horizon stays at k = 1.16 tests. Each of 10 disables turns its test red: revise-everything, no path guard, no joint ablation, ignore source ablation, ungrounded productive, no step budget, no figure check, greedy selection, no depth bound, no pool bound and no dominance filter.
Scope
Draft; not for automatic merge.
🤖 Generated with Claude Code
https://claude.ai/code/session_01XAqKMPWTwovjYJEUMuGPb4