Repository navigation
D-GSO-7: reasoning-band earning and downgrade (P7) - #1360
Conversation
The ReasoningBand on a CausalEdge64 (bits 61..63) is read only through band_reading::project_band (declared present, asserted provenance) and earned one rung per executed, passing proof under the shipped Pearl CausalMask: an SO observation a majority corroborates gives Association, a passed PO intervention on top gives Causal, a passed SPO counterfactual on top of that gives Counterfactual. Observations never lift past Association, a NotRun test never raises, and no rung is skipped. A failed test drops to the rung below the one it guarded; contradicting independent evidence (Quorum, silence excluded) caps at Association when a minority and drops to Surface when a majority; detected confounding (CausalMask::simpsons_paradox_risk) caps at Association. The same events replay to the same edge bits, and only bits 61..63 move. The ladder and thresholds are policy pins; Relation, Perspective, Meta and Transcendent have no obligation and are never reached. 6 tests (association cannot skip, executed-passing-only one rung, exhaustive no-rise-without-obligation sweep over 8 bands x every event, contradiction/confounding/failure lower, replay, unreadable band refuses), 7 disable runs red. Board: D-GSO-7 In PR; D-GSO-6 Shipped (#1358). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DgxrCafsJuAahpBs7oC14R
|
Warning Review limit reachedYour organization has reached its usage spending cap. Adjust your spending cap in the billing tab. Next included review available in 13 minutes. View limit detailsLimit details: You’ve used the included review currently available. Your 67 included PR review attempts over the past 7 days set your current allowance at 1 review per hour. Review configuration: ⚙️ Run configuration
📒 Files selected for processing (4)
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. Comment |
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: a89fe437-0332-41e1-b6a7-1989b9f80ee8) |
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. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 9a09c668a4
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".
Codex review: Tested { outcome: Passed } let any caller fabricate a pass
and raise the band, so the probe showed an asserted transition, not the
P7 invariant that the obligation was actually executed.
Events now carry trial data and run() executes it:
- intervention (PO): treated vs control counts; passed when both arms
were measured and the treated rate is higher, failed otherwise, not
run when an arm is empty;
- counterfactual (SPO): a removal attack on premise masks; passed when
the conclusion derives with the candidate and not without it, failed
when it never derives or the candidate is dispensable, not run when
there is nothing to attack.
There is no Passed input left. Replay re-executes the trials. New test
a_pass_is_derived_from_the_trial_data; disable runs D8-D10 (rate check
removed, removal never re-derives, empty arm counted as run) go red, and
the earlier ladder disables still do.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DgxrCafsJuAahpBs7oC14R
Supersedes the meaning of D-GSO-7's rungs; #1360's probe is unchanged. Under a reading declared per class, bits 61..63 hold the strongest relational statement the sealed model may assert: 0 Open, 1 Associated, 2 Related, 3 Contributes, 4 CausalCandidate, 5 Causes; 6 and 7 refuse. ReasoningBand is used only as the 3-bit carrier, and a class declared under the historical reading refuses the same bits. Contracts are integer folds over unit masks and causal_audit's SupportLedger: association needs two distinct sources, the robustness mask can only refute, Contributes is a stable effect in every declared stratum, Causes needs two distinct InterventionBacked sources plus executed randomized arms. entails is an explicit table, equal to code >= on 0..=5 only; historical Meta/Transcendent do not satisfy Causes. An exhaustive 2,500-model family shows the chain is monotone with association scoped to the declared populations (0 violations) and not with marginal association (10, Simpson). Sibling specificity and removal under overdetermination are shown not to be rungs; confounded observational data does not demote an intervention-certified relation. 12 tests, 11 disable runs red. Names and thresholds are policy pins. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01T8w7Ucm7NhitfTduFKmhYx
Supersedes the meaning of D-GSO-7's rungs; #1360's probe is unchanged. Under a reading declared per class, bits 61..63 hold the strongest relational statement the sealed model may assert: 0 Open, 1 Associated, 2 Related, 3 Contributes, 4 CausalCandidate, 5 Causes; 6 and 7 refuse. ReasoningBand is used only as the 3-bit carrier, and a class declared under the historical reading refuses the same bits. Contracts are integer folds over unit masks and causal_audit's SupportLedger: association needs two distinct sources, the robustness mask can only refute, Contributes is a stable effect in every declared stratum, Causes needs two distinct InterventionBacked sources plus executed randomized arms. entails is an explicit table, equal to code >= on 0..=5 only; historical Meta/Transcendent do not satisfy Causes. An exhaustive 2,500-model family shows the chain is monotone with association scoped to the declared populations (0 violations) and not with marginal association (10, Simpson). Sibling specificity and removal under overdetermination are shown not to be rungs; confounded observational data does not demote an intervention-certified relation. 12 tests, 11 disable runs red. Names and thresholds are policy pins. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01T8w7Ucm7NhitfTduFKmhYx
CodeRabbit / markdownlint: a line starting with '#1360' reads as an ATX heading without a space. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01T8w7Ucm7NhitfTduFKmhYx
What
P7 of
.claude/plans/2026-10-06-global-sudoku-replayable-orchestration-v1.md: the reasoning band rises only when its proof obligation was executed and passed, falls under contradicting evidence, and every change replays (§7, §8, §18 P7).crates/cognitive-shader-driver/examples/reasoning_band_probe.rs, no library code changes.Reused, unchanged
ReasoningBandinCausalEdge64bits 61..63, read only throughband_reading::BandDeclarations::project_band(declaredPresent, asserted provenance).CausalMask.ontology_warrant::Quorumfor independent evidence (silence is abstention).CausalMask::simpsons_paradox_riskfor confounding.The ladder (a policy pin)
AssociationSOobservation a majority of speaking sources corroborateCausalPOintervention trial, executed here and passedAssociationCounterfactualSPOremoval attack, executed here and passedCausalA pass is computed, never supplied (Codex review). An event carries trial data, and
run()executes it to derive the outcome:There is no
Passedinput, and replay re-executes the trials.Downgrades
Associationwhen it is a minority and drops it toSurfacewhen it is a majority.Association.The four P7 requirements
Association. An exhaustive sweep over 8 bands × every event kind finds exactly the three ladder steps as the only rises, each tied to an executed passing trial.Scope limits
Relation,Perspective,MetaandTranscendenthave no obligation and are never reached. No stored-row writes.Tests (7) and disable runs (10, all red)
D1 an observation lifts to
Causal· D2 aNotRuntrial raises · D3 an intervention skipsAssociation· D4 contradiction ignored · D5 silence counted as contradiction · D6 confounding ignored · D7 band read without the contract · D8 intervention passes regardless of rates · D9 removal never re-derives without the candidate · D10 an empty arm counted as a run.Board
STATUS_BOARD: D-GSO-7 In PR; D-GSO-6 Shipped (#1358).
🤖 Generated with Claude Code
https://claude.ai/code/session_01DgxrCafsJuAahpBs7oC14R