Repository navigation
Co-research map: Simpson's paradox as an epistemic signal; P-pool probe - #1596
Conversation
Exploration map from the co-research council. Finds that NARS revision of stratum beliefs equals the union recount, so it reproduces Simpson's reversal, and that NAL's overlap guard does not stop it. Lists the seams that keep the revision gate pure set algebra and nine pre-registered probes. Ratifies nothing. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh
…union recount Three cases over isa::truth revision (f32) and CausalEdge64::revision (u8): the certification Simpson fixture reverses (0.6 vs 0.3 against both strata); balanced agreeing strata do not reverse (silence twin); agreeing but exposure-unbalanced strata still reverse, so 'pool only at Supports' is not sufficient on its own. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars. ⚙️ Run configuration
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. Comment |
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: 4241f541b8
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".
| //! beliefs equals the union recount, so it reproduces Simpson's reversal. | ||
| //! | ||
| //! A stratum belief "exposure → outcome" carries `f = hits / n` and | ||
| //! `c = n / (n + 1)`, so its evidence weight `c / (1 - c)` is `n`. Revision |
There was a problem hiding this comment.
For larger cells, this identity does not hold in the production arithmetic: causal-edge/src/isa.rs::evidence_weight replaces any c >= 0.999 with f32::MAX, so an f32 cell with n >= 999 no longer has weight n; on the packed path, n >= 509 already rounds confidence to 255. Revising two such cells overflows the summed weight and can produce NaNs (which the register packs as zeros), rather than the union recount. Because the probe only uses n <= 90, it incorrectly presents a size-limited result as the central general finding; add saturation-boundary cases and qualify the conclusion.
Useful? React with 👍 / 👎.
2 participants
What
These two commits from the 2026-10-17 co-research run were never pushed. They are cherry-picked here onto current
main.Board entry
2026-10-17-coresearch-simpson-epistemic.md(status OPEN, ratifies nothing). The question opened when D-CGW-1: CE64 bits 43..45 are the orientation #1594 removed the per-edge Simpson carrier. The entry holds:Central finding: NAL revision of stratum beliefs equals the union recount, so it reproduces Simpson's reversal. NAL's overlap guard does not stop it, because disjoint strata pass that check trivially.
Probe P-pool (
lance-graph-planner/tests/simpson_revision_pooling.rs). It revises the Simpson fixture's two stratum beliefs withisa::truth, both at f32 and through the packed u8 register. It pins the reversal (exposed 0.6 vs unexposed 0.3, equal to the union recount) and checks the twins: the one-stratum robust fixture does not reverse, and two raising strata with balanced exposure admit.Verification
cargo test -p lance-graph-planner --test simpson_revision_poolingon currentmain: 3 pass. The entries index was regenerated (entries_index.py --write, then--checkis green, 365 rows). The README conflict was resolved by taking main's index and regenerating it.Open
Which crosswalk rows go forward is a decision for the maintainer. Rows 9 (rename the misnamed
simpsons_paradox_risk) and 10 (strata declared before the seal) are marked ADOPT-NOW in the map. Neither is implemented here.Generated by Claude Code