D-PUZZLE-ATTN-0: entropy as the focus of attention on crosswords by AdaWorldAPI · Pull Request #1398 · AdaWorldAPI/lance-graph

AdaWorldAPI · 2026-10-07T23:36:43Z

What

The question: does entropy work as the focus of attention when solving crosswords?

Shared core. The D-PUZZLE-0 step 3: crossword propagation as Cartesian-addressed masking, four arms, cui-bono ablation #1387 crossword core moves unchanged into crates/cognitive-shader-driver/examples/shared/crossword_core.rs. D-PUZZLE-0 step 3: crossword propagation as Cartesian-addressed masking, four arms, cui-bono ablation #1387's own probe now uses it, and its 20 tests still pass.
New probe: crossword_attention_probe.rs (test = true).
- Propagation is a fixed point, so deduction needs no attention. The probe changes only which slot the search picks next once propagation stalls.
- Puzzles, propagation and value order (most frequent word first) are held fixed.
Board:
- entry 2026-10-07-puzzle-attn-0-entropy-attention-crosswords.md;
- STATUS_BOARD row D-PUZZLE-ATTN-0;
- D-CE64-CYCLE-0 + D-PUZZLE-ATTN-0: cross-cycle state, and entropy as the focus of attention on crosswords #1397 merge hygiene (PR arc, LATEST_STATE, D-CE64-CYCLE-0 set to Shipped).

Measured

Release build, 20 created puzzles and 20 empty grids, sides 5 and 7, 200,000 nodes per run.

policy	prove: solved	prove: nodes	open: solved	open: nodes / to first	fill: solved
fixed	16/20	852,535	7/20	2,858,517 / 2,488,660	5/20
random	15/20	1,049,801	5/20	3,151,741 / 2,572,628	0/20
widest	16/20	916,437	5/20	3,186,550 / 2,653,695	10/20
popcount (`log2 count`)	20/20	58,729	20/20	570,450 / 91,930	10/20
Shannon (COCA-weighted)	19/20	348,578	16/20	1,156,793 / 550,562	10/20
dom/wdeg	20/20	53,892	20/20	539,520 / 40,938	10/20

Uniform entropy focuses the search. On "prove", popcount needs 14.5× fewer nodes than fixed order.
Frequency-weighted Shannon misdirects it. It needs 5.9× popcount's nodes, and picks a different slot on 36–50% of decisions.
- A frequency prior measures how predictable a slot is. A search pays for how many candidates it must try.
Earlier contradictions help the search reach a first fill sooner. dom/wdeg gets to its first fill about 2× sooner on open puzzles. Within a search, the contradiction rate stays flat (0.73–0.85).
A slot with no candidates has no entropy under either reading. A test pins this.

Gates

9 tests run by the shader-driver cargo test step.
7 disable runs, each red on its named falsifier.
The dom/wdeg falsifier was vacuous on first write. It now compares only the search path.
clippy and fmt clean.
entries index and supersession index current.

Open

Shannon is untested as value ordering.
Coverage: one prior, one vocabulary, sides 5 and 7 only.
No CE64 register in this loop.

🤖 Generated with Claude Code

https://claude.ai/code/session_0177nSSyQaXqbNuGR1vMXXN6

Generated by Claude Code

Summary by CodeRabbit

New Features
- Added a crossword-solving comparison that evaluates six slot-selection strategies across generated, partially filled, and empty puzzles.
- The evaluation found that Popcount and DomWdeg solved all 20 generated and open puzzles tested. Frequency-weighted Shannon solved fewer and used more search nodes than Popcount.
Documentation
- Added experiment notes and updated project records with the comparison’s findings and limitations.

Move the #1387 crossword core (codebook, lexicon, grids, compile, token masks + fixed point, creation) unchanged into examples/shared/ crossword_core.rs; crossword_mask_propagation_probe includes it (its 20 tests pass unchanged). New crossword_attention_probe (test = true) varies only the search's slot choice: fixed, random, widest, popcount (log2 count), Shannon under the COCA frequency prior, dom/wdeg. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0177nSSyQaXqbNuGR1vMXXN6

…h only Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0177nSSyQaXqbNuGR1vMXXN6

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0177nSSyQaXqbNuGR1vMXXN6

…ped) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0177nSSyQaXqbNuGR1vMXXN6

coderabbitai · 2026-10-07T23:36:51Z

Note

Currently processing new changes in this PR. This may take a few minutes, please wait...

⚙️ Run configuration

Configuration used: Organization UI
Review profile: CHILL
Plan: Essentials
Run ID: ab3aebc7-b7dd-4b5e-b09a-3b01bb860183

📥 Commits

Reviewing files that changed from the base of the PR and between 4a0a72c and ce63247.

📒 Files selected for processing (9)

.claude/board/LATEST_STATE.md
.claude/board/PR_ARC_INVENTORY.md
.claude/board/STATUS_BOARD.md
.claude/board/entries/2026-10-07-puzzle-attn-0-entropy-attention-crosswords.md
.claude/board/entries/README.md
crates/cognitive-shader-driver/Cargo.toml
crates/cognitive-shader-driver/examples/crossword_attention_probe.rs
crates/cognitive-shader-driver/examples/crossword_mask_propagation_probe.rs
crates/cognitive-shader-driver/examples/shared/crossword_core.rs

 _____________________________________________________________________________________________________________________________________________________
< Progress in a fixed context is almost always a form of optimization. Creative acts generally don't stay in the context that they are in. - Alan Kay >
 -----------------------------------------------------------------------------------------------------------------------------------------------------
  \
   \   (\__/)
       (•ㅅ•)
       / 　 づ

✨ Finishing Touches

📝 Generate docstrings

Commit to this branch
Create a new PR

Autopilot · 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.

_{Comment @coderabbitai help to get the list of available commands.}

chatgpt-codex-connector · 2026-10-07T23:41:50Z

Codex Review Summary

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

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-07T23:44:46.076047Z	`ce63247`	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: ce63247ca6

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 · 2026-10-07T23:44:43Z

+        let pc = by_count(&|s| st.count(s) as f64);
+        if self.policy == Policy::Popcount {
+            let sh = by_count(&|s| self.prior.entropy(st.mask(s)).unwrap_or(-1.0));


Exclude used words before scoring candidate domains

Once any word has been placed, another open slot's mask can still contain that word because the all-different rule is enforced only later by go's st.placed.contains(&w) skip. Consequently this count—and the analogous Shannon and dom/wdeg scores—includes illegal values that the search will never try, so these policies do not actually select by legal domain size or entropy; this can change their ordering and undermines the reported comparisons, especially on empty grids with many same-length slots.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-10-07T23:44:43Z

+            let failed = match propagate_token(self.hot, self.puz, &mut next, &mut q) {
+                Err(Stop::Contradiction(t)) => {
+                    self.weight[t as usize] += 1;
+                    true
+                }
+                Err(Stop::Language) => unreachable!("one language"),
+                Ok(()) => !no_repeats(&next.placed),


Separate duplicate-word failures from empty-domain contradictions

When propagation succeeds but forces a word already placed elsewhere, this branch sets failed through !no_repeats and then records it as a contradiction and in fail_trace, even though no slot ran out of candidates and no dom/wdeg weight was incremented. Because the probe defines contradictions as empty domains and reports contradiction rates as evidence about dom/wdeg, any such forced duplicate contaminates the reported rates with a different class of failure; track these separately or attribute them consistently.

Useful? React with 👍 / 👎.

…pped) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0177nSSyQaXqbNuGR1vMXXN6

…-bits-3oe0dq Board: #1398 merge hygiene

claude added 4 commits October 7, 2026 23:34

crossword attention probe: dom/wdeg falsifier compares the search pat…

d9fa765

…h only Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0177nSSyQaXqbNuGR1vMXXN6

Board: D-PUZZLE-ATTN-0 entry and status row

fbe7c56

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0177nSSyQaXqbNuGR1vMXXN6

Board: #1397 merge hygiene (PR arc, latest state, D-CE64-CYCLE-0 ship…

ce63247

…ped) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0177nSSyQaXqbNuGR1vMXXN6

AdaWorldAPI mentioned this pull request Oct 7, 2026

D-CE64-CYCLE-0 + D-PUZZLE-ATTN-0: cross-cycle state, and entropy as the focus of attention on crosswords #1397

Merged

AdaWorldAPI marked this pull request as ready for review October 7, 2026 23:41

AdaWorldAPI merged commit 89694cc into main Oct 7, 2026
8 of 10 checks passed

chatgpt-codex-connector Bot reviewed Oct 7, 2026

View reviewed changes

AdaWorldAPI pushed a commit that referenced this pull request Oct 7, 2026

Board: #1398 merge hygiene (PR arc, latest state, D-PUZZLE-ATTN-0 shi…

f9683d3

…pped) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0177nSSyQaXqbNuGR1vMXXN6

AdaWorldAPI mentioned this pull request Oct 7, 2026

Board: #1398 merge hygiene #1399

Merged

AdaWorldAPI added a commit that referenced this pull request Oct 8, 2026

Merge pull request #1399 from AdaWorldAPI/claude/epistemicstate5-ce64…

061c57e

…-bits-3oe0dq Board: #1398 merge hygiene

Navigation Menu

Repository navigation

D-PUZZLE-ATTN-0: entropy as the focus of attention on crosswords - #1398

D-PUZZLE-ATTN-0: entropy as the focus of attention on crosswords#1398
AdaWorldAPI merged 4 commits into
mainfrom
claude/epistemicstate5-ce64-bits-3oe0dq

AdaWorldAPI commented Oct 7, 2026 •

edited by coderabbitai Bot

Loading

Uh oh!

coderabbitai Bot commented Oct 7, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector Bot commented Oct 7, 2026 •

edited

Loading

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Oct 7, 2026

Uh oh!

chatgpt-codex-connector Bot Oct 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

AdaWorldAPI commented Oct 7, 2026 • edited by coderabbitai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What

Measured

Gates

Open

Summary by CodeRabbit

Uh oh!

coderabbitai Bot commented Oct 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chatgpt-codex-connector Bot commented Oct 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codex Review Summary

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Oct 7, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot Oct 7, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

AdaWorldAPI commented Oct 7, 2026 •

edited by coderabbitai Bot

Loading

coderabbitai Bot commented Oct 7, 2026 •

edited

Loading

chatgpt-codex-connector Bot commented Oct 7, 2026 •

edited

Loading