Repository navigation
Conversation
… groups, reachability Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…nd short group chips Fade ILP edges on the map, frame the dense core, and lead with the finding that most reductions are at most quadratic. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…n hover Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Explore is a map-style search over problems and rules with place cards, a layered view by distance from SAT, and a growth legend. Directions is its own mode. Drop italics and the mutually-reducible default panel. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…sk at a time Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… panel, one toolbar, Layers view and route objective Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…rate levels, overhead chips, a dedicated reduction view Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Witness and Value rules share one view (Value is a subset of Witness); Turing rules, which call the target several times, are shown on request as dash-dot edges. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The static hero illustration becomes a small force simulation over problem families, drawn with the existing hero classes and palette. Hovering lights a problem's direct reductions; nodes can be dragged. It pauses off-screen and settles without animation under reduced motion. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Layout: one deterministic force model (lib/force-layout.mts) replaces the fCoSE build layout. Rules into ILP and QUBO pull weakly, so the map shows clusters instead of a star around ILP. The page reuses the same model live: dragging a problem pulls its neighbours, and the "Without ILP & QUBO" and "NP-hard only" views settle in place. - Colour: nodes show whether a problem is proven NP-hard (reached from SAT) and whether it reaches ILP, with a legend and a status line on the card. Problem categories are no longer shown. - Panel: folds to its search bar when idle; a labelled Directions button and a labelled Growth filter replace the bare icon and chips. - Details: rules stay neutral until in focus, labels try four sides to avoid collisions, formulas print as tidy polynomials, the intro unfolds the map, and phones get a one-row header with a peeking search bar. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every route step, including variant changes inside one problem, now shows its rule's growth badge in the timeline, and each route edge on the map takes the same growth colour. The selected route is laid out as an evenly spaced chain from start to end, with the rest of the map relaxing around it; labels sit across the chain, and other best routes are no longer drawn. Returning to explore restores the published map. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Defer the path engine until Directions opens, warm it before search, and settle graph layouts in cancellable batches. Fix the mobile panel transition and replace oversized focus outlines with a thin circular ring. Remove unused layout dependencies and homepage highlights, and update existing browser and build checks for the current behavior.
…ucture' into fix/simultaneous-incongruences-complexity
Each ILP rule's num_nonzeros upper bound was num_constraints * num_vars, a dense-matrix bound that the corpus measured at under half of its prediction on 55 rules. Derive the bound from each reduce_to's constraint families instead, summing the terms each family contributes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add SetSplitting → NAE-SAT clause-per-subset reduction
Review fixesPushed The fixed parent UI is integrated. Sixteen Node tests, TypeScript/ESLint checks, production bundling, and the three focused desktop/mobile browser regressions pass, including K2/K3 separation, incompatible family variants, excluded Turing/aggregate-only routes, and visible load failures. The combined stack passed 7,194 native tests and the paper build at |
[Critical] Remove KSatisfiability KN variant and short-clause constructors
Declare exact parameter counts that constructions always produce
Tighten ILP num_nonzeros bounds to per-constraint counts
Tighten StaffScheduling and SWRTD QUBO size bounds
Add binary decision ILP to SAT and fix solution validators
Add staff-scheduling QUBO and tighten direct-reduction bounds
Add Phase 3 conditional direct QUBO reductions
Add Phase 2 direct QUBO reductions
Add Phase 1 direct QUBO encodings for scheduling and assignment
…nces-complexity Fix SimultaneousIncongruences complexity to include the search period
GiggleLiu
left a comment
GiggleLiu
left a comment
There was a problem hiding this comment.
Agentic Review Report: integration follow-up
The previous blocking findings are fixed: concrete-variant witness-route indicators replace false hardness claims, both views show loading failures, and the inherited growth labels now describe conservative size bounds. The full production artifact passed cross-browser testing; the newly found WebKit inline-math rendering issue also has an independently reviewed CSS fix and regression.
The semantic source tree is 0cf54de5aa06ad17aae5a480e77ad62747ef308c. Backend child PRs retain their merge records. The previously reviewed integration resolutions and this PR's original source are preserved; the GitHub ancestry merges have identical trees. #1219, #1220, and #1221 remain excluded.
Validation: the combined Rust 2024 integration passed 7,194 native tests, formatting, all-target/all-feature Clippy, the paper build, and 23 pipeline tests. The subsequent website-only changes do not alter native sources. The production verifier passed 336 browser cases before the last visual fix; that fix passes focused Chromium/Firefox/WebKit dark/light tests and independent review. Its fresh complete website run and final main CI are pending and remain required before merging the aggregate into main. This approval is for folding into the feature parent only.
Rechecked PR comments, inline comments, reviews, linked conversation, and coverage feedback. No new material concern. The independent quality and harm-signal gates passed. The previously documented compatibility breaks were accepted by the maintainer.
Stacked on #1196 (base:
feat/polynomial-comparator-v1).What
P → Decision<P>edges, solved by binary search on the bound) sit behind an off-by-default Turing toggle and are drawn dash-dot. Directions never routes through them.activeRulesaccepts one scope or a list, with a test for the combined view.Checks
tsc,eslint,npm run test:frontend-contracts(41 pass) andscripts/build_website.pysucceed. The Playwright site tests were not run.🤖 Generated with Claude Code