Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions .github/workflows/verify.yml
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,8 @@ name: Reproduce and verify

on:
push:
branches: [main]
tags: ['**']
pull_request:
workflow_dispatch:

Expand Down Expand Up @@ -74,6 +76,7 @@ jobs:
python -m unittest discover -s scripts -p 'test_atlas*.py' -v
node scripts/check_atlas_ui.cjs
node scripts/check_atlas_map.cjs
node scripts/check_atlas_selection.cjs
node scripts/check_tm_atlas_ui.cjs
python scripts/check_tm_retrieval.py evidence/tm-retrieval-v1/answers.json
- name: Check N5 and Turing source context bindings
Expand Down
74 changes: 40 additions & 34 deletions README.md
Original file line number Diff line number Diff line change
@@ -1,40 +1,44 @@
# Finite Congruence Checks

![Finite Congruence Checks — project identity banner: small algebras, explicit witnesses and reproducible evidence](charts/project-banner.svg)

[![Reproduce and verify](https://github.com/Maverick0351a/finite-congruence-checks/actions/workflows/verify.yml/badge.svg)](https://github.com/Maverick0351a/finite-congruence-checks/actions/workflows/verify.yml)
[![Recorded Lean 4.34.1 checks passed: concrete N5/B3 and targeted upstream modules](charts/lean-checks-badge.svg)](docs/LEAN_VERIFICATION.md)
[![Release v0.1.0](charts/release-badge.svg)](https://github.com/Maverick0351a/finite-congruence-checks/releases/tag/v0.1.0)
[![License: Apache 2.0](https://img.shields.io/badge/license-Apache%202.0-315B78)](LICENSE)
[![Python: 3.10+](https://img.shields.io/badge/python-3.10%2B-315B78)](#quick-start)

A focused research toolkit for finite congruence lattices. Search colored graphs,
verify every quantified condition with an independent implementation, recover
the corresponding unary algebras, and find their smallest operation lists on a
fixed representation. The core experiment uses only Python's standard library.

In plain language, we build small systems whose rules permit exactly a chosen
pattern of consistent groupings. Start with a three-object example in the
**[illustrated introduction](docs/PLAIN_LANGUAGE_GUIDE.md)** to see what that means.

The starting specification is the graph-witness criterion in
[OpenAI's math corpus, family 206](reference/README.md), pinned to an exact commit.
This is an independent companion project. The original graph experiment covers
eight specified lattices on carriers of size **one through four**. The atlas
classifies all representable grouping families through four objects, yielding
22 abstract lattice shapes.

**[Interactive demo](https://huggingface.co/spaces/Maverick03511/finite-congruence-checks)** ·
**[Understand the math](docs/PLAIN_LANGUAGE_GUIDE.md)** ·
**[Why the saved N5 rules work](docs/N5_SOURCE_CONTEXT.md)** ·
**[Readable atlas](https://maverick03511-finite-congruence-checks.static.hf.space/atlas.html)** ·
**[Quick start](#quick-start)** · **[Evidence gallery](docs/index.html)** ·
**[Exact claims](docs/CLAIMS.md)** · **[Lean status](docs/LEAN_VERIFICATION.md)** ·
**[Agent repair experiment](docs/REPAIR_BENCHMARK.md)** ·
**[TM translation checks](docs/TM_TRANSLATION_CHECKS.md)** ·
**[Compare Turing examples](docs/TM_ATLAS.md)** ·
**[Snaky on finite boards](docs/SNAKY_FINITE_BOARD.md)** ·
**[Changelog](CHANGELOG.md)** · **[Support](SUPPORT.md)**
Research tools for exact mathematical claims: finite constructions, independent
checks, and scoped Lean proofs. Compare representations, inspect their hypotheses,
and trace a result back to its source and reproducible evidence.

The project builds on selected results from **[OpenAI Math](reference/README.md)**,
starting with the graph-witness criterion in family 206 and extending to exact
Turing-machine examples and Snaky on finite boards. Source revisions and claim
boundaries are recorded separately for each result. This is independent work;
source attribution does not imply OpenAI affiliation or endorsement.

The broader goal is to discover useful technical capabilities and foundational
technology through sustained mathematical work. AI architectures are one possible
direction. Mathematical progress, failed approaches and reusable methods inform
what to investigate next; novelty and performance gains remain research goals.

**[Explore the research atlas](https://huggingface.co/spaces/Maverick03511/finite-congruence-checks)** ·
**[Inspect exact claims](docs/CLAIMS.md)** · **[Reproduce a result](#quick-start)** ·
**[Read the Lean evidence](docs/LEAN_VERIFICATION.md)** · **[Research roadmap](docs/ROADMAP.md)**

The repository is the canonical source. The hosted Space is a separate deployment;
see the [publication record](docs/HUGGING_FACE_DEMO.md#publish) for its recorded revision.

| Research capability | Evidence and limits |
|:--|:--|
| [Enumerate finite constructions](docs/ATLAS.md) | 39 representatives and 22 abstract shapes through four objects; independently checked finite computation. |
| [Compare exact claims](docs/ATLAS_QUERY.md) | Stable entry IDs, fixed-carrier operation minima, hypotheses and source citations. A shared shape name does not transfer a proof. |
| [Inspect proof correspondence](docs/CONTEXT_QUERY.md) | Exact saved N5/B3 constructions, proved N5 renamings and selected Turing claims; each has its own scope and binding. |
| [Follow finite-board Snaky](docs/SNAKY_FINITE_BOARD.md) | The existing paper result is formalized for its 251-cell envelope and containing 17 × 17 square; an example play is separate from the 21-claim guarantee. |

The original graph experiment searches eight specified lattices on carriers of
size **one through four**, checks every quantified condition independently, and
recovers minimum unary-operation lists for fixed representations. It uses only
Python's standard library. Its totals below are separate from the atlas counts.

| Candidate colorings | Full witnesses | Vertex classes | Saved examples |
|:--|:--|:--|:--|
Expand All @@ -44,12 +48,14 @@ The HTML gallery is self-contained and works offline. Download the repository
and open `docs/index.html` in a browser; GitHub displays HTML source in its file
viewer. Each figure is also available as PNG and editable SVG in `charts/`.

## Understand the math visually
## Compare constructions and inspect evidence

Start with **[the readable atlas](https://maverick03511-finite-congruence-checks.static.hf.space/atlas.html)**:
three objects, one rule, and five possible groupings. Try a grouping, see why
it works or fails, then explore the complete catalog through four objects.
Every diagram explains what its nodes and lines mean; technical data is expandable.
Use the **[interactive atlas](https://maverick03511-finite-congruence-checks.static.hf.space/atlas.html)**
to move between the construction map, exact entry records and claim citations.
The current source keeps map, catalog, detail and export selection synchronized.
Its claim inspector separates saved finite computation from formal or model
checks that are not supplied for that entry. The yellow/black interface emphasizes
objects, rules and evidence; a worked example explains the diagram notation.

The catalog contains **39 grouping-family representatives** after renaming
objects, covering **22 abstract lattice shapes**. Compare two four-object
Expand Down
119 changes: 119 additions & 0 deletions WORK_LOG.md
Original file line number Diff line number Diff line change
@@ -1,5 +1,124 @@
# Finite graph search work log

## 2026-10-20 - Repair contrast-checker false passes after independent review

Independent controls exposed CR-R1 (first declaration incorrectly winning an
equal-priority tie inside one rule) and CR-R2 (an unsupported `:not` color
selector silently treated as inactive). Both failures were reproduced before
repair. Cascade priority now includes declaration position after rule order.
Selectors for modeled color/custom-property rules are validated before component
matching, so unsupported syntax fails explicitly even without a matching fixture.
Only the existing metrics separator pseudo-element is explicitly excluded.

The original reviewer controls now pass. The CI selection checker retains all
282 assertions and runs 224 color pairings, eight contrast regressions, four
unsupported-selector rejection controls and two valid cascade controls, including
earlier `!important`. Presentation CSS and selection code are unchanged from the
first contrast handoff. The checker remains a bounded source fixture; its explicit
subset and exclusions are documented in the repair handoff, with no browser or
full CSS/accessibility acceptance claim.

The original 87-file evidence manifest and every frozen artifact are preserved;
four reviewed source snapshots and before/after control logs are retained in
`runs/atlas-contrast-20261020/repairs/CR-v2/`. Focused syntax, source/selection checks
and the reviewer's controls pass. No unchanged mathematical suite was repeated,
and no proof/model, browser or remote action occurred. Independent repair review
remains separate.

## 2026-10-20 - Verify the full Atlas cascade and repair keyboard-link contrast

The PR19 review named six pale surfaces, but later unconditional rules in
`atlas.css` already replace those fills with theme tokens. Reading only the
earlier minified rules reproduced neither their final colors nor the reported
general failure. A source-level cascade audit instead found a real residual
problem: the keyboard skip link inherited yellow text over a literal white
background (1.293946:1). It now uses the dark surface and text tokens. SVG order
edges also receive an explicit system-color fallback under forced colors.

Added `check_atlas_contrast.cjs`, invoked by the existing CI selection checker.
It resolves the checked stylesheet subset in actual load order against static
HTML and generated claim/operation/SVG markup, including normal, hover, selected
and keyboard-focus states at two widths and both color-scheme preferences.
It retains the separate system-color checks and seven deliberately injected
regressions. This closes the gap in the earlier token-only contrast checks.
Unsupported solid-color values fail; this is a bounded source fixture, not a
complete CSS engine, browser rendering, native-widget or accessibility audit.

Author validation: 224 component pairings pass (minimum text 6.73:1 and checked
SVG/focus 7.83:1), as do all 282 existing selection assertions and the other two
Atlas checkers. All 33 root tests pass; the required finite reproduction and
evidence comparison agree with the recorded results. Original sources, failing
skip-link control, cascade diagnosis, commands, logs and exact handoff hashes are
retained under `runs/atlas-contrast-20261020/`. No proof/data, remote publication,
model execution or denied browser-preview workaround occurred. Independent
review of these exact bytes remains a separate gate.

## 2026-10-20 - Repair runtime reproduction commands after independent review

Independent review found OR-R1: Atlas initialization replaced the corrected
static reproduction list with an older string, so the displayed list and Copy
commands omitted `node scripts/check_atlas_selection.cjs`. The earlier source
checks did not cover this user path. Added the missing runtime command and four
regression assertions for the runtime/static lists, complete clipboard API input,
and reading the current displayed text on each copy action.

The review also identified that filtering push branches alone suppressed tag
pushes. Added `push.tags: ['**']` so release-tag verification is retained while
feature-branch/PR duplication is removed. Every existing job and check remains.

The new regression failed against the preserved original runtime source, then
passed after repair. All three affected Atlas checkers pass; the selection suite
now has 282 assertions. The reviewer's original focused control also passes.
Clipboard writes are mocked in the DOM fixture: no actual browser clipboard,
rendering or download acceptance is implied. No Python reproduction, proof or
model job was repeated.

Versioned repair evidence is in
`runs/online-atlas-repair-20261020/repairs/OR-R1-v2/`. The original 97-artifact
manifest, report, source snapshots and handoff remain byte-for-byte unchanged;
failed-control output and original source are preserved. This repair awaits the
same independent reviewer's clearance. No commit or remote action occurred.

## 2026-10-20 - Prepare the research atlas for online delivery

The map, catalog, detail and export previously maintained inconsistent selected
entries. Reused the independently reviewed five-file Atlas integration on a
separate worktree at main `0102845fc6a91c46e4b2def7455b6d418e671c01`; that tree
matches the patch's original base. The implementation owner's retained dirty
checkout, helper and log were not changed. The new source shares selection and
shows exact construction/minimum records, hypotheses, citations and separate
unknown formal/model evidence gates.

Added the 278-assertion selection check to CI and reproduction guidance. All
existing CI jobs, required contexts, commands, timeouts and permissions remain;
push now targets main while pull-request and manual triggers remain, avoiding
duplicate feature-branch and PR proof pipelines. Updated the research-first
README, yellow/black Atlas and Space copy, and roadmap. The aim is mathematical
progress and useful foundational technology, without predetermining AI
architectures or asserting novelty. HF card colors use documented yellow/gray;
the atlas stylesheet is yellow/black.

All three Atlas checkers passed, including 39 entries, 518 grouping cases,
304 rejection witnesses, 117 claim links, 115 map-filter combinations and 278
selection/inspector assertions. All 50 Atlas query tests and 33 ordinary root
unit tests passed. Standard finite reproduction also passed, including its own
33-test stage; its totals match saved evidence: 143,685 candidates, 217 witnesses
and 22 vertex classes. Atlas data/claim freshness, independent classification,
saved-evidence validation and 111 local documentation targets passed.

Evidence is retained in `runs/online-atlas-repair-20261020/`, including exact
patch/base identities, original-checkout preservation hashes, command logs,
source snapshots and review handoff. The first setup command was rejected by a
transport disconnection before process creation; a read confirmed no directory
or source mutation before retry. No product check failed in this allocation.

These are local source/data/fixture checks. This change does not publish the
Space, establish rendered/assistive-technology acceptance, execute Lean/model
work, or alter mathematical evidence. At the recorded 20 October read, the
Space was still at `63a85954e68f4e0202db2382c875267f9ae03f40`, predating Snaky.
Exact-byte independent review, hosted CI and publication acceptance remain
separate. Substantial AI assistance was used.

## 2026-10-19 — Fix a Linux process-disappearance race in the cleanup test

The push run for `4102fc1ec98a8516182270765e730a04ec626295` failed its
Expand Down
2 changes: 2 additions & 0 deletions docs/ATLAS.md
Original file line number Diff line number Diff line change
Expand Up @@ -76,6 +76,8 @@ python scripts/check_atlas.py --check-receipt evidence/atlas-review.json --out r
python scripts/query_atlas.py build-claims --check
python -m unittest discover -s scripts -p 'test_atlas*.py' -v
node scripts/check_atlas_ui.cjs
node scripts/check_atlas_map.cjs
node scripts/check_atlas_selection.cjs
```

The exporter starts from the complete small-family enumeration used by the
Expand Down
14 changes: 9 additions & 5 deletions docs/ATLAS_MAP.md
Original file line number Diff line number Diff line change
Expand Up @@ -67,15 +67,19 @@ either minimum a universal lower bound for that abstract shape.
```sh
node scripts/check_atlas_ui.cjs
node scripts/check_atlas_map.cjs
node scripts/check_atlas_selection.cjs
python scripts/query_atlas.py build-claims --check
```

The new check loads the real vendored Cytoscape build in headless mode, checks
The map check loads the real vendored Cytoscape build in headless mode, checks
all graph components and entry-to-claim links, exercises seven rejection
controls, verifies library/license hashes, and checks light/dark theme contrast
for the specified text, controls, and graph colors. These automated checks are
supplemented by desktop and narrow-screen interaction review; they are not a
complete accessibility certification.
controls, verifies library/license hashes, and checks the base stylesheet's
light/dark contrast tokens. The selection check exercises the actual Atlas
scripts together in a DOM/headless fixture, including exact claims and export
selection. Neither check establishes rendered behavior or contrast under every
stylesheet override. The yellow/black presentation still needs browser, narrow-
screen, keyboard and assistive-technology acceptance; no complete accessibility
certification is claimed.

Cytoscape.js 3.34.4 is vendored with its MIT license and integrity manifest in
[`space/vendor/`](../space/vendor/README.md). No CDN or package installation is
Expand Down
17 changes: 15 additions & 2 deletions docs/HUGGING_FACE_DEMO.md
Original file line number Diff line number Diff line change
Expand Up @@ -7,8 +7,10 @@ seven passing witnesses, four connectivity obstructions, equivalence classes,
and minimal unary operations for each saved positive representation. Its
canonical source is `space/` in this repository.

The Space now opens the [readable representation atlas](ATLAS.md), beginning
with a three-object lesson. The original graph explorer remains at `index.html`.
The configured entry point is the [representation atlas](ATLAS.md), with a
construction map, exact claim inspector and worked example. The original graph
explorer remains at `index.html`. The publication record below identifies the
last recorded deployed revision; current source changes may be newer.
The atlas catalog covers all representable congruence families through four
objects and groups them by object renaming and abstract lattice shape. It has
separate generated data, exhaustive verification, and browser checks.
Expand Down Expand Up @@ -40,6 +42,7 @@ python scripts/query_atlas.py build-claims --check
python -m unittest discover -s scripts -p 'test_atlas*.py' -v
node scripts/check_atlas_ui.cjs
node scripts/check_atlas_map.cjs
node scripts/check_atlas_selection.cjs
python scripts/query_atlas.py build-tm --check
node scripts/check_tm_atlas_ui.cjs
python scripts/build_n5_reuse.py --check
Expand Down Expand Up @@ -101,6 +104,16 @@ independent checks and formal extensions of selected results. This banner was
published through PR #17 from GitHub commit
`3632a58d80320e56a7a44c017a74746008b920d8` to Space commit
`63a85954e68f4e0202db2382c875267f9ae03f40`; all 44 published file hashes matched.
A read of the Space README on 20 October 2026 still returned that commit in
`X-Repo-Commit`. This recorded deployment predates the Snaky source merge and the
yellow/black Atlas selection repair. No newer deployment is claimed here.

The current source adds shared selection and exact claim inspection. Its local
DOM/headless checks are distinct from browser rendering, responsive layout,
keyboard/assistive-technology behavior and actual browser downloads. Complete
those acceptance checks in a permitted context, retain exact-head independent
review and required CI, then publish the accepted merged `space/` files and
record their source/deployment revisions and file hashes.

Publish only the contents of `space/` to the static Space
`Maverick03511/finite-congruence-checks`. Its README declares `sdk: static` and
Expand Down
Loading
Loading