OPENMATH

Publication and maintenance

One feedback board, reviewed changes, and permanent publication history.

This repository is the source of the public atlas. Project text, paper editions, feedback, and deployments have different roles. A successful website build is not a review of a mathematical proof.

One feedback record

Use this repository's GitHub Issues for all reader discussion. General feedback is a separate category on the same board, without a required theorem or version. Project-specific forms preserve optional project, result, page, and edition context. A maintainer may clarify missing context in a reply; do not invent it.

Keep one thread per issue. When a duplicate appears, link the earlier thread and explain the closure. Add a resolution comment rather than rewriting the reader's original argument. Keep the contributor's attribution where permitted. Do not republish private messages or attachments without their consent.

The normal sequence is:

  1. Reader opens an Issue or comments on an existing one.
  2. Maintainer identifies its scope: general discussion, exposition, reference, mathematical correction, proposed proof, counterexample, or formalization.
  3. A proposed change is developed in a branch and Pull Request linked to the Issue.
  4. A reviewer checks the exact change and its affected dependencies.
  5. An authorised maintainer merges after the relevant checks pass.
  6. The revision is published, its live page is checked, and a resolution comment links the Issue to the Pull Request and permanent publication version.

Closing an Issue, labelling it, or accepting a Pull Request does not by itself establish a theorem. A proof status changes only with appropriate proof evidence. Labels are helpful but optional: the forms and website also identify categories from their title prefixes. No new editor permissions are granted by feedback.

Project acknowledgements and adopted feedback

Every current project has its own /projects/<project-id>/acknowledgements/ page. /acknowledgements/ is only a directory of those separate pages. Their sources have their own editorial history and do not change a mathematical edition, source receipt, theorem status or unsigned-paper policy.

For each feedback adoption, the publishing maintainer must:

  1. Inspect the original Issue and record the actual scope and explicit acceptance. Check any mathematical change and affected dependencies under the existing project review policy.
  2. Incorporate the change in a reviewed Pull Request or commit. The same release adds the credit record; acceptance for possible future work is not yet incorporation. A Pull Request link may be used when the implementing commit is the one currently being prepared; verify that it merges and deploys.
  3. Check the contributor's public attribution preference in the Issue or its comments. Do not infer a real name from an account profile or email address. named and username credit require a public consentUrl in that same Issue. Without that permission, use unattributed and omit the display name. The original public Issue link remains visible; this is not anonymous feedback.
  4. Append a stable ledger entry with the fields below, update the ledger's actual updatedAt date, run the build and checks, and inspect the generated page. Each record has its own stable ID. One Issue may support separate adopted contributions, including contributions by different people; keep each person's permission and the evidence for that contribution on its own record.
  5. After publication, link the live acknowledgement and incorporating revision in the original Issue's resolution. Keep private moderation notes, contact details and unpublished correspondence out of both public JSON sources.

Each adopted entry has these exact fields:

Use only supported feedback categories: general, question, correction, reference, suggestion, proof, counterexample, formalization. The validator rejects unknown projects, duplicate identities, missing acceptance or incorporation, inconsistent dates, unsafe or unrelated links, unknown fields, and named credit without a same-thread permission reference. These are structural checks: the maintainer must read the linked evidence. They do not establish mathematical correctness, novelty, authorship or endorsement.

The publishing Agent follows the same protocol. The website build renders the accepted records automatically, but it does not monitor or adopt future feedback on its own. No new scheduled review service is implied.

Corrections to attribution use a new reviewed commit with the same stable item ID and a concise explanation. Changing a current display name does not erase earlier Git history. Handle privacy requests under the separate privacy-removal procedure when ordinary public history would be inappropriate.

Editing and review

The repository owner controls write access. CODEOWNERS identifies ownership; it does not itself enforce approval. A branch ruleset requiring Pull Requests, code-owner review, and passing checks is a recommended owner setting, not an assumed or silently enabled feature. Never share an account or commit a token.

Publication versions

There are two kinds of identifier:

Do not conflate these numbers with mathematical truth status. A layout-only or privacy correction can require a new publication edition while leaving every mathematical claim unchanged.

For an atlas revision, save the previous complete bundle in content/history/<project>/v<revision>.json, then increment the current revision and record publishedAt and changeSummary. Existing history is append-only. Before increasing a formal edition, preserve the previous public edition as a frozen source snapshot with its manifest and exact hashes. Add a new directory rather than changing files beneath an old version URL.

Every release note records the date, affected project and result IDs, the change, its mathematical or editorial character, and related Issues or Pull Requests when they exist. Never invent issue numbers, review evidence, or a release date. Give each paper a matching HTML, TeX, and PDF edition. Keep stable current links and explicit version links; preserve old anchors or supply visible redirects.

A statement later found incorrect keeps its permanent ID, receives a correction or withdrawal notice in a new revision, and links to the previous statement and the reason for the change. Historical versions describe their historical status; the current page must not present a withdrawn result as valid.

Mathematical agent releases

The independent mathematical tools on AI & Math have a separate version track from mathematical results. The current pointer is public/agents/current.json; fixed archives and their SHA-256 manifest are in public/agents/releases/<release>/. No agent update changes a research theorem or formal paper edition by implication.

For a new agent release, first review the installed skill sources and their diagrams, exclude machine caches and personal Site bindings, then run:

python3 scripts/package-agents.py --source-root <absolute-installed-skills-parent> --release YYYY-MM-DD-short-name
python3 scripts/package-agents.py --verify
npm run build
npm run check

The packaging tool refuses an existing release directory. Inspect the public archives, including SKILL.md, diagram scripts, fixed SHA-256 hashes, name/privacy screen, installation guidance and all three site links. Publication requires a normal reviewed repository update. Previous ZIP URLs remain fixed; roll back the current pointer in a new Git commit only after confirming the intended release and preserving history. No background sync from personal agents to GitHub Pages has been configured.

Agent feedback uses the same GitHub Issues board as the mathematical atlas. General usability, documentation and feature comments need no theorem or edition; a source-specific mathematical objection should use the relevant project, permanent result ID and exact formal edition instead.

Release checklist

  1. Review the source diff and status/dependency changes. Check other projects are unchanged unless the Pull Request explicitly covers them.

  2. Build and test with the repository's scripts:

    npm ci
    npm run build
    npm run check
    
  3. Inspect the rendered overview, Progress, route pages, formal papers, equations, and PDF downloads. Check a narrow layout and keyboard navigation.

  4. Verify version links, permanent result links, canonical URLs, and the sitemap. Confirm paper hashes refer to the exact public bytes. For Beilinson–Bloch, rebuild and review papers in a new draft, then run npm run freeze:release. For Bedford–Taylor, use the audited reading-receipt import procedure below. Rebuild, then run npm run check and npm run check:release; the release check verifies both projects. Include each new edition’s frozen files and checksums, plus its source receipt where applicable, in the publication change. Do not copy old review or freeze markers into a draft.

  5. Audit public text, PDF metadata, TeX, JSON, and archive members for unintended names, private paths, email addresses, credentials, or private correspondence. Preserve legitimate external bibliographic attribution.

  6. Check general feedback needs no result ID, and mathematical feedback preserves the exact optional context. Opening a GitHub form must not pretend to submit it.

  7. Merge through the authorised GitHub workflow. Confirm the Pages deployment succeeded; then check the live project, version, feedback, and PDF URLs.

  8. Record the published revision and resolve the linked Issue with that URL.

The Pages workflow builds from main. Browser-visible success, automated checks, deployment success, and mathematical review are separate observations. Report only checks actually performed.

Beilinson–Bloch editions

The current project address is /projects/strong-green/; exact editions live under /projects/strong-green/editions/<edition>/. The current pointer is content/strong-green/current.json, and the corresponding Atlas project card must name the same edition.

Maintain mathematical records and full article HTML in content/strong-green/releases/<edition>/. Follow the paper rebuilding workflow to compile and inspect matching TeX, PDF and HTML in a new unpublished draft. Reconcile their hashes and affected mathematical records, then use npm run freeze:release for the completed Beilinson–Bloch edition. Keep the previous pages and downloads unchanged under their old version addresses.

The ordinary website build consumes the checked-in paper artifacts; it does not execute TeX or recompile contributor uploads. strong-green/views/ controls future page presentation; it must not change an already frozen edition.

Bedford–Taylor editions

The public project has a stable current address at /projects/bedford-taylor/ and exact editions below /projects/bedford-taylor/editions/<edition>/. The initial unsigned snapshot is v1.0.0; current publication v1.0.1 adds the completed feedback and source-scope review of 18 November 2026. Both preserve the recovered research statuses. It does not claim that the research programme or every recorded proof has been completed.

The source of a mathematical correction remains the maintained mathematical project. Revise the relevant definitions, statements and proofs there, check affected dependencies, and rebuild matching HTML, TeX and PDF reading copies. Do not correct a PDF or a rendered theorem in isolation. Run the unsigned reading-export audit before importing the next publication.

The import adapter uses an exact file allowlist, not a whole-workspace copy. For a newly audited reading export, capture a new receipt containing the source commit, edition, audit identity, paths, sizes and SHA-256 values:

node scripts/import-bedford-taylor.mjs --source <audited-reading-dist> --capture-receipt <absolute-new-path-in-content/bedford-taylor/source-receipts> --source-commit <exact-source-commit> --audit-report <passing-reading-export-audit.json>

Capturing a receipt records bytes; it does not approve them. Inspect the source diff and completed audit with the proposed publication change. Then create a new edition using the accepted receipt:

node scripts/import-bedford-taylor.mjs --source <audited-reading-dist> --edition <new-edition> --receipt <accepted-source-receipt.json>

The importer refuses to replace an existing edition and does not move an existing current pointer. After inspection, update content/bedford-taylor/current.json, preserve the previous Atlas bundle in content/history/bedford-taylor/, and increment the publication record in content/projects/bedford-taylor.json. Record the actual change and any related Issues or Pull Requests; leave those references absent when none exists.

Run the full build, tests and release check. The Pages workflow checks frozen file membership and hashes, source receipts, links, and append-only history before deployment. It does not execute uploaded contributor files or compile unreviewed TeX. Review PDF text and metadata when preparing the reading edition; the public build additionally verifies that it serves those exact PDF bytes.

The website's Versions page links each fixed edition and its checksums. The older Atlas outline is retained as an explicitly unsigned historical reading copy. The two inherited ZIP filenames both provide the current edition's unsigned reading-source package, as their README explains; their filenames do not certify a historical snapshot. Never describe them as untouched old archives.

Very large rendered HTML pages may be stored losslessly in a release's packed/ directory, indexed by packed-assets.json. The ordinary build restores their exact public paths and checks both compressed bytes and the original frozen manifest before using them. This storage choice does not alter the published file, its URL or checksum. Keep the packed bytes and index immutable with the release. A clean checkout must run the build before the read-only release check; do not hand-edit the restored, ignored HTML copies.

Bedford–Taylor research map

The current /projects/bedford-taylor/map/ presents a logical research map with three sections and 17 focused diagrams. Its source index is content/bedford-taylor/research-logic.json, with the typed node and relation data in logic-core.json, logic-descent.json and logic-energy.json beside it. The complete thematic register from research-map.json remains a secondary reading index, covering all 74 topics, 14 formal programme items and 26 definition groups. Its branch membership is navigation, not an implication.

Maintain the logical data against the actual statements and proofs in the selected mathematical edition. Nodes link directly to source statements or proof locations. Each relation records its premises, conclusion, kind, additional hypotheses, mechanism and source evidence. Inputs within one relation are joint requirements (AND). Distinct sufficient relations with the same conclusion are alternatives (OR), each retaining its own hypotheses. A proof dependency does not become a sufficient condition merely because it is cited. An implication with an open premise remains conditional; an open-step or obstruction is not a proved implication. State exactly which shortcut a counterexample excludes. An open question may have a linked context node without an incoming arrow when no established logical connection is available.

Keep all required hypotheses readable in the relation's text, rather than available only through hover. In particular, preserve actual analytic-zero premises, the rank bound on a union tuple, the supplied normal-three model, complete protected fibers, the specified background and domain, and coupled tail or coefficient restrictions. Retain the distinction between capacity-zero and pointwise conclusions, and between a completed weak solution and an actual classical solution. Source statuses and permanent result identities remain those of the edition. Diagram review does not supply new proof acceptance.

The checked-in diagrams live in content/bedford-taylor/logic-diagrams/ as clickable SVG files and their Graphviz .dot sources. Its manifest.json binds each SVG by SHA-256 to its bytes, selected edition, logical panel input renderer and shared logical-schema module. After changing the logical data or renderer, regenerate the artifacts in an environment with Graphviz's dot available:

npm run build:bt-logic
npm run build

Commit the regenerated SVG, .dot and manifest together with the reviewed source change. The ordinary website build validates the input bindings and SVG hashes and consumes these checked-in artifacts; it does not require or invoke Graphviz. A stale diagram must be regenerated, not accepted by copying an old input hash onto new content.

When selecting a new mathematical edition, review both logical data and the complete register against that edition before moving their edition bindings. The build rejects stale edition bindings and incomplete primary register coverage, and rejects any source topic or formal open item missing from the logical nodes. In the secondary register, keep each topic and formal programme item in exactly one primary branch; other nodes may cross-reference it. Map-only changes increment the Atlas project revision and preserve the exact previous bundle in content/history/bedford-taylor/. Git retains diagram, data and layout revisions. Frozen reading editions, their sources and PDFs retain their existing addresses and bytes.

Map feedback continues to use the same repository Issue board. Preserve the actual map URL and its precise diagram, relation or register fragment together with the selected mathematical edition. A map fragment is page context, not a new theorem ID. Each logical step also has a direct contextual link to this shared board. General feedback remains available without a required result or edition.

After changing the map, run the full publication checks and inspect desktop and narrow reading layouts. Verify arrow meanings, visible joint conditions, source links, keyboard access, exact feedback fragments, the secondary register, and the project navigation and homepage entry. Record the checks and live deployment actually observed in the release note; leave unfinished checks explicitly pending.

Overall Bedford–Taylor logical graph

The same map now starts with #total-logic. The principal-route view selects canonical relations through logic-overview.json; the complete view includes every node and relation. Selecting a relation always includes all its premises and its conclusion. The view never creates a direct implication by suppressing intermediate requirements. The 17 focused diagrams remain the stable home of their original explanations. Graph nodes link to #logic-node-<id>; connections link to the single canonical explanation and its exact source references.

logic-bridges.json records the separately inspected connections between branches, including explicit additional inputs. Each bridge has the same premises, kind, hypotheses, mechanism and source contract as an existing relation. General descent and finite energy currently meet at an open Green-product interface. A bounded-calculus proof input is not an assertion that bounded calculus alone supplies every geometric or analytic hypothesis.

lib/bedford-taylor-overview.mjs checks global identities, complete coverage, source-edition bindings and cycles across the separate components. The checked-in artifacts in overview-diagrams/ have their own manifest binding the exact view, full node and relation records, ambient section scopes, renderer and schema files. npm run build:bt-logic regenerates both the focused diagrams and the overall graphs. The ordinary build verifies both manifests without requiring Graphviz. Keep generated bytes and their reviewed inputs together.

The complete graph is a large canvas. Preserve readable labels, an exact node locator and the connected text view; fitting the entire graph to a screen is for orientation. On narrow screens, the text view exposes the same joint inputs, typed connections and complete additional hypotheses. Test the locator, fit/readable scaling, full-height mode, keyboard links and hash disclosure. Overview feedback uses the existing board with #total-logic; cross-branch feedback retains its precise relation fragment. Neither creates a theorem ID.

An overall-graph update advances the Atlas presentation revision and preserves the previous project bundle. It does not change the mathematical edition or the source acceptance status. Record the actual source review and live UI verification in docs/releases/, separately from the formal mathematics.

Beilinson–Bloch Progress proof diagrams

The current /projects/strong-green/results/proof-diagrams/ and its per-result pages are source-bound graphical reading views, separate from frozen paper editions. Their coverage includes every case and application in strong-green/views/content/pairing-progress.mjs. The progress-proof-*.mjs source modules provide the proof trees; progress-proofs.mjs joins them to the existing Progress inventory and validates coverage.

progress-proof-required.mjs is an independent minimum inventory of the reviewed proof trees and the source files required by each result family. Extend it after reviewing new units; do not regenerate or shrink it from the candidate graph merely to make a deletion pass. It protects detailed trees which are not themselves application entry points. Source locators are nonblank string arrays: existing descriptive entries are reading notes, not automatically parsed TeX labels. Their exact mathematical targets still need source review. Scope/context boxes cannot contain an inference; put every derivation in a drawn root, premise or related argument.

Maintain the exact source file hashes and proof locators when revising a graph. Do not recompute a binding merely to bypass a stale-source check: reread the changed statement and argument, check each affected inference, then record the new binding. Preserve ALL-premise bundles, alternative routes, equivalences, open bridges and blocked inferences. A source's conditional status remains conditional in its diagram. Intermediate diagram nodes are not newly accepted theorems.

Every application has a stable diagram anchor and a link from Progress. Preserve those anchors when refactoring; give each diagram its correct SVG download, title, canonical URL, search record and feedback page context, including a valid linked diagram fragment. A fragment is not a new theorem ID. Never overwrite an old mathematical edition to add these reading pages. Run the complete repository tests and release/history checks after changing the renderer or the proof data, and inspect representative long diagrams and their source links.

The current interface follows the Bedford–Taylor logical map: parallel inputs appear above a numbered AND/OR/IFF junction and its conclusion, with open and blocked links explicitly distinguished. Each authored tree has an always-visible graphical panel. Keep node labels compact; the complete node notes, hypotheses, rule notes, connection explanations and sources belong in linked HTML details immediately below that panel. These details supplement the drawing, not replace or hide its branches. Preserve the old tree, occurrence and junction IDs; append -detail for their new explanation targets.

Graphviz creates the checked-in coordinate cache in content/strong-green/logic-diagrams/. Regenerate it with npm run build:sg-logic after changing a bound layout or logical source. The normal build verifies its manifest, source hashes and geometry hash without requiring Graphviz. A stale cache must fail the build; do not silently use a fallback layout for an existing published tree. Synthetic test fixtures may use the deterministic top-down fallback. The cache stores geometry, never a second copy of mathematical prose or statuses.

Keep the SVG's DOM reading order aligned with its visible top-down order. Diagram fragments must move keyboard focus as well as the viewport and open any containing explanation details. Each panel has independent two-axis scrolling and zoom. Preserve its reading coordinates when changing scale, leaving other panels unchanged. Fit-width mode may make text small, so retain a 100% readable-size control and keep the diagrams visible on narrow screens. Explicit fit may go below the manual zoom range; label such a view as an overview. Reserve a stable scrollbar gutter so height-driven scrollbar changes cannot oscillate the fit width. Automatic resize must not restore the page's scroll position, and a user scroll must win over a queued automatic canvas adjustment. Test these cases independently from explicit zoom anchoring. Standalone SVG downloads pin paper and permanent-record evidence to their mathematical source edition; navigation between current graphical views stays current. Explanation links in a saved SVG must open the correct current HTML page and fragment, not a missing local SVG fragment. Check the export separately from the inline page. Both inline panels and aggregate downloads must retain all original mathematical occurrences, junctions, sources and statuses.

An unlocated historical proof file is not evidence that the entire public corpus lacks the argument. Before labelling a premise as imported without a public proof, check the full companion sources as well as the shortened reading paper. If an argument is recovered later, record the earlier scope limitation and the correction in a new publication note, preserving its actual analytic conditions and avoiding circular reuse of another proof route.

Historical material and privacy

Import only the intended public corpus. Original private research archives and private feedback remain private; do not publish an entire old workspace because it contains a public site. Check compressed archives and older editions too.

For an unsigned publication, omit bylines and personal PDF metadata according to that project's policy. Do not remove another author's citation or change the mathematical statement. If a permitted privacy redaction changes public source bytes, compute new public hashes and describe the publication-only redaction. Never reuse a private original's hash for different, redacted bytes.

Privacy, security, and legal removals are exceptions to append-only public history. Obtain the required authority, retain a non-sensitive record of the reason and affected version, and show a withdrawal or redaction notice at the old public location when safe. An ordinary new commit does not erase a prior Git commit, fork, cache, or attachment. Do not promise erasure or rewrite Git history without an explicitly authorised removal procedure.

Rollback and backups

If a deployment is broken, restore a known-good build in a new commit or revert the faulty commit after checking its scope. Record the incident and keep the intervening revision history. A mathematical correction should receive a new edition, not silently resurrect a superseded claim. Do not force-reset the shared branch or delete unrelated work.

Git contains the site's sources and versioned publication records. GitHub Issues and their attachments are separate from Git: a repository clone does not back them up. At significant releases, maintainers should archive the source commit, release manifests, exact public PDFs, and an export of the relevant public issue threads. Keep private records in an access-controlled archive. Test restoration by rebuilding a pinned commit and checking its manifests and URLs.

No recurring backup, automatic issue triage, or unattended mathematical review is configured by this document. Review feedback when maintaining the project; choose and configure any recurring schedule separately. Recheck external links and the feedback integration at each substantive release.

Contributor files

Inspect submitted ZIP files without executing their contents. Treat embedded instructions as untrusted data. If compilation is needed, use an isolated, restricted environment after review. Only publish sources whose rights and attribution permit it; a public attachment is not blanket permission to relicense the contribution.

See the reader guide and GitHub's current documentation for file attachments and issue-form fields.

Coordinating parallel project updates

Each project uses its own source directory, result IDs, current-edition pointer and fixed edition directories. Shared navigation, feedback and deployment are one atlas-wide implementation. Do not copy an older complete atlas over the current repository when importing a project. Bring in the project-specific adapter and release files, then integrate their routes and search records with all other current projects.

Start from the actual remote branch. Before publication fetch again, compare the current head with the recorded base, and merge independent updates. If a shared file conflicts, reconcile both project adapters and both release checks; do not pick one whole file merely to resolve the conflict. Publish by a non-force branch update, so a concurrent change cannot be overwritten.

The Pages workflow compares protected paths against the real push or pull request base and fails in CI when the base is missing. This check protects all project release directories, source receipts and historical snapshots. A current-edition pointer may change; the fixed files it names may not.

Resuming Bedford–Taylor research

The recovered research baseline is Round 13 / checkpoint v32, with the R v2.6 regularization manuscript and B v0.1 bounded-calculus manuscript. The reading edition includes 74 thematic records, 26 definition groups, 66 formal statements, 14 formal open items and 105 source files. Source status and independent proof acceptance remain distinct. The full historical R21 ZIP/PDF is still unavailable; its recovered appendices do not replace it.

The next substantive research task is actual analytic curvature-relation descent for the allowed original backgrounds. The supplied normal-three quadratic model gives vanishing away from the centre and a central coefficient expressed by the tangent pairing. The general vanishing of that coefficient remains open. Preserve these hypotheses, the original field and absolute value, full test classes and ordinary locality in future work. Update an exact source and its affected dependencies before changing a public status.