Publication control

A visible gate.
Not a hidden checklist.

This owner-only edition is a publication rehearsal. The ledger distinguishes machine evidence from the rights, privacy, accessibility, archival, and editorial decisions that still require human judgment.

9 gates0 machine pass9 open or partial0 author decisions pendingnot public ready

Claim boundary

This ledger records technical evidence and unresolved publication decisions. A machine pass is not a legal opinion, privacy consent, accessibility certification, or independent scientific validation.

partial

Durable domain and archive

Evidence now. The author selected a personally controlled canonical domain and approved immutable external deposits for every dated PDF. The exact provider-neutral packets are staged and verified.

Before public release. Complete the external deposits and record each resulting persistent identifier before announcement.

complete

Rights and licensing

Evidence now. The author selected the existing-notice policy: Categories for AGI retains its printed Apache notice; LINCS, Infinitesimal Creativity, and Discovering Compositional Worlds from Interaction retain their printed all-rights-reserved notices. PDF and HTML reading editions may be published; source TeX remains private. Asset-specific rights remain a separate visual-review gate.

Before public release. No book-level policy action remains; complete the separate checksum-keyed asset review.

partial review

Visual privacy

Evidence now. The four HTML editions contain 249 raster assets: 133 in Categories for AGI, 43 in LINCS, 41 in Infinitesimal Creativity, and 32 authored ORACLE diagrams. The prior checksum-controlled triage covers the first 217 assets; the new ORACLE assets are deterministic rasterizations of authored categorical diagrams.

Before public release. Extend the checksum-controlled visual register and sign-off packet to the 32 ORACLE diagrams, then complete the previously recorded rights, identity, and privacy decisions. The triage is not consent or approval.

partial

Machine accessibility checks

Evidence now. The semantic-edition pipeline supplies purpose-specific descriptions from authored captions or deterministic diagram provenance. The previous complete crawl covered the three-volume site; the 265-unit ORACLE edition and its 32 diagrams have passed conversion-level structural checks but require a refreshed rendered accessibility audit.

Before public release. Extend the machine-assisted rendered accessibility checks and representative-route packet to the fourth book, including its HTML contents, chapter pages, diagrams, search results, and PDF.

pending manual

Manual accessibility review

Evidence now. The existing manual-review packet identifies representative HTML surfaces and the first three PDFs. It predates the fourth volume, and none of its machine-assisted evidence is represented as a human accessibility certification.

Before public release. Add representative ORACLE surfaces and its PDF to the packet, then execute the full review with named reviewers using keyboard traversal, a screen reader, zoom, contrast inspection, reduced-motion settings, mathematical speech, figure interpretation, and PDF accessibility software.

complete

Corrections and moderation

Evidence now. The public four-volume repository https://github.com/sridharmahadevan/categorical-ai-books-corrections provides a structured issue form keyed by book, dated edition, stable tag, and correction type, plus an immutable-edition moderation contract and machine-readable accepted-correction register.

Before public release. Maintain the append-only correction register and incorporate accepted reports only in later dated editions.

partial

Experiment code publication

Evidence now. All 38 experiment families resolve to 625 family-scoped code records across 20 packages and expose 148 concrete, machine-checked Start here files; 15 package destinations are public. Public GitHub package and file links are pinned to full audited commit hashes rather than moving branches, and one representative code surface for each of the 13 public code-bearing packages returned HTTP 200. Every family has a stable HTML page and checksum-controlled Markdown and JSON records, with a reproducible ZIP archive. The five local companions contain 352 unique staged source files. Every private candidate also embeds checksum-controlled Markdown and JSON indexes that map each dependent experiment family to its staged entry points, source records, bounded result records, and evidence boundary. The Categories packet now executes the Chapter 6 DB energy notebook, verifies exact plot and producer hashes for the two Chapter 8 commutator-proxy figures under an explicit plot-only boundary, and retains its GT architecture and 87-cell table checks. Fresh Python 3.11 environments also pass the ATLAS/CSQL DuckDB fixture, the Deep URL CPU plus byte-for-byte report checks, and all seven staged Judo Calculus table artifacts without representing dataset reruns. A clean Python 3.11/NumPy 2.3.5 environment passes the Infinitesimal Creativity DIAL-core and CLIC CPU smoke contract: six exact core worlds, eighteen admission worlds, and eighteen held-out confirmation worlds satisfy every registered bounded metric.

Before public release. Apply the explicit disposition in localCodePackages to the five remaining local packages.

partial

PDF preservation

Evidence now. Four current PDFs have immutable dated paths, page counts, byte counts, and SHA-256 digests: the 478-page Categories, 304-page LINCS, 448-page Infinitesimal Creativity, and 292-page Discovering Compositional Worlds artifacts.

Before public release. Choose an archival mirror and preserve every later edition without overwriting.

partial

Production-origin verification

Evidence now. The four-volume release candidate builds successfully and contains 1,588 scholarly reading units, four current PDF records, 95 chapter explorations, and a searchable ORACLE edition with 265 stable units and 32 diagrams. The previous authenticated custom-domain and experiment audits remain valid only for the three-volume deployment they tested.

Before public release. Refresh the full production-origin crawl, visual register, accessibility packet, and authenticated custom-domain audit for the four-volume build before public announcement.

Rights exception queue

No active asset-specific exception.

The previously identified COCO-derived panels were removed at manuscript source. The collection-wide manual privacy and rights review remains open.

Owner decision matrix

Six choices cannot be inferred from a passing build.

Each recommendation is a conservative default, not an action already taken. Public access, licensing, archival deposits, and repository creation remain owner decisions.

decided · blocks archive-domain · production-origin

Choose the durable canonical public address and redirect policy.

Recommended default. Use a personally controlled custom domain as the canonical scholarly identity and retain the Sites origin as a recovery endpoint.

Alternatives
  • Make a custom domain canonical and redirect the UMass page to it.
  • Use the Sites origin temporarily, with an explicit migration and redirect commitment.

decided · blocks archive-domain · pdf-preservation

Choose an archive that preserves every dated PDF and its checksum independently of the website host.

Recommended default. Deposit immutable dated releases in Zenodo or an institutional repository and record the resulting persistent identifier in release_history.json.

Alternatives
  • Use an institutional repository as the primary archive and Zenodo as a mirror.
  • Use another preservation service that supplies immutable records and persistent identifiers.

decided · blocks rights-license

Confirm the rights notice for each book and whether source TeX is part of the public release.

Recommended default. Preserve the currently observed per-book notices, publish PDF and HTML reading editions, and keep source TeX private until publisher, employer, and third-party rights are settled. License code separately in each repository.

Alternatives
  • Adopt one explicit open-content license for all four books after a rights review.
  • Reserve all rights for every book while keeping public reading access free.

decided · blocks code-publication

Decide whether the five curated local code companions are launch requirements or may be labeled forthcoming.

Recommended default. Publish the high-priority Categories and Infinitesimal Creativity companions first; expose the three medium-priority packages as explicit, non-downloadable provenance records until their clean-environment and data-rights gates pass.

Alternatives
  • Delay the website until all five companions are public.
  • Launch the website with all five companions labeled forthcoming and publish them incrementally.

decided · blocks correction-channel

Choose a permanent correction and moderation home for the trilogy.

Recommended default. Use one public editorial repository for all four books, with issue forms keyed by book, edition, stable tag, and correction type.

Alternatives
  • Maintain one correction tracker per book.
  • Keep the current per-project trackers and add a dedicated Infinitesimal Creativity tracker.

decided · blocks visual-privacy · accessibility-manual · production-origin

Authorize the change from owner-only rehearsal to public access after manual review is signed off.

Recommended default. Keep the current deployment owner-only until visual privacy, keyboard, screen-reader, contrast, zoom, mathematical accessibility, and deployed-origin checks are recorded as complete.

Alternatives
  • Publish a limited beta URL with an explicit accessibility and correction notice after privacy and rights review.
  • Wait for every code companion and archival deposit before enabling public access.

Code-publication queue

Five archives remain local by design.

A local directory is never presented as a reproducible public package. Each archive has an explicit publication destination and a bounded preparation contract.

high priority · split release

categories-gt-archive

new curated Categories for AGI experiments repository or versioned release assets

  • retain the Chapter 8 plots as archived plot-only evidence unless their per-step raw logs are recovered
  • preserve Chapters 13--14 as archived-only unless complete raw artifacts are recovered
  • dependency lock
  • license decision
  • exclude raw corpora, checkpoints, caches, credentials, and unrelated parent-repository files

medium priority · merge or snapshot

kan-do-dcdi-archive

curated release within Causal-Discovery-on-Sheaves or a separately versioned companion

  • distinguish Judo Calculus runs from the public sheaf package
  • document Sachs, LINCS, and PISA dataset provenance and redistribution rights
  • repair and clean-rerun the PISA calibration driver
  • license and dependency review

medium priority · curate separate

atlas-csql-archive

small ATLAS/CSQL companion repository

  • choose a license
  • decide whether to retain the corpus-specific notebook and pin its separate environment

medium priority · curate separate

deep-url-archive

Deep URL reproducibility companion

  • choose a license

high priority · dedicated repository

synthetic-creativity-archive

new Infinitesimal Creativity code repository

  • separate reusable library code from one-off experiment drivers
  • rights and privacy review for visual assets
  • publish, vendor under an explicit compatible license, or exclude the six-file alfworld_bridge dependency
  • license and model-access decisions
  • manual security and privacy review beyond the narrow passing pattern scan

Reproducible companion audit · 2026-08-25

Private packets are coherent; publication is still a separate decision.

The audit recomputes every registered source, support, result, and archive checksum. It also verifies each package's experiment-family links against the documentation closure register. A structural pass is deliberately narrower than independent replication or permission to redistribute the software.

5 packages5 structurally verified4 bounded result packets5 archives verified26 experiment families15 open decisions

verified · blocked author decisions

categories-gt-archive

35 source files · 4 experiment families · 15 bounded result files

Artifact boundary. verified bounded artifacts.

3 open decisions
  • Choose the companion repository boundary and license
  • Recover per-step raw logs for the Chapter 8 plots if they still exist; otherwise retain their explicit archived plot-only status
  • Preserve Chapters 13–14 as archived-only evidence unless complete raw run artifacts are found

verified · blocked author decisions

kan-do-dcdi-archive

21 source files · 1 experiment family · 7 bounded result files

Artifact boundary. verified bounded artifacts.

4 open decisions
  • Decide whether the packet belongs in Causal Discovery on Sheaves or a dated snapshot repository
  • Choose a license compatible with the public companion
  • Document redistribution rights and provenance for the Sachs, LINCS, and PISA inputs
  • Repair the PISA runner's diagnostic formatting defect and repeat the retained experiments in a clean environment

verified · blocked author decisions

atlas-csql-archive

3 source files · 1 experiment family · 0 bounded result files

Artifact boundary. no result packet.

2 open decisions
  • Choose a license for the public companion
  • Decide whether the corpus-specific notebook belongs in the public packet and pin its separate environment

verified · blocked author decisions

deep-url-archive

5 source files · 1 experiment family · 33 bounded result files

Artifact boundary. verified bounded artifacts.

1 open decision
  • Choose a license for the public companion

verified · blocked author decisions

synthetic-creativity-archive

288 source files · 19 experiment families · 629 bounded result files

Artifact boundary. verified bounded artifacts.

5 open decisions
  • Separate reusable library code from one-off experiment drivers
  • Choose a license, dependency policy, model-access policy, and media provenance format
  • Complete manual security, privacy, and publication review beyond the passing narrow path, credential-pattern, media-type, and file-size checks
  • Publish or vendor the six-file alfworld_bridge dependency
  • Choose a publication or vendoring boundary for external `alfworld_bridge` from `GT_ALFWorld/alfworld_bridge`.

Author decision packet · 2026-08-25

Fifteen choices are explicit; none is silently inferred.

Every publication blocker in the structural audit now has a stable decision ID, a conservative default, two alternatives, and an empty author selection. Recommendations remain visibly distinct from authorization to publish.

Suggested sequence. Publish the two high-priority companions only after their package decisions and manual gates are complete. Keep the three medium-priority companions as non-downloadable provenance records until their individual gates pass.

5 packages15 registered decisions15 awaiting author

high priority · 3 decisions

categories-gt-archive

Categories for AGI learning experiments

1. Choose the public repository boundary and software license.

A · Conservative default. Create a dedicated Categories for AGI code companion and license the curated software under Apache-2.0; keep book text, raw corpora, checkpoints, and unsupported artifacts outside that repository.

  • B. Publish the curated packet as immutable release assets in the existing Categories repository, with an explicit software license and the same exclusions.
  • C. Keep the packet private and label the affected experiment families forthcoming.

categories-repository-and-license · awaiting author

2. Choose the evidence status of the two Chapter 8 commutator-proxy figures.

A · Conservative default. Publish the exact figures and producer hashes with an explicit archived plot-only boundary unless the original per-step logs are recovered and verified.

  • B. Delay this portion of the companion while searching for and auditing the original per-step logs.
  • C. Exclude the two figures from the public companion and retain only their manuscript provenance record.

categories-chapter-8-evidence · awaiting author

3. Choose the public evidence status for the Chapters 13–14 experiments.

A · Conservative default. Retain the experiments as archived-only evidence with no independent reproduction claim unless complete raw run artifacts are recovered.

  • B. Delay public release of these records until complete raw artifacts are recovered and audited.
  • C. Omit these records from the code companion and keep only the chapter-level evidence boundary.

categories-chapters-13-14-evidence · awaiting author

medium priority · 4 decisions

kan-do-dcdi-archive

Judo Calculus supplementary experiments

1. Choose whether the Judo Calculus packet is merged or independently versioned.

A · Conservative default. Publish a dated standalone snapshot that links to Causal Discovery on Sheaves but keeps the Judo-specific runners, evidence boundaries, and version history distinct.

  • B. Merge the packet into a clearly isolated Judo Calculus directory in Causal Discovery on Sheaves.
  • C. Keep the packet private and expose only its provenance record.

judo-repository-boundary · awaiting author

2. Choose a software license after checking compatibility with reused public code.

A · Conservative default. Use Apache-2.0 only after the dependency and provenance review confirms compatibility; otherwise keep the packet private.

  • B. Use the compatible license required by the selected upstream repository and document the reason.
  • C. Remove reused components and license only the independently authored snapshot.

judo-license · awaiting author

3. Choose the data-distribution boundary for Sachs, LINCS, and PISA inputs.

A · Conservative default. Exclude third-party datasets; publish provenance, retrieval instructions, checksums where lawful, and generated or rights-clear fixtures only.

  • B. Include only datasets whose redistribution terms have been reviewed and recorded file by file.
  • C. Keep the entire packet private until all three data sources have signed provenance records.

judo-dataset-rights · awaiting author

4. Choose the PISA experiment boundary for the first release.

A · Conservative default. Exclude PISA execution claims from the first public snapshot until the formatting defect is repaired and the retained experiment is rerun cleanly.

  • B. Delay the whole Judo packet until the PISA runner and clean rerun pass.
  • C. Publish the non-PISA portion and a non-executable PISA provenance record that states the unresolved defect.

judo-pisa-runner · awaiting author

medium priority · 2 decisions

atlas-csql-archive

ATLAS/CSQL worked examples

1. Choose the software license for the rights-clear ATLAS/CSQL companion.

A · Conservative default. Use Apache-2.0 for the curated SQL, generated fixture, and smoke-test code; apply no license to excluded source documents or unreviewed derived data.

  • B. Use MIT for the small standalone code packet.
  • C. Keep the packet private and expose only its provenance record.

atlas-license · awaiting author

2. Choose whether the corpus-specific notebook is part of the public packet.

A · Conservative default. Exclude the corpus-specific notebook from the first release; publish the generated synthetic fixture and deterministic DuckDB example as the executable core.

  • B. Include the notebook only after its corpus provenance, outputs, and separate environment are reviewed and pinned.
  • C. Keep the entire ATLAS/CSQL packet private.

atlas-notebook-boundary · awaiting author

medium priority · 1 decision

deep-url-archive

Deep URL reproducibility packet

1. Choose the software license for the curated Deep URL packet.

A · Conservative default. Use Apache-2.0 for the five audited source files and deterministic reproduction support; preserve the result manifest and evidence boundary independently of the license.

  • B. Use MIT for the standalone reproducibility packet.
  • C. Keep the packet private and expose only its provenance and checksum record.

deep-url-license · awaiting author

high priority · 5 decisions

synthetic-creativity-archive

Infinitesimal Creativity code companion

1. Choose the source boundary for the first Infinitesimal Creativity repository.

A · Conservative default. Publish the reusable DIAL library, registered experiment drivers, manifests, and bounded textual result records; place one-off exploratory drivers in a separately labeled archive or omit them.

  • B. Publish the full staged source tree with explicit experimental and archival labels.
  • C. Keep the companion private until the reusable API has been separated completely.

creativity-source-boundary · awaiting author

2. Choose the software, dependency, model-access, and media policy as one coherent release contract.

A · Conservative default. Use Apache-2.0 for authored code; pin dependencies; identify model APIs and checkpoints without redistributing weights; exclude image and video payloads until each item has an approved provenance record.

  • B. Use MIT for authored code while retaining the same dependency, model, and media boundaries.
  • C. Publish no code until every dependency, model, and media record is individually approved.

creativity-release-policy · awaiting author

3. Choose the manual-review gate that must pass before repository publication.

A · Conservative default. Require named manual security, privacy, confidentiality, and media-provenance sign-off on the exact candidate archive before any public repository is created.

  • B. Prepare a smaller source-only archive and run the same named review on that reduced boundary.
  • C. Keep the companion private and publish only its experiment-to-code metadata.

creativity-manual-review · awaiting author

4. Choose whether the six-file ALFWorld bridge is published, vendored, or excluded.

A · Conservative default. Exclude the bridge from the public archive until its authorship and compatible license are documented; preserve its checksum manifest and installation boundary.

  • B. Vendor the six files after recording an explicit compatible license and upstream provenance.
  • C. Publish the bridge as a separately versioned dependency and require it from the main companion.

creativity-alfworld-license · awaiting author

5. Choose the long-term dependency boundary for the external ALFWorld bridge source.

A · Conservative default. Treat the bridge as an optional external dependency with a documented version and checksum; do not copy it into the main repository without a compatible license.

  • B. Vendor the reviewed bridge under a clearly separated third-party directory with its own notice.
  • C. Remove ALFWorld-dependent experiments from the first public companion.

creativity-alfworld-dependency-boundary · awaiting author

Inspectable records

Download the gate itself.

The human checklist and machine ledger are published with the same artifact as the site, so release decisions can be reviewed rather than reconstructed from memory. The external-link packet separates 22 resolver-confirmed DOI links from nine completed direct-destination reviews and two verified TAC source corrections; five of the reviewed destinations have canonical replacements queued for the next dated edition. Human-readable gate → Machine ledger → Visual review queue → Machine asset register → External-link review → Machine link register → Experiment closure register → Machine experiment closure register → Code-companion audit → Machine companion audit → Author decision packet → Machine decision packet →

← Return to the edition registryInspect the experiment-to-code atlas →