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
2 changes: 1 addition & 1 deletion README.md
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,7 @@ Optional provider configuration is described in `.env.example` and `app/llm/`. K

Use **Settings → Check inputs** or `POST /api/validate` with `{"model_id":"..."}` or `{"model":{...}}`. Inspection covers names, references, syntax, initial values, flow endpoints, tables, settings and a short numerical run. Units and provenance are retained. Scientific validity is not established by these checks or by finite simulation output.

The curated public library contains **995 examples: 992 dynamic models and three explicitly labeled static calculators**. Curation retired 155 failing, incomplete, degenerate or redundant definitions from the served library while preserving their exact bytes as test evidence. Five flowless examples were reconstructed from attributable equations and checked independently. Every active example must complete its full default run; numerical completion is not empirical validation. See [curation decisions](docs/LIBRARY-CURATION.md), [reconstructions](docs/MODEL-RECONSTRUCTIONS-2026-10.md), and the historical [failure register](docs/FAILURE-REGISTER-1.9.md). Additional user models are preserved during release.
The curated public library contains **996 examples: 993 dynamic models and three explicitly labeled static calculators**. Curation retired 155 failing, incomplete, degenerate or redundant definitions from the served library while preserving their exact bytes as test evidence. Five flowless examples were reconstructed from attributable equations and checked independently. Every active example must complete its full default run; numerical completion is not empirical validation. See [curation decisions](docs/LIBRARY-CURATION.md), [reconstructions](docs/MODEL-RECONSTRUCTIONS-2026-10.md), and the historical [failure register](docs/FAILURE-REGISTER-1.9.md). Additional user models are preserved during release.

## API and MCP

Expand Down
9 changes: 9 additions & 0 deletions docs/FOURTH-ADMISSION-HISTORY-REVIEW.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,9 @@
# Independent fourth-admission historical selector review

2026-10-24. Reviewer ci_repair_resume independently approves the scoped historical-test correction after the complete sanitized public run exposed the old current-count assumption. No runtime, model definition or historical receipt changes. All29 original retirement/replacement/archive/source/review assertions remain intact; the immutable follow-up receipt explicitly remains993 active.

Exact reviewed test SHA2561c60f4431011284e9a391d1a7234d277605db24b1a11178215d01d32b8043bc3; added tests/fixtures/fourth-admissions-20261024/ledger-before.json SHA2569b28aa91078e2f151a4be915c62672ba80640f7a7c6cb9000f9d5c0751f77010. Independently verified that fixture is byte-exact from the pre-adoption ledger at research commitcea243259744772eea866de0d1e465c8469c141e, not a reconstructed or edited historical receipt.

Historical77admitted/993active accounting is tested against that fixed-hash ledger and every original993 current file hash. A separate new test requires current cumulative rows to be exactly the historical77 plus the three reviewed fourth-admission rows, current80/996 and923+80-7 accounting, each new source/hash/review binding, disjoint old/new path sets and the exact complete996 current model path set. Arbitrary extra files, altered new bytes or changes to original models cannot be silently discounted. Shared baseline selection still yields916 and no historical scientific assertion is weakened.

Independent focused run:26passed4.23s, covering quality follow-up, cumulative admissions, new mechanism/adoption tests and public validation. A complete sanitized public rerun remains required before publication. The prior391-test historical selection omitted this follow-up file; it must not be described as a complete public-suite certification. The earlier public corpus-path failure was caused by test setup before new paths were staged, independently passes once tracked, and requires no test change.
9 changes: 6 additions & 3 deletions docs/LIBRARY-ADMISSIONS-2026-10.md
Original file line number Diff line number Diff line change
@@ -1,8 +1,8 @@
# Reviewed model admissions — October 2026

The public source library contains 993 models: 990 dynamic examples and three explicitly static calculators. Three batches add 77 independently reviewed examples to the 923-model curated baseline. Source-quality reviews retire seven unsupported or redundant definitions in total, including two May parameter variants, and replace reviewed canonical examples: 923 + 77 - 7 = 993. The 26 later citation corrections change descriptions only. Historical admissions, the earlier five quality retirements and the original 155 retirements retain their exact evidence and archives. Model count does not justify duplicates, unsupported equations or nominal variations. Local source adoption is separate from public deployment; see the May migration document and quality-follow-up manifest.
The public source library contains 996 models: 993 dynamic examples and three explicitly static calculators. Four batches add 80 independently reviewed examples to the 923-model curated baseline. Source-quality reviews retire seven unsupported or redundant definitions in total, including two May parameter variants, and replace reviewed canonical examples: 923 + 80 - 7 = 996. The 26 later citation corrections change descriptions only. Historical admissions, the earlier five quality retirements and the original 155 retirements retain their exact evidence and archives. Model count does not justify duplicates, unsupported equations or nominal variations. Local source adoption is separate from public deployment; see the May migration document and quality-follow-up manifest.

The additions comprise five mechanistic numerical benchmarks, twenty biological systems, thirty-one feedback/control examples, ten environmental process-law adaptations and eleven further physical/scientific examples. Each admitted file is an exact copy of its independently reviewed candidate, with a SHA-256 approval and source path in `operations/library-admissions-20261024.json`. The second twenty are separately recorded in `operations/library-admissions-second-20261024.json`. Candidate source, license notices, numerical evidence and independent review are retained alongside each batch. Admission tests enforce exact bytes and IDs, and cross-batch duplicate checks exclude only the candidate being checked.
The additions comprise five mechanistic numerical benchmarks, twenty biological systems, thirty-one feedback/control examples, ten environmental process-law adaptations and fourteen further physical/scientific examples. Each admitted file is an exact copy of its independently reviewed candidate, with a SHA-256 approval and source path in `operations/library-admissions-20261024.json`. The second twenty are separately recorded in `operations/library-admissions-second-20261024.json`. Candidate source, license notices, numerical evidence and independent review are retained alongside each batch. Admission tests enforce exact bytes and IDs, and cross-batch duplicate checks exclude only the candidate being checked.

The numerical checks use handwritten independent derivatives, SciPy reference solvers, matrix exponentials, exact recurrences, analytic equilibria, conservation identities and step refinement where applicable. These tests check the described mathematics and implementation; they do not establish empirical accuracy or suitability for forecasting. Illustrative parameters, simplified closures, inaccessible original sources and incomplete unit metadata remain disclosed in the batch documentation.

Expand All @@ -11,4 +11,7 @@ Five short chaotic/neural examples remain outside the active library: their curr
Batch evidence: [mechanistic](LIBRARY-CANDIDATES.md), [biology](LIBRARY-CANDIDATES-BIOLOGY.md), [applied control](LIBRARY-CANDIDATES-APPLIED.md), [environment](LIBRARY-CANDIDATES-ENVIRONMENT.md). The historical curation manifest is unchanged; the admissions manifest extends it explicitly.


The final batch and quality actions are recorded in `operations/library-admissions-third-20261024.json` and `operations/library-quality-20261024.json`. Five retired definitions and the replaced original remain byte-exact regression fixtures. The quality review takes priority over reaching exactly 1,000; finite execution alone did not justify their scientific claims. See [review](../operations/remaining-library-quality-review.md) and [independent repair/retirement approval](../operations/legacy-quality-repairs/PEER-REVIEW.md).
The third batch and its quality actions are recorded in `operations/library-admissions-third-20261024.json` and `operations/library-quality-20261024.json`. Five retired definitions and the replaced original remain byte-exact regression fixtures. The quality review takes priority over reaching exactly 1,000; finite execution alone did not justify their scientific claims. See [review](../operations/remaining-library-quality-review.md) and [independent repair/retirement approval](../operations/legacy-quality-repairs/PEER-REVIEW.md).


Fourth-batch source admission: three distinct independently reviewed examples add the free Chaplygin sleigh, gyroscopic Foucault pendulum and driven optical Bloch atom. Exact approved hashes/source destinations and all993 pre-admission hashes are retained in operations/library-admissions-fourth-20261024.json. The candidate package retains primary attribution, rights limits, analytic/conservation/limiting tests and independent review. No old model was changed or retired; no empirical calibration is asserted. This local adoption is separate from any public release.
22 changes: 22 additions & 0 deletions docs/LIBRARY-FOURTH-ADMISSION-2026-10.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,22 @@
# Fourth science admission: three reviewed additions

Local source now contains996public definitions:993dynamic and3explicit static calculators. Historical accounting is923+80admissions-7retirements=996. This adds only the exact reviewed free Chaplygin sleigh, Foucault small-angle Coriolis pendulum and optical Bloch two-level atom. All993previous active bytes remain unchanged. Existing editor/library/navigation and saved/import/export/user-model paths are preserved; the two public interfaces change their count text only. This source transaction is separate from deployment; production is not certified by this document.

The committed candidate research package cea243259744772eea866de0d1e465c8469c141e retains primary source attribution, explicit illustrative parameters, units, rights/access limits, exact input/output evidence and independent mathematical review. No publication PDF/text/figures/code is bundled. These are useful different mechanisms, not parameter variants to force1000. Admission ledger operations/library-admissions-fourth-20261024.json binds the three approved candidate-to-active copies and every previous model hash; cumulative ledger now records80admitted/996active. Historical retirement/admission receipts and candidate approval files are unchanged.

All996Git-tracked defaults passed against the frozen22-module typed-reduction engine in96.03s (fallback seed42). Full audit evidence is retained separately and bound by SHA256dd56704cbf144c338dde400ea8211bfada7107491314239c33bf7ff515dd1a63; source compact hash-bound index SHA25640d2c08fc402302ef9807c366d40ff8038c42e01db28ec7e9f0922cb111a7ca7. Public static/library-validation.json SHA25621e1c4d114e7b6ab6cd3781e0da74841ac7847bd75f74546d7af956e7b63c9ac extracts only public model id/hash/status and engine hashes in the existing schema. The compact audit export omits only its private full-report location; all evidence hashes and counts are retained. Finite completion is not scientific/empirical validity.

Focused new admission/public-index checks:23passed4.09s. The nineteen original mechanism tests are collected into the ordinary test suite without changing their frozen research copy. Independent candidate evidence additionally includes36parameterized cases/156series/positivity checks and consistent symbolic units. All prior candidate/admission/curation suites, including new cases, passed391tests in122.03s; historical selectors and hash-bound archives remain compatible. Independent source-adoption review is recorded separately after its final verification. No publicworktree/push/remote activation is part of this transaction. A future PR needs its own exact allowlist, CI, staged production-dependency/API/MCP checks and approved release, preserving all existing/private user definitions and database bytes.

The sanitized public compact audit at `operations/library-fourth-audit-20261024.json` has SHA-256 `7eb5f2c89a6d79783c78e33d3267947cbeb51970729ea3aba6235bad43322e48`; it differs from the reviewed source compact receipt only by omission of the full-report location field.

## Sanitized public package verification

The complete public suite passed3,110 tests and18subtests, with3expected failures,
in353.32s. This includes the fixed-hash historical ledger regression and current
996-model accounting. Actual local HTTP, legacy SSE and stdio MCP checks passed
996-model inventory and existing analytical/error/resource/prompt behavior.
Each transport retrieved and simulated all three additions against independent
analytical trajectories (1,001 sleigh samples,2,000 sampled Foucault points,
1,001 Bloch samples), without saves. All993 prior model bytes and the PR9 engine
remain unchanged. Exact-head CI and production staging remain separate gates.
27 changes: 27 additions & 0 deletions docs/LIBRARY-FOURTH-ADMISSION-REVIEW.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,27 @@
# Independent fourth science source-adoption review

2026-10-24. Reviewer ci_repair_resume, independent of adoption author legacy_repair_resume. **Approve the exact15-path local source-adoption package plus this additive review for scoped private commit.** No public branch copy/push, staging or deployment is approved by this review. Candidate mathematics/source/rights review remains operations/library-candidates-fourth-science/PEER-REVIEW.md; original research commitcea243259744772eea866de0d1e465c8469c141e is unchanged.

## Exact byte and preservation fences

The independently frozen source-adoption package binds all15 implementation/evidence/document/test paths. Public export retains those reviewed bytes except explicitly documented audit-location and prose omissions; the complete operator report remains separately hash-bound. Any numerical or source change requires renewed review.

Independently snapshotted all993 model hashes before author adoption (map digest16f80f0c8763b82119547e99b84186109859d4410113cc16dd0ded763c4428e9). Every original path/byte remains unchanged; the new admission manifest's complete before map exactly equals that independent snapshot. Exactly three new tracked paths exist, and each is byte-identical to its reviewed candidate with the approved SHA256:

- models/physics/chaplygin-free-sleigh-fourth-science.json:11ecc9e03478a01f5e635a25d74025d3795f9ad45e76e39c50b2d6eb58c0b73e
- models/physics/foucault-coriolis-pendulum-fourth-science.json:d732b3244a66b361aae4c56ec121397314932f0cb1aa8de949b4a8054f3dd45b
- models/physics/optical-bloch-two-level-fourth-science.json:597cc3eb8b4bfb7745b1fb45fa65f91ca0c61ff3278b6cd84b83c30fd5686401

The cumulative ledger preserves all77 old admission rows exactly and appends only these3; all other historical metadata except admitted/active totals is unchanged. Accounting923+80-7=996, with993 dynamic examples and3 static calculators, is consistent. Prior retirement/replacement receipts and source fixtures are unmodified. Both interface diffs are strictly993-to996 count text substitutions: navigation/features/content and runtime code remain intact. README also corrects its stale prior count. No private/user production definitions or database are in this source transaction.

## Numerical receipts and tests

Independently verified the full audit SHA256dd56704cbf144c338dde400ea8211bfada7107491314239c33bf7ff515dd1a63 and every ordered row against all996 current tracked paths/IDs/hashes. Every status is passed; evidence reports96.03s with fallback seed42. All22 core module keys/hashes match current frozen source. This reviewer checked extraction/hash bindings rather than rerunning the complete96-second audit.

The minimal public receipt SHA25621e1c4d114e7b6ab6cd3781e0da74841ac7847bd75f74546d7af956e7b63c9ac exactly extracts each ordered id/hash/status row and the same engine hashes into the existing schema. Source compact operator receipt SHA25640d2c08fc402302ef9807c366d40ff8038c42e01db28ec7e9f0922cb111a7ca7 retains full-audit digest and model/engine bindings. Public export omits only the full-report location field; the complete report is separately hash-bound.

Independently ran new candidate/adoption and public-index regressions:22passed3.73s, plus the cumulative-admissions test1passed0.15s. Author's combined23test run4.09s includes those same three files. Independently inspected the completed391-test historical candidate/admission/curation log:391passed122.03s. This historical run is author-executed evidence, not an independently repeated391-test claim. Shared tests/model_history.py and tests/library_admission.py remain byte-identical; no historical assertion/selector needed weakening. The shared helper automatically excludes only exact hash/source/review-bound cumulative admissions. Current public count assertions appropriately change to996. The new collector executes the frozen19 scientific tests unchanged and adds exact active/source/before-baseline fences.

Finite defaults and consistent symbolic units are numerical/metadata evidence, not empirical validity. The prior independent source review's36parameterized cases/156trajectory-or-positivity checks and explicit mechanism/domain/rights limits remain applicable to the unchanged candidate bytes. No publication/APIs/browser/production behavior is certified here. A future public PR must bind reviewed artifacts and sanitized transformations, run its own complete checks/CI/staging and preserve all existing/private definitions and database bytes. Production release is a separate decision; this source approval changes no live inventory.

The sanitized public compact audit at `operations/library-fourth-audit-20261024.json` has SHA-256 `7eb5f2c89a6d79783c78e33d3267947cbeb51970729ea3aba6235bad43322e48`; it differs from the reviewed source compact receipt only by omission of the full-report location field.
Loading
Loading