Skip to content
Closed
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
The table of contents is too big for display.
Diff view
Diff view
  •  
  •  
  •  
13 changes: 13 additions & 0 deletions cbz/contributions/math514/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,13 @@
# Math514 CBZ contribution: exact-module kernel replay and project coverage

This contribution supplies a completed, bounded adapter to the installed Lean 4.34.1 `LeanChecker.replayFromImports`, plus a public-safe map of the eight Math Lake projects and Main core. Canonical federation integration, GOVERNANCE/TOOLING registration, full G2/G3/G4 coverage and final G0–G11 acceptance remain with main#684/#685. This directory declares no CBZ-0 result.

The kernel adapter runs one exact named module per fresh process, serially, with explicit source/object/runtime/context bindings and resource bounds. It preserves upstream import-only replay and all available exported/server/private parts. It rejects symlinked artifact/search paths and project-local checker shadowing. See [portable usage](kernel_replay/README.md), [mechanical scope](kernel_replay/SPEC.md), [registry payload](kernel_replay/REGISTRY.json) and [independent review](kernel_replay/independent-review.md).

Observed self-verification: 27 distinct applicable tests pass (25 from the final production-source run plus two newly added tests; not a repeated 27-test run). All 16 actual source mutations were killed by assertions, with no errors/skips/survivors. The original four mutation survivors, resource failures and the reproduced pre-fix symlink false-green remain retained and expressly invalidated as acceptance. No project proof replay occurred in building this tool. The canonical947 handoff already records a successful same-source complete Math-core gate with two threads at8151. Reuse that result; this adapter remains an exact-scheduling option for independently justified future coverage, without requiring another Math-core replay. G2 axiom policy is independent and remains NOT_ASSESSED by this adapter.

The negative fixture sources intentionally contain forged self-referential public/private theorem bodies or an unapproved axiom. They are isolated test inputs, never CLOSED mathematical declarations or production imports. Their expected rejection distinguishes a real checker from green-by-construction output. Preserve them and their historical failed-tool evidence.

[Project inventory](inventory/README.md) and [coverage gaps](inventory/TOOLING_GAPS.md) retain exact observation cuts and available recipes. This is discovery, not complete-scope certification. Stage67 mathematical source was published separately in Math#514 at86c00d4c08c6c2af0529ec1744693de3ca6ac6e4: one conditional kernel-closed theorem, with its unproved ordinary realization/controls precisely quarantined. Conventional prose is not promoted to a kernel proof.

For canonical registration, import the exact tool facts/commands/self-verification/limits/ledger payload from `kernel_replay/REGISTRY.json`. The registry payload is contribution evidence, not an edit to the federation's canonical rulebook. The separate manifest/isolation auditor is delivered in Main#697 with70 passing tests/53 assertion-killed mutants and focused postfix review; it is not included in this branch and actual repository G1 remains NOT_RUN.
52 changes: 52 additions & 0 deletions cbz/contributions/math514/inventory/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,52 @@
# Proof-project coverage contribution

Prepared for the canonical single writer, [main#684](https://github.com/d6g8k5htny-coder/main/issues/684)/[main#685](https://github.com/d6g8k5htny-coder/main/issues/685). This is a bounded contribution handoff, not a registry, integration decision or federation baseline declaration.

Original inventory cut: **2026-10-29T17:05:21.411546+00:00** across five Math worktrees and one Main checkout. Current contribution STATUS and the packaged stage67 closure overlay that cut. No new research, checks, tests, builds, kernel replays or source-digest runs were performed. Completed applicable evidence remains reusable; a new agent/session alone never warrants rechecking.

[proof-projects.json](proof-projects.json) maps exact repository-relative locations, recorded command/cwd recipes, evidence pointers and all G0–G11 gaps. [TOOLING_GAPS.md](TOOLING_GAPS.md) gives the remaining coverage boundary. No raw private code, local branch census or original absolute-path metadata is copied.

## Concrete stage67 contribution

Math- package `frontiers/source_derivative_gaussian_20261025` supplies `SourceActualLowerSeparation.lean`: **CLOSED as a conditional kernel theorem**, exactly one public proposition and four private theorems, with only `propext`, `Classical.choice`, `Quot.sound`. Literal global `IsMorseDistinct` remains an explicit hypothesis. No genuine fixture, nonvacuity witness, unconditional source application or Parent result is claimed.

Reused source identity:14899B / `e72ed4f92e7867dc210b771e6afb4548cfad679e8b38795110fa6fd76a15e121`. Reused own-object identity:409960B / `c001bad5446351036112f0c78b264fd3ffc423dc7df7fc09f9431bdbdcac43cc`. Strict006, exhaustive ALL5/std3 and proper own-module kernel exit0 are retained under `cbz/stage67/evidence/`. Kernel interval:2026-10-29T16:47:24.817525–16:47:29.785445Z; entry completed16:47:29.806374Z. The changed Lake registration subsequently built exit0 at17:33:58.953065Z. These observations are scoped evidence, not final federation gates.

The exact boundary is `cbz/stage67/PROOF_INDEX.md` and `CLOSURE.json`. `quarantine/Q-MATH-SDGA-67-ORDINARY/QUARANTINE.json` retains failed normalization source/diagnostics and absent genuine global-hm/positive/collision/mutation/False-control obligations. It is outside Lake/default targets, has no ordinary object and is not a production dependency. Preserve it unchanged for integration.

From the **Math- source-derivative package directory**, the already documented reproduction recipes are:

```sh
lake build SourceActualLowerSeparation
lake env lean -DwarningAsError=true cbz/stage67/evidence/Inventory.lean
lake env leanchecker SourceActualLowerSeparation
```

These recipes are provided for the canonical writer's mapping. No repeat of completed unchanged checks is requested. The installed prefix CLI is unambiguous for this module, which has no matching submodules; all-project replay still requires its own exact-module census/tool admission.

## Exactly eight Math projects, plus Main core

| Math- project path | Scope |
|---|---|
| `formal` | ResearchFormalCoreR1 scalar companions; registered targets do not alone establish ALL-declaration coverage. |
| `specifications/lean_statements_20261010` | Definitions, interface structures and registered Prop specifications; not theorem closure by definition. |
| `specifications/lean_statement_proofs_20261011` | Theorems of exact sibling specification Props; historical source records describe78 checked theorems, not proof of every registered Prop. |
| `frontiers/cap_first_exit_lean_20261002` | Cap/first-exit Lean package. |
| `frontiers/source_smooth_realization_20261011` | Actual smooth Gaussian source realization and source carrier. |
| `frontiers/source_lifetime_realization_20261017` | Literal source field/geometry and Lifetime interface bridge. |
| `frontiers/determinant_moment_20261025` | Determinant/moment package; shares package name with formal but is a distinct root. |
| `frontiers/source_derivative_gaussian_20261025` | Source derivative/conditioning/topology development, including exactly scoped stage67 contribution. |

Main's ninth distinct project location is **main:`formal/`**, package `UniversalLaw`, dependency-free Lean core with an empty package manifest. Math configs declare Lean v4.34.1 and Mathlib `d13f23b723b8a846827a245b89c10fc7d3f11612`; the proof packet additionally uses its sibling statement package. Declared pins are not installed-dependency or cross-repository resolution evidence.

At the cut, these give41 actual root occurrences plus five copies of a historical ten-module published lakefile,46 configs in13 variants. That historical metadata is **not a ninth active Math project**. Source-derivative default-target counts were14/18/29/66 on different cuts; the later stage67 registered-build evidence supersedes only the earlier unregistered-stage67 note, without claiming a fresh all-project census.

Non-Lake coverage includes the torus-filtration and three D2 companions, regular-superlevel project/negative-project modules, asymmetric elder, compact adapter inputs and review scratch. Their staged `formal/` environment, generated sources and LEAN_PATH must be represented explicitly; nearest-ancestor lakefiles do not prove build membership.

## What the inventory establishes

The original lexical inventory retained2842 Lean file occurrences in1976 deduplicated groups,39853 declaration candidates,807 Python math/float candidate groups,2745 status/source/pin surfaces and2763 conventional-claim documents. It retained archive/evidence/incoming material and exact locations privately with the canonical writer. These are discovery counts, not soundness or absence proofs.

Conventional prose and exact-certificate evidence remain valuable in their stated scope. They are separate from kernel-proved statements. Definitions/interfaces are specifications, not unverified axiom injections merely because their Props have no theorem. Missing geometric Kac–Rice/elder/density/Parent formalizations require explicit boundaries; impossible underdetermined universal statements must not be repaired by assuming their desired conclusions. Source-explicit binary64 certifying use needs A3 isolation; diagnostic float tokens and exact rational-grid calculations require their own dataflow classification.

All final G0–G11 remain **not established by this handoff**. Canonical admission, final input binding, full-tree coverage, regression, pins, status synchronization and clean-Git/cold-start acceptance stay with the canonical writer. Older tooling's kernel-pending and nine-active-project wording is dated and superseded as described above.
46 changes: 46 additions & 0 deletions cbz/contributions/math514/inventory/TOOLING_GAPS.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,46 @@
# Tooling and all-project coverage gaps

Contribution to main#684/#685; no integration or new-registry decision. Original lexical observation cut:2026-10-29T17:05:21.411546+00:00. Current contribution STATUS and packaged stage67 evidence provide the later explicit overlay. No new execution or re-verification occurred.

## Existing recipes to map

The following exact package-local drivers were recorded in math-tooling discovery. **cwd: Math- repository root.** Source-only mode is a scoped source/manifest check; execution mode may overwrite evidence/build directories and must retain historical failures under the canonical writer's preservation policy. This table records capabilities and does not request unchanged reruns.

| Source-only command | Execution command |
|---|---|
| `python -B -S formal/gate.py` | `python -B -S formal/gate.py --execute` |
| `python -B -S frontiers/determinant_moment_20261025/gate.py` | `python -B -S frontiers/determinant_moment_20261025/gate.py --execute` |
| `python -B -S frontiers/cap_first_exit_lean_20261002/gate.py` | `python -B -S frontiers/cap_first_exit_lean_20261002/gate.py --execute` |
| `python -B -S specifications/lean_statements_20261010/check.py` | `python -B -S specifications/lean_statements_20261010/check.py --execute` |
| `python -B -S specifications/lean_statement_proofs_20261011/check.py` | `python -B -S specifications/lean_statement_proofs_20261011/check.py --execute` |

Stage67's three exact package-directory recipes and receipts are in README and JSON. Torus filtration has the recorded repository-root workflow command `bash companions/torus_filtration_20261011/replay.sh` (`.github/workflows/torus-filtration.yml:48`). The source-lifetime static recipe is `python -B -S frontiers/source_lifetime_realization_20261017/validate_static.py` from the repository root. The recorded generic `lake build` applies in each exact package directory; it does not supply an observed receipt here.

Source-smooth's historical hosted workflow and non-Lake staged drivers still need explicit scope/evidence mapping. Main core's specific gate/selftest/kernel driver was outside the Math tooling census and is left **NEEDS_MAPPING**, without guessed commands. No top-level source-derivative federation auditor was established by the old discovery; stage67's owned Inventory proves only its five owned constants.

## G0–G11 unmet scope

| Gate | Exact remaining boundary |
|---|---|
| G0 | Current contribution STATUS records an earlier56/56 selftest and39/39 checker-source mutants killed, followed by reported escaped-name/native-option defects under investigation. Changed-tool coverage/admission and federation-wide auditor selftests remain unresolved; the old run stays exact-scope evidence. |
| G1 | No complete repository/federation Layer-A run is recorded. Lexical inventories do not establish absence, soundness or G1 green. Cover embedded/generated source, archives, dependencies and actual CLOSED/quarantine isolation with an admitted auditor. |
| G2 | Map every CLOSED declaration, including private/data/generated constants, to owning module and approved transitive axioms. Stage67 ALL5/std3 evidence is reusable; registered target lists do not prove whole-federation coverage. |
| G3 | Cover all eight Math roots, Main core and non-Lake staged companions under one final source/pin binding. Stage67 registered build exit0 covers changed registration only, not final federation G3. |
| G4 | Stage67 proper own-module replay is completed evidence. Federation all-CLOSED module census and bounded exact-module replay remain to be admitted; current STATUS describes a separate wrapper still being verified. Skip escapes/prefix assumptions cannot supply missing replay. |
| G5 | Map all retained suites/frontier and companion tests to actual execution. Existing scoped passes remain reusable; final coherent full-regression run not established. |
| G6 | Map semantic and auditor mutation obligations to actual tools/scopes; preserve failure evidence. Historical scoped mutation passes do not establish final federation threshold. |
| G7 | Resolve installed dependencies and cross-repository pins at final inputs. Config-declared Lean/Mathlib pins and source-ID strings are discovery evidence, not resolved remote consistency. |
| G8 | Canonical single writer must reconcile exact dispositions, evidence and status surfaces. Dated availability labels, specified Props and conventional review acceptance cannot be silently promoted to kernel CLOSED. |
| G9 | No final tagged cache-free cold-start acceptance is recorded. Canonical STATUS must supply exact usable commands; no guessed runner is proposed here. |
| G10 | Canonical writer owns full branch/PR/worktree/stash/tag cleanliness and integration. Original local census is an observation cut, not a current or final clean-Git claim. |
| G11 | Canonical admission/documentation/self-verification and resource/version coverage for every mission tool remain required. This handoff is an inventory contribution, not a new tool registry. |

## Limits that must survive integration

- Reuse completed applicable mathematical verification. New agent/session/handoff alone is not a reason to rerun it. Pending process metadata never erases that evidence.
- The original discovery scripts are unverified inventory collectors. No absence, soundness or complete G1 conclusion follows from lexical counts, including zero code-token matches.
- Imported project inputs, generated/embedded test source, private/data/generated declarations, archive failures and external dependency closure need admitted mechanical coverage. Mathlib bodies were not audited by this task.
- Keep the original exact-boundary stage67 ordinary quarantine and source-explicit binary64-certifier candidates isolated from CLOSED claims. Preserve failed tests/attempts; never delete for green.
- Statement-only interfaces and conventional prose do not become kernel CLOSED through a green build, review label, hash or source register. Conversely, completed theorem/conventional evidence remains reusable in its exact scope.
- Historical external-review/quorum/custody ceremonies are superseded. Keep actual scientific evidence and observed failure outcomes; do not reinstate those ceremonies through old driver defaults.
- Final same-input federation execution/admission, documentation and Git/pin reconciliation belong to the canonical writer. This handoff supplies no final green result or broad publication of the private initial census.
Loading
Loading