Skip to content

ADR-0017: Adopt CMDG — Certified Reconstruction of the Mathematical Dependency Graph

Status

Council consensus: RATIFY_WITH_CORRECTIONS on 2026-08-08.
Human Steward disposition: HUMAN_STEWARD_RATIFIED_WITH_COUNCIL_CORRECTIONS.
Human Steward evidence: PR #289 top-level comment 5225270603, binding exact reviewed head 11ed37d35825a381ce75bcc935bb87d2273eed53.
Protected admission: PR #289 merged as 076bcdf1631b6f39b8ac84046c301344a43310da on 2026-08-08.
Protected authority: active subject to the stage-bounded corrections CMDG-C01 through CMDG-C08; documentary closure is recorded by CMDG-COUNCIL-AUTHORITY-CLOSURE-001.

Context

The Euclid end-to-end exemplars demonstrated that MATH-PROGRAMME can carry a bounded mathematical statement from source reconstruction through formalization and certification. The Human Steward then approved a broader successor conception: rather than immediately select another isolated advanced theorem, reconstruct and certify the dependency architecture that permits modern mathematics to stand.

The motivating conception is preserved verbatim in records/CMDG_GRAND_CHALLENGE_PROGRAMME_MEMORIAL.md, with a rendered pointer at docs/CMDG_GRAND_CHALLENGE_PROGRAMME_MEMORIAL.md. Council docket #288 requested ratification of the programme architecture, foundation policy, typed dependency graph, certification semantics, execution order, cross-foundational natural-number experiment, and Condensed Mathematics frontier.

The Agent Council reviewed the docket under docs/AGENT_COUNCIL_GOVERNANCE.md, docs/AGENT_COUNCIL_WORK_PACKAGE_CHECKLIST.md, and schemas/agent_review.schema.json. All fifteen schema-required offices recorded findings. The complete historical deliberation is docs/CMDG_COUNCIL_DELIBERATION_001.md; the machine-readable pre-ratification candidate record is governance/cmdg_council_review_candidate.json.

Council finding

Quorum was defined for this docket as completion of all fifteen schema-required office reviews plus Referee synthesis and absence of blocking dissent against presentation to the Human Steward.

  • offices reviewed: 15/15;
  • offices supporting adoption: 15/15;
  • RETURN_FOR_REVISION: 0;
  • REJECT: 0;
  • Referee synthesis: RATIFY_WITH_CORRECTIONS.

Council found CMDG coherent, technically feasible, aligned with the existing MATHFORGE → MATHSOLVE → MATHCERT authority split, and suitable as an overarching MATH-PROGRAMME Grand Challenge.

Human Steward decision

The Human Steward ratified the Council disposition at PR #289 exact head 11ed37d35825a381ce75bcc935bb87d2273eed53 and adopted:

CMDG — Certified Reconstruction of the Mathematical Dependency Graph

with the mission to construct a machine-readable, machine-checked, provenance-bearing reconstruction of the dependency architecture of modern mathematics, from formal logic and foundational systems through structural mathematics, category theory, topology, analysis, sheaf/homological machinery, and a demanding modern frontier such as Condensed Mathematics.

The binding disposition is:

HUMAN_STEWARD_RATIFIED_WITH_COUNCIL_CORRECTIONS

Adoption preserves the following core decisions:

  1. CMDG is not a project to re-formalize all mathematics from scratch.
  2. CMDG separates semantic mathematical dependency, checked-proof dependency, implementation/import dependency, and provenance dependency.
  3. The canonical mathematical object is a typed directed multigraph; only certified equivalence-generating edges may be collapsed for acyclic dependency projections.
  4. Lean dependent type theory is the operational proof substrate; ZF/ZFC and other foundational systems are represented object theories or realizations, not silently identified with Lean's metatheory.
  5. The existing Level 0–5 theorem certification ladder remains in force.
  6. GRAPH_CERTIFIED is an orthogonal dependency-certification status and is not synonymous with machine_checked.
  7. CMDG proceeds by thin vertical demonstration spines followed by horizontal closure.
  8. CMDG-NAT-CONCORDANCE-001 is the first cross-foundational experiment, followed by CMDG-EUCLID-BRIDGE-001.
  9. Condensed Mathematics is the first major modern load test, beginning with exact bounded CM0–CM2 targets before later homological, solid, and liquid stages.
  10. MATH-PROGRAMME owns ontology and authority; MATHFORGE owns source reconstruction/candidate dependency evidence; MATHSOLVE owns mathematical reconstruction; MATHCERT owns proof replay, dependency extraction, concordance checking, and certification.

Council correction register

The Human Steward adopted the following stage-bounded corrections as binding obligations.

CMDG-C01 — Manifest-relative graph certification

Before CMDG-SCHEMA-001 finalization or any GRAPH_CERTIFIED status, define graph certification relative to a versioned manifest recording root, ontology version, direct semantic dependencies, closure policy, boundary nodes/trust classes, proof environment, axiom/classicality footprint, and reviewed semantic-edge evidence.

CMDG-C02 — Cross-layer realization semantics

Before CMDG-SCHEMA-001 finalization, add an explicit semantic-to-formal realization relation such as REALIZES_AS or FORMALIZES_AS; distinguish direct reviewed edges from computed transitive closure; restrict equivalence quotienting to certified equivalence-generating edges.

CMDG-C03 — Exact ZFC and NNO profiles

Before CMDG-NAT-CONCORDANCE-001, specify whether the set-theoretic realization is syntactic, semantic, or both with an interpretation bridge; define categorical natural-numbers objects by universal property in a named ambient category with explicit universe conditions.

CMDG-C04 — Exact Condensed Mathematics target

Before CM0–CM2 promotion, pin the exact condensed-object definition and state the cardinality/concordance boundary between the chosen formal implementation and Clausen–Scholze formulations. Any equivalence across presentations is a separate concordance obligation.

CMDG-C05 — Solid-module scope

Before any general-ring CM4 claim, restrict to a formally supported ring regime or reconstruct the intended general solid-module definition; do not overstate an implementation whose own documentation carries a generality caveat.

CMDG-C06 — Adversarial and replay gate

Before the first production GRAPH_CERTIFIED artifact, install retained fixtures for hidden classicality, import/semantic conflation, equivalence laundering, boundary laundering, alias inflation, transitive omission, universe issues, source mismatch, stale pins, and clean-environment replay.

CMDG-C07 — Documentary integration

Before protected CMDG authority activation, preserve the memorial, full Council deliberation, machine-readable review, Human Steward disposition, this ADR and index entry, artifact-ledger and terminology updates, exact-head review evidence, protected merge receipt, and post-merge readback.

CMDG-COUNCIL-AUTHORITY-CLOSURE-001 is the dedicated closure operation for this correction. Its protected receipt binds the Human Steward disposition, PR #289 exact head, protected merge, protected-main readback, and pre/post-merge workflow evidence. The historical Council deliberation and candidate review remain unchanged as pre-ratification records.

CMDG-C08 — Spine terminology

Use demonstration spine or certified spine for V0/V1/V2 unless minimality or uniqueness is separately proved.

Consequences of ratification

  • CMDG-CHARTER-001 is authorized immediately and must incorporate C01, C02, C08, and the controlled terminology identified by Council.
  • Schema and validator work may proceed only after the charter fixes the manifest-relative certification contract.
  • Natural-number concordance may not promote until C03 is discharged.
  • Condensed frontier promotion may not proceed past the corresponding C04/C05 gates.
  • No theorem becomes GRAPH_CERTIFIED until C01, C02, and C06 are operational.
  • The I–XIX memorial remains a stable motivating reference; later architectural departures require explicit decision deltas rather than silent rewriting.

Protected admission evidence

PR #289 was admitted from the exact reviewed head named by the Human Steward disposition:

  • reviewed PR: #289;
  • exact reviewed head: 11ed37d35825a381ce75bcc935bb87d2273eed53;
  • Human Steward top-level comment: 5225270603;
  • pre-merge Programme policy: run 31247672054, success;
  • pre-merge GCL conformance: run 31247672257, success;
  • pre-merge administrative dispatcher: run 31247672063, success;
  • ready-for-review dispatcher: run 31247891459, success;
  • protected merge commit: 076bcdf1631b6f39b8ac84046c301344a43310da;
  • protected-main readback: 076bcdf1631b6f39b8ac84046c301344a43310da;
  • post-merge Programme policy: run 31247906697, success;
  • post-merge GCL conformance: run 31247906944, success;
  • post-merge administrative dispatcher: run 31247906696, success;
  • post-merge protected synchronization: run 31247924635, success;
  • post-merge workflow-run dispatcher: run 31247924653, success.

The machine-readable closure record is governance/cmdg_council_authority_closure_001.json.

Rejected alternatives

Return CMDG for redesign

Rejected. No office identified a defect requiring abandonment or fundamental restructuring. All identified corrections are stage-bounded refinements of the approved thesis.

Treat import graphs as mathematical dependency graphs

Rejected. This would collapse the programme's central semantic distinction.

Treat machine_checked as GRAPH_CERTIFIED

Rejected. Kernel checking establishes proof validity under a formal environment; CMDG additionally requires a reviewed dependency manifest and declared trust boundary.

Require exhaustive bottom-up formalization before a frontier test

Rejected. The Council supports a thin vertical demonstration spine followed by horizontal closure.

Treat current mathlib Condensed/Solid implementations as automatically identical to every source formulation

Rejected. Exact target identity and any concordance must be stated and checked.

Claim boundary

This ADR authorizes a programme architecture and controlled implementation route. It does not:

  • prove a new mathematical theorem;
  • establish consistency or relative consistency of ZFC or another foundation;
  • certify any existing mathlib theorem as semantically dependency-complete;
  • claim formalization of all mathematics;
  • claim independent reproval of Clausen–Scholze results;
  • authorize novelty, priority, publication, patentability, product, deployment, or commercial claims.

Authority record

The Human Steward disposition, exact-head validation, protected merge, and protected-main workflow evidence are bound by CMDG-COUNCIL-AUTHORITY-CLOSURE-001. The original Council deliberation and candidate review remain historical evidence of the pre-ratification state and are not rewritten by this closure record.