Skip to content

M2 implementation brief: Corpus authority alignment

Work contract

Field Value
Milestone M2
Status verified; independent verification passed at PR #13 head 395fbd2
Primary gaps G5, G6, G9
Definition items D-09, D-10, D-11
Depends on M0 verified; M1 verification requested
Mutation rule Binding corpus edits only through proposal → Hector approval → PR
Tracking issue #6

Objective

Make the live corpus compile the Initiative, concurrent-workstream, stage, role, authority, and known/adapt/novel model already established by M0 and physically proven by M1.

M2 changes policy representation. It does not build the plan-review or authorization runtime; that is M3.

Locked inputs

  • Conductor is a supervisory compiler, not the project-state writer.
  • Notion Initiative and workstreams are the organizational state spine.
  • Workstreams may run concurrently; there is no Initiative-wide active line.
  • One execution authorization covers one workstream, stage, role, and exact plan.
  • GitHub Issues are optional repository evidence, not the Initiative contract.
  • Core model, organization profile, adapters, state, and secrets are separate layers.
  • Generated documentation is a derived mirror, never design authority.
  • Binding corpus changes follow the governed proposal workflow in AGENTS.md.

Deliverables

M2.1 Corpus architecture

  • Define unit identity, status, version, dependencies, precedence, and deprecation.
  • Define core versus organization-overlay packaging.
  • Define how stage, role, inventory, standard, pattern, and example units compose.
  • Define conflict and missing-dependency behavior.

M2.2 Complete stage contracts

For every active stage, encode:

  • entry conditions;
  • required bindings;
  • permitted roles;
  • expected artifacts and evidence;
  • exit conditions;
  • allowed next stages;
  • deterministic checks versus judgment;
  • rollback or recovery behavior.

Incomplete stages must emit an explicit gap.

M2.3 Authority migration

  • Remove active_line and attending_lines as live policy.
  • Replace single-line ownership with persistent Portfolio plus concurrent line workstreams.
  • Remove mandatory GitHub Issue requirements.
  • Align role ownership and handoffs with the authority model.
  • Preserve superseded material only as archive/recovery evidence.

M2.4 Pattern mode

Represent each material pattern choice as:

  • known — apply accepted precedent and report deviations;
  • adapt — reuse precedent with explicit differences and validation;
  • novel — preserve exploration, assumptions, risks, and review points.

No successful execution automatically promotes a pattern into binding policy.

M2.5 Deterministic generated documentation

  • Regenerate the corpus mirror from live approved units.
  • Restore generated pages to navigation.
  • Record source unit and version provenance.
  • Add checks for stale generated output.

M2.6 Consistency enforcement

Add automated checks that fail on:

  • obsolete Initiative-wide line fields;
  • mandatory GitHub Issue policy;
  • unknown stage or role IDs;
  • missing stage-contract sections;
  • conflicting authority owners;
  • unresolved unit dependencies;
  • organization-specific vocabulary placed in core packages;
  • generated documentation differing from approved corpus units.

Execution sequence

  1. Approve the decisions in decision-package.md.
  2. Inventory contradictions without modifying live corpus.
  3. Group changes into small conductor-proposal issues.
  4. Hector comments exactly Conductor-Approval: approve on each accepted proposal.
  5. Implement approved proposals as PRs; never merge automatically.
  6. Regenerate documentation.
  7. Run schema, compiler, consistency, snapshot, and documentation checks.
  8. Request independent M2 verification.

Acceptance conditions

  1. Every active stage has a complete contract.
  2. Corpus and canonical documents agree on Initiative/workstream concurrency.
  3. No active unit requires obsolete line fields or a GitHub Issue.
  4. Core and organization-specific policy are mechanically distinguishable.
  5. Unit dependency and precedence behavior is deterministic.
  6. Pattern mode supports known, adapt, and novel work.
  7. Generated docs reproduce deterministically from the approved corpus.
  8. Consistency checks detect each named superseded-policy fixture.
  9. Historical archive cannot become live authority.
  10. Every binding change has proposal, exact human approval, and PR evidence.

Non-goals

  • Plan bundle and authorization-envelope implementation (M3).
  • Real-world Asset triage or decisive EstateComparison execution (M4).
  • Learning-loop implementation (M5).
  • Second-organization onboarding (M6).
  • Routine Notion or production mutation by Conductor.

Handoff

Report proposal/approval/PR links, migrated and deprecated unit IDs, consistency results, generated-doc reproducibility, known gaps, and M2 status as review requested or a named blocker. Do not self-mark verified.

Owner implementation approval was recorded on 2026-07-26 in PR #13. The first independent verification returned BLOCKED. Its four findings were remediated without merging:

  • mutation tools and webhook writeback were removed from the live app surface;
  • all 73 YAML and Markdown corpus sources now carry typed metadata and participate in dependency/layer validation;
  • stage inventory bindings now use resolvable unit IDs with a failing unknown-ID fixture;
  • all 40 stages now carry line-specific entry, evidence, exit, checks, and recovery contracts, with both deterministic and judgment modes.

The remaining gate is a clean independent re-verification; owner approval alone does not self-mark M2 as verified.

The second verification confirmed those four remediations and found three additional fail-open edges. The current revision also:

  • rejects malformed, unsupported, and incompatible exact or caret dependency constraints in schema, CI validation, and live corpus loading;
  • resolves known/adapt precedents against the active accepted pattern-unit set;
  • excludes archived-status units from live corpus snapshots and rejects active dependencies on archived units.

The third independent verification returned VERIFIED with no blockers. Evidence is recorded in PR #13. The PR remains unmerged.