Check collection contract coverage

Result and boundary

The split draft covers the collection definition's authoring surfaces after the repairs below. The shortening draft preserves the existing layout and generated review behavior while removing duplicated instruction and analytical detail. Neither is a deployed contract. Their explicit collection-read dependency and conditional maintenance route still need implementation. This check does not adopt the split, migrate evidence, or establish improved analytical quality.

The source is the current collection contract. The standard is the collection definition and authoring composition. Rules about a kind belong to its type; operational procedures need an actual required consumer. Moving a rule into an optional document does not preserve it. The file is a coverage audit, not an independent agent review.

Repairs made during this check

  1. The split purpose now includes run state and local types. They are owned by the new collection, although they are not analytical report prose.
  2. Title rules apply to titled reports. Run-state/type titles and exact-heading worker fragments keep their own formats. The synthesis and verification jobs do not produce full titled reports; a universal report title rule contradicted their literal output grammar.
  3. The split link grammar now permits local type specs to link the governing shared method contracts in agentic-systems. Mechanical relocation creates cross-collection links: suppressing this destination would make moved types violate their own collection contract.
  4. Both split owners and the shortening contract require method authors to load the maintenance draft before changing instructions/shared contracts/types. Promotion must rewrite its draft link to the canonical instruction. The draft route binds the applicable author, without adding the procedure to analyst read-first lists. The maintenance draft alone previously supplied no upstream mandatory route.
  5. The split retains the native account independently of mappings, routing for theory/raw captures, and the prohibition on prospecting unrelated web sources.

Clause dispositions

Each row covers the substantive clauses of the named incumbent section. S = split analysis contract; O = split existing-owner contract; H = shortening contract. Operational and type destinations are named below.

Incumbent clauses Split disposition Shortening disposition / verified consumer
Whole-system scope; native explanation; fidelity/economy S and O; native account survives rejected mappings H retains these collection-wide rules
Memory/knowledge lens; legacy corpus excluded from new comparisons O scope; S is general analysis scope H scope; comparison/synthesis readers exclude legacy evidence
Memory report copied unchanged, owns comparison profile; trial grants no semantic clearance Memory type and assembly/completion checks; O readers H keeps concise invariant; memory type, workflow run_job/close_round, run-state type and analysis checks already supply details
Root operating material; reviews/comparisons placement S/O split placement H preserves current areas
Method/shared-contract/synthesis/taxonomy ownership; general transfer scan O and S exclusion H placement; no global machinery move
Executability, precision, goal first, authoring metadata Maintenance draft, instruction type; O conditional authoring rules H plus mandatory maintenance read; existing instruction type already owns composition/metadata
Verified inheritance; explicit packet constraints; parent authority; unstated consequential choice; clean-context benefit Maintenance draft and instruction type Same; worker rules already supply concrete authority boundaries
Instruction change consumer/channel; caller/callee and baseline cohort checks Mandatory maintenance route in S/O Mandatory maintenance route in H; deployment still needs promotion
Shallow links, no reciprocal mirroring, search scope, declared execution dependencies Maintenance draft; O method link grammar H grammar and mandatory maintenance draft
No reviews/archive/workshop method inputs; canonical files; projections/promotion O plus maintenance draft H plus maintenance draft; instruction deployment checks remain explicit
Ignored state, cleanup, explicit validation, opening method and earlier-run recovery S area roles; worker authority in worker-rules; skill/code/marker own operations H short lifecycle; skill/run-state/code/marker own details
Frozen retained sets; distinct other-report names; archive producing contracts/pins; no current-schema/comparison participation S maintenance/placement; proposed migration must preserve marker and reader exclusions H retains invariant; current marker and reader path population enforce exclusions
Bounded migration only; old/new hash map; no finding/source/identity/provenance change S says coordinator-owned; decision and migration drafts spell bounds and map H keeps bounded exception; ADR 099 supplies only the earlier migration's authority
Mandatory generated publication; workflow ownership; no hand-editing; new-run correction; ordinary artifacts unaffected O replaces projection with reference; S frozen-output rule H preserves all current invariants; generated-review type and skill own details
Review metadata fields, private candidate validation, incumbent preservation, run completion Intentional replacement in reference/publication drafts; code not implemented Existing review/run-state types, skill and publication/recognition code; H need not duplicate fields/procedure
Exact accepted set membership, byte copies, manifest/member pins, reproducible comparison input Set/member types and publication draft; selection through O references H short invariant; existing set/run-state types and comparison reader verify pins without local state
Evidence basis and capture date; independent tiers and per-field basis S member format; O comparisons; source/member types H authored one-line basis and typed format; H comparison tiers; sources contract and memory profile own basis
Theory conditions, independent learning/persistence/addressability/reflection, no absence from opacity Shared records already in analytical/judging role dependencies; ordinary theory accounts stay in O and need an explicit shared-record read H explicitly requires Theory account read for ordinary theory assessments; analytical jobs already receive it
Current transfer separation; keep unresolved state to disposition then replace/delete S exclusion; transfer skill owns state operations H exclusion/state rule; transfer workflow owns operation
Descriptive titles; argumentative title-as-claim S reports use member formats; O retains ordinary claim titles H retains ordinary conventions, typed report and fragment formats
Sources, external evidence, theory links S narrow grammar; O keeps broader author grammar H retains original destinations/labels in compressed wording
Related subsystem accounts, retained evidence, Commonplace analogues and procedure mappings Stay in O for ordinary/comparative accounts; not analytical worker discovery tasks H retains their destinations/labels
Type eligibility and implicit text S/O local owners H unchanged eligibility
Theory, source, Commonplace, transfer, generic procedures, WIP routing S/O scope and boundaries H routing; generic procedures covered by method placement

Explicit loading gaps and fixes

  • All analysis jobs: AnalyseAgenticSystem.job currently supplies the job, worker rules and role extras, but not COLLECTION.md. Add the selected contract to read-first and hashed inputs for either alternative. The root writing rule already makes it mandatory; do not count current omission as a legitimate saving.
  • Method authors: the repaired candidate contracts provide the conditional maintenance route. Promote its target with the contracts. Do not put the whole maintenance instruction into every worker packet.
  • Ordinary theory-account authors: H provides a conditional shared-record route. The repaired O draft now supplies the same route; analyst jobs already have the shared records dependency. A definition link alone would not supply all operational distinctions formerly present in COLLECTION.md.
  • Split coordinator/code: add both contracts, new types and publication instruction to the method-change guard and explicit maintenance inputs. Existing METHOD_PATHS names only the old collection/types. The publication instruction draft is not currently loaded or enforced. Until this consumer change lands, its gates cannot replace live generated-review checks.
  • Archives and migration evidence: preserve bytes and markers; declare any exceptional migration authority and map before retyping current sets. Historical bytes do not silently acquire the new contract's current-schema requirements.

A Markdown-link scan of the two current retained sets found 24 links, all within the current analysis owner; they are supporting member links after relocation. The current local types contain 15 within-owner links and two notes links. The within-owner count includes shared instruction links, which cross owners after the split, and the generated-review type, which stays historical. Remove/update current overview publication references as the plan requires. This bounded sample supports the narrowed worker grammar; it is not evidence about every future source or historical artifact. Citation anchors in code spans are covered by the source contract, not counted as Markdown outbound links.

The role instruction templates confirm that boundary/synthesis/verification outputs are fragments. Runtime, memory, epistemic and assembled reconciliation are typed reports. All write somewhere under working collection state, so all currently need the collection read even when they produce no retained member.

Interference check

Count rule groups by the output decision they govern, rather than counting sentences or bytes. The following groups have no authoring decision in any of the eight current worker outputs. The count is specific to those jobs; it is not an estimate of model interference. Mixed paragraphs still contain relevant rules and cannot be removed wholesale.

Rule group without a worker-output decision Shortening Split analysis contract Required consumer
Canonical method editing and skill projection/promotion ownership 1 0 Method author/maintenance
Generated-review publication obligation and maintenance 1 0 Coordinator/publication
Ordinary account's opening evidence-line convention 1 0 Ordinary account author; typed workers have their own formats
Ordinary argumentative account's title-as-claim convention 1 0 Ordinary account author
Conditional method/type-maintenance read 1 1 Method/type author
Transfer-state retention and disposition 1 0 Transfer-scan owner
Related subsystem-account link policy 1 0 Ordinary/comparative account author
Separately retained general-research link policy 1 0 Ordinary/comparative account author
Commonplace-analogue link policy 1 0 Ordinary/comparative account author
Mapped Commonplace-procedure link policy 1 0 Ordinary/comparative account author
Instruction-composition link policy 1 0 Method author
Run-state/local-type title/description convention 0 1 Coordinator/type author; shortening inherits their type contracts
Historical archive preservation rule 1 1 Archive maintainer; current workers have no archive authority
Total groups in this audit 12 3

The old shared collection still needs these authoring/maintenance conventions for its other consumers; removing them requires routing applicable kind rules through types or explicit author instructions. The split removes most of that input, but also owns local types, run state and archives. It therefore does not yet satisfy a literal requirement that every clause govern a current analysis worker's output. Its three residual groups should be reduced or routed further if that is the adoption threshold. Do not silently drop the conditional maintenance route or historical preservation rule to obtain a zero count.

Complexity check

Neither design changes the fact that current packets omit the mandatory collection read. Both can supply it as one literal read-first path and hashed dependency. The role constructors already supply job, worker, source, record and member-type paths directly; the split does not inherently simplify that part further.

The split creates additional cross-owner type-to-method links. Workers already receive the needed shared contracts literally, so those links must not become new execution-time searches. Ordinary authors still conditionally read the Theory account and maintenance instruction; candidate contracts name those files directly with links, and promotion must preserve resolvable paths. Relevant linked definition reads remain under root vocabulary doctrine and are not eliminated by either layout.

The candidate routing is specified and locally resolvable. Dependency hashing and literal contract insertion are not deployed or exercised by this documentation check; verify them with the eventual packet-builder change. The split's type relocation and publication rewrite remain additional consumer work.

Decision implication

Both designs can remove most of the incumbent's mandatory reading cost. The split retains a smaller shared input and a boundary against method/comparison rule growth. Shortening avoids relocation, retyping and pin/redirect migration. Publication by reference is independent of either layout: it can be implemented in the existing collection if adopted. Do not credit that publication change as an input saving from the split.