info-tech-canon/history/2026-09-05_002437+0200-scope-against-intent.md

Scope against intent — repository review

Review date: 2026-09-05, Europe/Berlin. Evidence timestamp: 2026-09-04T22:24:37Z (2026-09-05T00:24:37+02:00). Base commit: 78e917e366dfc26f9abdf92240006bbea7922751. Basis: current working tree, including pre-existing modified and untracked files; the commit alone does not reproduce this assessment.

Assessment

Implemented scope substantially advances the intent: there is a usable canon service, connected artifact corpus, generated retrieval assets, executable profile examples, and concrete assimilation results. The old SCOPE.md materially understated this by describing validation, schemas, indexes, profiles, and agent briefs as future work and showing obsolete root-level paths.

The principal remaining gap is assurance that declared semantic contracts support independent integration. Passing repository checks demonstrates local consistency within their coverage; it does not establish exhaustive concept ownership, compatibility across versions, or consumer adoption. The Taxonomy boundary is coherent with the service: serving and checking artifacts supports that role. No reviewed implementation establishes ownership of consumer runtime state.

SCOPE.md was rewritten to describe the implemented baseline, current paths, ownership boundaries, maturity, dependencies, and limits. Existing implementation changes were preserved. Recommendations below were not implemented by this review.

Changes reviewed

Recent commits include the NetKingdom layering note and source reference, work-record identifiers, intake creation, and index maintenance. The working tree incorporates the layering clarification into the body of INTENT.md. The original interoperability commitments remain intact.

The main uncommitted implementation adds Emission Cadence: a generic standard, JSON Schema, adapted example, source-preserving assimilation workspace, artifact registration, retrieval assets, validation, and negative tests. canon.yaml and CHANGELOG.md advance the canon to 0.7.0 while retaining draft status. The contract supports two cadence forms and namespaced profile extensions. Its worked example explicitly disclaims being the source-owned live declaration.

The other substantive addition completes the Small-SaaS capability illustration and marks ITC-WP-0014-T04 done. Workplan/index, intake, and .repo-manager/ metadata changes are work-management surfaces, not additional semantic capabilities. ITC-WP-0018 records local Emission Cadence work as finished. Local workplans scanned are marked finished; this does not close promotion questions retained in canon artifacts.

Verification and limits

Check run	Result
make test	29 passed in 6.47 seconds.
PYTHONPATH=src python3 -m info_tech_canon validate	Pass; 76 artifacts, 315 relationships, zero errors/warnings.
PYTHONPATH=src python3 -m info_tech_canon profile validate small-saas	Pass; 15 artifacts and 15 payloads, zero errors/warnings.
generation.concept_ownership(load_context())	129 entries, zero cross-owner conflicts; one same-owner duplicate for AgenticDrivesFunctional.

Validation ran without rewriting infospace/validation/latest.json. Ownership entries include artifact titles and explicitly declared concepts, rather than 129 independently extracted semantic definitions. The duplicate is not an ownership conflict.

The State Hub returned no active domain workplans. Its inbox contained a 2026-09-04 NetKingdom request awaiting the published Emission Cadence contract and stable locator (message 13f55e9c-5660-4a30-9bfe-4c546f5064f8), which was read and marked read. This shows a pending handoff at message time, not proof that adoption remains blocked today. No consumer repositories, remote release publication, or deployed integrations were verified. External standards were not audited; this is a repository-local assessment.

Coverage against intent

Intent commitment	Repository evidence	Assessment and gap
Clear owner for every canonical concept	Models, owned_concepts, generated ownership index	Partial. Titles and explicit frontmatter do not cover every concept in prose.
Import instead of redefine	Artifact relationships, import matrix, ownership boundaries, capability anchor mappings	Partial. Comprehensive concept-level import/redefinition enforcement is absent.
Explicit, versioned, scoped, justified external mappings	Mapping artifacts, assimilation comparisons, CARING RBAC benchmark	Partial. Mapping schema requires only id/title/source/target; versions, scope, and rationale are not mandatory. Candidate mappings are not a complete external-standard mapping corpus.
Assimilation produces gaps, conflicts, mappings, proposals	IT Capability Canon and Emission Cadence workspaces; CARING findings	Substantially present. Retained proposals and questions need follow-through.
Patterns explain practical combinations	PracticePattern scheme, two practice patterns, Intent–Scope–Purposes	Present but narrow; expand from recurring consumer evidence.
Profiles constrain concrete implementations	Executable Small-SaaS profile and capability illustration	Present for one local proof; independent behavior and compatibility remain unverified.
Useful to humans and agents	Markdown, CLI/API, graph/views, retrieval indexes, briefs, alignment kit	Strong local support; portability and freshness checks remain gaps.
Preserve provenance, compatibility, rationale	Frozen sources, changelog, versions, mapping rationale, draft gates	Partial. Policies exist; immutable distribution and cross-version consumer evidence were not demonstrated.
Initial kernel, models, formats, and subsystem interface examples	Kernel, models, schemas, profile proofs, interface-card template and expectation packs	Most structures exist; a template does not establish independently owned interface declarations.
Taxonomy ownership without runtime-state ownership	Updated intent, cadence ownership section, local read-only API	Aligned. Operational policy, scheduling, and observation stay with consumer owners.

Gaps and closure criteria

G1 — Semantic validation coverage (high)

Evidence: validation.py, generation.py, and mapping schema. Required schemas are generally checked for presence; selected families receive specialized checks. No general pass proves every registered payload conforms to its declared schema. Ownership generation reads titles and frontmatter; general Markdown links are not comprehensively validated by the reviewed local validator.

Closure: publish a validation-coverage inventory, bind supported artifact kinds to schemas, and enforce concept ownership/import references and mapping metadata with representative negative cases. Explicitly label unsupported checks so zero errors cannot be mistaken for full semantic conformance.

G2 — CARING conformance (high)

The benchmark findings explicitly identify a missing CARINGAccessDescriptor schema, effective-access derivation rules, tenant-boundary evidence profile, and DerivedCapability ownership pressure. Structured examples do not prove deployed effective access.

Closure: formalize descriptors and representative derivation rules, resolve ownership, and exercise workload-mediated access, secrets, and tenant-boundary cases. Keep operational evaluation with the relevant consumer.

G3 — Consumer adoption and contract distribution (high)

The Emission Cadence standard distinguishes example validity from operational truth. canon.yaml requires two independent source implementations before draft promotion. The validator checks the bundled example; there is no general Emission Cadence CLI command accepting arbitrary consumer declarations analogous to capability-review.

Closure: provide an immutable, version-addressable contract bundle and reusable validator covering both JSON Schema and unique source_id semantics. Capture two source-owned declarations and importing-consumer results, including unsupported-version rejection. Retain the evidence owner and tested canon revision. Do not infer publication from a local changelog entry.

G4 — Portable service consumption (medium)

pyproject.toml declares an absolute local infospace-bench dependency; bench.py loads modules from a sibling source checkout. The service locates canon data relative to its code. Local tests do not prove a wheel installed elsewhere can find its dependency or corpus.

Closure: use a reproducible dependency source, normal package imports, and an explicit corpus location/package-data contract. Demonstrate installation and inspection in a clean environment without the workstation directory layout.

G5 — Freshness and lifecycle visibility (medium)

The old scope contradicted implemented features. WORK-RECORDS.md still lists ITC-WP-0014-T04 as todo, while its working-tree workplan says done. Package and canon versions differ without explanation in the previous scope. The generated validation report has no source revision or execution timestamp.

Closure: regenerate work-record projections from owner files, document version surfaces, and add deterministic generated-asset freshness checks and validation revision/timestamp evidence. The scope rewrite resolves the scope-description part; projection maintenance remains a recommendation.

G6 — Remaining semantic pressure (medium, demand driven)

IT Capability Canon open questions retain canon-wide state qualifiers, continuity/recovery ownership, and structural models behind commerce/intelligence capabilities. Capability anchor mappings record low-confidence recovery/continuity alignment and unmapped domains.

Closure: resolve owners where consumers need structural contracts, record rationale and compatibility impact, and preserve explicit deferral where only a capability identifier is needed. Intent does not require inventing every possible domain model now.

Possible optimizations

1. Prioritize reusable conformance over corpus expansion. Start with G1 and the cadence validator/bundle in G3. This advances integration by declared contract and makes draft promotion evidence tangible.
2. Reduce metadata drift. Define authoritative inputs and derive navigation, work-record projections, and a small scope inventory from them. Check that regeneration produces no unexpected diff; retain human-authored judgments.
3. Organize validation by contract. The roughly 2,040-line validator embeds many pack-specific ID sets and checks. Factor reusable shape, reference, and provenance checks into common code and keep specialized semantics alongside contracts. Preserve meaningful negative tests during refactoring.
4. Make consumption reproducible. Fix the sibling-loader dependency and prove clean installation before expanding the HTTP service. Immutable artifacts let consumers pin inputs without a live canon runtime dependency.
5. Use consumer evidence to select the next profile or pattern. Complete the two-source cadence proof and target CARING's recorded gaps before adding speculative catalogs. Route needs through demand and utility results through feedback, preserving ownership.
6. Measure before caching. service.load_context() reloads the infospace for many operations. If repeated API reads become material, benchmark parsing cost and consider revision-keyed snapshots with explicit invalidation. No performance bottleneck was measured in this review.

Suggested order: conformance coverage and adoption proof, freshness and portability, then targeted semantic expansion. These are follow-up options, not changes to the repository's runtime responsibilities.