Guide Changelog

Material changes to scope, terminology, maturity, architecture, and review status are recorded here. Minor punctuation and formatting corrections may be grouped.

2026-09-04 — Six-infographic architecture cross-check

Added

• Added a concise independent-axis configuration rule in Chapter 23 and the full lens in the architecture communication appendix, so trigger, control behavior, capability, knowledge mode, topology, authority, and autonomy are specified rather than hidden behind an "agent type" label.
• Added a channel-adapter contract in Chapter 37 for chat, email, voice, portals, APIs, and agent-to-agent intake, including identity, source and transformation provenance, idempotency, confirmation, and shared status.

Changed

• Reviewed all six supplied visuals against their canonical chapter owners. Recorded overlapping material as corroboration, rejected the volatile product catalog and mixed-axis taxonomy as architectures, and imported only the two missing contracts above.
• Preserved the eight-stage value stream and six-area architecture; no new stack, lifecycle, vendor map, or agent taxonomy was introduced.

2026-09-04 — Ontology engineering and maturity

Added

• Expanded Chapter 19 with an ontology boundary model, implementation contract, task-scoped semantic slices, governance workflow, operating metrics, and agent-specific failure modes.
• Added a five-level, domain-and-workload-scoped ontology maturity model: Foundation, Owned Domains, Executable Contracts, Governed Operations, and Governed Adaptation.
• Added W3C OWL 2 and SHACL to the research canon alongside SKOS and PROV-O.

Changed

• Reframed the supplied visual's final level so automated detection and proposal do not imply autonomous promotion. Semantic changes remain subject to validation, impact analysis, authorization, measured rollout, and rollback.
• Kept ontology inside the existing semantic-engineering owner instead of introducing another top-level architecture or standalone chapter.

2026-09-04 — Practitioner-source review

Added

• Added an intent-consistency check immediately before external effects, so a valid action against a drifted goal cannot pass as successful execution.
• Separated content safety, schema validity, authorization, and transactional correctness at the tool boundary, with deterministic invariant enforcement before mutation.
• Extended the existing mastery scorecard with five observable coding-agent operating skills: workflow direction, autonomy calibration, review, environment stewardship, and agent foundations.

Changed

• Resolved seven supplied redirect links to their canonical public posts, reviewed eight unique sources, updated one previously incomplete source record, and documented accepted, corroborating, and rejected material in the editorial intake.
• Kept the eight-stage value stream and six-area architecture canonical; no additional agent taxonomy, stack, or lifecycle was introduced.

2026-09-03 — Technical-vision integration

Added

• Added the execution trajectory to the trusted delivery artifact while preserving the distinction between forensic telemetry and criterion-bound evidence.
• Added a falsifiable architecture-bet ledger with disconfirming signals and predefined design responses.

Changed

• Sharpened the opening thesis around abundant code generation and scarce trusted delivery.
• Integrated one technical-vision artifact into existing canonical owners instead of adding another top-level model, and recorded its provenance and evidence limitations in the research canon and editorial intake.

2026-09-03 — Structural reading and navigation pass

Added

• Split eight overloaded chapters along their existing topic boundaries, taking the guide from 36 to 44 numbered chapters without dropping their underlying material.
• Added Mission Control operator surfaces as Appendix G, moving the eleven-surface catalog out of the case-study chapter while keeping direct links to every surface.
• Added compatibility redirects for all 28 renumbered production chapter URLs and all 72 anchors whose sections moved into a new second-half chapter.

Changed

• Designated the eight-stage value stream and six architectural areas as the two canonical Chapter 2 models; other frames are now explicitly labeled as lenses, references, or implementation views.
• Reduced every chapter's "Retain this" section to five to seven durable points.
• Marked 32 high-priority entries among the glossary's fifty core concepts with a key badge and explained the marker in the front matter.
• Recomputed chapter numbers, ranges, citations, navigation, search content, coverage surfaces, and the visual atlas for the 44-chapter structure.

2026-09-03 — Public-source integration and consistency pass

Added

• Folded four rounds of public source material into the existing chapters: a public taxonomy of skills, loops, and the factory maturity model; the Mission Control repository's own lexicon (genericized); and a synthesis of six public sources on loop engineering, harness engineering, context engineering, and factory economics (Uber Engineering's published Uber-scale figures, two public "layers of an agent system" explainer posts, a public loop-engineering explainer, and Warp's closed-loop factory post).
• A provenance table in the research canon for every practitioner talk, livestream, and public engineering post the guide draws on, including what the recurring "public practitioner talks, 2026" citation covers and why it does not name speakers or venues.

Changed

• Resolved a terminology collision on model adapter: the glossary's trained-weights sense is now trained adapter; model adapter is reserved for the per-provider integration layer defined in Chapter 11 and used in Chapter 21.
• Merged duplicate glossary entries for loop engineering (three entries collapsed to two: the bounded feedback path within one execution graph, and the wider discipline) and workload identity (two near-identical entries merged to one).
• Follow-up review caught three issues in the pass above: an "example" repository URL in Chapter 31 that still resolved as a real dead link (now example.com, the domain reserved for documentation); a "Retain this" line in Chapter 40 that still stated learning writeback reads only after success without the surrounding section's caveat for separately labeled failed-run signals (reconciled); and the research-canon provenance table, which named sources without linking most of them — now carries a confirmed-live link for every source where one exists, and says plainly where none could be found rather than leaving the gap implicit.

Resolved in the structural pass

• The word-count review found eight chapters running past a comfortable single-sitting length, competing "canonical" models in Chapter 2, and "Retain this" sections that restated rather than summarized. The structural reading and navigation pass above closed those three findings.

2026-08-30 — Reference architecture depth release

Added

• A detailed architecture ownership and coverage matrix.
• Four synchronized reference-architecture views: lifecycle, logical components, deployment/trust boundaries, and authority/evidence.
• Governed factory-system inventory, governance operating model, decision rights, ten-family control catalog, and emergency-control semantics.
• Twelve orchestration component families, runtime envelopes, state, stop, failure, recovery, capacity, cost, and compatibility contracts.
• AI systems foundations, permission-aware knowledge and retrieval pipeline, tool/skill/integration contracts, multi-agent collaboration contracts, and a minimum-sufficient autonomy ladder.
• Enterprise operations and FinOps reference plus a control-tower response lifecycle from observation through verified improvement.
• Four failure-injection labs covering authority, orchestration, knowledge, and external capabilities.
• A public architecture hub with lifecycle, plane, component, governance, inventory, pattern, monitoring, data-flow, and evidence views.

Changed

• Deepened knowledge engineering, scheduling/capacity/cost, resilience/disaster recovery, and observability/forensics with specification-level contracts.
• Expanded all navigation, learning, coverage, glossary, reviewer, and search surfaces while preserving Markdown as the source of truth.
• Kept all new chapters review ready: documentation completeness does not advance implementation evidence to validated or operationally proven.

2026-08-30 — Autonomous factory coverage expansion

Added

• Agent Factory capability supply chain, registries, packaging, dependency resolution, certification, promotion, deprecation, and revocation.
• Repository onboarding, codebase intelligence, readiness, and drift.
• An explicit autonomous engineering workflow catalog with change and operational workflow patterns.
• Complete testing strategy, CI/CD, artifact, migration, compatibility, progressive-delivery, rollback, and production-verification coverage.
• Developer portal, service catalog, golden paths, scheduling, capacity, cost, fairness, resilience, disaster recovery, and human-agent control surfaces.
• Agentic threat modeling, workload identity, secrets, privacy, licensing, and compliance coverage.
• Evaluation science, controlled experimentation, capability optimization, learning from success, and regression control.
• Seven executable lab specifications for certification, repository onboarding, adversarial defense, delivery rollback, incident response, continual improvement, and disaster recovery.
• Public capability coverage, maturity, reviewer guidance, and feedback paths.

Changed

• Expanded Builder and Deep Study paths beyond pull-request generation into capability management, delivery, production verification, and learning.
• Reworked topic discovery around section, persona, lifecycle, maturity, and risk filters.
• Made document maturity visible in discovery views.
• Clarified that review-ready guide does not imply an operationally proven implementation.

Known evidence boundaries

• The original golden-path execution lab remains blocked until its exact runtime prerequisites and accepted browser evidence exist.
• New executable labs are review-ready specifications; accepted lab runs are not yet claimed.
• The complete mission-to-production and governed-improvement paths remain architectural targets until versioned operational evidence is retained.