# Executable Product Projections Last updated: 2026-07-26 The three product contracts now produce concrete, generated navigation artifacts rather than only describing intended audiences. They still share the canonical Quarto chapters, source inventory, claim records, proofs, tests, and release state. No projection is allowed to become a second authority. ## Narrative technical book `products/narrative_product_spine.json` defines a 22-unit thesis-to-method route. Each unit uses one representative canonical chapter while routing its compound and specialist owners through the complete reference. Each selected chapter has one canonical core-claim reference and three explicit projections: a plain-language thesis for comprehension, a normative engineering rule for design responsibility, and a bounded machine contract with a noninheritance boundary. It also carries one reader question, running example, strongest objection, failure story, evidence that would change the conclusion, and handoff. The reader generator places all three projections in the opening orientation instead of asking one overloaded core-claim sentence to serve human, normative, and machine interfaces. It can materialize the bounded candidate with: ```bash python3 scripts/build_reader_edition.py \ --narrative-spine products/narrative_product_spine.json \ --output build/narrative_product ``` The generated orientation is derivative editorial navigation. It adds no evidence and changes no support state. The other 65 chapters remain visible in the architecture reference and are not rejected or deleted. The candidate is not a reviewed reader release; here, "reviewed" means that the final artifact gates have not yet been run, not that external-human review is required. Final-candidate continuity, repetition, accessibility, figure, and format gates still apply. No external-human review is a prepublication requirement. `products/narrative_unit_crosswalk.json` assigns every canonical chapter to exactly one unit. The generated product page now expands each unit into its specialist reference-owner list and gives every one of the 65 owners a direct Human-view chapter link. Each route also states the owner's canonical distinct responsibility, claim identity, claim label, and current support ceiling. This changes routing, not ownership: representative chapters do not absorb the sources, claims, evidence, or authority of their specialist owners. The root projection manifest binds the crosswalk digest, and product validation rejects omitted, hidden, stale, support-inflated, or multiply assigned owners. `products/narrative_running_example_trace.json` makes the running example cumulative rather than thematic. It begins with one human repository-change request and adds one distinct artifact at each unit, including the responsibility map, cost ledger, authority/failure map, claim record, constitutional disposition, replacement contract, command and plan, world-model reconciliation, semantic IR, context and memory records, verification and claim review, typed work and observed effects, substrate and development records, readiness/liveness decision, lifecycle-cost ledger, integrated trace, and book-change record. It is regenerated by `python3 scripts/build_narrative_running_example.py`; every step consumes only artifacts already carried forward and hands off to the next unit. `python3 scripts/validate_narrative_running_example.py` checks all 22 editorial contracts, including the three projections and each machine contract's noninheritance boundary, plus canonical selection order, scenario continuity, 23 cumulative artifact identities, prior causes, handoffs, and the no-promotion/no-release boundary. Six controls reject order drift, cumulative-state reset, an unknown consumed artifact, a skipped handoff, duplicate output identity, and editorial continuity presented as support or release approval. The earlier 15-chapter candidate's 20-page/40-view browser receipt remains historical. The new candidate has its own 27-page HTML render and passed 54 desktop/mobile page-view pairs, recorded in `docs/c1_c8_22_unit_reader_render_review_2026_07_25.md`. Authorial meaning-preservation, synthesis-figure, and automated accessibility-preparation checks are complete. Final line editing, full accessibility conformance, reader comprehension, and EPUB/PDF/DOCX/audio quality remain publication work. ## Architecture reference specification The generated architecture route is a complete 87-chapter lookup index in canonical manifest order. Every row carries the canonical chapter and core-claim identity, distinct responsibility, claim label, support ceiling, and assignment to one of the three defended contributions. Protocol schemas, implementation horizons, and the glossary remain supporting routes. Completeness of the index does not establish a deployed implementation or architecture quality. ## Evidence, proof, and release registry The evidence route begins from the canonical status and groups claim, source, proof, test, replay, release, review, and residual surfaces. Repository-backed JSON and Markdown records are copied into the projection as content-addressed snapshots with SHA-256 digests; rendered Quarto pages remain links to their owning public surfaces. A snapshot proves byte identity only, not claim truth, reviewer competence, or open-world record faithfulness. ## Build and release integration After a clean Quarto render and canonical-status build, release CI runs: ```bash python3 scripts/build_product_projections.py \ --output _site/products \ --status _site/status/canonical-public-status.json ``` The product pages are therefore created before the moving `/latest/` mirror and tested-site bundle are hashed. The deployment workflow still consumes only the exact tested bundle; it does not rebuild product pages. ## Validation boundary `scripts/validate_product_projections.py` regenerates all projections and the bounded reader source in temporary workspaces. It checks canonical order, complete routing, all three generated orientation projections, explicit machine-contract noninheritance, source digests, canonical counts, and unchanged `argument` support. Eight mutation controls reject a 23-unit narrative, a missing machine contract, a substantial-looking machine contract without a noninheritance boundary, out-of-order selection, a changed evidence snapshot, a specialist owner hidden from its unit, a stale responsibility, and an inflated support state. With `--site _site`, it also checks the rendered canonical-status binding, product manifests and pages, every chapter target, every visible specialist-owner link, every rendered evidence route, content-addressed snapshots, the current crosswalk digest, and the presence of the product manifest inside the moving `/latest/` mirror. These controls establish projection consistency. They are not independent editorial review, artifact approval, empirical architecture evidence, or a chapter-core support transition.