# Validation record Validated on 2026-09-26 in Linux x64 with .NET SDK 10.0.101, .NET runtime 10.0.1, Node.js 22.21.0, and authenticated Codex CLI 0.155.1. Runtime-generated reports and compiled outputs are intentionally ignored; the useful generated project files are tracked. ## Completed checks | Check | Result | | ------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Main build | `dotnet build --configuration Release` passed with 0 warnings and 0 errors. The root project now explicitly includes only `Program.cs` and `SuperIntelligenceBuilder.cs`, so nested example sources are not accidentally compiled into it. | | Example builds | All three example projects restored and built in Release with 0 warnings and 0 errors. | | Maintained HTML | `html-validate` 10.4.0 passed all 9 HTML pages: the manual, LLM Recipes, compatibility redirect, About, examples landing page, all three example detail pages, and the generated repair-cafe page. The repository's existing lowercase-doctype, XHTML-void, and explicit-region style preferences were disabled; the run found no semantic or ARIA errors. | | Local links and anchors | A read-only Node.js check resolved every local `href` and fragment across the manual, LLM Recipes, redirect, About, examples, three detail pages, and generated repair-cafe demonstration: 9 pages checked, 0 failures. | | Help and default dry run | Main help ran. The root and all three examples defaulted to a plain-text `HYPER INTELLIGENCE / DRY-RUN MODE` plan. Each used normal harness discovery, displayed the selected harness and exact fluent statement, emitted no ANSI color, created no files or reports, requested no confirmation, and performed no CLI probe or launch. With Codex removed from `PATH`, dry run still exited 0 and reported that no supported harness was found. | | Explicit execution gate | The main program and all examples pass `Options.Proceed = true` only for the exact `--yes,please,proceed` option. A live-path test with the flag and Codex removed from `PATH` reached `AgentNotFound` without creating the missing instruction file. | | Authenticated agent integration | Three independent calls through `SuperIntelligenceBuilder.WithFile` launched Codex CLI 0.155.1 using its workspace-write adapter. Every process exited 0 and produced a matching `super-intelligence-builder/v1` completed report. | | Specification example | Generated `AGENTS.md`, constitution, index, product specification, plan, and tasks. The agent validated 12 local Markdown links and traceability across 17 requirements, 15 acceptance criteria, 8 plan decisions, 11 named tests, and 12 ordered tasks. | | Static-design example | Generated agent policy, five design documents, and a working local HTML/CSS/JavaScript site. HTML structure, local links, anchors, ARIA references, JavaScript syntax, CSS balance, responsive/reduced-motion rules, and remote-dependency absence passed. The final review found and corrected a 2.80:1 volunteer-eyebrow pairing; white on orange now measures 4.72:1. Local callouts no longer create nested complementary landmarks. | | Enterprise C# example | Generated `AGENTS.md`, `.editorconfig`, `Directory.Build.props`, and four engineering guides. XML, EditorConfig structure, Markdown links, internal consistency, and a no-warning build passed. | | Existing-file refusal | Inputs `y` and `yes` were rejected as insufficient. A following exact `no` returned `ResultState.Declined`, exited 0, launched no agent, and created no run directory. | | Existing-file approval | Exact `yes, please proceed` advanced past confirmation to agent resolution. Codex was intentionally removed from that test process's `PATH`, so it then failed safely with `AgentNotFound` without launching an agent. | | Fail-closed input | End-of-input at the confirmation prompt returned `ConfirmationRequired`, exited nonzero, and launched no agent. | | Civilised diagnostics | Builder failures retained their typed `ErrorCode` and technical cause while being wrapped in the Bespoke-inspired `A Regrettable Circumstance` format. | | Browser behavior | The original Playwright 1.55 full-site and interaction passes remain recorded in `FINAL-WEBSITE-REVIEW.md`. The LLM Recipes integration added a 7-test rendered pass: Manual, Examples, Recipes, and About returned 200; all exposed the same four-item navigation and correct current-page state; Manual and Recipes had no overflow at 1440×900, 390×844, or 320×720; Recipes filtering, anchors, copy controls, syntax spans, outline, and keyboard entry worked; console and page errors were empty. Axe reported 0 violations on all four peer pages at desktop and mobile widths. The pass found and fixed About's broad `nav a` section selector before succeeding. | | Width and masthead follow-up | The annotated `review-evidence/bug.md` report was reproduced and fixed. Manual rail selectors are now scoped to `.rail nav`; all four global links share one row on every documentation page. The 790px main cap was removed, ordinary prose now permits up to `120ch`, and desktop frame padding is exactly 30px. Playwright measured all 7 documentation pages at 1494×939, 390×844, and 320×720: desktop main width 834px with a 30px right gap; mobile main widths 354px and 284px with 18px side gaps; 0px tab-top spread; 0px horizontal overflow; no console errors. Axe found 0 violations on Manual, Recipes, About, and Examples at desktop and 320px. | | Screenshot review | The 21-image pre-fix set under `review-evidence/final-pass/` and matching post-fix set under `final-after/` preserve the earlier whole-site comparison. The LLM Recipes pass added 8 full-page captures under `review-evidence/llm-recipes/`: Manual and Recipes at 1440×900, 390×844, and 320×720, plus Examples and About at 390×844. Direct comparison confirmed the shared shell, content-driven differences, readable four-link navigation, and absence of unintended breakpoint changes. PNG evidence is intentionally ignored. | | Final editorial review | `FINAL-WEBSITE-REVIEW.md` preserves the prior whole-site content audit. For LLM Recipes, `editorial/llm-recipes-draft.md` is the 626-line expansive pass and `editorial/llm-recipes-line-review.md` is the 1,115-line separate review of every sentence or logical command unit. The HTML applies the accepted keep, rewrite, remove, move, and structural decisions; it omits the unused clone-refresh branch and one redundant slogan. | | About integration | About's historical body remains unchanged. Its global navigation now includes Recipes and no longer presents Guardrails as a peer page. The existing broad section-navigation CSS and JavaScript selectors were scoped to `.rail nav`, preventing the semantic global `