A clearer window into Habitat

What makes a small AI village understandable at first glance? The homepage now introduces the idea, four authored starting characters and one real, dated test before the daily-life pictures and latest build news. Pip’s catch-to-diary receipt is explicitly a controlled historical fixture from 15 October, not an invented autonomous decision or private memory. Native disclosures keep profiles, sources and the complete settings guide nearby. Five navigation destinations group the six existing routes; English and German keep matching addresses. World and Residents lead the homepage and default RSS, while Website changes stay in their own journal section. Native calendar-day dividers group recent, earlier and Website notes without merging entries. Stable entry IDs, individual permalinks, full original notes and historical RSS GUIDs retain the earlier history. The Design library adds search, type/environment, preview and animation filters, sorting, counts and shareable queries. Models start paused and load only when opened. Eight web-only texture derivatives retain decoded pixels, geometry, materials and animation payloads; their exact newly rendered posters and manifests were individually reviewed. Source models and original recorded prompts remain unchanged. Static cards, all journal entries and bilingual offline routes remain readable without JavaScript. Compact startup metadata enhances the existing HTML; full history and provenance load only on request. No simulator or resident data is fetched by the overview. Four measured values, all from 16 October 2026: (1) static integrity improved from 14 broken-anchor occurrences across 12 baseline pages to zero errors across 304 pages and 10,966 references on the reviewed candidate (static-integrity.json and integrity-candidate.json; filesystem references and metadata only). (2) On release snapshot index f1e24d4422ee521338190286995498838cc03ba0b926ff035ca61fe0a7d5e6cc, median mobile Home transfer was 384,665 bytes versus 1,644,948 bytes on the earlier reviewed-story snapshot cf9f3acc, from three Lighthouse 13.5.0 runs each over the uncompressed local static server. This is lab transfer, not field speed. (3) Fourteen selected optional model downloads total 23,574,916 bytes versus 27,657,108 before (14.76% smaller), from exact candidate GLB receipts measured 17:21–17:30 UTC. (4) The release axe-core 4.14.0 audit found zero automatic violations across 48 language/page/width/theme conditions; all 48 startup checks retained static content and made no initial full JSON, viewer or model request. Automated checks are not a screen-reader certification. The website-focused Python suite passed 207 tests with one expensive archive test deselected; that actual archive passed separately in 75.78 seconds. The later individual-capture publication boundary passed 52 focused checks. Actual startup/menu/chronology/journal/motion Node tests passed. The in-memory offline audit checked 304 bilingual pages and 9,712 local references with zero broken paths. Browser inspection covered six routes at 320, 360, 768, 1024 and 1280 pixels in both themes, plus the paired English/German 390/1440 matrix. Model pause/play, library filter counts and URL state, journal filtering, matching language permalinks and menu Escape/focus were exercised. Screenshots revealed a hidden-state styling defect and an unfinished smooth scroll; those exact failed frames stay private and were recaptured after correction. The initial five visual iterations and the matched baseline remain in immutable evidence. The first square artwork crop, restored full composition, compact startup, contrast/filter corrections and final library ordering each have distinct source identities. Supporting reviewed captures include 20261016T173219529000-1b61f7533754, 20261016T173642952000-6f5d17bced7e, 20261016T175407078000-33ad84a32217, 20261016T175913379000-43acbb61918a and 20261016T180239230000-3738983e4b43. These are website captures, not new gameplay evidence. The local mobile Performance medians passed (Home 93, Assets 91), Accessibility100 and CLS under0.1, but local LCP remained above2.5 seconds (Home 3.067 seconds, Assets 3.372 seconds) before the final library poster-order/priority correction. Preserve those failures. The compressed live edition needs a separate post-publication audit. Ordinary gameplay footage, human comprehension sessions, a full screen-reader/text-zoom/touch pass, external feed-validator checks and field INP remain pending. The complete 46-item adaptation register records implemented and deliberately adapted suggestions. Publication and live byte verification are recorded separately in the website-sync receipt. Next is to verify the deployed delivery and gather those remaining human/footage checks. Follow-up fallback review kept the first no-JavaScript footer failure as evidence: Refresh is now hidden without JavaScript, and static German model-download actions are localized. The day-divider/chronology/static checks passed 66 tests, preserving stable IDs and within-day ordering. Final integration: a parallel native build changed the delivered game, so the first finalizer correctly blocked stale showcase photos. The producer supplied separately reviewed current village/bonfire/orchard/lake frames from unity-social-ui-31ef8556e74d-20261016181124. The website agent independently inspected pixels and sidecars, pinned each ID, selected distinct matching views and reviewed their 12 responsive derivatives. Sixth-iteration website captures 20261016T181713417000-d11c2a80f709 and 20261016T181714446000-ce15b5fd84ee show actual day dividers and this refreshed world section; their exact 5e3e28e33fde993e website snapshot is retained. No social memories, private transcripts, new gameplay claims or model calls were added to the website. Follow-up no-JavaScript 24/24 and focused axe/startup 8/8 checks passed; all 147 historical entry texts remain available. Current source/footer/localization/static grouping checks passed 80 tests. The final compressed live performance check remains separate. Live delivery and final acceptance, measured 16 October 2026: website commit 2ad4cec7f7d9182b84638bfa1782024227e907ab delivered through the normal finalizer; at 18:21:14 UTC all 1,185 checked deployed files matched. The final static snapshot had 306 HTML pages and 11,081 references with zero errors. Seventh-iteration capture 20261016T182235540000-93bb12b0aa6b shows the actual deployed introduction. Three live-host simulated-mobile Lighthouse 13.5.0 runs per route at 18:24 UTC all scored Performance 100/Accessibility 100. Medians: Home LCP 1.673 seconds, CLS 0.00152 and 203,258 bytes; Design library LCP 1.590 seconds, CLS 0.01975 and 254,675 bytes. All six used actual HTTP/2+Brotli for HTML/CSS/JS, no initial full JSON/viewer/model request and no runtime warning. Eight selected decoded sources matched before and after, against website index 6965969b20fe5de1373dd168ede93ff275ff8f94125e1c9177493b83ab9fe965. Every accepted numeric lab target now passes; the earlier local failures remain preserved for those server/source conditions. These are lab values, not field INP. Ordinary gameplay footage, human comprehension and the full manual accessibility/touch/zoom checks remain pending. The local overview also renders only the requested page or permalink before creating its matching language edition. On the current 699-capture/151-entry local input, one in-memory comparison measured 98 ms for the selected homepage editions versus 8.89 seconds for all 314 pages, with byte-identical output. This excludes asset-review/disk/network time. The selective-render and overview checks passed 113 tests; full publication/export behavior and unknown-route 404/privacy boundaries remain intact. Both existing read-only overview processes were reloaded and returned the current heading/compact record; no game/runtime process was restarted.