Local preparation evidence — 2026-10-11

Branch: landing; isolated worktree .worktrees/landing-refresh. Opening source: 954a8d800576fec24acc9edfd3b93aad2ac15450. Evidence covers the local font/preparation diff, not a remote published revision. Primary dev was occupied by unrelated community work and was not changed by this task.

Implemented and verified

• Local variable Manrope + Source Sans 3, four content-hash WOFF2 files and two original OFL notices, recorded official HTTPS provenance and bounded build copy.
• Meaningful font regression observed RED (expected four emitted font references, got zero), then GREEN after CSS/build integration. Existing tests retained.
• npm --prefix landing run typecheck: PASS.
• npm --prefix landing run build: PASS; after the final CSS provenance-comment correction, emitted CSS styles.c083a948.css, JS main.72d6a45f.js.
• npm --prefix landing run check: PASS, including final artifact.
• npm --prefix landing test: 25/25 PASS after font behavior integration. Later source edits were documentation, reference tooling and a CSS comment only; build/check and actual browser font loading were reverified, without repeating an unchanged aggregate suite.
• bash scripts/tests/test-gate-selection.sh: 18/18 PASS in the permitted execution environment. An initial sandbox-limited invocation could not run subprocesses; that attempt was not accepted as test evidence.
• FontTools inspection: valid variable Manrope and Source Sans 3 axes; required Portuguese accented glyphs available. Four WOFF2 files total 128,916 bytes; actual browser on the Portuguese page fetched the two Latin files (53,708 source bytes, excluding HTTP overhead). Performance follow-up must include font bytes; the existing checker is not a full Core Web Vitals measurement.
• Actual HTTP landing/dist browser: both requested font families loaded from local 200 responses, correct computed body/H1 families, no CSS injection.
• Reference browser: original images decoded, Next/Previous changed between the genuine captures, no page errors or horizontal overflow at 320/375/768/1024/ 1440px, desktop/mobile DPR2 renders inspected. This is bounded prototype QA, not completed LR32 accessibility or full tutorial acceptance.
• Eight original PNG hashes and dimensions validated; every accepted capture visually inspected for private data. Rejected music-overlay/history attempts were not included. Original dark theme restored and confirmed through the Ativar tema claro control; only the created temporary device UI XML removed.
• LR00–LR39 and TU01–TU30 IDs verified, new document links resolve, diff check and repository secret scan pass. Scoped new documentation/manifest/HTML secret surface separately reviewed because the repository scanner excludes docs.

Still pending

Production redesign, complete tutorials, populated/private/conditional lesson captures, device/source SHA proof, owning app acceptance, LR performance/accessibility exit, public domain/server responses, webmaster ownership and indexation evidence. No backend/mobile runtime code, storage reset, account/data submission, app font change, deployment, external URL submission, issue/PR/merge or wiki publication. SEO research is a dated design input, not proof of indexation or ranking.

Next action: LR00 reconciles these prepared artifacts and actual branch state, then proceeds with the next authorized bounded phase. Recheck the installed app and changing official SEO documentation at implementation time.