landing: restructure spiritwriter.ai around the "memory you own" narrative

[aaronmarkham]

Summary

Restructures the spiritwriter.ai root landing (templates/landing.html, rendered by zeitghost) around the local-first / "memory you own" narrative from the website-copy draft. Keeps the existing editorial design system intact — only the body content + a little component CSS changed.

Narrative (new spine)

• Hero: "Agent memory you own." — pip install + GitHub CTAs, a "see it running → frio · news" line, and the marginalia now carries the sharpest line: "Show everything, or prove nothing leaked — same code."
• I. The ache — "You've built this before." (the re-glue-it-every-app problem)
• II. One substrate, trust is a dial — with the layer diagram, the page's key asset: your orchestrator / retriever sit on spiritwriter ("not an agent framework, not a vector DB — what those sit on").
• III. Proven twice, at the extremes — two cards: frio (maximum privacy) + news (maximum transparency, now "every fact signed and traced" — accurate after the provenance work).
• IV. Adoption — additive, local-first, agent-readable, cheap to leave.
• V. The numbers — measured precision/recall, honestly framed.
• VI. What it ends — problem → solution grid.
• Closing: "Own your agents' memory."

Design + safety

• Design system preserved — same fonts (Fraunces / Newsreader / JetBrains Mono), parchment + terracotta palette, paper grain, reveal-on-scroll, section folios. New CSS is additive only: .stack, .proof, .ends-table, .section-lede, .see-running.
• Test-critical bits untouched — version-chip {% if %} guards, footer commit-link + local fallback, and og:image/twitter:card social meta all preserved. Title/description/OG/Twitter copy updated to the new positioning.
• Accuracy (carried from the copy review): shards described as JSON on disk + one JSON index (not "one SQLite file" / "JSON-LD"); entity resolution stated as 100% precision, 0 false merges, 100% of same-entity matches surfaced (the recall@any-tier = 1.000 number), not a bare "catches 100%". Phalanx / CMC-Lite / ESS jargon kept off the surface; footer spec link relabeled "entity resolution".

Test plan

• python -m pytest tests/ — 67 passed
• Landing template tests green: version chip omitted when version unknown (no X.Y.Z literal leaks), commit link wired with full SHA href + short SHA text, social meta present
• Full render with realistic context — 16/16 structure checks (all sections, stack diagram, 2 proof cards, 6-row ends-table, both copy buttons, no jargon, no JSON-LD)

Note

Visual review recommended on the deployed preview (zeitghost build renders this at the root). The three new components (stack diagram, proof cards, ends-table) reuse the established type/color tokens, but a human eye on spacing/responsive is worth it before this fronts the brand.

🤖 Generated with Claude Code