landing: demote Phalanx from marketing, lead with precision (backed by spiritwriter-core PR #58)

[aaronmarkham]

Summary

Marketing-copy rewrite of templates/landing.html to demote insider jargon ("Phalanx") from the homepage and replace the wishy-washy ≥85% recall sidebar stat with the precision-first claim now backed by spiritwriter-core PR #58's measured campaign numbers.

Driven by the decisions captured in spiritwriter-core's docs/cleanup/cmc-phalanx-canonicalize.md (decisions A/B/C: Phalanx is a deep-dive system name, not marketing). Backed by the multi-corpus benchmark campaign in docs/benchmarks/runs-log.md.

Changes

All 6 Phalanx mentions demoted to capability-shaped language (per cleanup doc §3 table):

Location	Before	After
Nav (line 898)	Phalanx	Resolution (anchor unchanged)
Manifesto §I (line 965)	"And Phalanx resolves entities..."	"And entity resolution works by defining fields, not surface forms..."
Primitive vi header (line 1031)	*Phalanx* — entity resolution	Entity *resolution* (matches header form of other primitives)
Primitive vi tag (line 1033)	cmc-lite · ess digest · tiered	sqlite-backed · domain-agnostic · zero-infrastructure
§IV folio (line 1124)	folio iv — phalanx	folio iv — resolution
Bear prose (line 1133)	"Phalanx hashes..."	"The resolver hashes the defining fields..."

Primitive vi body also extended with the competitive frame from the cleanup doc rewrite: "No graph database to operate, no embedding service to call — define your identifying fields, hand in records, get canonical IDs back."

Sidebar stat (line 1153) replaced with the precision-first claim now backed by measured numbers:

Before: ≥85% recall on semantic duplicates with ≤5% false-merge rate
After: 100% auto-merge precision across 5 benchmark corpora · 0 false merges · 12/12 hand-curated collision pairs correctly distinguished

The ≥85% was the cmc-spec's target for the full CMC pipeline (with LLM clustering). CMC-Lite is the deterministic subset and doesn't claim that number. PR #58's falsification battery confirmed the precision invariant holds across 5 corpora and hostile collision pairs — that's the defensible headline.

New homepage entry ix — Shingled extraction added per cleanup doc §2:

Turn long-form text into atoms without losing facts at chunk boundaries. Overlapping windows + multi-pass extraction; only atoms that appear across multiple passes survive. The result feeds the shard store and the entity-resolution engine: extract once, resolve continuously.
● overlapping windows · n-of-k voting · checkpoint-resumable

Rationale: shingled extraction was previously bundled under Phalanx by accident. It's a distinct primitive (text → atoms vs. shards → store vs. records → canonical) and it closes the loop the other primitives describe (now the reader knows how atoms get made).

What's NOT touched

• Footer Specs section still has a cmc-lite link to the spec file — appropriate audience for that name (deep-dive readers), per cleanup doc decision A.
• The Bear Problem deep-dive section's structure is preserved; only the proper-noun "Phalanx" is replaced with capability language ("the resolver").
• Visual design / CSS / typography unchanged. Only content edits.

Test plan

• All 6 Phalanx mentions removed (grep -n Phalanx returns nothing)
• cmc-lite only appears in the footer Specs link target (deep-dive territory, by design)
• ≥85% removed
• New entry ix added after entry viii (grep "<div class=\"num\">ix" finds the new primitive card)
• Manual visual review: 9 primitive cards in the 2-column grid handles the orphan card gracefully (CSS already supports odd counts)

Cross-repo references

• spiritwriter-core PR #58 — multi-corpus benchmark campaign that produced the precision claim (merged 2026-05-28)
• cleanup/cmc-phalanx-canonicalize.md — coordination doc with the decisions + the §3 per-repo action items this PR implements
• docs/benchmarks/runs-log.md — the measured campaign report; "What this means for marketing claims" section is the source for the new sidebar stat copy

🤖 Generated with Claude Code