nyc-ride-market

L11 · Publish and findings

Wave 3 · Phase 3–4 · Spec §9.2, §11 (phase 3), CLAUDE.md “Language on the site” · Feeds L12–L17

Goal

make publish turns metric outputs into the contract files the site reads, including every title, subtitle and finding sentence, all generated from computed values and never typed.

Owns

• pipeline/publish.py, pipeline/findings.py, tests/test_publish.py, tests/test_findings.py
• Committed output: site/src/data/** (real data only)

Tasks

• Run metrics through pipeline.metrics; shape each file exactly to contracts/site/; as_of_month and generated_at on every file
• Refuse to write unless validation_summary.json passes. --dev writes locally with "validated": false, which deploy refuses
• Copy taxi_zones.simplified.geojson; zone files 1–263 each under 50 KB; thin zones produce nulls, not errors
• findings.py: one builder per question (Q1–Q15), plus page titles and chart subtitles. Numbers formatted per contracts/README.md. Language: “estimated effect”; “per hour with a passenger”; taxi tips “card-only”
• Changes vs a year earlier compare the latest month with the same month of the previous year
• Remove stale files the contract no longer lists

Tests

• Factory aggregates → every file matches its contract example
• Every route stays within 1.5 MB
• Three or four finding sentences checked against values worked out by hand
• Generated text never contains “caused” or “causes”; taxi tip text says card-only; pay-per-hour text says “with a passenger”

Done when

• On validated aggregates (G3), make publish writes every contract file
• Q1–Q15 each has a generated finding (G4)

Not this lane

• Metric formulas (L09, L10); page layout (L12–L16)

---

Status log

Only this lane writes here. Newest first.

Date	State	Note
	not started

Blocked: none

Hand-off

• Built:
• Checked:
• Known limits:
• Follow-ups: