India Roadmap Instrument

Longitudinal research instrument: India's condition and trajectory, UPA baseline (frozen May 2014) through the Modi-era terms. Private audience. See CLAUDE.md for the rules that govern the data and the build — they are binding, not advisory.

Next.js, static export, no server, no database, no auth.

Commands

npm run validate           # gate: schemas + cross-references + instrument rules over /data
npm run validate:incoming  # merge gate: also checks /data/incoming drops
npm run validate:strict    # warnings become errors
npm run validate:selftest  # proves the gate is closed (broken fixtures must fail)
npm run dev                # local development
npm run build              # validate, then static export to out/
npm run typecheck          # tsc --noEmit

npm run build runs the validator first, so an invalid repository cannot compile and therefore cannot deploy. Vercel runs the same script.

Layout

data/
  series/*.json        indicator time series, India + peer panel (BGD, VNM, IDN, CHN)
  ledger/*.json        reform / event / episode records (L-xxxx)
  provenance.json      measurement-dispute records (P-xx), first-class citizens
  incoming/            drops from research sessions — validated, then merged
schemas/               the contract; research sessions author against these
tools/                 validator CLI
tests/fixtures/        deliberately invalid records proving the gate rejects them
lib/                   build-time data access and formatting
app/                   routes (static export)
components/            status marks, tier tags, series tables with break seams
docs/verification-log.md   what was pinned this cycle, and the open queue

A data file may hold one record or an array of records. The layer is decided by path, not by contents.

Routes

Route	What it holds
/	what is loaded, by layer, term and domain
/domains/, /domains/[domain]	the fourteen fixed domains
/series/, /series/[id]	series with points, breaks as seams, source + tier + provenance
/ledger/, /ledger/[id]	records with assessment, both cases, tiered sources
/provenance/, /provenance/[id]	disputes, competing accounts, what they touch
/terms/, /terms/[term]	counts of assessments per term — never a grade
/peers/	peer panel, vintage shown with the values (P-09)
/counterfactual/	both methods stated; no counterfactual computed yet
/method/	status marks, tiers, splicing policy, the gate

Views beyond Phase 0 (charting, timelines, filtering, the counterfactual computation) are marked as scaffolds in place, so the gap is visible rather than implied.

Adding data

1. Research session drops files under data/incoming/, mirroring the layer shape.
2. npm run validate:incoming — nothing merges until this is clean.
3. Move the records into the canonical layer files.
4. npm run validate, then log what was pinned in docs/verification-log.md.

/data is additive-only: corrections are made by editing the record and noting the change in the verification log, never by silent deletion.