LincPayandClaude Opus 5
9d307c2d20
The Reports page carried a BENCHMARKS block of invented figures. It now reads GET /admin/model/metrics and the block is deleted, with a comment recording what it held and why it is gone. No fabricated number remains anywhere in the product. The comparison chart sorts by value rather than by kind, which puts the obligation_ratio_only baseline at 0.768 above the LightGBM model at 0.760. Grouping models above baselines would have hidden that, which is the original problem wearing better clothes. Baselines are drawn grey, models blue. The synthetic-label share travels with every figure, so accuracy cannot be read here without the caveat. fmt() prints an em dash for an absent metric, never 0.00 — a missing measurement and a measured zero are different claims. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
UdyamLinc — UI
Three surfaces in one Vite/React app, matching the three institutional tiers. Full project documentation, including the demo path and what is real versus seeded, lives in ../UdyamLinc-Backend/README.md.
npm install
npm run dev # http://localhost:5182
The backend is expected at http://localhost:8082/api/v1/udyam. Override with VITE_UDYAM_API_BASE_URL in .env.local.
The three surfaces
| Route | Tier | Shell |
|---|---|---|
/nabard |
NABARD — aggregates only | features/nabard |
/officer |
Bank console — one institution's book | features/officer |
/app |
Enterprise PWA — offline, Hindi-first | features/enterprise |
Both staff tiers share one sign-in form and land in different portals; homeFor() picks the destination from the role. That routing is a convenience only — the backend re-derives the tier from the signed JWT and refuses the other surface with a 403, so a tampered session gains nothing.
The NABARD portal is deliberately a different chrome from the bank console. An apex user and a branch officer are not two ranks of one job, and a portal that looked identical would invite the assumption that the apex user is a bank user with more rights.
The house pattern
Every screen follows the same shape, and new ones should:
const { data, isError } = useQuery({ queryKey: [...], queryFn: () => get<T>('/path') })
const demo = isError
const rows = data ?? demoRows() // in-file fallback from lib/demo/
…
{demo && <StatusPill status="Demo data" />}
A screen must render with the backend down — field devices are on patchy 2G — but it must say so. The "Demo data" pill means that screen's call failed, not that the whole app is a mock.
Demo fallbacks deliberately include failure states: the data-entry fallback contains rejected rows, and the NABARD fallback contains a bank below the suppression threshold. A demo where everything is clean teaches the wrong thing — a reviewer who only ever sees green rows will not know the rejection panel exists.
Screens that carry the argument
/officer/rules— every trip-wire with its outcome, including the ones that were skipped and why. This is the screen that shows the system refusing to fire on stale data./officer/feeds— provenance. Mandi rows carry a source:Agmarknet, or an amber Hand chip for a price somebody typed. By the time a price reaches a feature the two are indistinguishable, so the distinction has to be visible here./officer/data-entry— upload, preview, commit. Accepted, rejected and overwriting rows stay in three separate groups./officer/integration— core banking / LMS. The mapping table is the argument: each lender's core names the same eleven fields differently, and a per-bank mapping is what absorbs that. Simulated adapters are badged as simulated on every screen they appear on, and unmapped vendor columns are shown being dropped at the boundary./nabard— the India choropleth and the zone → state → district drill-down. States with no portfolio are drawn in a neutral grey that is nowhere on the stress ramp, so absence is never read as health.
Checks
npx tsc -b # types
npx oxlint src # lint
npm run build # production build
All three are expected to pass before a commit.
Attribution
State boundaries by the DataMeet India community (datameet/maps), used under CC BY 4.0. Simplified for display and not an authoritative depiction of any boundary. The credit renders on the map itself, as the licence requires.