Every Trinetra dataset at a stable address, with its licence.

Trinetra's products publish their data as open files, from river discharge and heat outlooks to official alerts and treaty texts. Reusing one file meant knowing its path, reading its product's manifest to learn where the data came from, and working out the terms of each source by hand: 153 sources under 18 different sets of terms, from CC BY 4.0 and ODbL to official releases and news reports cited for their figures.

Without that, open data gets reused without its credit, simulated data loses its label, and a change of format breaks someone's code without warning.

• Builds one catalogue from every product's manifest: each dataset's stable /api/v1/ address, size, rows, when it was written, how fresh it is against its product's schedule, and whether it is real, simulated or partly simulated.
• Infers a compact schema for every JSON, JSON Lines, GeoJSON, CSV and XML feed file: fields, types, examples, ranges, formats, time zones and coordinate order, with a small sample.
• Groups per-place and per-day files into families with one address pattern, such as /api/v1/varuna/station/{station}.json, and lists every member.
• Lists every source with its stated licence and the credit line its provider asks for, and builds a credit line and a citation, in plain text and BibTeX, for each dataset.
• Checks the data contract on every build: UTC times, [lat, lon] order, units in field names, file sizes, manifests, and Sakshi's personal-data scan.
• Records every format change it detects between builds, and states what version 1 of the data API promises.

curl -s https://trinetrahq.com/api/v1/varuna/board.json | jq '.stations[0].today'

1. Read manifestsEach product's manifest lists its files, sources, licences and retrieval times.
2. Check every pathRefuse anything outside the product folder, dotfiles, caches and key files. Read nothing else.
3. Infer schemasVisit every value once; record types, formats, ranges, coordinate order and time zones.
4. Credit sourcesClassify each licence statement and attach the provider's own credit line.
5. PublishWrite the catalogue, one schema file per product and the change log; the portal reads them.

Formulas, rules and known failure modes are in the methods note. The generator runs offline: cd pipeline && python3 -m trinetra.setu run.

Setu adds no data of its own and simulates nothing. Its catalogue is CC BY 4.0, like the rest of Trinetra's derived data.

• Developers and integrators: addresses, schemas, samples and code that runs, so an app or a dashboard can use river, heat, sea and alert data directly.
• Researchers and journalists: where each number came from, when, whether it is simulated, and how to cite it.
• Agencies and partners: the terms of every source before reuse, and a catalogue they can run on their own exports, on their own servers.

• Only data that is already public. Setu reads the export folder and nothing else: no pipeline caches, no .env files, no keys, no network.
• The console's sign-in is a demo gate in the browser, not access control, and the portal says so.
• Values that look like credentials are shown as [redacted]. If Sakshi's scan flags a file, its sample is withheld.
• Every source keeps its licence and credit, and simulated data keeps its label, in the catalogue and in every citation.
• Events, places and conditions, not people. Reusers are asked not to combine the data to identify, profile or score individuals.

• Schemas are inferred from today's files. They describe the data; they are not a contract for every value.
• Sources are declared per product, so a dataset lists all of its product's sources even when it uses only some.
• The licence pages are a guide to each provider's stated terms, not legal advice. Open-Meteo's free API, behind much of the weather data, is for non-commercial use.
• Static files on a CDN, with no query interface and no uptime commitment. Products still in build can change their formats without notice; the changelog records each change Setu detects.

Building on Trinetra data, or need a dataset we do not publish yet? Tell us what you are making. Agencies can run the same catalogue over their own exports on-premise. contact@trinetra.in