What was seized on the frontier, where, and who reported it.

Interdiction news is scattered. Ministry releases, force posts and regional papers each carry part of it, in prose, with quantities in kilograms, tablets, bottles, heads and crores. Punjab alone reported 313 drones and 385 kg of heroin recovered in 2025, and 161 drones and 392 kg in the first seven months of 2026 (The Tribune, 31 Jul 2026). Analysts rebuild these tallies by hand.

Tools that try to help often turn reports into lists of people with risk scores. That is neither lawful targeting nor useful intelligence: the people named in a press release are accused, not convicted, and court records name petitioners and witnesses. Our first version did this. Version 2 removed it.

• Counts frontier events in 13 categories, from narcotics, gold and cattle to drones, wildlife and border incidents, by day, district and source.
• Reads the quantities each report states, in normalised units (kilograms, tablets, bottles, head of cattle, rounds, rupees), and keeps one figure per item per report so a total and its parts are not counted twice.
• Places each report at a town, district or state, and says which.
• Flags districts and categories above their own history, places that keep appearing, stated routes, and places reported by more than one kind of source.
• Answers plain questions ("gold seizures West Bengal last year") over the event store, in the browser.
• Publishes profiles of places, agencies and categories, and a daily brief in which every line links to its reports.

1. FetchPIB releases of three ministries, 22 news feeds, GDELT, and the BSF and CBI listings, with an honest status for each source.
2. Keep eventsA report must name a category and an interdiction. Speeches, ceremonies and statistics are dropped.
3. ExtractPlaces, agencies and quantities. Names of people are found only so they can be removed.
4. MergeThe same seizure reported by two outlets becomes one event that lists both.
5. PublishA privacy gate scans every export before it is written. Every number carries its source and time.

Formulas, thresholds, the category list and known failure modes are in the methods note. The console's Sources page shows each source's last check and why it failed when it did.

Console follows the Trinetra doctrine: events, not people. Private persons named in a source (accused, convicts, petitioners, victims, witnesses, relatives, junior personnel) are replaced with a role label in every headline, summary, page, export and API response. Public officials acting in their office may appear with that office, in the report view only. They are never scored or graphed.

Each build counts the private persons it withheld and scans every export for names that leaked before; the result is published in privacy.json.

Nothing in Console is simulated. Blocked sites and bot challenges are reported as found and never bypassed. Status refreshes from the latest build.

• Analysts at frontier and sector headquarters: this week's seizures by district and category, with the reports behind each number.
• State border cells: which districts and categories are above their own history, and which places keep coming back.
• Researchers and journalists: a sourced, de-duplicated record of what agencies announced, by place and quantity.

• No person entities, profiles, lists or scores, in any view, export or API response. Tests fail the build if one appears.
• No selection or filtering by religion, caste, ethnicity, language or nationality. Community and nationality words attached to people are neutralised in exported text.
• No private messaging, intercepts, personal social media, court text, or citizenship-adjudication records.
• Every operator action is logged; the tamper-evident record is kept in Sakshi.
• Software recommends; trained people decide. Nothing here is grounds for action against any person.

• Coverage is what is published and reachable. Routine BSF seizures are posted on social media, which we do not ingest, and the BSF and CBI sites are blocked from our network. Counts are a sample of activity, not a measure of it.
• Person detection is pattern based. It can miss a name with no cue around it, or redact a place name as [person].
• Quantities are as stated by the source. The per-report maximum undercounts reports that list separate seizures without a total.
• Dates are publication dates; places are town or district reference points, about 10 km.
• Not an official record. Confirm with the issuing agency.

Pilot with one frontier or one state border cell: add your own release feeds and places, and get a sourced daily brief that never profiles people. contact@trinetra.in