Where the data comes from.
Every filing on Sanket originates from a public regulatory source. We do not purchase data, scrape competitor products, or invent entries.
Sanket pulls from five primary sources, in this order of priority:
We pull twice each trading day — once mid-session (12:30 IST) and once after market close (16:30 IST). Best-effort latency from exchange publication to Sanket appearance is 30 minutes, not milliseconds.
We do not currently scrape SEBI's portal directly. Their PDF disclosures lag exchange filings by 1–3 days and require OCR — both add error risk we don't want compounded into a signal score.
How we map a shell company back to its parent.
This is the hardest, most opinion-driven part of Sanket. It's also the difference between a feed and intelligence.
SEBI requires the legal entity to be named in the filing — not the parent group. So a Reliance promoter buy might appear under any of approximately 45 different LLP, trust, or holding-company names. To a retail investor scrolling raw filings, "Tattvam Enterprises LLP" and "Reliance Industries" look unrelated.
We maintain a curated mapping table covering the Nifty 500 plus actively-traded midcaps — currently around 95 promoter families, growing weekly. New entities trigger a manual review flag before they enter the live feed. No unmapped filing is published with a parent attribution we cannot defend.
This mapping is the single most labour-intensive part of the operation. It is also the moat. We expect to maintain it manually for the first year, then partially automate using filing-cluster analysis.
The four-factor signal score.
Every filing is scored from 0 to 100. The score is deterministic — same inputs, same output, every time. The commentary on top is the editorial layer.
Score tiers and what they mean:
- 80–100 · High signal. Reviewed by a human before going to the daily Telegram broadcast. Eligible for the Saturday Digest.
- 50–79 · Moderate signal. Auto-published to the dashboard with commentary. Surfaced in watchlists. Not in the digest.
- 0–49 · Low signal / noise. Visible in the dashboard with a "likely administrative" tag. Routine ESOP exercises and small trust transfers live here.
The commentary line you see attached to every event ("Damani has added for three consecutive quarters...") is generated by an LLM trained on our editorial style guide, then reviewed by a human for any score above 80. Lower-scored events use the LLM commentary directly.
The human review layer.
An algorithm can score. Only a human can decide whether something deserves a sentence in front of a serious reader.
Every event scoring 80 or above passes through editorial review before it's published in the daily Telegram broadcast. The review window opens at 7am IST each day. Lower-scored events are auto-published.
The Saturday Digest — three stories per week, 800 to 1,200 words — is human-curated and human-written end to end. The LLM produces draft commentary; the editor decides which patterns are real, which are coincidence, and which deserve reader attention.
This single point of human dependency is acknowledged. We are designing the workflow so that one editor can sustainably handle weekly production for the first 18 months. Beyond that, we add an editorial second seat.
What we deliberately exclude.
More data is not always better data. Some things create noise without signal.
- ESOPs and trust transfers below ₹5 Cr — administrative, not directional
- Foreign-promoter intra-group transfers with no economic effect on holding
- Mutual fund disclosures — different regulatory regime, different signal class
- FII/DII transactions — institutional flow, not promoter signal
- Suspicion or rumour — we only publish filed transactions
- Insider tip-offs — we will not accept or publish them, ever
Every excluded category is an explicit choice. The list above will evolve. When it changes, this page is updated and the change is announced in the Saturday Digest.
What Sanket is not.
A useful product is precise about its claims. These are ours.
- Not investment advice. We do not recommend buying, selling, or holding. We surface, score, and contextualise public filings. The decision is yours.
- Not real-time. Best-effort latency is 30 minutes from exchange publication. If you need millisecond data, you should be on a Bloomberg terminal.
- Not predictive. A high signal score is not a prediction. It is a measure of how unusual or pattern-fitting a filing is — that is correlation with past patterns, not a forecast.
- Not exhaustive across all listed stocks. Coverage is currently Nifty 500 + active midcaps. Full small-cap coverage is roadmap, not present.
- Not a substitute for primary diligence. Promoter activity is one signal among many. Sanket is a lens, not a verdict.
What we're still figuring out.
Listing what we don't know is a trust signal. So is keeping the list short.
- Fixed-income promoter activity — NCD pledges, debenture transfers — currently out of scope for V1. Likely added by Q1 2027.
- Derivative-based promoter exposure — visible in some filings but not yet integrated into the score.
- Cross-border MNC promoter trimming patterns — early signal class, methodology evolving as data accumulates.
- Backtest depth — first formal case studies covering 100+ historical high-signal events will publish before paid tier launch.
- Score weight calibration — the 30/30/25/25 split is our starting point, not a final answer. We will revise based on observed predictive value over time, transparently.
If you are a serious user with thoughts on any of these, write to us. The methodology is opinionated but not closed — and the people who will most help us refine it are the ones already paying attention.