BhavRadar grades every signal it fires against the median stock in the same universe over the same days. Beating that 50% of the time is what a coin does. Anything a signal is worth has to show up as the excess on top.
This page is generated from the live measurements. Nothing on it is typed by hand.
None of the 12 measured families shows an edge we can prove. Not one interval sits clear of the 50% null. We publish that because it is what the measurement says, and a number we cannot defend is worth less than nothing.
| Signal family | Independent bets | Beat the median stock | Median excess | Verdict |
|---|---|---|---|---|
| Delivery spike | 531 | 50% (45.9–54.3%, 43.9–56.3% once the bar is widened for 12 families) | +0.00% | no edge proven |
| Relative strength | 469 | 51% (46.4–55.5%, 44.4–57.5% once the bar is widened for 12 families) | +0.13% | no edge proven |
| Stealth accumulation | 459 | 49% (44.5–53.6%, 42.4–55.7% once the bar is widened for 12 families) | -0.09% | no edge proven |
| Follow-through after a footprint | 212 | 50% (43.3–56.7%, 40.3–59.7% once the bar is widened for 12 families) | +0.12% | no edge proven |
| Multi-institution consensus | 106 | 63% (53.7–71.8%, 49.3–75.2% once the bar is widened for 12 families) | +1.98% | no edge proven |
| Promoter stake rose | 65 | 49% (37.5–61.1%, 32.6–66.1% once the bar is widened for 12 families) | -0.44% | no edge proven |
| Two or more stacked | 64 | 37% (26.7–49.7%, 22.5–55.3% once the bar is widened for 12 families) | -1.85% | no edge proven |
| Buyers who accumulate together | 53 | 59% (45.1–70.7%, 39.2–75.5% once the bar is widened for 12 families) | +1.25% | no edge proven |
| Day-high break | 33 | 45% (29.8–62%, 24.1–68.6% once the bar is widened for 12 families) | -1.19% | no edge proven |
| Score crossing | 33 | 53% (35.2–67.5%, 28.9–73.6% once the bar is widened for 12 families) | +0.48% | no edge proven |
| Volume surge | 32 | 50% (33.6–66.4%, 27.4–72.6% once the bar is widened for 12 families) | -0.20% | no edge proven |
| VWAP reclaim | 32 | 46% (30.9–63.6%, 24.9–70.1% once the bar is widened for 12 families) | -1.04% | no edge proven |
| Fresh 52-week high | 11 | — | — | still accruing |
| Whale buy | 10 | — | — | still accruing |
2 families are still accruing. They are measured off their real fires as those fires mature, which is the only honest way to grade a signal that cannot be reconstructed from history — the intraday ones come off live feeds nobody archived. Until a family reaches 30 independent dates we show how many it has and nothing else. There is no rate to report yet, so there is none here.
Entry is a price you could have paid. NSE publishes bulk and block deals after the close, and delivery percentages arrive with the evening bhavcopy — so for any signal that cannot be known before the bell, entry is the NEXT session's close. Measuring from the signal-day close instead flatters every one of these numbers; we did exactly that until we caught it, and delivery's apparent edge disappeared entirely when we fixed it.
The benchmark is the median universe stock over the same window, so a gain in a rising market is not counted as skill.
The sample is INDEPENDENT DATES, never raw fires. Ten names lighting up on one day share one market backdrop, which is closer to one bet than ten — every confidence interval on this page is computed on those dates. Counting raw fires instead makes every interval look far tighter than the evidence supports.
We also looked harder than this table shows. We searched 72 conditional cells — each signal family split by whether the stock was already in an uptrend, a downtrend or going nowhere over the previous 60 sessions, at 20, 40 and 60 holding periods, plus three price-only swing patterns (pullback, 20-day high, reversal) at 5, 10 and 20 sessions graded net of a 0.60% round trip in costs, one measured retail rate. Testing many combinations guarantees some clear a 95% bar by luck, so the bar was widened for the number of tests. Nothing survived. 2 cleared the uncorrected bar, which is almost exactly the 3.6 that chance predicts from 72 tries.
Corporate-action jumps are excluded on both legs — a 1:5 split otherwise reads as −80%. A family needs 30 independent dates before any verdict is offered at all, and below that we show the sample rather than a rate.
Names where two or more counterparties that carried their disclosed positions bought inside a trailing 30 sessions. Market makers are excluded: 62% of the disclosed tape is a counterparty buying and selling the same name in the same session, which is quoting, not a position.
This pattern's own record: names entering 2+ institution consensus went on to beat the median universe stock 63% of the time at +20 sessions (CI 53.7–71.8%, n=163 events over 106 independent dates, priced at the NEXT session's close). That clears the 50% null: the whole interval sits above chance. Single-buyer control over the same tape: 54% (CI 46.3–61.6%, n=396 over 161 dates) — the cluster reads higher, though both carry intervals.
BhavRadar is a research instrument, not a signal service. It tells you who is on the tape, whether they are real money or a quoting desk, what their own disclosed record looks like, and honestly how much any of it has been worth.
Every statistic here carries the sample that earned it. When we cannot measure something, the number is absent rather than caveated — there is nothing to misread.