Methodology

1. Who gets read

Every NSE-listed company with price history — 2,613 names — that is expected to report in the next seven days, whether the exchange has confirmed the date or we have projected it from the company's own filing habit. Companies under ASM/GSM surveillance are left out. A company is read only once it has 16 completed results with a clean price reaction on record — the full 16-quarter window every read is built from. Younger companies still have a page, a price and a results date here, but no read: with less history than that there is no habit to measure, only noise, and we would rather show nothing than a guess.

2. The six layers

Each layer is computed from the company's own price history and turned into a percentile among that week's reporters (0–100). They are blended with the weights below; when a layer cannot be computed for a company — most often the sector layer — its weight is shared across the others rather than filled with a neutral score. A layer at 0% is still computed and shown on every page; it simply does not move the rank until the evidence says it should (section 3).

3. Where the weights come from

The weights above were fitted from the post-mortems, not chosen. On 1 Oct 2026 every judged read on file — 5,425 of them — was split by season; for each season the weights were fitted on the seasons before it and scored on that one alone, so no read was ever scored by weights that had seen its outcome. The measure is the rank correlation between the morning's ordering and the ten-session excess move that followed (the information coefficient). The fitted weights scored 0.086 out of sample over 28 unseen mornings (t = 4.5, positive on 89% of them); the original hypothesis weights — reactivity 30%, momentum 20%, vs market 30%, vs sector 20% — could not be told from zero on the same mornings.

A fit goes live only if it clears that bar (t ≥ 2 over at least 20 unseen mornings, and better than the hypothesis); one that later fails it is withdrawn and the read goes back to the hypothesis weights. Every fit, applied or not, is kept. Reads already made are never rescored: each page shows the weights it was actually ranked under.

4. The read

The blended score ranks every eligible reporter for the week. The read is the company's position in that ranking, in five bands: strong top fifth, favourable next fifth, neutral middle, unfavourable, weak bottom fifth. It is relative to that week's field: the same company can read strong in a quiet week and neutral in a crowded one.

Alongside it the page shows the company's own record over the two weeks after past results — the mean ten-session return, the share of results after which it was higher, and the same against the market. Those numbers are the inputs to the own-record layer, at whatever weight it currently carries.

5. The post-mortem

Ten sessions after each result, the last read made before it is judged. The outcome is the share's excess return over NIFTY 50 across those ten sessions: a strong or favourable read is right if the excess is above +2%, wrong below −2%, flat between; weak and unfavourable the other way round; neutral reads make no call. Results we never read are judged too, as the base rate.

For every judged result the post-mortem also records the things we expect to matter, so that after a few seasons the question "which inputs separate winners from losers" can be answered from the table rather than argued:

• The result itself — revenue and profit growth, margin change, and whether the quarter was better or worse than the company's own last four.
• Where the share stood — the 30-session run-up, distance from the 52-week range, delivery and volume in the fortnight before, and whether its usual pattern held.
• The context — the market and sector over the same window, how many peers reported that week, and whether the announcement came during trading hours.
• Our read — the label, rank, each layer's score, and how much history stood behind it.

5b. The daily read — every name, every session

The results read only covers companies reporting that week, so between seasons it has nothing to say. The daily read covers the whole exchange every trading session: every active name with 260 sessions of history and a median turnover above ₹5 crore a day (names under exchange surveillance are left out). Each is scored on the same layers, normalised across that day's field, blended and ranked; the top five are names to look at and the bottom five names to stay away from. It is keyed by the close it stands on — one read per session the exchange traded, so a weekend or a holiday is never a gap.

The starting weights are momentum 0.40, market divergence 0.35, sector 0.25; reactivity, the company's own post-result record and the last reported quarter start at zero. Every read is judged ten sessions later against NIFTY 50 and kept; the day-by-day page shows the hit rate by pick and by tenth of the field, and the "chance" beside a name is simply the historical rate at which names in its tenth went that way. Weights are refitted monthly and replace the hypothesis only when they beat it on unseen sessions — with a harder gate than the results model (t ≥ 3.5 over ≥ 120 sessions), because consecutive daily windows overlap and would otherwise flatter the statistic.

The entry reference is the close; the upside and downside references are the name's own median ten-session rise and fall over the past year — arithmetic on its history, stated so the outcome can be checked, not a level to act on. A name with no sectoral index has no sector layer and its weight is redistributed; peer baskets are not built for the daily read.

6. Windows

T-1

the last close before the result — the reference price for every move

T0

the reaction session: the announcement day if filed before 15:30 IST, otherwise the next session

T+1

the core reaction: gap, announcement session, one day of follow-through

T+10

two calendar weeks of trading — the window the site judges itself on

All prices are adjusted for splits and bonuses, so an archived read stays comparable with today's chart.