Methodology
This page explains, in plain language, exactly how Street Verdict decides an analyst call is a HIT or a MISS — and where the process is imperfect. It is written for skeptics, not for marketing: every number below is queried from the live database at the moment this page renders, not typed in by hand. If a claim here can't be independently checked, it doesn't belong on this page.
1. What gets graded
An analyst opinion starts as a quote pulled from a tracked news source (see Section 5 for the current source list), tagged with a direction — bullish, bearish, or neutral — and, where possible, a specific tradeable instrument. Not every opinion we log is gradable: before grading ever starts, an AI pass checks whether the claim is specific and resolvable against real price data. Vague or non-directional commentary (general market color, macro opinions with no price call, valuation musing with no target) is excluded at this stage — it never gets a HIT or MISS, it simply isn't the kind of claim that can be checked.
4,624
opinions logged in total
664
currently graded (HIT or MISS) — 16.6% of non-suppressed opinions
2,043
still pending their resolution window — 51.0% of non-suppressed opinions
Most logged calls are still waiting for their window to elapse, not stuck or ignored — see Section 3 for how that window is set. A further 32.5% of non-suppressed opinions were reviewed and found not-gradable (Section 2) rather than published as a shaky HIT or MISS. The remainder were removed from the public record entirely — see Section 4 on suppression.
2. How HIT / MISS is decided
Once a call's resolution window has passed, we pull the instrument's real closing price and compare it against the price on the day the call was made. We use a 1.5% noise floor: a bullish call needs the price to have actually risen more than about 1.5% to count as right; a bearish call needs it to have fallen more than about 1.5%; a neutral call is right if the price stayed within roughly that band either way. Moves smaller than 1.5% are treated as “too close to call” — excluded from the record entirely, never counted as a win for anyone, directional or not.
The quality gate. Every verdict is self-rated by the AI as high, medium, or low confidence. A HIT or MISS the AI itself is not confident about is never published as a HIT or MISS — it is downgraded and excluded instead. A wrong public verdict on someone's track record is worse than a slow one, so when the signal is genuinely ambiguous we would rather show nothing than guess.
3. Resolution window
A call isn't graded early, and it isn't graded on some fixed company-wide schedule — it's graded once the analyst's own implied timeframe has actually passed, inferred from their language. A few examples of how that's read: “the next few weeks” grades in about 3 weeks; a bare price target with no stated timeframe defaults to about 2 months out; an explicit year-end or named-quarter call grades on that real calendar date (e.g. “by FY27” grades March 31, 2027). Multi-year, “long term” calls are tracked but not auto-graded — a claim that far out isn't something we can honestly check on a short window.
4. Suppression
Opinions can be manually removed from the public scorecard by the editorial/moderation team. This is a simple boolean fact about a row — suppressed or not — independent of whatever grading status it has underneath; a suppressed opinion doesn't disappear from our database, it's just excluded from every public number on this site.
13.3%
of all logged opinions (616 of 4,624) are currently suppressed
Why opinions get suppressed. Based on what our own cleanup tooling actually does, the likely reasons are: duplicate or conflicting extraction (the same analyst quote pulled twice, or two directly contradictory reads of the same article — we suppress rather than guess which one is right), misattributed or unresolvable text (a quote that doesn't map to a real tradeable instrument), and junk or too-short extractions that don't clear a basic quality bar. Most suppressed rows — 95.5% of them — were still awaiting grading when they were suppressed, so suppression is mostly a data-quality cleanup step, not a way of hiding unfavorable verdicts after the fact.
Honest limitation: we do not currently store a reason code per suppressed opinion — suppression is a single removed/not-removed flag, with no itemized breakdown of which specific reason applied to which call. The categories above describe what our cleanup process does in general, not a verified reason for any one specific suppressed row.
5. Data sources
Price data comes from Yahoo Finance — end-of-day/delayed quotes, not tick-level or real-time data. Historical end-of-day price data for grading older calls is cross-checked against NSE/BSE bhavcopy archives. Neither source is instantaneous, and we don't claim otherwise anywhere on this site.
Opinions are extracted only from a tracked allowlist of news sources — currently 4 domains:
- economictimes.indiatimes.com
- cnbctv18.com
- livemint.com
- moneycontrol.com
This list will grow as coverage expands — it is queried live from the same allowlist constant the extraction pipeline itself gates on, so it can never silently go stale on this page.
6. Known limitations
How we tag direction, and why bullish still leads
Of 4,008 non-suppressed opinions, 68.3% are tagged bullish, 15.4% bearish, and 16.2% neutral. Some of that gap is real: Indian brokerage research genuinely skews toward initiations, upgrades, and buy calls more than outright sell calls, and analysts hedge positive views in print far more often than negative ones — independent sentiment baselines run bullish-leaning too, just not always this wide. We work to keep our own extraction and counting process from adding to that real-world skew on top of it: the model is shown equally rigorous worked examples of bullish, bearish, and neutral calls, so it isn't primed to reach for one direction more readily than another; when a single article's signals are genuinely mixed, we tag whichever direction is actually dominant, with a true tie landing on neutral rather than being awarded to conviction; and every sentiment number on this site counts each analyst's latest stance on an instrument once, no matter how many times they've repeated that same call across different articles — a restated view is still one opinion, not several. We validate this quote-by-quote against real published calls, not just by checking whether an aggregate percentage looks better. The number above will keep drifting as new opinions replace older ones in our rolling windows — it won't land at an even three-way split, and shouldn't, because the underlying commentary itself isn't evenly split either.
The sample-size caveat
A hit rate computed from only a handful of calls can look impressively high (or low) purely by chance. Take our own worked example: an analyst with 4 hits out of 6 graded calls has a headline rate of 67% — but the real 95% confidence interval on that number, computed with a Wilson score interval, is roughly 30–90%. That is a huge range — wide enough that this analyst could plausibly be a genuine coin-flip or a genuinely strong performer, and 6 calls alone can't tell you which. Anywhere you see a percentage on this site with a Low sample badge next to it, this is why — look for the confidence-interval range next to the number, not just the number itself.
Have a question this page doesn't answer, or spot something that looks wrong? Every call on the scorecard links back to its original source article — that's the ultimate check on any number here.