Pundit Rumble is a referee, not a critic. We don’t argue about who is smart. We write down what people actually said, attach the deadline they gave, and check the result against the original source when the date arrives.
A prediction gets scored only if it clears five bars: it names a specific thing, it has a number or a clear yes-or-no outcome, it has a deadline, it comes from a source you can click, and it was written down here before the outcome was known. If a call misses a bar, we can’t fairly grade it, so it doesn’t get a verdict. We still show what was said; we just won’t pretend it’s scoreable.
| Verdict | Means |
|---|---|
| Correct | The call hit its bar by the deadline. |
| Missed | It didn’t. |
| Too vague | No clear bar or deadline to grade against. Shown, not counted. |
| On the clock | Written down and dated; the deadline hasn’t arrived. Never counted until it has. |
A record reads as a fraction, like “4 of 15 correct.” We don’t show a percentage until someone has at least 20 resolved calls, because a percentage on a handful of calls misleads. A correct call and a missed call get the same prominence: the hits are what make the misses credible.
Close calls are read in the way most favorable to the person who made them. A target like “ends the year at 5,000” counts as correct within 5% either way, so precision isn’t punished. A range resolves at its friendliest end. A prediction that depended on a condition that never happened is set aside entirely; it neither helps nor hurts. Quotes in curly quotation marks are verbatim, checked against the original video or article. Anything else is a faithful paraphrase, never dressed up as a quote. We never invent a number. Judgment calls are noted in the open, so the ruling can be argued with, not the data.
Plenty of what pundits say is directional but not scoreable: bullish on AI, bearish on oil, no number, no date. We keep those on the record too, in their exact words with sources, grouped by topic. They are never counted in anyone’s score. When a view grows a number and a deadline, it moves to the scoreboard.
Every bullish or bearish call about something with a price is read at the same three points in time: one week, one month and three months after the first market close that followed the broadcast. The windows are fixed in advance and are the same for every call. All three are shown together, every time, including the ones that make a call look bad. Hiding a window would be the same kind of mistake as inventing a quote.
| Window | Counted as | Role |
|---|---|---|
| 1 week | 5 trading days after the anchor close (7 calendar days for markets that trade every day, and for series that print weekly or monthly) | The immediate read. Noisy by design, shown anyway. |
| 1 month | 21 trading days (30 calendar) | The primary window for pattern observations. |
| 3 months | 63 trading days (91 calendar) | The window most calls are actually about. |
The clock starts on the day the show aired, at the first close after it, because an upload time is not an air time and a sentence may have been said after that day’s close. A Friday evening interview is read from Monday’s close. A window that has not arrived yet says “not yet”. Nothing is read early because it looks good, and nothing is stretched because it looks bad. Treasury yields, credit spreads and mortgage rates move in basis points; everything else moves in percent.
A call is with the market when the price moved the way the speaker said. Bullish on gold and gold rose: with the call. Bullish on Treasuries and yields fell: with the call, because a bond’s price rises when its yield falls. Where a phrase has meant different things to different speakers (“rates” or “yields” on their own), the call is kept on the record but not read against a price. A wrong arrow next to a named person is worse than no arrow.
A call is read against the windows when a named person, identified on the tape, said something bullish or bearish about a thing that has a price we track. Hedged views are read too and marked as hedged. Views with no direction, tone about the Fed (hawkish, dovish), and subjects with no honest price are kept on the record and never read. A call with a number and a deadline can also be scored, under the stricter rules above, once the deadline arrives; level targets resolve at their deadline, never by window.
The site says what happened and stops there. It never says buy, sell, follow or fade. Counts are counts: “with the call in 7 of 12” is written as a fraction, not a percentage, whenever there are fewer than twenty. A pattern seen four times is an observation, not a sample, and the page says so where it happens. A “lean” on the Insights page is a week in which at least 8 named voices put a call on one market and 60% or more pointed the same way; those two numbers were fixed before the data was looked at and do not move.
Errors are corrected publicly and prominently, and every correction stays on the record in the corrections log. Mistakes reach us fastest by comment or DM to @punditrumble. This policy is the product: a scoreboard that hides its own errors is worthless.
Samples are small while the project is young, which is exactly why records stay fractions. We track prominent, frequently-quoted figures, not a random sample of all forecasters. Yes-or-no scoring is hard on almost-right calls; we accept that trade for consistency. Nothing here is investment advice.
This is a media project supported by a newsletter and, eventually, disclosed sponsors. There is a hard firewall: no money from any firm whose people we score, and every sponsor is disclosed. The scoreboard is not for sale.
Every Sunday, one short email in plain words: who called it, who missed, and which big calls face their deadline next. Free.