A standup is a prediction market (that nobody is running)

Try this at your next standup. Don't say anything. Just listen for the verbs, and notice how many of them point at the future.

"I'll have the timeout fix in by Thursday." "I think it's the retry logic, not the queue." "That refactor should be low risk." "We're probably going to drop the Redis cache." Nobody in the room thinks of these as forecasts. They think of them as updates — small, disposable, courteous noises made so the meeting can end. But read them again, closely, the way you'd read a term sheet. Every single one of them is a claim about a state of the world that does not yet exist, made by someone with incomplete information, who will not be around when the world arrives to check it.

That is not a status report. That is a bet. Somewhere between "good morning" and "okay, next person," your team runs a small, chaotic prediction market — dozens of times a week, at full volume, with real money riding on the outcomes — and then, uniquely among prediction markets, nobody settles it.

Picture how strange this would look in literally any other setting where people make forecasts for a living. A meteorologist who said "rain, probably, at some point" every single day, for years, and was never once compared against what actually fell from the sky, would not be a meteorologist. They would be a person who enjoys weather-themed small talk. Nobody in that field would tolerate it, because the entire discipline exists to be checked. Software somehow built an entire daily ritual — attended by the same people, forever, at the same time, about outcomes that genuinely matter to the business — that nobody thought to check even once. Not out of malice. It just never occurred to anyone that the checking was the missing ingredient, because the checking was never the point of the meeting. The point of the meeting was to feel synchronized. Feeling synchronized and being right are not the same experience, and only one of them shows up on next quarter's incident report.

The other places forecasts get scored

This is not a hypothetical failure mode. We know, in extraordinary empirical detail, what happens to forecasts that never get scored, because someone eventually got tired of guessing and built the scoring system.

Start with Philip Tetlock's two-decade research program on expert judgment, summarized in his book Expert Political Judgment. Tetlock collected tens of thousands of predictions from political and economic experts — the kind of people paid handsomely to say what will happen next — and then did the one thing nobody in their professional lives had ever done to them: he wrote the predictions down, attached dates, and checked. The results were not flattering. Aggregate expert forecasting performance barely beat chance on many questions, and — this is the detail that matters for a standup — the experts who did worst were disproportionately the ones who spoke in unfalsifiable hedges: "probably," "could go either way," "a real possibility." Vague verbiage wasn't caution. It was a defense mechanism, and it worked, because nobody was checking closely enough to catch it. It only stopped working the moment someone kept score.

Now look at what happens when an organization builds the scoring mechanism on purpose. Bo Cowgill and Eric Zitzewitz studied internal prediction markets run inside three real companies — Google, Ford, and an anonymized firm they called "Firm X" — where employees traded on questions like whether a product would ship on time or a metric would hit target. The striking finding wasn't that the markets were perfectly efficient (they weren't; Google's markets showed a persistent optimism bias, for reasons that will sound familiar to anyone who has sat through a standup). It's that even an imperfect, internally traded market on employee forecasts still outperformed the unscored alternative — because trading forces a forecast to have a price, and a price is a claim nobody can quietly walk back later. The market didn't make employees smarter. It made their forecasts accountable, which turned out to matter more.

Tetlock scored the pundits. Cowgill and Zitzewitz scored the employees. Nobody has ever scored the standup — which is strange, because the standup is where the forecasts that actually run your company get made.

Six ways to make a claim without anyone noticing you made one

Here's a thought experiment. Picture the sentences an engineer can say in a two-minute update, sorted by how easy they'd be to disprove three weeks from now.

1. "I'll ship the timeout fix in charge() by Thursday." Checkable. A commit either lands, on that function, before Thursday, or it doesn't.
2. "The flakiness is coming from the retry logic, not the queue." Checkable, differently — this is a claim about cause, and production either agrees or it doesn't once the retry logic changes.
3. "That refactor is low risk, nothing calls it." Checkable instantly, against the thing that already knows the answer: the call graph.
4. "We're dropping the Redis cache." Checkable — either it's gone in a later commit, or someone reverses the decision and that's checkable too.
5. "Still chipping away at it." Not checkable. There is no future state of the world this rules out.
6. "Yeah, going fine." Not checkable. This is the sound a claim makes when it has been surgically removed of content.

Every engineer who has run a standup has heard all six in the same fifteen minutes, and — this is the part worth sitting with — has treated them as equally valid contributions. They are not equally valid contributions. Items 1 through 4 are forecasts with a settlement date. Items 5 and 6 are the conversational equivalent of a trader shouting a price and refusing to say for which stock.

Why nobody built the exchange

The honest answer is that scoring a forecast is expensive, in a very specific, un-glamorous way: somebody has to remember what was said, wait for the future to arrive, and then go check. A trading floor solves this with a ticker and a clearing house. Tetlock solved it with a research team, a spreadsheet, and years of patience. Your standup solves it with nothing, which is another way of saying it doesn't solve it — the claim gets made, the meeting moves on, and the only "settlement" that ever happens is a vague, retroactive feeling about whether so-and-so is "reliable," assembled from vibes rather than checked commitments. That feeling is itself an unscored forecast, laundered one level up.

And notice the asymmetry this creates. The engineer who says "still chipping away at it" every day for three weeks never has a forecast on record to be wrong about. The engineer who says "Thursday" and misses it by a day has made themselves the only checkable person in the room — which means the scoring system we don't have quietly punishes precision and rewards vagueness, exactly backwards from what you'd want, and exactly the pattern Tetlock found among the pundits who hedged their way to a permanent reputation for wisdom.

What a price actually reveals that a status update hides

Here's the part of the prediction-market comparison that's easy to skip past because it sounds like finance trivia and isn't. A market price isn't just a checked forecast — it's a compressed one, forcing every trader's private information into a single number that can be watched moving in real time. Cowgill and Zitzewitz's internal markets did exactly this inside real companies: when Google's ship-date market started drifting a week before a launch, that drift was itself information, visible days before anyone would have said so out loud in a meeting. Nobody had to wait for someone brave enough to announce "actually, I don't think we're going to make it." The price said it first, because the price didn't need anyone's social courage — it only needed people quietly updating their private bets.

A standup has no equivalent instrument. There is no number drifting in the background that tells you, three days before anyone says it in a meeting, that the Thursday commitment is slipping. There is only the sentence, said fresh each morning, exactly as confident-sounding as yesterday's, right up until the day someone finally says "actually" — usually the day it's already too late to route around. A cleared market moves gradually and visibly. An unscored standup moves suddenly and only in retrospect, which is the single biggest reason "we didn't see it coming" gets said about problems that, somewhere in the room, several people had actually already half-seen.

What settling the market actually requires

The fix isn't a better meeting. It's a settlement mechanism, and the raw material for one already exists in most engineering orgs, just never connected to the thing people say out loud: the commit log knows whether the timeout fix shipped. The telemetry knows whether the fault rate moved after the retry logic changed. The call graph knows, right now, whether anything calls the function about to be refactored. None of that evidence has ever needed a human's memory of a Tuesday morning meeting — it just needed someone to keep the claim around long enough to check it against the record.

That's the whole idea behind treating a standup as a market instead of a status report: every checkable sentence — a commitment, a prediction, a stated risk, a decision — gets kept, dated, and eventually resolved against evidence that already exists, independent of anyone's memory of who sounded confident. (This is the specific, narrow thing Cadence — the standup module of CodeNSM — does: it keeps the claim and waits for the evidence, the same way a market waits for settlement day.) Claims that never named anything checkable don't get scored as wrong. They get counted as unresolvable, which is its own honest number, and the subject of a post two parts from now.

Here's the reframe worth carrying into your next standup: you are not listening to updates. You are watching a market open, trade, and close, uncleared, fifteen minutes a day, forever — unless somebody, finally, keeps the tickets.