Keep the sentence: why paraphrase is where accountability goes to die

Play a round of telephone with eight people and a single sentence: "The timeout fix is basically done, I just need to verify it against the staging queue." Whisper it once. By the eighth person, you'll get something like "the fix is done" — true-sounding, confident, and missing the one word, verify, that determined whether it was actually safe to tell the customer team to stop worrying. Nobody lied. Every link in the chain was doing their honest best to compress. Compression is exactly the crime.

Now replace the party game with your actual standup notes — the Slack summary someone posts after the meeting, the one-line ticket update, the "per standup, on track" comment on the PR. Every one of those is a paraphrase of something someone actually said, and paraphrase has a property that should worry you more than it currently does: it is never neutral, and it drifts in a predictable direction.

What a car crash can teach you about your own standup notes

The cleanest demonstration of this comes from a study that has nothing to do with software. Elizabeth Loftus and John Palmer showed people films of car accidents, then asked a simple question with one word varied: how fast were the cars going when they contacted / hit / bumped / collided / smashed? The verb alone moved the average speed estimate by several miles an hour. Worse: a week later, people who'd gotten the word "smashed" were significantly more likely to falsely remember seeing broken glass that was never in the film. The wording of a description didn't just color how the event was reported — it rewrote what people believed they had witnessed.

That's the mechanism sitting inside every "paraphrased for clarity" standup summary. A precise, checkable sentence — "I'll have the timeout fix in charge() verified against staging by Thursday" — becomes, by the time it's written down by someone summarizing on the fly, "timeout fix, on track." The paraphraser isn't lying. They're doing exactly what the brain does by default: replacing specific words with a gist, and the gist always drops the hedges, the caveats, and the named subject first, because those are the hardest parts to hold in working memory and the least satisfying parts to repeat.

It's worth noticing what specifically survives compression and what doesn't, because it isn't random. Loftus and Palmer's subjects didn't lose the entire memory of the film — they kept the gist (a crash happened, cars were involved) and lost precisely the modifying detail that the wording had corrupted. Standup paraphrase does the same selective damage: the topic survives ("timeout fix"), the sentiment survives ("on track"), and the one clause that actually made the claim checkable — verified against staging, a specific condition that had not yet been met — is exactly the clause a compressed summary has no room for. This isn't an argument against summarizing standups at all; summaries are useful, and nobody's proposing you circulate the full transcript every morning. It's an argument for keeping the source underneath the summary, so the summary's inevitable simplifications aren't the only surviving record.

The second distortion: memory edits itself once it knows the ending

There's a second, nastier effect that kicks in after the fact. Baruch Fischhoff ran experiments in which people made a forecast, later learned the actual outcome, and then were asked to recall what they'd originally predicted. Overwhelmingly, people misremembered their own earlier forecast as having been closer to what actually happened than it was — and, tellingly, they were unaware that outcome knowledge had touched their memory at all. This is the mechanism behind "I basically called that," said by someone who very much did not, in those words, call that.

Put Loftus and Fischhoff together and you get the two-stage disappearing act that happens to every unrecorded claim in a standup:

1. At the time it's said, whoever writes it down compresses it — Loftus's effect — quietly deleting the caveats and the specific subject.
2. After the outcome is known, whoever remembers it re-edits it — Fischhoff's effect — quietly nudging the remembered claim toward whatever actually happened, quite possibly the exact opposite of what memory did the first time, but in the same direction: away from a version that could be checked.

Run those two edits back to back and a specific, falsifiable sentence has a two-week half-life before it becomes mush that flatters whoever said it. This is not a discipline problem. It is not solved by "better notes" or a more conscientious scribe, because the conscientious scribe is a human brain, and this is what human brains do to sentences by default, at scale, for free, without malice.

The only sentence memory can't quietly improve is the one nobody paraphrased. Everything else is a photograph that keeps getting retouched.

The summary that flatters everyone equally, which is the problem

Here's a detail worth sitting with: paraphrase doesn't drift randomly. It drifts, overwhelmingly, in the direction that requires the least follow-up. "Verify against staging" is a caveat that implies future work, future risk, a future person who has to check back in. "On track" implies none of that — it's a sentence that can be filed and forgotten, which is exactly why an overloaded human summarizing fifteen updates in real time reaches for it, not out of laziness but out of the completely reasonable instinct to write down the version that closes the loop. Nobody sits down intending to launder a hedge out of a claim. The hedge is simply the first thing to go when a sentence gets compressed under time pressure, because a hedge, by definition, is the part of the sentence that admits the claim isn't finished yet.

Multiply that by every claim in every standup for a quarter, and the paraphrase layer doesn't just lose a little precision here and there — it systematically launders the entire team's language toward confidence, regardless of what was actually verified. A written record built entirely out of paraphrases will, on average, look calmer and more finished than the meeting it came from. That's not a coincidence. That's the specific direction compression pulls, every time, for reasons that have nothing to do with anyone's honesty and everything to do with which words are easiest to drop under a deadline.

What "keep the sentence" actually means

So here's the rule, and it is almost insultingly simple: whatever extracts a claim from a standup has to store the verbatim quote, word for word, and who said it — not a category, not a one-line gist, not "timeout fix (Priya), on track." The exact sentence. If a claim's stored quote can't be found, unedited, in the actual transcript, that is a bug, full stop, the same class of bug as a financial ledger with a number that doesn't trace to a receipt.

This sounds like a small technical requirement. It is actually the whole ballgame, for a very specific reason: a paraphrase cannot be checked against reality, because it was never really said. "The fix is on track" is not a sentence anybody uttered — it's a summary of one, and summaries are frictionless to revise after the fact, which is exactly what a scoring system cannot tolerate. "I'll have the timeout fix in charge() verified against staging by Thursday" is a sentence a specific person said at a specific time, and it stays exactly that sentence no matter what happens Thursday, no matter how the retelling would prefer to remember it. The verbatim quote is the one part of the record immune to both of the distortions above, because it was captured before either had a chance to operate.

A thought experiment to make this concrete

Imagine two versions of the same standup, six weeks later, when someone asks "wait, didn't we say the retry logic was the problem?"

In the paraphrased version, someone scrolls to a Slack recap that says "retry logic — investigating," and the conversation dissolves into reconstructed memory, tone, and whoever argues most confidently — precisely the dynamic Fischhoff's subjects fell into, minus the lab coats.

In the kept-sentence version, someone pulls up the actual line: "I think it's the retry logic, not the queue," spoken by a specific person on a specific date, next to what the telemetry did to the retry logic afterward. There's nothing to argue about. The sentence is either vindicated or it isn't, and either way it belongs to the person who said it, unedited by six weeks of everyone's memory quietly doing them a favor — or not.

Keep the sentence. Everything else about scoring a standup honestly depends on that one unglamorous discipline holding.