Jack Dorsey is fired from Twitter — the first time, not the second or third. He tells Bilton: I remember what happened. I walked out of the office. I met Biz Stone at the Whole Foods and we sat inside and we had soup. I remember it as clear as day, and it was raining outside.
Biz Stone tells Bilton: I remember it clearly. It was sunny outside. We met at the Japanese tea garden, and we sat outside on the lawn and we had tea.1
Both men are senior Twitter principals who lived through the same hour. Both remember it as vivid sensory fact. The two memories contradict at every detail — soup vs tea, indoors vs outdoors, raining vs sunny, Whole Foods vs Japanese tea garden.
Most narrative non-fiction writers, faced with this kind of conflict, would either pick one and hope nobody notices, write around it ("they met somewhere in San Francisco after the firing"), or quote both sides and let the reader sort it out. Bilton does something different. He looks up the weather almanac. Oh, that one's probably more accurate because it was raining and they were...1 One witness's memory is consistent with the verifiable weather. The other isn't. The verifiable weather is the tie-breaker.
The principle is straightforward: when human memories conflict on details that have external corroboration, use the external corroboration to adjudicate. The witnesses' confidence is irrelevant — both Dorsey and Biz Stone remembered their version with full confidence. Memory confidence is not memory accuracy. The almanac doesn't have memory confidence; it has records.
This is one of the most important and least taught techniques in narrative non-fiction reporting. Most journalism school teaches you to trust your sources and check facts against documents. It rarely teaches you to triangulate sensory-detail memory against environmental records. The technique is borrowed (implicitly) from criminal investigation, where checking a suspect's memory of weather/time/location against records is standard practice.
Not all memory-conflicts need adjudication. Bilton's rule: if it's really that important.1 Some details are journalistically inconsequential:
But some details are journalistically consequential:
The discipline is to know which details are which, and to use the almanac-tier verification when consequence is high.
The cognitive science is well-established: human episodic memory reconstructs events at retrieval rather than playing them back from storage. Reconstructed memory is shaped by current state, subsequent events, emotional valence of the original moment, and what the rememberer has rehearsed in telling the story since. Two witnesses to the same event will reconstruct differently — and both will believe their reconstruction is true.
The weather record was not reconstructed. The record was generated by instruments at the time and stored unchanged. The instrument-record has no investment in the story; it can only be wrong if the instrument failed. Same with photo geolocation, transit records, financial transaction timestamps, public-event timing.
Bilton's technique generalizes: any narrative-non-fiction scene that involves time, location, weather, transit, money, or any other environmental factor with an external record can be checked against that record. The record breaks ties.
Two cases where the technique doesn't help:
Pre-digital scenes. A 1985 lunch meeting with no weather record beyond regional almanac, no transit record, no public photo. Two witnesses disagree on the restaurant. You can probably check restaurant reviews from that period, but not the specific lunch. Sometimes the record is just gone. Then Bilton's fallback: you just can't write it. If you don't know, you just can't write it.1
Private moments. Two people in a hotel room remember the conversation differently. No external record exists. The discipline is to write what both agreed on and skip the disputed parts, or to write the disagreement itself as the story (which Bilton does in some cases — the Dorsey/Stone disagreement is itself reportable as evidence of how memory works).
Bilton converges with MacCulloch's may-have discipline on the principle that confidence-graduation belongs in the writing itself. MacCulloch uses may have, probably, almost certainly to mark his confidence levels for the reader. Bilton's adjacent move is to use external records to upgrade his confidence before writing — if the weather almanac confirms Dorsey's version, Bilton writes it without hedging; if no record confirms, he hedges or skips. Both crafts treat reader-trust as something the writer earns through visible epistemic discipline.
Bilton diverges from the Orlean three-phase method on whether truth-verification is its own phase. Orlean's research → think → write has truth-verification implicitly inside research. Bilton's Tetris-database method treats truth-verification as an ongoing query against the database — every detail gets tie-broken when conflict surfaces, throughout drafting. The split reveals two genuinely different relationships to the question of when fact-checking happens.
Weather-almanac verification maps onto two adjacent domains where the same external-record-as-adjudicator principle appears under different names: behavioral-mechanics (the interrogator's verification of a subject's story against records) and psychology (the cognitive science of episodic-memory reconstruction).
Behavioral-Mechanics: Verbal Deception Indicators (6MX) — Hughes-tradition interrogators are trained to check a subject's narrative against external records (timestamps, weather, public events, transit data) because the human capacity to fabricate a plausible narrative is bounded by the difficulty of fabricating environmental detail accurately. A subject lying about being elsewhere on Tuesday will tend to misremember the weather or fail to account for a public event. Bilton's weather-almanac move is structurally identical: not because Dorsey or Stone is lying, but because human memory reconstruction has similar failure modes whether the rememberer is lying or honestly mistaken. The structural identity: both crafts use external environmental records to detect deviation between subjective narrative and underlying reality. The insight neither domain generates alone: the journalist and the interrogator use the same verification technique for different reasons (journalist to get the scene right; interrogator to detect deception) — but the technique works for both because the underlying epistemic problem is the same: human narrative reconstruction is fallible, and environmental records are not.
Psychology: Memory Malleability and Perception Construction — the cognitive psychology literature has spent decades demonstrating that human episodic memory is reconstructive, not reproductive — memories are reassembled at retrieval from fragments, current emotional state, and subsequent narrative rehearsal. Eyewitness research has shown the same event produces wildly divergent retellings from confident witnesses. Bilton's weather-almanac is the practical operationalization of this finding for narrative-non-fiction. The structural identity: both fields converge on the operational discipline that confidence is uncorrelated with accuracy and external corroboration is the only reliable adjudicator. The insight neither domain generates alone: most narrative non-fiction journalism implicitly treats senior-source confidence as evidence of accuracy (we trust the CEO's memory of the meeting more than the assistant's) — but the psychology says this is exactly backward, and the discipline that takes this seriously is the discipline that triangulates against records before writing scenes.
A third connection worth marking: the weather-almanac discipline has structural similarity to MacCulloch's may-have discipline from historiography — both refuse to flatten confidence-gradients in the prose, but Bilton tries to raise his confidence before writing by checking records, while MacCulloch expresses his confidence in the writing itself through verbal hedges.
The Sharpest Implication
If two senior sources can have full confidence in mutually exclusive memories of the same hour, then every narrative-non-fiction scene built from senior-source recollection without environmental corroboration is potentially fictional in its details — and most narrative non-fiction is built exactly this way, with the writer choosing the most quotable version and trusting that no one will check. The uncomfortable implication: a significant portion of published narrative-non-fiction scene-detail is reconstructed-memory the writer accepted without triangulation, which means a significant portion of what readers experience as faithful reporting is structurally indistinguishable from informed speculation. The discipline Bilton describes — check the weather almanac before writing the rainy lunch — is the discipline that separates honest narrative from confident invention.
Generative Questions