Read Your Draft Aloud to Someone Who Was There
In 2003 I read a draft paragraph about a landslide in Andahuaylillas to Delia Huamán, who had lived through it, and she nodded through the whole thing. Two days later I read the same paragraph to her cousin Teodoro, who had stood forty meters further up the slope that night, and he stopped me at the third sentence to say the mud had not come from the direction I had written. Delia had not caught the error because she had not been looking that way when it happened. Teodoro had been looking at exactly that ridge for eleven minutes, waiting for his brother to climb down it.
The habit I had built up to that point was the one most testimonio editors are taught: take the finished passage back to the person whose story it is, read it to them, watch their face. That habit catches a great deal, and I still use it. What it cannot catch is anything the primary speaker did not personally witness, because a speaker's memory has the same blind spots the recording had, and reading a flawed sentence back to the person who supplied the flaw usually confirms the flaw rather than exposing it. Delia's nod was not agreement with the facts. It was recognition of her own account, error included.
The corroboration map
What Teodoro gave me instead, without either of us naming it at the time, was a second, independent line of sight on the same fifteen minutes, and I now build that check into every draft with more than one living witness. I call it the corroboration map. Before a passage goes to print I find a second person who was present for the same event, someone other than my main speaker, and I read the passage aloud to them without warning which sentences concern them. I mark, on a printed copy in front of me, exactly where they interrupt.
The marks cluster. In the Andahuaylillas passage, four sentences produced no reaction at all from Teodoro, three produced a nod, and one produced the correction about the mud's direction. That single interruption told me something the nod count alone could not: three sentences in that paragraph were corroborated by two independent memories and could stand in the final text without hedging, one sentence rested on Delia's account alone and needed either a source note or a rewrite, and the mud sentence was simply wrong and had to change regardless of whose version it favored.
Over the four testimonies where I have run this properly, the interruption rate has never been zero, and it has never been higher than one correction per six sentences. That range is itself useful. A passage that produces no interruptions from a second witness has either been checked too gently or genuinely happened exactly as told, and I have learned to distrust the first case more than the second.
Why this is not the return pass again
This looks, at a glance, like the practice I have described elsewhere of reading an edited testimony back to its own speaker before publication. It is a different check answering a different question. Reading a passage back to the person it belongs to tests whether the writing matches how the experience felt to them. Reading the same passage to someone else who was there tests whether the writing matches what happened at all, independent of any single person's felt sense of it, including the main speaker's. A testimony can pass the first test completely, because a speaker will always recognize their own remembered version of events, and still fail the second, because remembered versions are not audits.
The objection
You could argue that finding a second witness for every scene in a testimony is not always possible, since many of the events testimonio records were witnessed by exactly one surviving person, and demanding corroboration where none exists would silence accounts that deserve print regardless. I agree with this completely, and I want to be precise about what the corroboration map is for. It is not a requirement for publication. It is a check applied only where a second witness happens to be alive and reachable, and its absence in a single witness account is not a mark against that account, only a limit on how many of its details I can independently confirm. Where no second witness exists, I say so directly in the text rather than pretending the single source carries the weight of two.
What changed permanently after Teodoro was my assumption about what a nod means. I had spent years treating a speaker's agreement with their own transcript as close to the end of verification. It is the end of one kind of verification, the kind that asks whether a page sounds like the person who is supposed to have said it. It answers nothing about whether the mud came from the east ridge or the west one, and for that question I need someone else who was standing in the dark that night, looking where my main speaker was not.