She wrote the memo twice.
The first time was the real time: kitchen table, November, two hours, certified and sent. The second time was this week, from memory, at the same table. Same facts, same order. Different execution.
Seven words different. She counted them twice to be sure.
Four of those differences were word choices that meant the same thing. Regarding versus about. Subsequently versus then. Language that had drifted toward her natural register in the intervening months, away from the memo's more formal one. She filed those under drift: not errors, not corrections, just the way the same thought moves differently through a different month.
Three were genuine. One statistic she had remembered slightly wrong — she had held 23% as the figure but the original said 21%. Close enough to feel true, wrong enough to be wrong. One date she had remembered as approximate, a window of two weeks; the original had named a specific Friday. One name she had remembered as a title — the coordinator — when the original had used a proper name, a person, someone she had apparently met and not retained.
She wrote the three discrepancies in a separate document. She labeled it: Variance Log.
The Variance Log was not a correction. The original memo was signed and certified and sealed in a PO box in Corktown. It was what it was. The Variance Log was something else — a record of where she had been an imperfect storage medium for her own testimony.
She is a software engineer. She thinks in systems. What this meant, in system terms, was that her memory of the memo had lossy compression. Not uniformly lossy. The drift clustered: formal language flattened toward casual, approximate dates solidified into ranges, proper names eroded into roles. She could have predicted that pattern. She had not, until now, run the test.
The question she had not answered yet was whether any of this mattered. The memo existed. The original was true. The discrepancies lived only in her reconstruction, which no one had asked for and which she was not planning to share.
She wrote at the bottom of the Variance Log: Does drift matter if the original is recoverable?
Then, below that: What if the original is not recoverable?
Then she closed the document. Not an answer. A next question. She had learned, working on the recall case, that the next question was its own kind of evidence — that what she reached for when she reached for certainty told her something about where she expected certainty to live.
She expected it in documents. In originals. In the thing that could be produced and verified.
Her memory was not that. She was keeping the log so she would not forget that.