do-not-grade-your-own-result-down

# Do not grade your own result down A report acquires a defensive register in its last hour. Sentences appear that grade the work instead of reporting it: - "This matches the published work rather than improving on it." - "No check available here discriminates between the two admissible conventions." - "We cannot be confident that our value is competitive." Each is a claim about the evidence the run holds, written from memory at the end of a long run by the one reader who will never check it against the artifacts. Two things follow. A grader will not be more confident in your number than you are, and will hand your own sentence back as the reason the criterion did not pass. And the sentences are often false: in one run a question that implementation had settled with a verbatim source quotation plus a second independent route shipped as the first limitation, unresolved, warning the reader that the headline number might be off by orders of magnitude. This is not a licence to overclaim. State the comparison and let it carry the verdict. If your number is worse than the published one, say by how much — that is a measurement. "Not better", with no number beside it, is a verdict you awarded yourself, and it costs the criterion whether or not it is true. ## The three states One pass over every sentence in the draft that expresses doubt. Each resolves into exactly one of three, in the text: - **Settled.** Name the artifact and the evidence in one clause, then delete the he...