evidence-discipline

Reference for any writing that turns data into claims: a report, a check-in, a summary, a "done" message. The failure it closes is overclaiming: prose that asserts more than the evidence supports - a direction stated without its numbers, a stat with no source, a guess dressed as a finding. The discipline is one rule applied everywhere: a claim is only as strong as what you can show a skeptic.

Every claim names its source

A claim earns its place by naming what produced it: the query, the command, the file, the tool result. A number is false until the command that regenerates it is named - not suspect, false, because an unregenerable number can never be checked and never corrected.

• Overclaimed: "Organic pageviews grew strongly this week."
• Evidenced: "Organic pageviews: 113 for Aug 11-17 (analytics dashboard, organic segment, weekly view)."

If the source is a tool call you made, name the tool and the parameters that matter (segment, date window, filter). If it is a file, name the file. A reader should be able to rerun every number in the report.

Comparisons carry both operands

A comparison is four facts: the baseline value, the current value, and the source of each - same units, same window. Write the direction only beside the two numbers, because direction alone is where reversals hide. A real failure shape: 113 current against a 116 baseline, written as "first week above the baseline" in three places, because the direction was asserted and the operands never sat next to it.

• Overclaimed: "Traffic is above baseline."
• Evidenced: "113 pageviews this week vs the 116 weekly baseline - 3 below. Both from the same organic segment, weekly windows."

When the two windows or units differ, the comparison is not yet a comparison; say what still has to be normalised, or drop it.