Build an evidence table
Prose hides weak links; a table exposes them. This skill lays every supporting claim beside the evidence behind it, so a conclusion cannot lean invisibly on one thin source.
Procedure
State the conclusion in a single line at the top of
evidence.md.Create a table with columns:
# | claim | source | type | confidence | falsifier.Set
typeto exactly one ofprimary(you ran it or read the artefact directly),secondary(a report of the primary), orinference(derived from other rows).Give each row a confidence from a fixed rubric —
verified/corroborated/inferred/assumed— so the column does not drift between reviewers.Fill
falsifierwith the command or observation that would disprove the row. A row without one is unfalsifiable and should be deleted.Read the table back and ask: does the conclusion still hold if the lowest-confidence row is wrong? If not, gather one more primary source before reporting.
Check the ratio of primary to secondary rows:
grep -c "| primary " evidence.mdversusgrep -c "| secondary " evidence.md.
Pitfalls
- Twenty secondary sources traced to one origin are one piece of evidence, not twenty; count origins, not copies.
- A confidence column without a rubric drifts toward optimism as rows are added.
- Padding rows to make the table look thorough dilutes nothing but the reader's trust; keep it tight.
- An
inferencerow that cites only otherinferencerows is speculation; ground it or cut it. - A source you cannot reopen later (a chat scrollback) is weaker than one with a URL or path.
Verification
grep -c "^|" evidence.md; grep -ci "secondary" evidence.md
Report the primary-to-secondary ratio and name the single weakest row the conclusion depends on.