When sources disagree

Disagreement between sources isn't a problem with your research — it usually is the research. This lesson is the discipline of handling it honestly: reading conflicts for information, keeping your own hopes from picking the winner, and reporting splits without smoothing them. It's the questions lesson's warning — the question that contains its answer — grown into a full practice.

First, diagnose the disagreement. Conflicts have kinds, and the kind tells you what to do: different data (two surveys, two populations — find what each measured); different dates (both true, at different times — the newer isn't automatically righter, but the gap is the finding); different interests (a vendor's blog versus an independent review — weigh accordingly); genuine uncertainty (the field itself is split — your report should be too). Ask directly: "these sources disagree on X — what explains the split: data, dates, or interests?"

Your thumb is on the scale — plan for it. You already know the mechanism from the questions lesson: hopes leak into phrasing, phrasing steers sourcing. The counter-practice is structural, not willpower: run the disagree-move on every conclusion you like ("find the strongest sourced case against this"), and notice which verdicts you accepted fastest — speed of acceptance is the tell. The claims you double-check should include, by design, the ones that made you feel good.

Steelman with sources. When a conflict matters, make the desk argue both sides properly: "the strongest sourced case FOR, then the strongest sourced case AGAINST — best evidence each, no averaging." Averaging is the enemy dressed as fairness: "sources are mixed" tells a reader nothing, while "retail data says growing, wholesale operators report flat — split traces to channel" hands them the actual shape of reality.

Report splits as splits. In briefs and scans, conflicts survive compression intact: the finding states both positions, their evidence classes, and your diagnosis of the split — plus which side your recommendation leans on and why. A reader who sees the split can re-weigh it when new evidence lands; a reader given the smoothed average can only be wrong with you. This is the confidence-grading discipline meeting its hardest case.

Bias hunts in both directions. The disagree-move guards against loving a conclusion; the mirror trap is dreading one — the research you keep extending because the answer so far says something expensive. Under-researching what you hope and over-researching what you fear both bend the record. The tell is the same in both directions: notice which findings changed your behavior instantly and which you keep "looking into." Time-box the dread topics the way you time-box scans — a verdict by Friday beats a vigil — and let the flip-fact carry the residual uncertainty like it carries everything else.

One mistake to skip: resolving disagreements by volume. Ten sources against two feels decisive — but you've met the one-origin funnel: ten repetitions of a press release versus two independent studies is a 2–0 game, not 10–2. Count origins, weigh interests, date everything — and when the honest answer after all that is "genuinely split," say so and act under uncertainty like an adult: smaller bets, faster checks, the flip-fact named.