Update thesis confidence without false precision
Update thesis confidence by explaining what changed in the evidence, not by attaching an unexplained percentage to your conviction. Record direction, relevance, reliability and independence. Confidence can rise while remaining low if an important uncertainty is still unresolved.
Distinguish a new observation from a new copy
A company release, a transcript quoting it and a news report repeating both may share one information origin. Reading all three can improve understanding without adding three independent confirmations. A source can also be reliable but cover the wrong customer population.
The 10-K includes both management discussion and financial statements. Treat the type of statement as part of the update. A reported number, management’s causal interpretation and your own extrapolation should not carry the same evidentiary status.
Hypothetical confidence log
Assume your thesis concerns whether a new product can retain customers after a trial. You initially have only management commentary. A subsequent report discloses paid renewals, but only for the earliest small cohort.
| New information | Effect on mechanism | Confidence update | Remaining uncertainty |
|---|---|---|---|
| Management repeats optimism | No new observed behavior | Little or no change | Renewal evidence absent |
| First cohort renews at full price | Supports willingness to pay | Increase modestly | Small, possibly selected cohort |
| Later cohort has weaker renewal | Challenges generalization | Reduce or narrow claim | Cohort mix and service changes |
| Multiple articles repeat first cohort | No independent origin | No extra increase | Same underlying observation |
A useful entry reads: “Confidence increased from tentative to moderate because paid renewal is now observed; generalization remains limited by cohort size.” The labels are internal categories, not statistically calibrated probabilities.
Consider how surprising the evidence is
Ask whether both the thesis and its rival would predict the new observation. If both predict an early pilot succeeding, pilot success should cause only a small update. A less visible observation that differs sharply between explanations can matter more.
If you use probabilities, preserve the prior, the evidence model and the assumptions linking the observation to alternative hypotheses. A numerical update without defensible likelihoods creates false precision. Qualitative updates with explicit reasons are often more reproducible than an unsupported move from 63% to 71%.
Verify that negative evidence is allowed to reduce confidence. Set a review rule that requires writing the strongest challenge alongside the best support. Do not average several weak items into certainty or count absence of a warning as proof that a risk is gone.
This method does not calibrate a forecaster by itself. Calibration needs a collection of clearly resolved predictions, recorded before their outcomes. It is also not a position-sizing system; confidence in a business mechanism is different from certainty about price or suitability.
A further check on the example
After writing an update, ask how much of it comes from familiarity. Repeatedly hearing the same explanation can make it feel more credible without adding observation. In the renewal example, compare the exact information available before and after the release. If only the wording became more confident, leave the evidence assessment unchanged. If a new cohort was added, record whether it was independently selected or subject to the same favorable pilot conditions as the first.
FAQ
Is qualitative confidence too subjective?
It can be, so define the labels and attach reasons. Precision in the explanation matters more than extra decimal places.
Should every new filing change confidence?
No. A filing can leave the central uncertainty unchanged. “No meaningful update” is a valid result when the new material does not discriminate between explanations.