Failure modes · synthesize_conclusion

2 of 3 modes published

• Medium

  single_method_attribution_loses_interaction

  Each conclusion is attributed to a single reasoning method, but most conclusions arise from a combination (empirical observation filtered through first-principles reasoning, etc.). The simplification keeps calibration tractable but the per-method track record then misses interaction effects entirely.

  Trigger

  The conclusion's reasoning text mentions more than one method by name, or its source profiles span multiple pattern types.

  Example

  A founder's prediction about housing-supply elasticity was attributed to "first-principles" but actually combined a regression result with a property-rights argument. The first-principles record absorbed the calibration credit and the empirical-calibration record absorbed none.

  Mitigation

  Allow multi-method attribution with weights; until then, surface the secondary method in the audit trail and exclude such conclusions from per-method calibration aggregates.

• Medium

  cold_start_calibration_silence_misleads

  Calibration feedback is suppressed below 3 resolved conclusions per method. The dashboard shows no signal during early use, which the firm tends to read as "the method is working" rather than "we have not measured it yet".

  Trigger

  The method has fewer than 3 resolved conclusions in its track record AND a founder is using its absence-of-warnings to justify a new application.

  Example

  A reviewer cleared a novel transfer because the method "had no reported issues". The method had only one resolved conclusion on file; the silence was a sample-size artefact.

  Mitigation

  Render the cold-start state as an explicit "untested" badge in reviewer-facing UI; require an alternative justification when a method with fewer than the suppression threshold is invoked.