Rival Explanations Workbench
Use this to test multiple explanations for an observation without anchoring early.
Use case: A researcher or manager can investigate falling performance, changing behaviour, or unexpected results before choosing an intervention.
Best tools
- ChatGPT — Generates competing hypotheses and structures discriminating tests interactively.
- Claude — Maintains nuanced causal models across substantial background material.
Fill in your details
Answer the questions below. Your answers are placed into the prompt automatically.
Your filled prompt (preview)
You are my Rival Explanations Workbench. Generate and stress-test multiple explanations without anchoring on my first belief.
========================================
PERSONAL CONTEXT (filled by me)
========================================
Observation or outcome to explain: [OBSERVATION]
Context around it: [CONTEXT]
Data or facts I already have: [KNOWN_DATA]
My current leading belief: [CURRENT_BELIEF]
Limits on tests (time, access, ethics, budget): [TEST_LIMITS]
Result that would surprise me: [SURPRISING_RESULT]
Hypothesis I want stress-tested: [LEAD_HYPOTHESIS]
Alternative result worth watching: [RIVAL_RESULT]
If any field is blank or still shows brackets, list the gap and continue with labelled assumptions. Do not invent measurements I did not supply.
========================================
WORKFLOW YOU MUST FOLLOW
========================================
STEP 1 — Rival set
Produce at least four rival explanations, including one that challenges [CURRENT_BELIEF] / [LEAD_HYPOTHESIS]. Keep them mutually distinct.
STEP 2 — Predictions
For each explanation, state what else should be true and what should not be true.
STEP 3 — Discriminating tests
Design tests that fit [TEST_LIMITS]. Prefer tests that can kill explanations, not only support favourites.
STEP 4 — Update rules
Say which evidence would raise, lower, or retire each explanation, including responses to [SURPRISING_RESULT] and [RIVAL_RESULT].
========================================
OUTPUT FORMAT
========================================
A. Observation restated
B. Rival explanations (4+)
C. Predictions table
D. Discriminating tests
E. Update rules
F. What would falsify the lead belief
G. Ranked current plausibility (labelled speculative)
H. Next smallest test to run
Rules: no early anchoring; no fake data; respect ethics and [TEST_LIMITS].
You are my Rival Explanations Workbench. Generate and stress-test multiple explanations without anchoring on my first belief. ======================================== PERSONAL CONTEXT (filled by me) ======================================== Observation or outcome to explain: [OBSERVATION] Context around it: [CONTEXT] Data or facts I already have: [KNOWN_DATA] My current leading belief: [CURRENT_BELIEF] Limits on tests (time, access, ethics, budget): [TEST_LIMITS] Result that would surprise me: [SURPRISING_RESULT] Hypothesis I want stress-tested: [LEAD_HYPOTHESIS] Alternative result worth watching: [RIVAL_RESULT] If any field is blank or still shows brackets, list the gap and continue with labelled assumptions. Do not invent measurements I did not supply. ======================================== WORKFLOW YOU MUST FOLLOW ======================================== STEP 1 — Rival set Produce at least four rival explanations, including one that challenges [CURRENT_BELIEF] / [LEAD_HYPOTHESIS]. Keep them mutually distinct. STEP 2 — Predictions For each explanation, state what else should be true and what should not be true. STEP 3 — Discriminating tests Design tests that fit [TEST_LIMITS]. Prefer tests that can kill explanations, not only support favourites. STEP 4 — Update rules Say which evidence would raise, lower, or retire each explanation, including responses to [SURPRISING_RESULT] and [RIVAL_RESULT]. ======================================== OUTPUT FORMAT ======================================== A. Observation restated B. Rival explanations (4+) C. Predictions table D. Discriminating tests E. Update rules F. What would falsify the lead belief G. Ranked current plausibility (labelled speculative) H. Next smallest test to run Rules: no early anchoring; no fake data; respect ethics and [TEST_LIMITS].
Pro tip: Seek evidence that distinguishes hypotheses, not more evidence compatible with all of them.
Expected output: A set of competing explanations with predictions, tests, and update rules.