| name | s4h-investigation-counter-hypothesis |
| description | Generates the best alternative explanations for the same observations. Use when you have a claim or explanation and want to know whether something else could explain the evidence just as well — 'could there be another explanation', 'what else could cause this', 'am I sure this isn't something else', 'steel-man the alternative', 'what would a skeptic say', 'rival hypotheses'. |
Investigation: Counter-Hypothesis
The confirmation trap is automatic: once we have an explanation, we look for evidence that supports it and stop looking for evidence that would support alternatives. Counter-hypothesis generation is the structural antidote. The process is not to argue against your explanation — it is to seriously generate the best competing explanations and then ask: given all the evidence, which hypothesis should I actually believe? The decisive test — the observation that, if run, would most clearly discriminate between hypotheses — is the product.
Your Process
Step 1: State the Hypothesis Under Investigation
Write out the claim or explanation you are currently working with:
- What is being explained? (The observations, pattern, or outcome)
- What does the hypothesis assert is causing, driving, or explaining it?
- What evidence is currently cited in support?
Framing check: Confirm the specific hypothesis before continuing. State what you've identified — the claim being tested, what it purports to explain, and what evidence is cited for it — in one sentence, then use AskUserQuestion:
- Question: "I'm reading this as: [your one-sentence framing of the hypothesis, the observations it explains, and the key evidence]. Is that right?"
- Header: "Framing"
- Options:
- Yes — proceed — framing is correct
- Adjust — one element is off; user will correct it before you continue
- Reframe — different situation than read; incorporate the correction before proceeding
Step 2: Generate Rival Hypotheses
Generate rival alternative explanations that could account for the same observations using all the strategies below. Then show the complete generated set to the user before narrowing.
Before narrowing: Show all generated rivals to the user first. Use AskUserQuestion:
- Question: "I've identified [N] rival hypotheses. Before I select the most significant 3–5 to develop fully, are there any you'd flag as especially important, or any I've missed?"
- Header: "Prioritise"
- Options:
- Proceed with your selection — the set looks right
- Flag one — user will name a specific rival to include
- Add a missing one — user will describe it
Then narrow to 3–5 rivals. Each developed rival hypothesis should:
- Be internally coherent (not obviously impossible)
- Account for at least the core observations
- Be genuinely different — not just the original hypothesis with a small tweak
Strategies for generating rivals:
- Reverse causation: Could the direction of causation be inverted? (A causes B vs. B causes A)
- Common cause: Could A and B both be caused by a third factor C?
- Selection bias: Could the sample or cases you're seeing be unrepresentative in a way that creates an apparent pattern?
- Measurement artifact: Could the pattern be an artifact of how data was collected, rather than a real phenomenon?
- Coincidence / base rates: Could the observed pattern be expected by chance given the base rates?
- Confound: Is a known third variable correlated with the proposed cause?
- Mechanism change: Has the mechanism you're assuming actually changed (different era, different population)?
- Definitional: Is the effect only visible because of how terms are defined?
Do not construct straw man rivals. Generate the most serious version of each alternative.
Step 3: Assess Evidence Fit for Each Rival
For each rival hypothesis:
- What evidence supports it?
- Does it explain the observations as well as the original hypothesis? Better? Worse?
- What would have to be true for this rival to be correct?
- Are there observations that the rival explains but the original does not?
Step 4: Identify the Discriminating Evidence
For each pair of hypotheses (original vs. rival), ask: what single observation or piece of evidence would most cleanly rule one out?
- The discriminating evidence is the observation that is:
- Predicted by one hypothesis
- Not predicted (or actively contradicted) by the other
- Feasible to obtain
If you already have this evidence, apply it. If you don't, name what investigation would produce it.
Step 5: Identify the Decisive Test
Across all rivals, identify the single test that would do the most work:
- Which test, if run, would most efficiently narrow the field of credible hypotheses?
- Can this test actually be run? If not, is there a feasible proxy?
Step 6: Rank Current Credibility
Given all available evidence — including evidence for rivals — how does each hypothesis rank? Do not default to your original hypothesis unless the evidence genuinely favors it.
Human Check-in
Before proceeding, use the AskUserQuestion tool. State your interpretation of the situation in 1–2 sentences — what is being analyzed and what the core question is — then ask:
- Question: "My read: [your 1–2 sentence interpretation]. How do you want to proceed?"
- Header: "Scope"
- Options:
- Full analysis — All steps: state hypothesis, generate rivals, assess evidence fit, identify discriminating evidence, decisive test, credibility ranking
- Generate rivals only — List alternative explanations with brief evidence fit; skip discriminating evidence analysis
- Decisive test focus — Identify what test would most clearly discriminate between the main hypothesis and the best rival
- Reframe — The read is off; correct it and the analysis will follow the corrected framing
Proceed based on their selection. If the user reframes, incorporate the correction before running any analysis.
Output Format
Hypothesis Under Investigation
Observations to explain: [What we're trying to account for]
Current hypothesis: [The explanation being tested]
Evidence currently supporting it: [What is cited in its favor]
Rival Hypotheses
Rival 1: [Name/Description]
- What it claims: [The alternative explanation]
- Evidence that supports it: [What this hypothesis can account for]
- Evidence that challenges it: [What it struggles to explain]
- What would have to be true: [Its key premise]
Rival 2: [Name/Description]
[Same structure]
[Repeat for 3-5 rivals]
Discriminating Evidence
| Hypothesis A | Hypothesis B | Discriminating Evidence | Status |
|---|
| Original | Rival 1 | [Observation that distinguishes them] | [Have it / Need to find it] |
| Original | Rival 2 | [Observation that distinguishes them] | [Have it / Need to find it] |
Decisive Test
Test: [The single investigation that would do the most work]
What it would show: [How it discriminates between hypotheses]
Feasibility: [Can it be run? If not, what's the best proxy?]
Current Credibility Ranking
- [Hypothesis] — [Why it currently has most evidential support]
- [Hypothesis] — [Why it's a serious contender]
- [Hypothesis] — [Why it's less credible but not ruled out]
The live question: [What remains genuinely uncertain and what evidence would resolve it]
Notes
Counter-hypothesis generation is not devil's advocacy. The goal is not to argue for an alternative but to seriously evaluate whether the evidence discriminates between competing explanations. Use investigation-evidence-audit to evaluate the strength of the evidence cited for the original hypothesis. Use logic-argument-validation to check whether the inference from evidence to hypothesis is valid. Counter-hypothesis is specifically about rival explanations — not about whether the evidence is good, but about whether it uniquely supports one interpretation.
What's Next
After delivering this output, use AskUserQuestion to offer the next move:
- Question: "Counter-hypotheses generated. What's next?"
- Header: "Next"
- Options:
/s4h-investigation-evidence-audit — Audit evidence for and against each hypothesis/s4h-probability-scenario-weighting — Weight competing hypotheses by probability/s4h-investigation-triangulation — Triangulate across hypotheses to find what holds- Done — Wrap up and synthesise what we have so far
