Generating Hypotheses

Read <problem-dir>/problem.md. This is the sole source of truth for what is being investigated.

Step 2: Read prior iteration context

For each existing hypothesis-*.md, read:

• ## Statement (what was claimed)
• ## Status (confirmed / refuted / inconclusive / pending)
• ## Conclusion section if present (what was learned)

Build a summary of what has already been tried:

• Confirmed -- do not re-propose
• Refuted -- do not re-propose; note what was ruled out
• Inconclusive -- may re-propose with narrower or different framing

Step 3: Determine next hypothesis numbers

Count existing hypothesis files. New hypotheses continue from the next number. If hypothesis-03.md is the last file, new hypotheses start at hypothesis-04.md.

Step 4: Generate hypotheses (minimum 3)

Synthesize the problem statement, success criteria, constraints, and prior conclusions into new, distinct, testable hypotheses (minimum 3). Generate as many hypotheses as the problem space warrants. More complex or multi-faceted problems should produce more hypotheses to ensure adequate coverage of the solution space.

Each hypothesis must:

• Be a clear, falsifiable claim that can be confirmed or refuted through experiment
• Contribute meaningfully to solving the problem if true (not a trivial observation)
• Not duplicate any previously confirmed or refuted hypothesis
• Build on inconclusive prior results by narrowing scope or changing approach

The set as a whole must:

• Cover different mechanisms or explanations (not variations of the same idea)
• Span a range of confidence levels -- include likely and less-obvious explanations
• Be ordered from most to least promising based on available evidence

Quality check each hypothesis:

• Specific enough to be experimentally testable or logically provable?
• Bounded -- avoids solving the entire problem in one claim?
• Respects constraints and scope from problem.md?

Step 5: Write hypothesis stub files

For each new hypothesis, write <problem-dir>/hypothesis-NN.md (two-digit zero-padded):

# Hypothesis NN: <Concise title (5-10 words)>

## Status
pending

## Statement
<1-2 sentences: a precise, falsifiable claim>

## Rationale
<2-3 sentences: why this hypothesis is worth testing -- connection to the problem, prior knowledge, or what prior conclusions suggest>

---
<!-- Sections below are added by downstream skills -->
<!-- refining-hypothesis adds: Literature, Refined Statement -->
<!-- designing-experiments adds: Experiments -->
<!-- running-experiments fills in: Results under each Experiment -->
<!-- drawing-conclusions adds: Conclusion -->

Only write files that do not already exist. Skip existing hypothesis numbers (idempotent).

Step 6: Report

Output a summary table:

State total hypotheses in the problem directory (existing + new).