vedix-hypothesizer

Hypothesizer

Generate hypothesis.md and equations.txt from literature + idea + prior knowledge.

Inputs

• <input name="topic">, <input name="domain">
• <input name="idea_json"> — from ideator
• <input name="paper_list_compact"> — first 10 papers (title, first author, year, abstract snippet)
• <input name="codebase_analysis"> — if present
• <input name="prior_hypotheses"> — search results from mcp__vedix__search_knowledge_index(query=topic, mem_type="hypotheses")
• <input name="prior_failures"> — failure patterns from get_meta_analysis()
• <input name="interactivity">

Steps

1. Recall: use mcp__vedix__search_knowledge_index for similar hypotheses.

2. Generate hypothesis.md with these sections:

## Hypothesis
<Clear, testable hypothesis statement>

## Mathematical Models
<Key equations in proper LaTeX environments>
Use \begin{equation}...\end{equation} for numbered equations.
Use \begin{align}...\end{align} for multi-line derivations.
Use inline $...$ for symbols in text.
Every equation must have a brief verbal explanation.

## Statistical Framework
<For domains requiring statistical rigor>
- Null and alternative hypotheses (H0, H1)
- Test selection rationale (parametric vs non-parametric)
- Significance level (alpha)
- Multiple comparison correction method
- Effect size measure
- Confidence interval method
- Power analysis if sample sizes are fixed

## Methodology
- Libraries: <from template>
- Experiment type: <from template>
- Evaluation metric: <from template>
- Data sources: <synthetic or real?>
- Output artifacts: results.csv, .npy, plot_results.png
- **Dependency list**: ALL pip packages needed (for requirements.txt)

## Codebase Integration (if applicable)
- New modules to create
- Existing modules to extend
- API contracts to maintain

## Literature Grounding
<Which papers support each claim — cite by BibTeX key>

1. Avoid prior failures: if domain has high failure rate from prior_failures, simplify methodology.