# Problem Statement

<!-- FILL IN: replace everything below with your actual problem description -->
<!-- This file is read by agents at the start of every IDEATE session -->

## What we're trying to solve

{FILL_IN: 2-3 sentences describing the research problem}

## Why it matters

{FILL_IN: why does solving this matter? What's the impact?}

## Current best approaches

{FILL_IN: brief description of SOTA and why it's insufficient}

## Our approach / hypothesis space

{FILL_IN: what class of solutions are we exploring?}

## Metric

We optimize: **{metric name}** ({lower/higher} is better)

Baseline (untrained or current best): **{value}**

Baseline std across 3 seeds: **{value}** (must be < effect size of a meaningful improvement)

Train + eval wall time: **{X} minutes** (target: 5-40 min)

If variance is too high relative to effect size, use more eval data, a smaller proxy task,
or a lower-variance metric -- before running any experiments.

## Data

{FILL_IN: dataset, size, splits}
- Train: {description}
- Val (eval.py uses this, FROZEN split): {description}

## Constraints

- {FILL_IN: compute budget, e.g. "5 min training on single A100"}
- {FILL_IN: any other hard constraints}

## Key papers

See `0_docs/papers/` for full summaries. Quick list:

| Paper | Key claim | Trust signal |
|-------|-----------|--------------|
| {FILL_IN} | {claim} | {citations / code / self-report} |

## Lessons so far

{start empty, accumulate as experiments run}

## What has NOT worked

{start empty, accumulate}