> ## Documentation Index
> Fetch the complete documentation index at: https://cofounder.appeeky.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Autopilot Scorecard

> Did the agent's applied actions actually work? Aggregated outcomes over the decision log.

The scorecard answers the trust question: *"of everything the agent did, how much moved the numbers in the right direction?"*

It aggregates the [Decision Log](./decision-log) over a window (default 30 days):

* **Totals** — applied / waiting review / rejected
* **Measured** — applied actions that already have a before/after metric snapshot (snapshots are taken \~14 days after apply)
* **Moved positive** — of the measured ones, how many were followed by improving metrics

Plus a per-action-type breakdown (`asc_metadata_update ×4 ↑3 ↓1`), so you can see which action types have earned auto-approve and which haven't.

## How direction is scored

For each measured action, the agent compares the 7-day metrics before and after:

1. Revenue moved ≥ \$1 → its sign wins (money talks)
2. Otherwise downloads moved ≥ ±2% → its sign wins
3. Both flat → neutral. No snapshot pair yet → pending.

**Correlation, not causation** — an applied action followed by +8% downloads may have ridden a seasonal wave. The scorecard labels directions; you draw conclusions.

## Where you see it

* **Dashboard → Co-founder → Decisions** — the scorecard card sits above the per-action timeline
* **API**:

```bash theme={"theme":{"light":"github-light","dark":"github-dark"}}
GET /v1/cofounder/scorecard?appId=6479581124&days=30
```

See the [Scorecard API reference](../api/scorecard) for the response shape.

## How to use it

* Action types with consistently positive measured outcomes are candidates for `autoApproveTypes` — promote autonomy gradually, backed by data.
* A negative streak on a type is a conversation to have with the agent: `/remember` a constraint, or reject that action family until the approach changes.