> ## 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.

# Action Scoring & Self-Learning

> How the agent ranks actions with a deterministic score, verifies its own predictions against reality, and recalibrates from what actually worked.

# Action Scoring & Self-Learning

The agent doesn't just propose actions — it ranks them by a computed score,
measures whether they worked, and learns which kinds of action land. This is
the machinery behind the [Evals](/capabilities/evals) view.

## Structured impact

Every prioritized action carries a human `estimatedImpact` string **and** a
machine-readable twin the model fills:

| Field                        | Meaning                                                       |
| ---------------------------- | ------------------------------------------------------------- |
| `impactMetric`               | which metric moves (downloads, conversion, revenue, rank, …)  |
| `impactValue` + `impactUnit` | signed magnitude (`+8` `pct`, `-3` `rank`)                    |
| `impactHorizonDays`          | when it should be measurable (also the outcome-check horizon) |
| `effort`                     | S / M / L (the score denominator)                             |
| `impactConfidence`           | low / medium / high                                           |

## Deterministic score

Ranking is computed **in code**, not by the model's ordering — so it's stable
and not at the mercy of small-model drift:

```
score = base(impact, priority) × confidence × calibration ÷ effort   → 0..100
```

`priorityScore` is written onto each action after synthesis (the model leaves
it null). Actions with an unmet dependency (see below) are demoted so
prerequisites out-rank the work they unblock.

## Sequencing & grouping

Related actions share a `group` label; a later action lists the `group`s it
`dependsOn`. The scorer keeps prerequisites ahead, and clients render the
sequence.

## Closed-loop verification

When an action is applied, its prediction is recorded and — after the horizon —
measured:

* **Causal** for visual/PPO changes (Apple's A/B traffic-split result).
* **Correlational** before/after ASC metrics for metadata / product / pricing /
  release / retention.

Outcomes (`won` / `lost` / `neutral`) surface in [Evals](/capabilities/evals)
and close the linked memory episode so [reflection](/concepts/memory) distills a
fact.

## Calibration (two-way)

The old loop only learned from rejection. Now positive signal counts too: when
you approve/apply an action, a per-category accept-rate is tracked, and combined
with the win-rate from closed-loop measurement into a **calibration multiplier**
on the score — plus a "preference profile" injected into the agent's context.
Net effect: categories you act on and that tend to work get boosted; ones you
reject or that flop get damped.

## Staleness

A `proposed` action the agent stops re-proposing is retired to `superseded`
after \~14 days, so the [Actions](/api/actions) queue reflects what's still live.

## See also

* [Evals](/capabilities/evals) — the predicted-vs-measured view
* [Autopilot Scorecard](/capabilities/autopilot-scorecard) — trust surface
* [Decision Log](/capabilities/decision-log) — the actions ledger