> For the complete documentation index, see [llms.txt](https://docs.mediafier.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.mediafier.ai/overview/spend-and-outcome-evidence.md).

# Spend & Outcome Evidence

Cost per successful outcome: verified-not-estimated spend, mastery scorecards, human touches, and declared, versioned baselines.

> **Cost per successful outcome, not cost per call.** Mediafier attaches cost and outcome evidence to governed execution so an organization can see what a result cost, whether it improved, and what a person still had to touch.

***

## What is measured

* **Cost per successful outcome.** Spend is settled against what actually ran and what it actually produced. A failed outcome is reported as a failed outcome — it is never dressed up as a partial success.
* **Spend verified, not estimated.** Where a figure cannot be verified it is reported as unavailable rather than estimated. Reconciliation of estimated-versus-actual spend is reviewed by the platform; adjustments are not automatic and are visible when they happen.
* **Quality.** MediaClaws carry mastery scorecards — a measurable record of how well they do the outcome they were built for — and a non-regression rule: a change that lowers outcome correctness does not ship.
* **Human effort.** Counted as the number of times a person had to touch a run (approve, correct, override, intervene), so "less human work" is a number, not a claim.

## Baselines

An improvement is only an improvement against something. Baselines are **declared by the organization**, **versioned**, and **attributable**: when a MediaClaw or an agent claims to have reduced cost, time, or human touches, the claim is evaluated against the baseline the organization approved, and the record shows which version of the baseline was used. Changing a baseline is itself a recorded, human decision.

## What the evidence looks like

Every governed run produces a receipt-style record: what was requested, what ran, what it cost, what it produced, and the trace identifier that links it to the audit record and to the Agent Operations view. Figures shown in product marketing are illustrative; the figures in your organization's records are the ones that are settled.

## Spend controls

* Chargeable work is checked before execution and refused when payment requirements are not satisfied; an agent receives a structured refusal it can act on (re-plan, or hand off to a person to top up).
* Topping up credits, accepting terms, and revealing credentials are human-only actions — an agent cannot perform them autonomously.
* Every charge carries the trace identifier of the call that produced it.

See [Credits & Billing](/overview/credits-and-billing.md) for how credits, wallets, and metering work.

## What this page is not

* It does not quote prices. Current commercial terms come from call-time responses and your agreement, never from cached documentation.
* It does not describe how settlement is implemented. The authenticated product and your organization's records are the authority for your figures.

## Where to go next

| If you're…                                   | Read next                                                                         |
| -------------------------------------------- | --------------------------------------------------------------------------------- |
| Supervising runs and deciding on exceptions  | [Agent Operations](/platform/agent-operations.md)                                 |
| Understanding credits, wallets, and metering | [Credits & Billing](/overview/credits-and-billing.md)                             |
| Following a charge back to a call            | [Audit, Observability & Compliance](/trust/audit-observability-and-compliance.md) |