# Reports

Every completed run renders a markdown report. It can additionally be published as a
branded, client-facing HTML document.

## Report sections, in order

1. **Header** — brand, category, domain, generated-at
2. **Brand type** — the inferred archetype and the crawl evidence for it
3. **What we understood you to sell** — the brand profile, or a warning that generic questions were used
4. **Score** — either the Agent-Readiness score with its arithmetic shown, or an
   *Insufficient evidence* notice with named blockers
5. **How you compare to your peers** — only when evidence passed
6. **Content coverage** — including a missing-role table
7. **Per-query results by engine** — one table per engine, plus any skipped engines
8. **Why you're losing recommendations** — a checklist with four states
9. **Top 3 fixes**
10. **Footer** — rerun guidance

## The checklist has four states

| Marker | Meaning |
|---|---|
| `[x]` | Signal present |
| `[ ]` | Signal verified absent — this one costs 10 points |
| `[?]` | Could not determine — excluded from scoring entirely |
| `[–]` | Not applicable to this archetype |

The distinction between `[ ]` and `[?]` is the whole evidence model in one glyph.

## The score arithmetic is shown

The report states the subtraction that produced the number, rather than presenting the
total as an oracle. A reader can check it.

## Fixes

Ranked by expected gain, descending:

| Channel | Trigger | Expected gain |
|---|---|---|
| `site` | Missing applicable, verified signal | 10 |
| `content` | Invisible observed query | 15 |
| `content_estate` | Missing expected page role | the role's weight |

"Expected gain" is **score impact if resolved** — the points that penalty is currently
costing. It is not a forecast that fixing it will move the score by that much, and given
the [noise floor](reliability.md#the-noise-floor), a single 10-point fix is not
independently detectable.

## Content coverage is not gated on evidence

Section 6 renders even when the score is withheld. The crawl and the engine panel are
independent failure domains — if the model panel failed but the site was read fine,
the coverage findings are still valid and still worth acting on.

## Branded client documents

A completed run can be published to a permanent link via
[`publish_report`](../mcp/tools/publish-report.md). These documents are:

- **Fully self-contained** — no external stylesheet, script, font, or image
- **`noindex`**, with `noarchive` and `nosnippet`

The robots directives are deliberate: these documents state how invisible a named brand
currently is. Letting a search engine index that would publish a claim about a customer
that the customer did not ask to have published.

Runs whose score was withheld are refused unless `force` is set, and publish with a
withheld notice in place of the scorecard.