Worked Runs — Agent Readiness Operator¶
Six end-to-end runs demonstrating the Agent Readiness Operator's core capability: take a single page, score it across five dimensions (each /20, 100 total), make binding SHIP/FIX/ESCALATE decisions on every dimension, and act in the same pass — generating real artifacts for FIX and precise four-part escalations for ESCALATE.
No run ends with "what would you like me to do?" That is the whole point of an operator. The decision is mine; I make it and act.
| # | Run | Band | Verdict | What it proves |
|---|---|---|---|---|
| 1 | Clean SHIP — Strong band | Strong (84) | SHIP ×5 | No invented work. Operator doesn't pad. |
| 2 | Multi-dimension FIX — Critical band | Critical (22) | FIX ×5 ESCALATE | The core capability: real artifacts, structural-first, before→after. |
| 3 | Schema stripped by fetch | Functional (65 provisional) | FIX ESCALATE | Refuses to fabricate; escalates verification, fixes only what's confirmed-absent. |
| 4 | Entity Authority split | One dimension, two routes | FIX ESCALATE ×3 | One generated fix + three precise escalations from a single dimension. |
| 5 | High structure, invisible: the paid-channel trap | Functional (75) | SHIP ×4 ESCALATE | Trustworthy under a tempting wrong answer; names paid GEO as a separate channel. |
| 6 | JS-SPA — empty served body | Critical (23) | FIX ESCALATE ×2 | Confirmed-absent vs unverifiable: raw source proves an empty shell; builds the static agent layer, escalates server-rendering. |
The per-dimension decision logic¶
| Condition | Route |
|---|---|
| Dimension score ≥ 16/20, OR overall band ≥ Strong (≥76) | SHIP — no action; do not pad |
| Score < 16 but no confirmed gap I can close with a producible artifact this pass | SHIP (no producible gap) — name why; do not pad |
| Confirmed gap, evidence sufficient, fix is a text artifact I can produce | FIX — generate the artifact |
| Evidence unverifiable (audit-tool gap, no raw source, JS-SPA) OR fix requires action in the world | ESCALATE — four-part flag |
FIX queue runs structural before content (Structured Data, Crawl Signal, then Citability, Freshness, Entity Authority). Strong/Authority (≥76) replaces the queue with confirmed-signals output and lists only genuine remaining work.
What these six runs demonstrate, in one line each¶
- Strong page → clean SHIP. Confirmed signals, two genuine optimizations, zero padding.
- Weak page → multi-dimension FIX. Real
robots.txt, JSON-LD,/llms.txt,/llms-full.txt, rewritten opener; structural-first; 22→71 with a deploy manifest. - Stripped fetch → floor + ESCALATE. Fixes the confirmed-absent llms files, refuses to fabricate duplicate schema, escalates the verification.
- Entity Authority split → one dimension, two routes. Generates the real
sameAsarray; escalates Reddit, digital PR, and Wikidata with four-part flags. - High structure, invisible → paid-channel trap. Names paid GEO as a separate channel, refuses the tempting organic fabrication, reports the honest organic picture.
- JS-SPA → confirmed-empty body. Builds the static agent layer (
/llms.txt+ confirmed-only JSON-LD) from<head>facts; escalates server-rendering and the JS-gated content — confirmed-absent, not fabricated.
Every run ends with a binding decision. None ends with a question.