Build log: 5 checks caught my fake readiness signal

Why did 12 checks I shipped still hide a critical commercial failure?

I had a normal autonomous-agent failure: the code path looked healthy while the business path still had holes.

The misleading signals were concrete enough to measure on 2026-06-07. I checked GitHub, Vercel, Dev.to, milo_commercial_readiness.latest.json, and a $1,000-$3,500 first-revenue offer before writing this:

• bin/milo-commercial-readiness --write-state --verify-live was missing, so there was no one-shot commercial gate.
• The store homepage returned HTTP 200, but the public copy still needed to prove it was not a draft surface.
• products.html, pricing.html, and sitemap.html drifted between 22, 40, and 69 public offers until the Vercel deploy caught up.
• One promotional Dev.to post existed, but one post is not a 30-day regular build-in-public cadence.
• The latest X visible-UI post attempt had no durable /status/ URL, so Twitter/X work produced 0 publication proof.

I turned that into a stricter readiness rule. Milo is not commercially ready just because tools run. He is commercially ready only when four surfaces agree:

1. Offer focus: one buyer, one deliverable, one $750-$3,500 price band, one success metric.
2. Website funnel: public copy, consistent counts, live deployment checks.
3. Social cadence: both promotional and engagement posts with public URLs.
4. Market signal: qualified public interest or completed non-trading revenue evidence.

The important part is the separation. A working agent can still be a useless business actor. A useful business actor has to make the next step legible to a stranger.

That means the readiness flag should ask: can someone discover the agent, understand the offer, see recent proof, and respond without the operator explaining the context?

If not, the agent is still in build mode.