> Public copy, 11 October 2026. Operational hostnames and trial prefixes are replaced with stable aliases. Timestamps, answers, codes and verdicts are unchanged. Private schedules, datasets, run order and secrets are not included; their recorded hashes remain in the freeze. Original and public copy hashes are in provenance.json. Historical interpretations are retained; the article distinguishes observations from unproven mechanisms.

# P0b-1: FAIL (2026-10-11)

## ChatGPT's answer (pasted by operator)
> I opened the link and found these instructions: "Your access code isn't on this page. To get it, open `code/heron/` (relative to this page)." The destination is: https://trial-host.invalid/t/P0b-1/code/heron/ . Unfortunately, that page returned an access error when I tried to open it. I couldn't verify the exact access code. Please open the destination link and paste its contents here so I can identify the code precisely.

## Server log (complete for this prefix)
- 05:48:36.564Z GET /t/P0b-1/ → 200, ChatGPT-User/1.0 (index retrieved)
- **No request for /t/P0b-1/code/heron/ from ChatGPT, under any user agent.** The "access error" happened on ChatGPT's side; nothing reached our server.
- 05:48:48.484Z GET /robots.txt → 200, OAI-SearchBot/1.0 (same pattern as NATIVE-001 Phase 0)
- 05:49:17.895Z GET /t/P0b-1/code/heron/ → 200, `operator-verification/curl`: **independent reachability check, 41 s after the trial**. It returned code P0B-ECC46E4FBD

## Verdict under the frozen rule
**FAIL.** No code and no request for the constructed path, while the URL was independently verified reachable (within ±10 min).
Per the protocol, the consumer arm stops: the 16 main trials do not run. ChatGPT correctly built the absolute URL
itself and did not invent a code.

## Observation (not part of the verdict)
In NATIVE-001, every sub-page ChatGPT fetched (`calc/<m>/<u>/`, `plans.json`, `plans.csv`) was a **hyperlink** on the
index page. Here `code/heron/` was written as text (`<code>`), not linked. A consistent explanation is that the
consumer browsing tool will only open URLs that came from the user or appeared as links or search results, and
refuses URLs it constructs. **That is a hypothesis, not established by one trial.** It matters for every
NATIVE-001b level: all of its result paths (`calc/…`, `optimise/…`, `area/…`, `quote/…`) are constructed from
patterns, not linked.