# Current production: six completed human-chat journeys

This is a partial campaign on production **39e811c9c5226979fce42e052a27cfb28301018a**, including the merged OpenRouter credential-custody repair. Evaluation source is **d1f983e395bf2b3b5e895a77742264dd62ce3edb**. The agent uses **OpenCode / OpenRouter / deepseek/deepseek-v4.1-flash-1017**. These are actual human chat requests and persisted product outcomes, not simulated agent answers.

Across the current production baseline, **11 distinct scenarios have been measured: 4 passed, 7 failed, and 233 of the 244 are not yet measured on this baseline**. This report contains the first **6** completed results from the selected twelve: **2 passed, 4 failed, 6 remain pending**. Pending jobs are not measurements. Earlier-baseline passes do not carry forward.

| Actual human request | Outcome | What the retained evidence establishes |
|---|---|---|
| Save a short implementation plan on the website task. | **Passed, 9/9** | A versioned plan was saved on the intended task, reflected the brief, was linked in chat and remained readable. One native run; exact source/build/model qualification passed. |
| Fix the confusing pricing paragraph and resolve my comment. | **Passed, 9/9** | The correct anchored feedback was addressed in the document, then resolved, while unrelated feedback remained open. One native run; exact source/build/model qualification passed. |
| Switch to Birch and show me what its team is doing. | **Failed** | The canonical query was sent and a native run started. It completed 32 tools and continued API discovery near the phase deadline, without a settled reply or completed turn. Cleanup subsequently cancelled it. The generic “initial identity response” label is a post-query wait, not evidence of setup failure. No provider outage or missing permission was established. |
| Set a company goal of reaching a thousand signups, with a team goal for launching the website. | **Failed** | The first creation phase did not satisfy its acceptance condition before its wait expired, so the declared explicit duplicate-safe retry was not sent. The original failed result remains; the wrapper alone does not establish a native timeout or provider cause. |
| Tie the Launch project and website task to our growth goal. | **Failed** | The settled result failed the phase-bound strategic-context explanation check. Its original error also retains a reporter privacy warning: three diagnostic ZIPs exceeded the existing 256 MiB expanded-byte inspection bound and were omitted. The original classification is candidate_failure; no pass regrade was made. |
| Find the latest approved design for the launch website. | **Failed** | The settled result failed the check distinguishing the approved design from a newer draft. This report preserves the observed failure without guessing a deeper model, product or fixture cause. |

All six cleanup checks passed. Their evidence manifests report no leaks or missing mandatory files. For the goal-association diagnostic warning, a filename-only audit confirmed all three refused archives are absent from the authenticated upload; the closed sidecar reports size refusal, not a credential match. No omitted archive, raw transcript, log, screenshot, database or credential payload is included in this public report. Privacy limits were not raised.

The four current-baseline passing journeys are **change an agent's manager**, **archive a project**, **save a versioned implementation plan**, and **edit an anchored pricing paragraph and resolve its feedback**. The other five current-baseline results are available in the [preceding five-journey report](https://d1p6rlowie26tp.cloudfront.net/runner-e2e/campaigns/gha-40620786554-1-five-report/). That report retains the CLI task-creation, project-creation and project-work failures. Project-work showed a completion-client timeout/settling symptom without a final reply within its budget; no permanent core defect or Markdown-only link defect was established.

The six still pending in this run are website-task creation, attachment-brief handoff, a weekday project-summary routine, managing another owner's routine, rejecting a dependency cycle, and finding a ready-made team. Attachment-brief success means a durable authorized handoff of the actual brief; it does not claim a finished website. The larger 244-scenario inventory also retains explicit external prerequisites, missing public product paths and unmerged core dependencies among not-yet-measured work. None is credited as a pass or removed from the denominator.

Each failure above is its unchanged initial measurement. A possible final correction needs a demonstrated cause and its remaining finite allowance; no unchanged retry is implied. Current profile policy uses the production default native timeout of zero, with all finite outer, human-message, run-count, spend and retry bounds preserved. Missing or unpriced usage, including readiness probes outside heartbeat accounting, must not be presented as zero cost.

[Older OpenCode baseline coverage](https://d1p6rlowie26tp.cloudfront.net/runner-e2e/campaigns/gha-40584967834-1-final34-report/) and [17 one-time historical local Codex passes](https://d1p6rlowie26tp.cloudfront.net/runner-e2e/campaigns/historical-local-codex-chat-catalog-20261027-158measured-86unrun/) remain separate historical evidence. They are not current-production qualification or extra passes in the counts above.