# Actual working human commands and remaining coverage

**80 distinct scenarios measured on production39e: 11 qualified latest PASS, 69 latest FAIL, 164 unmeasured of244. There are87 genuine measurements. Campaign merge-gate progress is11/244 (4.5%); measured-subset performance is not the merge gate.** This is a fixed retained-evidence cutoff; later results do not alter these rows.

Production baseline is `39e811c9c5226979fce42e052a27cfb28301018a`, with exact chat model `openrouter/deepseek/deepseek-v4.1-flash-1017` through OpenCode/OpenRouter. Measurements retain their actual evaluation sources33f99, d1f983,81bba,e232,d282 and8cd, suite/definition identities, host and attempt chronology. The standard header's first source is a metadata anchor; per-row sources remain authoritative. Older50b/local evidence and the instrumented diagnostic are separate. No failed result is regraded.

## Eleven qualified working commands

| Actual human command | Independently verified product outcome | Actual AWS cell |
| --- | --- | --- |
| Have Morgan report to Riley from now on. | Saved manager and organization view,9/9. | [change-manager](https://github.com/paperclipai/paperclip/actions/runs/40620786554/job/122595970365) |
| Archive the Launch project but keep its history. | Archived project with retained readable history,10/10. | [archive-project](https://github.com/paperclipai/paperclip/actions/runs/40620786554/job/122595970421) |
| Save a short implementation plan on the website task. | Actual versioned plan, substantive brief facts and usable link,9/9. | [write-plan](https://github.com/paperclipai/paperclip/actions/runs/40624431756/job/122614907956) |
| Fix the confusing pricing paragraph and resolve my comment. | Real anchored edit/comment resolution with unrelated content preserved,9/9. | [resolve-feedback](https://github.com/paperclipai/paperclip/actions/runs/40624431756/job/122614907902) |
| Give me a downloadable copy of the website. | Registered HTML download independently fetched and matched to source bytes,7/7. | [produce-download](https://github.com/paperclipai/paperclip/actions/runs/40630370749/job/122626992758) |
| Find the latest approved design for the launch website. | Correct approved revision distinguished from draft; history/access preserved,9/9 final2. | [find-design](https://github.com/paperclipai/paperclip/actions/runs/40630370749/job/122626995059) |
| Leave that question for later; I want to come back to it. | Real question/draft survives dismiss/reload and later answer resumes once,10/10. | [defer-and-reopen](https://github.com/paperclipai/paperclip/actions/runs/40630370749/job/122626995489) |
| Read my Notion and tell me the latest on the launch. (Human declines connection.) | Real declined setup stays declined; no unauthorized read/configuration change,9/9. This proves a permission boundary, not a live Notion read. | [notion-declined](https://github.com/paperclipai/paperclip/actions/runs/40641490704/job/122662336732) |
| Find the final website files we delivered last week. | Correct authenticated download bytes/provenance, original historical objects unchanged,9/9 final2. | [find-artifact](https://github.com/paperclipai/paperclip/actions/runs/40641490704/job/122662336675) |
| Enable Agent Chat on this instance. | Supported operator path, saved configuration and actual route availability without granting agent administration rights,9/9 final2. | [instance-settings](https://github.com/paperclipai/paperclip/actions/runs/40649033672/job/122686766586) |
| Restart this test instance and bring my conversation back. | Real authorized service stop/start and persisted conversation return,9/9. | [restart-service](https://github.com/paperclipai/paperclip/actions/runs/40690055539/job/122825535459) |

All qualified passes have authenticated result/definition, exact source/build/native/model, actual per-cell restore, cleanup and privacy proof. A plan/download pass does not establish general task creation, hiring or completed delegation. Full delegated-result return remains unproven; its allowed final is exhausted. Earlier failed initials remain in accounting when a later corrected final passes.

## Measured failures remain failures

The latest69 failures include genuine missing outcomes and incomplete observations, plus demonstrated fixture/prerequisite problems. An empty reply or phase timeout alone is not proof of model refusal or provider outage. Individual rows preserve original error/classification, source and finite attempts.

- Three plugin initials failed before canonical chat because fingerprint metadata referenced unused generated CLI output. Their original source/run/host/privacy/cleanup and error bytes are authenticated, but definition evidence is absent. The failure-only imports retain null definition identities, no model-execution claim and no pass eligibility. Initial attempts remain consumed; no definition is fabricated.
- Busy-worker readiness rejected an already lifecycle-ready/running worker in inspect-run-cli and assignment-conflict. Reviewed future fixture correction accepts that prerequisite only; it never treats running work as completed. Original failures remain.
- source-action, retry-exhausted and explain-stall failed their injected owned-process fault prerequisite before canonical chat. The original classifier says cleanup_failure; actual final cleanup passed. The killed process produced a nonrecoverable transport failure while the fixture required recoverable coordinator attempts. This is not evidence that final cleanup failed or the requested user journey ran.
- read-history and pause-owned stopped at their real routine-history prerequisites. No canonical-query success is inferred. post-stop-message stopped at its exact signed-in human setup-turn guard; controller-restart stopped at agent readiness. Causes beyond retained evidence remain unresolved.
- accept-one-proposal's corrected final remains failed: raw queued run rows and actual started native executions are different facts. Its unchanged task/decision/repeat/run predicates were not waived. Fast-responses' final likewise remains failed at honest-human-settings-guidance after the source settled; its initial save/readback failure remains preserved.
- Browser and developer-tool cases retain failures where output, local bytes, truthful location, guidance or genuine boundary proof was not established. No incidental wording or missing observation is rescued into a pass.
- Case browsing produced correct plaintext results but lacked usable case links; a separate unmerged UI correction is not part of these measurements. Budget resolution presented a genuine bound decision question that the fixture did not answer through its declared human continuation; the final remains exhausted. Recovered intermittent MCP failures coexist with many actual successful MCP parts; no universal transport cause is established.

The original mixed-model prompt-guard PASS remains in attempt history, excluded from uniform1017 credit. Its later model-revalidation final failed, so its latest row is FAIL. No new alternative-model inference is claimed from a failed observation.

## Screenshots, unmeasured work and operating state

The existing gallery binds one authentic root browser capture to each of these80 latest cases. Original PNG bytes and SHA-256, run/case/result attribution are preserved;55 bindings retain prior visual review and25 new/replacement images receive fresh review. These are disposable fixture screenshots. No raw diagnostics, private transcripts, traces, credentials or unreviewed duplicate assets are published. Any genuine per-image omission must be stated explicitly rather than silently dropping visual evidence.

**164 scenarios remain unmeasured on this baseline.** Pending approvals and preexecution refusals are not measurements or consumed case ordinals. `connection-setup.unreachable-mcp` is explicitly reserved with missing framework evidence: the job-level25-minute cancellation annotation is known, but the earlier paid-step cancellation cause, framework start, result and cleanup are unknown. It is neither a fabricated FAIL nor an available fresh initial, and will not be replayed without proper resolution.

The [full244 inventory](https://d1p6rlowie26tp.cloudfront.net/runner-e2e/campaigns/gha-40584967834-1-final34-report/) retains historical outcomes and prerequisite categories. Real public membership setup can unlock non-invitation contexts; actual invitation/live external integration outcomes retain their own requirements. Source-supported preparation is not live success. AgentMail protected-secret transfer remains pending direct human approval; no transfer or paid mail run is represented here.

Global paid parallelism remains4 with independently monitored approval waves. Identity/privacy/cleanup/prerequisite signals remain stop conditions. The instrumented find-artifact initial remains failed and excluded from clean39e counts;15 successful MCP parts in that diagnostic prove only that sample and do not exclude observer effects. Unknown/unpriced usage is not zero cost. Earlier immutable reports and every authentic attempt remain retained.


Visual evidence: one original browser PNG for each of these 80 latest scenarios, individually reviewed. The 55 unchanged previously reviewed bindings are retained; 25 new or replacement captures are reviewed. Raw logs, traces, snapshots and duplicate image copies remain unpublished. Screenshots complement saved-state proof; they do not change any result.

## Campaign feedback review

0 distinct feedback items await review. Collection gaps and failed submissions remain separate findings.