Execution & repair evidence / updated 2 November 2026
Follow the job. Inspect the result.
Explore recorded local runner jobs and the latest Jira-linked repair, retest and Slack merge. Each record identifies its source, outcome and qualification scope.
Recorded internal QA / 2 November 2026
A Jira-linked repair.
A confirmed merge.
The original push failed seven frozen tests. One repair passed the unchanged cases and fresh GitHub CI, received a separate QA-account review and merged through the real Slack approval.
Execution trail 8 steps
Push recorded
The original revision is retained, including its failure.
$ tbench-worker pushOriginal commit dd9b0a3b2c75GitHub run 41341232214Runner invoking computerA failing push. A tested repair.
Replay a real internal QA run: Jira requirements, a source-only AI patch, unchanged tests, fresh GitHub CI and approval in Slack.
The replay uses the recorded internal QA flow with an owner-operated free OpenRouter adapter. Public CLI provider qualification is tracked in the run evidence.
59492598a3275cc378d221b6Private QA repository · GitHub links require repository access. Both QA accounts are operated by one owner. This cohort used an internal free OpenRouter adapter; public provider transport and independent customer qualification remain separate gates.
Observed / native GitHub job intake
Two matrix jobs.
One enrolled laptop.
On 2 October 2026, GitHub queued a two-job matrix and Tbench supplied ephemeral runners on an operator-owned laptop using the existing workflow.
| Job | GitHub outcome | Job ID | Runner registration |
|---|---|---|---|
| alpha | success | 110847641788 | 39 |
| beta | success | 110847642157 | 40 |
Seven signed job deliveries returned HTTP 200. Numeric runner registrations remained correctly matched across the matrix. The temporary QA worker was revoked, and its runners, containers and networks were removed.
Measured / hosted control plane → local worker
From a browser request
to a verified result.
Three final scenario checks used the live Actions operator console, a dedicated private GitHub repository, and an operator-owned Docker host. Each scenario records dispatch, execution, its GitHub outcome and cleanup.
| Scenario | Observed outcome | Workload tests | Isolation probes | Cleanup |
|---|---|---|---|---|
| Passing workflow (private GitHub run, opens in a new tab) | success | 4/4 passed | 12/12 passed | Runner, containers and networks removed |
| Intentional failure (private GitHub run, opens in a new tab) | failure | 3/4 passed; one deliberate failure | 12/12 passed | Runner, containers and networks removed |
| Running-job cancellation (private GitHub run, opens in a new tab) | cancelled | 4/4 passed | 12/12 passed | Runner, containers and networks removed |
Execution controls
Authenticated dispatch, single-use worker enrollment, local execution, GitHub logs and artifacts, signed result ingestion, failure propagation, cancellation and scoped cleanup worked for this private pilot. The job used one CPU, a 1.5 GiB memory cap, 160-process cap, no host mounts and no Docker socket.
Traceable records
Download the exact GitHub run IDs, source commits, step conclusions, artifact hashes and timestamps. Repository links let an authorized reviewer inspect the original logs and artifacts alongside the recorded result.
Development run records · 28 September 2026. The cancellation scenario archives its evidence before the cancellation window; the private run history retains earlier integration attempts.
Primary sources
Explore the original sources.
Architecture documentation and provider pricing reviewed 28 September 2026. Follow each source for its current offering.
Published Linux x64 2-core rate: $0.006/minute. GitHub rounds each job up to a whole minute.
Self-hosted runner usage is currently free; infrastructure and associated services remain the customer's responsibility.
Listed Linux x64 2-vCPU rate: $0.004/minute. Consult its current price sheet for resource classes and billing terms.
Persistent snapshot-backed build state, with branch-protection controls for cache writes.
Managed CI on customer infrastructure using disposable microVMs. A direct alternative to evaluate.
Customer-owned AWS infrastructure with an annual software licence; cloud infrastructure is billed separately.
Runner code can access referenced credentials. Log redaction is not a security boundary.
Linux/KVM requirements constrain where this microVM backend can run.
Containers need explicitly configured resource limits; defaults do not reserve the host's resources.
15 days to put it to work
Bring your organization’s next workflow.
Apply with your company email. We review your organization, repositories and device scope. Your 15-day trial starts after organization approval. No automatic charge.