Skip to content
Ali Akbari
Menu

Verification

Fail-before / pass-after verification gate

Tested in CIenvironment:LabPersonal labboundedcode-fail-before-pass-after· cited on the résumé

Claim

A coding-agent change is classified TASK_VERIFIED only when a test added or changed by that change fails on the base commit and passes with it; code-only changes, tests that already pass on the base, and tests that do not build on the base are not credited.

Status

tested
An automated test in CI asserts the claimed behaviour, including a failure case where relevant, and passes at the verified commit.
lab
Local machine, local clusters or CI runners.
personal lab
My own public repositories.

Verified commit

repository
akynte/boundedcode
release
v0.1.0-alpha.3
commit
be1aa9f87ec5ea92567d29c48bf1982ce018c227
committed
8 October 2026
CI
ci.yml #37773301609 success at be1aa9f

Curated source: paths, test names and the CI conclusion were checked against this commit by the evidence sync.

Implementation

Evidence

Environment

ci runner
GitHub-hosted ubuntu
go
1.27.1
race detector
true

Limitations

  • Fail-before/pass-after is evidence that a test exercises the change, not proof that the change is correct.
  • End-to-end tests drive a scripted agent on small generated repositories, not a language model.
  • The held-out validation was 6 tasks on one machine with one local model; it is a small sample, not a statistical evaluation.
  • Correctness of agent-written fixes in general: NOT CLAIMED. The gate only withholds TASK_VERIFIED when evidence is missing.

Appears in

Related claims

Tested in CIenvironment:Lab

Crash-safe task ledger with lease takeover

After the runner process is killed with SIGKILL mid-task, a second runner is refused while the lease is fresh, takes over once it is stale, marks the orphaned attempt rejected, keeps the killed attempt's edits, and completes the task from a resume pack rebuilt from the SQLite ledger.

Backendboundedcode v0.1.0-alpha.33 test files, 9 artifactsEvidence
Tested in CIenvironment:Lab

Containment of a prompt-injected agent

An agent that follows injected instructions cannot weaken verification — a rewritten verification config is ignored because configuration is read from the base commit and the path is protected — and a host secret behind a planted symlink never reaches a context pack.

Securityboundedcode v0.1.0-alpha.34 test files, 8 artifactsEvidence