04Return to you
Only what needs you comes back.
Assent is designed so that only decisions, blocked stages, reviews and escalations come back, with what wasn’t checked.
Concept UI · illustrative data
What this shows
- Return to you: A pull request that says what wasn’t checked.
The usual way
Our hypothesisA green check reads as done. It says nothing about what wasn’t checked.
Enforced, checked and human-reviewed constraints are different things, and treating them as one creates false assurance. Every report has to say which is which.
- Passing checks presented as correctness
- Evidence from a different commit
- Approvals that outlived their policy
- Success reported by the agent itself
- Every notification treated as urgent
Not checked: Clock skew beyond the tests, Distributed behavior outside the contract, Security semantics beyond existing tests
Step by step
Synthetic exampleThe pull request, and the check that can withdraw it.
The chain of custody from decision to merge, the checks bound to the candidate and what wasn’t verified. Then merge eligibility, which a changed policy would withdraw even when the code is the same.
Check the exact candidate
At completion, the diff would be compared with the authorized paths and the approved target contract, and the registered checks would run.
Bind the evidence
Results would be pinned to the candidate and base commits, the decision digests, the current grant and the required reviews.
Say what wasn’t verified
Failed, unavailable, passed and unverified constraints would be shown separately. Completion would always list what remains unverified.
Bring back only what needs a person
Decisions, reviews, blocked stages and repair escalations would reach the attention list, and nothing else would.
What this shows
- Expose reservation expiry to Checkout
- Chain of custody: Decision rev 2, Grant GR-7f3a, Run 1 repair, Checks 3 of 3, Review pending, Merge people
- Checks · bound to: Path scope, Approved-contract comparison, Consumer contract tests, Provider and consumer review
- Not verified: Expiry behavior under clock skew beyond the exercised tests, Distributed-systems behavior outside the contract, Security semantics not covered by existing tests
What this shows
- Decision: Holds
- Policy: Changed
- Authority: Holds
- Approval no longer applies
- The code didn’t change. The policy did.
Capabilities
First buildall six capabilities
The design, in six parts.
Only what needs you
One list would show the decisions, reviews, blocked stages and repair escalations that need a person, and nothing else.
Evidence bound to the exact change
The pull request would carry the path-scope result, the approved-contract comparison, contract tests and repair history, tied to the candidate and base commits.
A check the runner can’t publish
The required check would come from a trusted publisher, separate from the runner. A result the agent publishes would never count, and stale evidence would be rejected.
What wasn’t verified, stated plainly
Every report would list what remains unverified. Passing checks would never be presented as complete correctness.
Stale approvals are caught
If a policy or governing decision changes before merge, merge eligibility would be withdrawn and revalidation requested, even when the code is unchanged.
People merge
In the first evaluation, authorized people would approve decisions and merge the result. Merge rules would stay with your Git provider.
Limits
What it won’t do, and what comes later.
Later
- Slack and Microsoft TeamsLaterAttention items delivered through existing channels, once approval friction shows it is needed.
- Team and system insightsLaterBottlenecks and decision latency at team and system level, above minimum group sizes. Never individual rankings.
- Steering by behaviorResearchReviewing by behavior instead of code, only for change classes whose verification coverage has been demonstrated.
A structural comparison does not detect every runtime behavior change.
Passing results cover defined rules and exercised behaviors only.
A successful check is a point-in-time result.
While people merge, a race between a policy change and a merge remains, and it would be disclosed.
Administrators may hold bypass powers.
The product doesn’t control changes made outside the configured workflow.
FAQ
Questions about what comes back.
How will the first evaluation be judged?
Read the noteAll questions27Return to you02
Will it merge or deploy on its own?
Not in the first evaluation: people approve decisions and merge. Policy-gated merge and release promotion would be later steps, enabled per change class only with evidence, explicit customer policy and an authorized policy owner’s adoption. A critical incident would step a class back immediately.
What does “unverified” mean in a report?
Constraints no registered check covers, such as transaction or security semantics. Every report would list them and route them to review; a passing check would never be presented as complete correctness.
Design partners
Bring one recent cross-team change.
We reconstruct it with you before showing a solution.
Each stage earned by the one before
Concept diagram
- 01Conversation
- 02Assisted comparison
- 03Shadow analysis
- 04Bounded evaluation
The program ends with a dated decision to continue, revise or stop.
Will it merge or deploy on its own? Not in the first evaluation: people approve decisions and merge. FAQ