You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Feedback that turns out to be wrong can be withdrawn by a person, and the developer who received it has a
way to say so from where they are reading it.
Problem
Hephaestus will be confidently wrong in public on someone's pull request. That is not a risk to mitigate;
it is a certainty to design for, and the first occurrence sets how much the team trusts everything after
it.
Today there is no product answer:
No retraction. There is no retract API and no retracted state. The pieces exist — FeedbackPlacement stores the posted comment reference, and the provider channels can edit and minimise
a comment — but they only fire from a subsequent review run. The only documented path is a manual
operator procedure in the runbook (docs: production operations runbook #1371): find the delivery in the ledger, delete the comment through
the provider API, mark the entry.
No reachable dispute. The reaction path is server-complete, including recurrence suppression that
stops a disputed observation being raised again, and it has zero call sites in the webapp. feat(webapp): developer reflection dashboard (per-practice observations and feedback) #1340
carries an in-app version as one clause of a whole dashboard; a developer reading a wrong comment on a
pull request has no path at all.
What
A retract action a workspace admin can take on a delivered unit, driving the stored comment
reference through the existing edit/minimise channels, with a retracted state in the ledger so the
history stays honest. Correct in place rather than deleting: anyone who already read it should see the
correction too.
An in-context dispute affordance on delivered feedback, so the developer can mark it wrong where
they are reading it, not only inside the web app.
Un-gate the existing suppression so a disputed observation is not raised again on the same work.
Outcome
Feedback that turns out to be wrong can be withdrawn by a person, and the developer who received it has a
way to say so from where they are reading it.
Problem
Hephaestus will be confidently wrong in public on someone's pull request. That is not a risk to mitigate;
it is a certainty to design for, and the first occurrence sets how much the team trusts everything after
it.
Today there is no product answer:
FeedbackPlacementstores the posted comment reference, and the provider channels can edit and minimisea comment — but they only fire from a subsequent review run. The only documented path is a manual
operator procedure in the runbook (docs: production operations runbook #1371): find the delivery in the ledger, delete the comment through
the provider API, mark the entry.
stops a disputed observation being raised again, and it has zero call sites in the webapp. feat(webapp): developer reflection dashboard (per-practice observations and feedback) #1340
carries an in-app version as one clause of a whole dashboard; a developer reading a wrong comment on a
pull request has no path at all.
What
reference through the existing edit/minimise channels, with a retracted state in the ledger so the
history stays honest. Correct in place rather than deleting: anyone who already read it should see the
correction too.
they are reading it, not only inside the web app.
Acceptance criteria
records who retracted it and why.
Out of scope