Skip to content

reach: a timed-out run reused after an edit depends on timeouts, so its coverage is no longer required (#90) #104

Open

Description

@KyleAMathews

Summary

PR #90 removed the reach state a timed-out run reused after an edit from the witness module's edit sequence. Its predicate compares the cut counts of a daemon check and a cold run (daemon.evidence.cut < coldRun.evidence.cut), and those counts depend on wall-clock timeouts. Determinism failed on the witness unit because the state was present in one run and not the other.

The state is no longer observed, required, or entered by its constructed history (an edit sequence over the witness module). That is a coverage loss: nothing now claims that the daemon reuses a timed-out run after an edit. Law 9 is still judged on the outcomes of that history, since the comparison runs on every check.

What restores it

A deterministic cut: the witness's hang is cut by a count the CLI records (for example, a call budget or an owner's run count), not by the clock. With that, the state's predicate depends only on the history, and the state can be observed and required again. #84 is the same kind of fix for the heartbeat state.

Authority

Register (2026-11-14): "Determinism under load covers witnesses too: an oracle driver must not rely on wall-clock ordering." The evidence counts are not observed, as for cli (note in the stability check).

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions