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
#735 is the umbrella: it says what the milestone contains and closes when its table is full. This issue is a different thing, and confusing the two is how a milestone ships on a feeling. It asks one question, once, at the end:
Has the dependency on the order of operations actually been eliminated, or has it only been described?
The release claim this gate defends:
The local cloud has one stable and executable semantics. For every runtime-backed effect, the desired topology, the provider mapping, the supported runtime modes, the observable witness and the lifecycle behaviour are explicit. Two accepted API histories that reach the same desired topology converge to the same normalised observable runtime.
A row that is described but not executed does not satisfy it.
The lifecycle is re-proved, not proved once. create, stop/start, reboot, a Day-2 gesture, destroy and cleanup each re-take the witness. A property proved at Day-0 and never re-proved is not in the contract.
The three providers each exercise at least one representative topology under incus-ovn, and the bridge mode is judged only on what it declares. A mode that does not claim an effect is not red for lacking it.
Nothing foreign is touched. During creation, permutation and cleanup, no runtime object outside feint's ownership is modified. This is a witness, not a review.
The gate is falsified before it is trusted
A release gate that has never been red is a gate nobody has tested. Before this issue closes, each of these must turn it red, and the run is recorded:
remove one route after create → red
reorder gateway and machine on a tree
reproducing #678 → red
blank one contract cell → red
make one witness unreadable and report deny → red
The fourth is the one that matters most: it proves the gate distinguishes "observed to block" from "we could not look", which is exactly what #754 defines and what four red nights were made of.
Not a promise about modes nobody measured. An unsupported, not applicable or control-plane only verdict is a valid answer here, and an unmeasured one is not.
#735 is the umbrella: it says what the milestone contains and closes when its table is full. This issue is a different thing, and confusing the two is how a milestone ships on a feeling. It asks one question, once, at the end:
Has the dependency on the order of operations actually been eliminated, or has it only been described?
The release claim this gate defends:
A row that is described but not executed does not satisfy it.
The gate
S → G → M,S → M → G,G(push=false) → M → PATCH push=true, and one permutation per contract row that has more than one accepted order.opn-fntwhile its plan claims no rule set, and the verification guard reds the runtime leg on it #741 are fixed under the contract, or the capability is removed, or the limitation is measured and written. A green claim over a broken witness is the one outcome this milestone exists to make impossible. (Under incus-ovn, platform-web-0 carries its public address on both eth0 and eth1 until the reboot verb takes it off eth0, and the stack gate reds on the difference #742 was closed on 2026-09-07; it stays in this list as the shape to re-run, not as outstanding work.)runtimejob of runtime-proof.yml never runsguard.sh verification, so a claim the emulator reported broken has never reddened a night #740: theruntimejob consumesguard.sh verification, so a runtime feint itself reported broken cannot pass.incus-ovn, and the bridge mode is judged only on what it declares. A mode that does not claim an effect is not red for lacking it.The gate is falsified before it is trusted
A release gate that has never been red is a gate nobody has tested. Before this issue closes, each of these must turn it red, and the run is recorded:
The fourth is the one that matters most: it proves the gate distinguishes "observed to block" from "we could not look", which is exactly what #754 defines and what four red nights were made of.
What this gate is not
unsupported,not applicableorcontrol-plane onlyverdict is a valid answer here, and an unmeasured one is not.Not established
incus-ovnleg alone took 32 minutes on 2026-09-07 (Runtime Contract v1: every runtime-backed effect has one normative definition, one accepted decision where needed, and one executable witness, and no provider path or operation order defines runtime semantics alone #735), and this gate is broader. Whether it runs in one workflow or in several is decided by the first measurement, not here.