Fabric proof rejected.
Digest mismatch.
03:14:27... consensus verification logs flipped the task from ACCEPTED to INVALID before I’d even scrolled past the payload header. The robot had already executed. Grip closed. Object moved. Clean motion. No hesitation in hardware.
Execution cleared. Fabric verification didn't.
Capability hash changed.
We pushed a minor skill module update thirty minutes earlier. Same outward behavior. Different internal path. Verifier recomputed. Digest boundary shifted. Proof no longer matched yesterday’s capability hash.
Consensus verification logs didn't blink.No... Just the label: proof invalid.
Seal event: missing.
Task indexed.
Settlement pending.
Reward state: none.
I pulled yesterday’s digest and lined it up against the new build. Tiny diff in the capability tree. An optimization patch. Faster branch resolution. Harmless in motion. Expensive in proof.
So I did the boring operator thing early... froze the skill chip version on the agent, pinned the Fabric capability hash, and blocked any further upgrades until this window cleared. It pushed the next job behind it. One slot.
Re-ran the same task.
This time the proof digest matched. Seal emitted on-chain. Settlement advanced.
I’m not keeping both versions live again. Not inside a @Fabric Foundation active verification window. Fabric makes that a scheduling constraint, not a preference.
Next time: patch after seal, or don’t patch.