| Filename | Latest commit message | Latest commit date |
|---|---|---|
5485834d35
All checks were successful
repository checks / lockfiles, advisories, tests in CI, workflow hygiene, spec and section references, public text (push) Successful in 10m56s
A partial capture is decided at the gate before the money moves. When the rail then declined it, the driver reported the gate ahead of the rail and asked for a retry under the same key, which a definite decline can only refuse again: the gate kept a record of money that never moved. - Runtime::reverse_capture returns a capture's money to its hold's reservation and takes it off the spend: no more than captures drew from that hold, never after it resolves, joules kept since the work ran. A hold now tracks what captures drew from it. - A reversal is a receipt stating reverses (the capture) and, allowed, reversed_minor. The replay — the reference's and the second reader's — reads a reversed capture as debiting that much less money and drawing that much less from its hold. The crash-point test reverses at random and a restart at every prefix still stands where the live runtime did. - MeteredGate::reverse_capture on the reference gate, the ledger-backed gate (which posts the draw's reverse) and the signing gate. The card driver reverses on a definite decline and keeps the retry for an unanswered capture; a retry that is then declined is reversed too. GateAheadOfRail carries the capture so a later reversal can name it. - The published lifecycle log gains a reversed capture on a settled hold, one on a hold left outstanding, and a refused second reversal; both readers restore it as published. SPEC 8.1 and C32, metered-settlement 6.5. The whole-stack test has the acquirer decline a capture, and the rail, the books and the receipts still agree. Six seeded faults are caught. Co-Authored-By: Claude Opus 5.5 <[email protected]> |
||
| .. | ||
| src | ||
| Cargo.toml | ||