Counter ASIC 3.0 status: main's ruling on the digest (an absent optional field contributes nothing; publish 1 no split, publish 2 the one sweep)

docs/plans/counter-asic-3-status.md

@@ -231,7 +231,7 @@ Two preconditions on the cut, from main (17:0x UTC, both from today's incident),
 
 Facts for the cut from the node lane (docs/plans/counter-asic-3-node.md): the devnet digest moves (c562d70e... to 3c505021...), so the cut is a one-sweep binary rollout and a 0.3.13 node is refused at the handshake afterwards (intended, fleet-wide); a 0.3.13 miner reads the v4 height as never (an optional proto field), so miners and nodes move together; `infra/fast-time/override-60x.json` as committed carried the proving-v1 block twice and lacked four 0.3.12 and 0.3.13 fields (the node refused the file; fixed, with a new CI check `override-json-check.sh`); the 48 GiB target clone `vendor/igneum-node-ca3v4/target-ca3v4` can go after the cut.
 
-THE ONE LINE FOR [user]: class v4 (`mx8+sh256x27`, 100,000 ops per hash in the latency shadow) has passed every gate on the fixed tree (the program-id blocker closed with its own failed case) and is ready for a whole-fleet one-sweep cut (the digest moves to ac8e60ce... at the floor N6) once P1 passes: P2, miner-signalled activation (95 percent of blue-block weight over a one-day window, the floor height after which it flips regardless), is designed, implemented and GREEN on the fast-time gate with its failed case and on G6; P1, the rehearsal on the rented fleet as a staging network, has its plan and objects written and waits for the fleet agent's run; the clean-day wait removed by the project lead ("can we run the v4 class now?", 18:2x UTC): the P1 rehearsal runs NOW on the fleet (15 prover boxes plus the four Devnet 2 boxes, the wave joining about 18:50Z) and the 0.3.15 cut (class v4 plus the fourteenth field, one digest move, miners first, hands last) goes tonight on the project lead's go the minute P1 passes; its cost is the 5090 at 431 W instead of 350 for 0.2 percent less rate (a rig pays about 23 percent more electricity), the M5 Max at 37 W instead of 21 for 1.5 percent, the 9070 XT no rate at all (its watts pending), the verifier +0.27 ms per warp; what it buys is the stored-dataset chip's per-joule edge over the 5090 falling from 5.6x to 2.1x at a chip core equal to the GPU's; proposed for a day when no other cut is in flight, not tonight (0.3.14 and the fleet night come first).
+THE ONE LINE FOR [user]: class v4 (`mx8+sh256x27`, 100,000 ops per hash in the latency shadow) has passed every gate on the fixed tree (the program-id blocker closed with its own failed case) and is ready for the cut once P1 passes (publish 1 the 0.3.15 binary with no handshake split under the digest-compat rule, publish 2 the sixteen-field object as the one sweep, its digest read on the fixed 0.3.15 binary): P2, miner-signalled activation (95 percent of blue-block weight over a one-day window, the floor height after which it flips regardless), is designed, implemented and GREEN on the fast-time gate with its failed case and on G6; P1, the rehearsal on the rented fleet as a staging network, has its plan and objects written and waits for the fleet agent's run; the clean-day wait removed by the project lead ("can we run the v4 class now?", 18:2x UTC): the P1 rehearsal runs NOW on the fleet (15 prover boxes plus the four Devnet 2 boxes, the wave joining about 18:50Z) and the 0.3.15 cut (class v4 plus the fourteenth field, one digest move, miners first, hands last) goes tonight on the project lead's go the minute P1 passes; its cost is the 5090 at 431 W instead of 350 for 0.2 percent less rate (a rig pays about 23 percent more electricity), the M5 Max at 37 W instead of 21 for 1.5 percent, the 9070 XT no rate at all (its watts pending), the verifier +0.27 ms per warp; what it buys is the stored-dataset chip's per-joule edge over the 5090 falling from 5.6x to 2.1x at a chip core equal to the GPU's; proposed for a day when no other cut is in flight, not tonight (0.3.14 and the fleet night come first).
 
 ## 6. Decisions for the project lead
 
@@ -297,7 +297,7 @@ A script never switches the installed app's cards: a POST of enabled false to /a
 |---|---|
 | The node: the signalling fork 0562a7f2 merged onto 0.3.14's node tip 4c6b129d = f86a33c0 | merged; the header processor byte-identical between 0562a7f2 and f86a33c0 (the class signal read after the cheap checks, as before) |
 | The suite on igneum-build-1 | one test, `cheap_checks_run_before_the_pow_engine`, failed at its LAST assertion (the genesis-day hint), not the four cheap-check assertions: a test race on two process-wide things (the engine captured at consensus construction inside the test's install window; the live day length another test can swap) that the box's 96 test threads hit and PC 2's 24 never did; the fix is tests only, fork commit 791ff22c on ca3-v4-order-fix from f86a33c0 (an install_engine_lock in kaspa-pow held across the window by both installing tests; the day assertion made the processor's own promise), with the shipper at 18:5xZ; the single test 1 of 1 on the box (43 s); the full kaspa-consensus run from the fix worktree on the box, its line pending |
-| The order, with the shipper's measurement on the 0.3.15 Mac binary (fork f86a33c0, no suffix, no peers) | THE DIGEST MOVES AT PUBLISH 1: on the live thirteen-field file 0.3.15 reads c09c3e48... where 0.3.14 reads b18ed271... (the two v4 fields fold into the digest at never when absent), so publish 1 (binaries only) is itself the one-sweep handshake split and every node moves inside one window there, miners first, the hand nodes and the seeds last; publish 2: the SIXTEEN-field object (the live thirteen plus 0.3.14's exec_restart_state_root, the floor 226,800 from DAA 209,458 at a 19:47Z publish, the window 86,400; digest 2c1162e2... on 0.3.15; the node's lines "Program class v4 from the override file: active from epoch 63 (DAA 226800)" and the 86,400 window at 9,500 bps) only once every node reports 0.3.15; a 0.3.14 node given it exits at parse ("unknown field program_class_v4_activation_daa"), confirmed on the Mac binary as the rehearsal found. The node lane's tests-only fix 791ff22c merges into the release node |
+| The order, with the shipper's measurement on the 0.3.15 Mac binary (fork f86a33c0, no suffix, no peers) | THE DIGEST MOVES AT PUBLISH 1: on the live thirteen-field file 0.3.15 reads c09c3e48... where 0.3.14 reads b18ed271... (the two v4 fields fold into the digest at never when absent), so publish 1 (binaries only) is itself the one-sweep handshake split and every node moves inside one window there, miners first, the hand nodes and the seeds last; publish 2: the SIXTEEN-field object (the live thirteen plus 0.3.14's exec_restart_state_root, the floor 226,800 from DAA 209,458 at a 19:47Z publish, the window 86,400; digest 2c1162e2... on 0.3.15; the node's lines "Program class v4 from the override file: active from epoch 63 (DAA 226800)" and the 86,400 window at 9,500 bps) only once every node reports 0.3.15; a 0.3.14 node given it exits at parse ("unknown field program_class_v4_activation_daa"), confirmed on the Mac binary as the rehearsal found. The node lane's tests-only fix 791ff22c merges into the release node. MAIN'S RULING (19:0x UTC): publish 1 must not be a handshake split. The node lane adds to ca3-v4-order-fix a digest-compat change: an ABSENT optional field contributes nothing to the digest, so 0.3.15 on the thirteen-field file prints b18ed271... and peers with 0.3.14; the binary rolls out with no window; the sixteen-field file at publish 2 is the one sweep (miners first, the hand nodes and the seeds last, only when every node reports 0.3.15); the sixteen-field digest is re-read on the fixed binary before publish 2. The rehearsal's flip (about 19:01Z) and floor (about 19:21Z) are the PASS clock main holds the project lead to |
 | the project lead's go | the minute P1 passes |
 
 ## 7a. The two fast-forwards to master (decision 5)