diff --git a/docs/plans/counter-asic-3-status.md b/docs/plans/counter-asic-3-status.md
index 8b6f4486d..a3eaf4b41 100644
--- a/docs/plans/counter-asic-3-status.md
+++ b/docs/plans/counter-asic-3-status.md
@@ -764,6 +764,37 @@ THE NODE LANE'S 06:30 WORD (06:27 UK): the stale-sink IBD trigger does not ride
 
 THE GATE'S 06:30 DECISION (reset-decision.txt, read by the coordinator at 06:33 UK): at 06:25 the override line "nothing written; RESET HELD: canary=none, reference moving=no (1 distinct DAAs in the last ten minutes of the feed; feed 24852 0x8c46c8a8 46800 at 05:24:55Z); the reference stands at the epoch-13 cut since 05:45:43 UK"; at 06:31:41 "GATE READ 06:30 UK: islands=27 (06:00 28, baseline 29) | hive-2 canary=FAIL | mini=PASS (kit 2's node 5d53a591 on a fresh datadir synced from genesis in 3,183 s past the fifth cut) | override='none' | CONVERGENCE distinct_tips=27" and "DECISION: RESET HELD for the morning line: hive-2 canary=FAIL, mini=PASS (both must be PASS); the islands keep serving; the serving gap and the chain's state are the founder's morning item" (the gate's own text; the record's wording is main's: the hubs hold and serve, the class the hive run or environment). The kit-binary canary on lp-4090-43 had no verdict by 06:25 (syncing to the standing tip); the feed at 06:33 UK 24,852 / 8c46c8a8 / DAA 46,800, standing since 05:45:43. Nothing done to any box or lock; the island count falling one an hour as mining boxes drop, not merging. Main told at 06:3x with the 08:00 shape.
 
+MAIN'S WORD (06:4x UK): the 08:00 shape stands. One preparation, no hand on any box: the fleet lane writes and stages (not runs) the morning's reset recipe in the decision file's folder so the founder's "go" turns into action within minutes: (1) the three reference hubs lp-4090-04, -05, -06 restarted on fresh datadirs with the 5d53a591 kit's igneumd (or c363ce90 if its 07:45 read is green, the fleet lane's stated call), the island datadirs set aside, SEED the mini, lp-4090-43's canary node and build-1's hand nodes (the 2.0.2 state holders), under pid files; (2) when the three hubs read within ten blocks of the mini's tip, every other fleet box the same way with SEED the three hubs; (3) lp-4090-43 last; (4) the home machines by their own entries. The recipe names the expected clock from "go" to one tip (the mini's 53-minute genesis sync on an M6 the slow bound; the 4090 boxes 14 to 15 minutes), the island count before and after, and the rule that no hub is touched before the founder's word; its path named in the 08:00 line as the default action with its clock.
+THE MORNING'S RESET RECIPE STAGED, NOT RUN (the fleet lane, 06:5x UK): /Users/joshm/igneum-fleet/morning-reset.sh, sha256 63a92452f84296a1 (tools/fleet/ on gpu-fleet), its path and sha in reset-decision.txt; at its head it refuses unless /Users/joshm/igneum-fleet/founder-go.txt holds "go" (main's relay of the founder's word creates it); run without it it printed the recipe and "REFUSED: no founder-go.txt ... Nothing touched." With the go file: pid file pids/morning-reset.pid, every line into reset-decision.txt: the islands before; STAGE HUBS: lp-4090-04, -05, -06 on fresh datadirs (dn4 moved aside as dn4.aside-roll-<time>, never deleted), igneumd fbf1d6e5 and miner 66fded0f from the 5d53a591 kit (the fleet lane's call: the binary the mini and the kit canary ran tonight; c363ce90's 123301a5 only when kit-choice.txt holds "c363ce90" after the node lane's 07:45 read, the reason printed; the coordinator's default: the file absent unless c363ce90 reads green against a moving chain), SEED the mini (its host:port from mini-p2p.txt, the shipper's line, missing at the staging and said so), lp-4090-43's canary node 47.47.180.12:22015 and build-1's 26631/26671; a 40-minute wait until the three hubs read synced=true on one tip (and within ten of the mini's tip if build-1:/srv/canary/mini-tip.txt exists); then with SEED the three hubs: DIAL (dn3-g1, lp-4090-01), HUBS2 (lp-4090-02/-03/-07/-08), MINERS (24 plus lp-4090-11, in tens), NODES (40, in tens), OUTAGE (23 plus dn2-2, in tens), LAST (lp-4090-43), the island count after each stage, RESET END, then reset-track.sh for the time every box reaches within ten; hub-1's nodes and the home machines not in it (the mini by its app, PC 1 and PC 2 by the Windows entry). The clock from go: the hubs about 15 minutes (a 4090's fresh sync 14 to 15 minutes tonight), the fleet about 35 more, one tip among the fleet about 55 minutes after go; the mini's 53-minute M6 genesis sync the slow bound on the home side. Its two inputs: the shipper's mini-p2p.txt (the mini's p2p host:port if reachable from the fleet, or NOT REACHABLE with the reason) and the node lane's kit-choice.txt. Kit 6's heal run on lp-4090-11 from 06:42 UK; the kit canary's bound on lp-4090-43 at 06:46 UK, then the hive-3 chain and the box-dn3 read.
+
+THE KIT-BINARY CANARY FAIL AT ITS BOUND, HIVE CANARY 3, KIT 6 (the fleet lane, 06:4x UK): the kit-binary canary on lp-4090-43 ended FAIL at 02-sync at 06:46:36 UK against a standing reference: NOT SYNCED after 5,404 s at blocks 21,756 / headers 25,197; it passed 3,687 at 05:19:47 but walled from 06:02 at lp-04's epoch-7 cut d6a925b1 ("of the chain to execute has no body on this node; the IBD's body sync brings it") and its last IBD from lp-06 ended on the lost-proof class ("the proofs of 6 carried records ..."); block sha 4f4cfbb872d0a612 under build-1:/srv/canary/1176efb9/lp-4090-43/. So no fresh 2.0.2 node on a fleet box reached lp-04's chain tonight from the hubs alone: the mini did (with build-1's hands holding the early bands, 53 minutes); the kit node got the early bands from the hands too and then walled at a later cut the hands do not hold (their DAAs 10,799 and 14,474). Hive canary 3 started 06:46:54 UK on the same box (cf939b77, the seven peers), verdict by about 08:00, the same label; the box-dn3 read after it. Kit 6 (bb8c0679) on lp-4090-11's walled datadir from 06:43:55 UK: fix 5's genesis capture of 3c3fab43 repeated (1,937 ms), then a new cut 1a87e870 "not captured (chain block ... of the chain to execute has no body on this node; the IBD's body sync brings it)" every 10 s, the staged-capture line for dac3dd04 not yet printed; its bound 07:44. The 06:30 decision stands (RESET HELD, 27 islands).
+
+THE RECIPE'S TWO INPUTS AND THE STAND'S NATURE (the shipper, read back 06:47 UK): mini-p2p.txt: "NOT REACHABLE: the mini's node listens on 0.0.0.0:26611 (LAN 192.168.68.70, public 2.121.2.18) behind the home NAT with no inbound map; a connect from build-1 to 2.121.2.18:26611 at 06:47 UK timed out; no port opened by hand. The mini dials out only"; so the hubs seed from lp-4090-43's node and build-1's hands in the recipe's default. build-1:/srv/canary/mini-tip.txt in the feed's shape, written every ten seconds by ~/mini-tip.sh on the mini under ~/mini-tip.pid, first line "24852 0x8c46c8a8 46800 2026-10-09T05:47:47Z", equal to lp-04's standing tip. The mini's log after 05:24:37: no PoW or rule refusal of any block, no new block at all; the only lines "PoW accepted 8c46c8a8 by igneum-lottery-v2-bound (daa 46799)" on every re-advertisement of lp-04's cut block (every 20 s), so the stand is no block at all after the epoch-13 cut, not a joiner-side refusal; the mini accepts the cut block and waits on it. The coordinator's consequence (06:5x UK, to the fleet lane and the node lane): the recipe's default (the hubs seeded from lp-4090-43's node at blocks 21,756 and the hands at DAAs 10,799 and 14,474) abandons lp-04's chain from about 21,756 to 46,800 including the mini's 252 blocks, a rollback (MODE=rollback, named so); a second variant staged beside it, MODE=continue: lp-04 not reset, kept up as the seed of its own chain to 46,800; lp-05 and lp-06 reset on fresh datadirs with the 5d53a591 kit, SEED lp-04 plus the hands plus lp-4090-43's node; when they read lp-04's tip synced they mine from it as 2.0.2 hubs (the mini showed a 2.0.2 node accepts that block), lp-04 reset last onto the continued chain; then the fleet with SEED the three hubs; A's risk: a fresh 2.0.2 node has not reached 46,800 from the hubs alone tonight (only the mini with the hands did), so A's first stage is itself the test, about 90 minutes for a 4090 through thirteen cuts if nothing walls; both behind the founder's go file ("go continue" or "go rollback"); the node lane asked by 07:15 whether a 2.0.2 node holding the chain to 46,800 can mine the next epoch from it (the seed derivable from its own executed state) and why a 2.0.0 hub cannot pass its own cut.
+
+THE RECIPE WITH BOTH MODES (the fleet lane, 06:5x UK; sha read by the coordinator): /Users/joshm/igneum-fleet/morning-reset.sh sha256 7f244e58b710c2d6 (tools/fleet on gpu-fleet, on the box remote), nothing of it running; founder-go.txt must hold exactly "go continue" or "go rollback" (main's relay of the founder's word); anything else or no file prints the two modes with what survives and what is lost and "REFUSED ... Nothing touched" (tested both ways). MODE=continue (A): lp-4090-04 not reset, the seed of its own chain to 46,800; lp-05 and lp-06 fresh (islands aside) on the kit's igneumd with SEED lp-04, build-1's hands, lp-4090-43's node; up to 100 minutes for both to read lp-04's tip 8c46c8a8 synced (a wall the recipe's STOP line and exit, nothing further touched), then up to 20 minutes for lp-05's DAA to pass 46,800 (the 2.0.2 hubs mining on from the cut block), then lp-04 reset last onto the continued chain seeded from lp-05/06, then the fleet with SEED the three hubs, lp-4090-43 last, the tracker; what survives: lp-04's chain to 46,800 with the mini's 252 blocks. MODE=rollback (B): the three hubs fresh from lp-4090-43's node (blocks 21,756) and the hands, the print "ROLLBACK ... lp-04's chain above about 21,756 abandoned, the mini's blocks included", then the fleet the same way. Both: no mini in any SEED (home NAT); the mini's tip feed at build-1:/srv/canary/mini-tip.txt; the kit fbf1d6e5 unless kit-choice.txt holds c363ce90 on the node lane's green read; every line into reset-decision.txt; stop by pids/morning-reset.pid only; hub-1 and the home machines outside it. Clocks from go: A about 90 minutes to the continued chain plus about 35 for the fleet; B about 15 for the hubs plus about 35 for the fleet.
+
+THE NODE LANE'S 07:00 LINE (06:56 UK): THE STAND RECONCILED: the invalid-PoW-under-a-diverged-seed class struck by the joiner reads: the kit canary (2.0.2) never saw 8c46c8a8; the mini (2.0.2) accepted it and refused nothing; the only refusals are four 2.0.1 fleet boxes (dn3-q05, dn3-x1, lc-3090-01, lp-3090-06) whose IBD from lp-04 timed out on the class v5 state wait and, on 2.0.1, marked lp-04's relay block refused and banned it under N6, a 2.0.1 class fixed in 2.0.2 (rule_refusal_marks_relay excludes the state wait). The stand is "no block after the cut": 8c46c8a8 (DAA 46,799, mined by dn3-q03 at 05:26:19) is the last block of epoch 12 on lp-04's chain; the first block of epoch 13 needs the execution state after the epoch's seed block (the chain block at DAA 46,200, the cut less the 600-block lead), and lp-04 (2.0.0) never reached it: its executor runs the pool-read state at proof-band speed under 80 peers, was 300 s short at 05:45, died on the fd class at 05:49, and since its 05:50 restart reads synced=false, "Processed 0 blocks"; a 2.0.0 hub cannot cross its own class v5 cut (none of the per-cut body sync e189afb4, the heaviest anchor, the foreign-seed capture or the store, its executor not reaching the seed state on time); it relays nothing past the cut, so the mini and every 2.0.2 joiner on its chain stand with it, the certificate eae27c3b a symptom. A 2.0.2 node holding the chain to 46,800 with its executor at or past 46,200 CAN mine epoch 13 (the seed derives from its own executed state): dn3-q03 did exactly that on 2.0.1 ("daa 49,549, epoch seed 284fb765", 23,480 blocks by its miner, mining on, peered to lp-04 which cannot validate its blocks): the chain's live continuation exists on dn3-q03 alone. Variant A sound on one condition: every miner of epoch 13 must hold the same executed state at the seed block 46,200 (the dataset derives from it); 2.0.0/2.0.1 and 2.0.2 states have differed before (the 52.74 IGN payout), so a 2.0.1 dn3-q03 and a 2.0.2 hub could split the chain at the cut into two seeds; the deciding read: eth_getBlockByNumber(0xb478 = 46,200).stateRoot on the mini (2.0.2, the shipper) and on dn3-q03 (2.0.1, the fleet lane) by 07:30: equal, A continues with both and dn3-q03's epoch-13 blocks are the chain; unequal, A is 2.0.2+ nodes only, dn3-q03 retired or upgraded, its post-cut blocks not the chain; the recipe to carry the branch on a roots file (the coordinator's staging ask). THE FD LEAK's path by shape: 9,808 inbound connects per 15 minutes (every 2.0.1 box re-dialling the hub it banned, every joiner's IBD retry every 30 s), 4,106 ban lines, the node pid's socket fds growing about 2 a minute: sockets accepted and closed by a reject whose fd the p2p server never releases; CLOSE_WAIT growing = the server side's stream dropped without the socket closed on a handshake/relay reject (protocol/p2p's connection handler, the reject-and-disconnect path); the 2.0.3 item that close, 13886a30's raised limit a delay only; the file and the clock in the 07:45 line. KIT-CHOICE: kit-choice.txt stays absent: c363ce90's fresh read on build-1 at 06:35 to 06:43 took a dead island (STALLED at 8,144/14,474) as dafc823f's and kit 6's did; against a standing reference no read is green; the hubs come back on 5d53a591 with c363ce90 the morning's upgrade once a moving chain exists (six suites, the record store and the fd limit in it, the fd CLOSE not). Kits: kit 5's kept read NOT HEALED (standing at 35,997 from 06:10; its node wedged at stop and went down by its pid file, the wedged-stop class the 2.0.3 rules bound and kit 5 lacks); kit 6 on lp-4090-11 from 06:43:55 (fix 5's capture again, then an island cut's body sync pending; the dac3dd04 staged line awaited by 07:44); kit 6's kept-datadir read with the relay flow at debug on build-1 from 06:41 (8,777 debug lines by 06:49) to name the stale-sink's silent relay by 07:45.
+
+THE ROOTS AT THE EPOCH-13 SEED BLOCK AND THE RECIPE'S BRANCH (the fleet lane, 07:02 UK): the seed block (DAA 46,200) by its hash 284fb765e18c... from dn3-q03's log: on dn3-q03 eth number 24,426, stateRoot 0x5344ff6d...; on lp-4090-04 eth number 24,425, stateRoot 0x98555df2...; UNEQUAL, the same hash at different chain heights meaning the two executed states parted before the seed block (2.0.1 against 2.0.0; the mini's 2.0.2 root asked of the shipper by 07:30 as the line that decides A1 against A2); the file roots-46200.txt on the Mac and build-1:/srv/canary/1176efb9/. The recipe now carries the branch (morning-reset.sh sha256 ead9dc1a959a8431, superseding 7f244e58, on the box remote, still refusing without the founder's exact words): under "go continue" it reads roots-46200.txt; EQUAL = A1 (dn3-q03's address joins lp-05/06's SEED, the wait "synced to dn3-q03's tip"); anything else = A2 (2.0.2+ nodes only, dn3-q03 retired, the wait lp-04's tip 8c46c8a8 synced then the chain moving past 46,800 from lp-05/06's own mining), the branch printed into the decision file. The socket states to the node lane verbatim: ESTAB growing about one a minute, no CLOSE_WAIT on the pid, 9,790 inbound connects per 15 minutes.
+
+THREE STATE ROOTS AT ONE BLOCK (the shipper's read on the mini's EVM RPC at 06:59 UK by hash): eth_getBlockByHash 0x284fb765e18c9420222ec22f4d0173a35f1b03e563871550a64d2806dd67d79c returns number 24,425, stateRoot 0x3fde4beb339ab818b0b9fd3da0994d179500bea0144b649a0e90861b690ea3a8, written as "mini 24425 0x3fde4beb..." into roots-46200.txt beside the fleet lane's two; so the one block hash carries three different state roots: dn3-q03 (2.0.1) 0x5344ff6d at eth 24,426, lp-4090-04 (2.0.0) 0x98555df2 at eth 24,425, the mini (2.0.2, kit 2) 0x3fde4beb at eth 24,425; the mini's executor at the cut block (igneum_getExecStatus: executedTip 24,852, executedTipDaa 46,799, executedTipHash 8c46c8a8, synced on the standing tip), so not below the seed block. A1 out (the mini unequal to dn3-q03); A2 "2.0.2 nodes only" at best; the node lane asked by 07:45 UK for the root of that block on every 5d53a591-or-later node that executed past it (kit 4's fresh node, the build-2 hubs-only node, lp-4090-11's and lp-4090-43's kit nodes) against the mini's: equal among 2.0.2 nodes means the executor is deterministic per version and differs across versions (the rule change between 2.0.0, 2.0.1 and 2.0.2 to be named: verified_by_acceptance and the payout rule the candidates), A2 standing and every mixed-version island a chain of its own by construction; unequal among 2.0.2 nodes means the executor is not deterministic within one version, the first fact of the 08:00 line after the heal, stopping every reset until named.
+
+THE EXECUTOR IS DETERMINISTIC WITHIN A VERSION AND DIFFERS ACROSS VERSIONS (the node lane, 07:04 UK; the table in roots-46200.txt): the epoch-13 seed block 0x284fb765e18c9420...: dn3-q03 (2.0.1) eth 24,426, 0x5344ff6d; lp-4090-04 (2.0.0) eth 24,425, 0x98555df2; the mini (5d53a591, 2.0.2) eth 24,425, 0x3fde4beb339ab818b0b9fd3da0994d179500bea0144b649a0e90861b690ea3a8, parent 0x3b5cfa3f; build-1's kit-4 kept-datadir node (9bdf7120 = the 2.0.2 source plus fixes 3 and 4; it executed lp-04's chain to DAA 46,253 at 05:47, restarted peerless at 07:02 for the read and stopped by its pid file): eth 24,425, stateRoot 0x3fde4beb339ab818..., parent 0x3b5cfa3f, EQUAL to the mini's: two hosts, two kits of the 2.0.2 line, byte-identical root and parent. Verdict: three roots for one block across 2.0.0, 2.0.1 and 2.0.2, one root within 2.0.2; A2 stands (a 2.0.2+ hub seeded from lp-04's chain continues it; dn3-q03 retired from the seed set, its epoch-13 blocks not the chain); every mixed-version island a chain of its own by construction. The cause between versions by the exec changes shipped: 2.0.1 to 2.0.2 the payout rule (verified_by_acceptance: a proof record paid on the acceptance verdict, not the pool read, the 52.74 IGN lp-04-vs-lp-06 class; dn3-q03's extra chain block below the seed block, eth 24,426 against 24,425, its chain merged differently, a record a 2.0.1 node paid that 2.0.2 does not); 2.0.0 to 2.0.1 the 7cfa422a roll's exec change (the shipper's changelog names it). A third 2.0.2 point syncing on build-2 (a fresh 5d53a591 node with lp-04 as its only peer from 07:01), its root appended by 07:45 if lp-04 serves its bodies through the epoch-7 cut, a wall there itself a fact. THE FD LEAK's kind, from the pid-accurate lines (06:10/06:25/06:40 UK: fds 168, 185, 199; sockets 103, 120, 134; ESTAB 96, 113, 127; CLOSE_WAIT 0; 9,788 inbound connects per 15 minutes; ban lines constant): ESTAB sockets on the node's own pid growing about one a minute with the reported peer count far below them, connections the node accepted and keeps open after the peer left, not a half-closed leak; the 2.0.3 fix a server-side idle/handshake timeout closing an accepted connection whose peer never completes the handshake or stops answering (the p2p connection handler's accept path in protocol/p2p; the ping flow's 120 s timeout covers only registered peers); the file and clock in the 07:45 line. kit-choice.txt absent.
+
+THE RECIPE SETTLED ON A2 (the fleet lane, 07:1x UK; the sha read equal by the coordinator): /Users/joshm/igneum-fleet/morning-reset.sh sha256 d2a8fd5566705736 (tools/fleet on gpu-fleet, on the box remote), refusing without "go continue" or "go rollback" (tested): A1 withdrawn 07:04 UK as refused text with its reason (the roots differ: 2.0.0 0x98555df2, 2.0.1 0x5344ff6d, 2.0.2 0x3fde4beb on the mini and build-1's kit-4 node; dn3-q03 is 2.0.1, its epoch-13 blocks not the chain); "go continue" is A2 only: lp-05/06 fresh on the 5d53a591 kit with SEED lp-04, build-1's hands, lp-4090-43's node; the wait lp-04's tip 8c46c8a8 synced on both, then lp-05's DAA past 46,800 from their own mining; lp-04 last onto the continued chain; then the fleet with SEED the three hubs, lp-4090-43 last. The rule at its head and in its print: one igneumd sha on every reset box (fbf1d6e5 unless one named kit for all); every 2.0.0 or 2.0.1 box left running an island by construction; dn3-q03, dn3-q05, dn3-x1, lc-3090-01 and lp-3090-06 verified present in the MINERS/NODES/OUTAGE lists; any box dn4-roll.py cannot reach named by its error line in the decision file. One more root for the table: kit 6 on lp-4090-11's island datadir published the epoch-9 cut's state at 07:04:10 UK with root 0x0254175e3f36..., byte-equal to lp-04's and lp-06's root for that cut, so a 2.0.2 executor reproduces the 2.0.0 state at epoch 9 and the cross-version split sits after it (at or after b0cd26dc, where lp-04 and lp-06 already differ from each other); the node lane has it.
+
+THE NODE LANE'S 07:45 READ (sent 07:16 UK). THE 2.0.3 READ, c363ce90 (release-2.0.3-node-k6 = 7153ef9a [bb8c0679 fix 6 + the exec 2.0.3 rules 44b26e03/c893a699/47b7e38d rebased] + the record store c90212e3 + finality a349ae7b + the fd limit 13886a30 + the bump; "igneumd 2.0.3" in the seed-class kit 8dd1497d, igneumd 123301a5): six suites green on build-8 (p2p 42, miner 33, exec 102, core 186, consensus 146, pow 19), pair placed, canary set 06:00; the fresh read on build-1 STALLED 06:43 at 8,144 / DAA 14,474 (the first island its dial list served), the kept-datadir read NOT HEALED 07:13 at 17,013 / 35,717 (dac3dd04's island end), both against a standing reference; both nodes exited cleanly on TERM (the 2.0.3 stop bound; kit 5's and kit 6's 2.0.2-line nodes wedged at stop tonight and went down by pid file with KILL; c893a699 in the 2.0.3 rules only, a reason for 2.0.3 on the hubs once a chain moves). Every fresh and kept read since 05:45 (dafc823f, kit 5, kit 6, c363ce90 twice) synced a dead island and stood: with lp-04 standing and lp-05/lp-06 on islands, a joiner takes whichever island answers first. kit-choice.txt absent; the hubs come back on 5d53a591's kit, c363ce90 the upgrade once a moving chain exists (its fresh-from-genesis read then the gate). THE VETO READ: the mini at the cut 0 vetoes, 0 refusals after 05:24:37; the mini's 03:26 class (lp-04's epoch-11 header hashed under the mini's own capture of b0cd26dc through 151 unwinds, 74 strikes, the reference banned) is fix 4's (9bdf7120, in kits 4/5/6 and c363ce90: re-derive before any strike) and did not recur. THE lp-04/lp-06 STATE-ROOT CALL: 2.0.0's state is NOT deterministic across nodes of one version (lp-04 0x99ab4fed against lp-06 0x959e3698 at the same cut block b0cd26dc, the 52.74 IGN pool-read-vs-verdict payout) while 2.0.2's IS (the mini and build-1's kit-4 node byte-equal at 284fb765) and reproduces 2.0.0's state up to epoch 9 (kit 6's scratch root at 9b980b2f 0x0254175e equal to lp-04's and lp-06's published root there); the divergence enters between 9b980b2f and b0cd26dc, the class the pool-read payout closed by verified_by_acceptance (2.0.2): 2.0.0 hubs cannot be references for any 2.0.2+ fleet past a cut where a pool-read payout differed; A2 (2.0.2+ only) the only continue. THE STALE-SINK: kit 6's debug kept read (the relay flow at debug, 18,797 lines) stood 07:00 to 07:11 with zero relay-inv or admit lines (only proving-relay records refused as "not chain block N on this chain" and pings): the read VOID for the silent relay, no new block existing anywhere to relay; repeated against a moving chain in the morning. In its place one new fact from build-2 (a fresh 5d53a591 with --connect to lp-04 as its only peer, 07:01 to 07:13): it connected and synced NOTHING (tip 0 for twelve minutes): Kaspa's IBD starts only from a relayed block with missing parents, so a fresh joiner whose only peers stand never syncs their 46,800 blocks; the 2.0.3 flows item (an IBD trigger from the peer's announced sink at handshake and on the N4 stall), the morning's first flows change, with its known-failed test first. THE FD FIX'S FILE: ESTAB sockets on lp-04's node pid above its peer count (96, 113, 127 over 30 minutes, CLOSE_WAIT 0, 9,790 inbound connects per 15 minutes under the 2.0.1 boxes' ban-and-redial churn): routers torn down on a reject (the flow error path) without the tonic transport closed; protocol/p2p/src/core/router.rs's close path and connection_handler.rs's accept (serve) path; the handshake's 30 s dequeue timeout in common.rs; the fix the transport close on router teardown plus an idle bound on accepted connections, a known-failed test on a stream left open after a reject, the morning's second p2p change, 13886a30's limit the delay meanwhile. KIT 6 on lp-4090-11: fix 5's capture (3c3fab43, 1,937 ms) and fix 6's staged capture of dac3dd04 (7,200 then 2,411 blocks, 27.6 s) where kits 2/3/5 refused for ever; 9b980b2f captured in two stages (37 s) with the hubs' root; headers 32,398 to 35,997; b0cd26dc next, bound 07:44; the HEAL lane's keyed scratch set (four held scratches by tip hash, one round per later cut on a two-island node) after kit 6. THE KITS TABLE: kit 1 490ebcfe (green, superseded); kit 2 5d53a591 (the fleet kit; the mini's fresh sync in 3,183 s; the hive walls on lp-4090-43 under h-run only, the box-dn3 read about 08:05 deciding); kit 3 09350682 (six green, fresh PASS 05:07, kept: a dead island); kit 4 9bdf7120 (six green, fresh PASS 05:16:57, kept: a dead island; the 2.0.2 root point); kit 5 685d72dd (six green; fix 5 confirmed on lp-4090-11; kept: a dead island; wedged at stop); kit 6 bb8c0679 (six green: p2p 40, miner 33, exec 95, core 185, consensus 143, pow 19; the seed-class kit 80d2246b, igneumd b8c272df; fixes 5 and 6 confirmed on the real island); 2.0.3: dafc823f (superseded) and c363ce90. Residue for the morning: the stale-sink relay read against a moving chain, the IBD-from-sink trigger, the transport close, the over-label at ingest, P22 f1f0fb77 not yet on the k6 base, the third 2.0.2 root point (needs a serving peer). THE COORDINATOR'S CONSEQUENCE (07:2x UK): A2 as staged (lp-05/06 fresh with SEED lp-04 plus the hands plus lp-4090-43's node) cannot reach lp-04's 46,800 from standing peers; the continue mechanism re-staged on a snapshot of the mini's datadir (the only 2.0.2-executed copy of lp-04's chain to 46,800), the shipper staging the snapshot script on the mini (the node stopped by its pid on the founder's go, the tarball with its sha to build-1:/srv/artefacts/mini-dn4-snapshot/) and the fleet lane the recipe's continue stage from it (lp-05/06 then lp-04 from the snapshot, then the fleet fresh from the three hubs); continue refused without the snapshot on build-1.
+
+THE RECIPE'S CONTINUE STAGE ON THE MINI SNAPSHOT (the fleet lane, 07:2x UK; the sha read equal by the coordinator): /Users/joshm/igneum-fleet/morning-reset.sh sha256 fb746667c2d28ef6 (tools/fleet on gpu-fleet, on the box remote), refusing without the exact words. MODE=continue: the newest build-1:/srv/artefacts/mini-dn4-snapshot/*.tgz with its .sha256 sidecar; none = "continue needs the mini snapshot; rollback is the only mode now; REFUSED, nothing touched" (exit 5); found = its name, sha and file time printed with the expectation that the mini's node was stopped by its pid for it, pulled to the Mac, refused on a sha mismatch; then STAGE SNAPSHOT lp-4090-05 and lp-4090-06 one after the other (the kit's igneumd and miner to the box, stop by pid files with the LOCK released, dn4 aside as dn4.aside-roll-<time>, the snapshot unpacked as dn4 with its meta/LOCK removed, NODE_BIN the kit's, MINE=1, SEED the other hub + lp-04 + build-1's hands, start through box-dn3.sh); a 30-minute wait for both to read synced=true at or past 46,800 (STOP and exit 4 on a wall, nothing further touched); a 20-minute wait for lp-05's DAA past 46,800; then lp-04 the same from the same snapshot with SEED lp-05 + lp-06 + the hands; then the fleet fresh with SEED the three hubs (DIAL, HUBS2, MINERS, NODES, OUTAGE with every 2.0.1 miner, lp-4090-43 last), the island count after each stage, the tracker; rollback unchanged; the snapshot the shipper's step under the founder's go; the recipe does nothing to the mini.
+
+THE MINI'S SNAPSHOT STAGED, NOT RUN (the shipper, read back 07:20 UK): ~/mini-snapshot.sh on the mini (sha256 63f6b7f78bc35c00), gated on ~/mini-snapshot.go which does not exist and is written only on the founder's go relayed through the coordinator; without it the script prints the plan and exits 2 ("no go file: staged only; nothing stops, nothing is tarred"). On the go, under ~/mini-snapshot.pid: api/quit bounded, the window host ended by its recorded pid, a refusal if the node is still up (a RocksDB datadir consistent only with the node stopped; the stop the founder's morning step), the LIVE devnet-4 datadir tarred (the island aside untouched) with sha256 and size, the app relaunched first (the three hubs env, the node resuming on its datadir), the tarball with its sidecar pushed to build-1:/srv/artefacts/mini-dn4-snapshot/<stamp>.tgz over the mini's build-1 key and read back there by sha256sum -c, every step clocked. The numbers: the datadir 7.5 GB on disk (one RocksDB tree igneum-devnet-4/); the mini's upload to build-1 7.7 MB/s on a 24 MB test at 07:20; the mini without its node for the tar only (about 3 to 5 minutes on the M6), the push after the relaunch about 11 to 17 minutes, the whole step about 20 minutes from the go, the mini's mining lost for the tar's minutes plus seconds of resync.
+
+THE RELAY LANE'S 07:45 LINE (read 07:41 UK from the intake and the PCs' agents): PC 1 (ae432dc7): accepted blocks since the devnet-4 genesis 0; first mined on 2.0.x never, still Igneum Miner 0.3.26 (the app restarted at 04:45 at its runner's cap) with mining paused and its node on Devnet 3 with no peers, 0.00 MH/s; the cause: the hash lane's jobs held its runner from 18:5x yesterday to 05:1x (the same-work OpenCL cells on the RX 7600), so the queued 2.0.1 update-now never reached the engine; expected: it installs the morning's Windows entry the minute its runner is free (or the founder's one click, Settings, Update now), MINING-ON within 15 minutes of the apply; PC 1's relay agent still down (the founder's bat step with the fixed zip owed). PC 2 (1ccfe586): accepted blocks since genesis 0; first mined on 2.0.x never; Igneum Miner 2.0.1 (applied by itself at 21:04:50 yesterday, restarted on an empty datadir at 21:28 as the fleet's probe); at 07:41 its node "behind", 2 peers, no new block for 44,899 s, every card waiting at 0 MH/s; the cause: a fresh 2.0.1 node walls at the class v5 epoch's wall (IBD to 3,598 then behind), the same read as the fleet's peerless boxes; no 2.0.2 Setup reached PC 2 (the Windows order's default on the 04:04 canary FAIL); expected: PC 2 mines again on the morning's Windows entry after the founder's decision on the chain; the Setup build, the rule-33 canary and the MINING-ON read armed in that order on the shipper's dl values.
+
+THE SHIPPER'S 07:45 MINI LINE (read 07:45:51 UK): blocks since the devnet-4 genesis 662 accepted by the mini's node across its 2.0.1 and 2.0.2 runs (29 in the 2.0.1 run from 00:34, 28 in the 00:35 2.0.2 run, 16 and 18 in the two 2.0.2 canaries at 01:10 and 02:14 on hub-1's then-chain, 325 on its island seed 9b980b2f from 02:23 to 03:53 (that datadir set aside), 252 on the reference chain from 05:01:36 to 05:24:37 on the fresh datadir); first 2.0.x mining 00:34 UK (2.0.1); first 2.0.2 mining 01:10 (kit 1's canary), on kit 2 since 02:14; the rate now 0.00 MH/s, the card waiting: the node "behind" on the standing reference at DAA 46,800 (lp-04's cut block 8c46c8a8, unmoved since 05:45:43), 9 peers, the last accepted block 05:24:37; no PoW or rule refusal after it; the chain the reference (lp-04's), seed b98d7c3e at the cut; its executor at the cut (executedTipDaa 46,799) with the 2.0.2 state root for the epoch-13 seed block (0x3fde4beb). What it means: the founder's home machine has mined nothing for 2 h 21 min and will not until the chain moves; it holds the only 2.0.2-executed copy of lp-04's chain to 46,800, which the morning's continue needs. Version 2.0.2 (kit 2, igneumd 5d53a591) on the live Mac entry; no hand on the mini overnight beyond the two named steps (the 02:23 host fix and the 04:08 reset).
+
+THE FLEET LANE'S 07:45 LINE (written 07:45:56 UK into /Users/joshm/igneum-fleet/morning-line.log with every decision-file line): THE CHAIN: the reference lp-4090-04's chain since 03:48 UK, standing at DAA 46,800 (its epoch-13 cut, tip 8c46c8a8, EVM 24,852) since 05:45:43, the feed unchanged at 07:45; lp-04 (2.0.0) cannot cross its own cut; its tip block mined by dn3-q03 (2.0.1), which crossed the cut on its own seed and mines at DAA 51,947; the three reference hubs hold no connection to each other (lp-05 at 30,998, lp-06 at 39,601 since the night); hub-1's own devnet-4 node on its own chain (48,275, 743 locks since the roll, the last at 07:47:16 UK checkpoint 1,839, one key 0x53fe9802 on its last 300 blocks); the fleet 26 distinct tips among 31 mining boxes at 07:45 (29 at the 04:10 baseline, 28 at 06:00, 27 at 06:30), every seed group one box, 60 tips across 98 nodes; no fresh 2.0.2 node on a fleet box reached lp-04's chain. THE GATE: RESET HELD at 06:30; the staged recipe morning-reset.sh sha fb746667c2d28ef6 refusing without founder-go.txt holding "go continue" (A2 on the mini's datadir snapshot) or "go rollback". THE ROOTS at the epoch-13 seed block 284fb765 as recorded; kit 6 reproduced the 2.0.0 hubs' root at epoch 9, the version split after epoch 9. THE FIXES on the real island (lp-4090-11, the kept datadir): kit 3 NOT HEALED (the walk bound), kit 5 captured the below-state cut from genesis (fix 5 confirmed), kit 6 captured the fifth and sixth reference cuts in staged rounds (fix 6 confirmed) and ended NOT HEALED at its bound on the body sync from standing peers; the node left running; lp-4090-11 free for the node lane's next kit. CANARIES: the kit-2 fleet canary FAIL (the walk bound), hive-1 FAIL (lost proofs, the three hubs), hive-2 FAIL (the hive build with the mini's peer set), the kit-binary canary on lp-4090-43 FAIL at 02-sync against the standing reference, hive canary 3 running (verdict about 08:00), then the box-dn3 read. THE FLEET LANE'S FAULTS tonight, each once: the canary sync bound, the hive config path, the in-place edit of a running harness, the cap guard's restart of lp-04 at 05:02, the 3090's verify rule; the seed guard's restart of lp-04 at 05:50 after its fd-class death (the leak: 256 fds, 184 ESTAB at 07:40, about one socket a minute, 9,800 inbound connects per 15 minutes, CLOSE_WAIT 0). SPEND: the UK day USD 423.22 of 3,000 (14 percent) at 07:02 UK, 115 instances at USD 60.15 an hour; 114 registry rows after the 05:27 reconcile; hub-1's disk 5.2 GB free and rising (6,200 of 13,038 aside files compressed, the aside kept per main). NOT DONE tonight: the two-hour declared-load hold on 2.0.2 and the F07 rows on 2.0.2 modes (no 2.0.2 fleet); every post-20:00 UK row labelled "on an island, not a network".
+
 A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's |
 | Main's rulings (7 October, morning) | no generator change to v4 on the live devnet; the record's null is the window model with numbers, sent by the hash lane to the attack-pass lane so AP-F8-1 re-gates against it; a fault beyond the model (a low-entropy source at site 15) stops at the coordinator with the two options priced (a 0.3.19 class amendment before the flip, or the flip held at the floor), nothing shipping without the founder's word; the tighter tail, an acceptance bound on the hot-set share, is a CLASS V5 item (sent to the v5 lane a6410f3b8abefb762 with the 64-seed census as its gate; the bound's number follows from the model) |