igneum-network/igneum
1
0
Fork 0
Code Issues Pull requests Releases Activity

release 0.3.22: route (A) staged and its pack-gate gap, the puller rule for 0.3.25, the node lane's three facts for the route decision

Browse source 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
igneum-labs 2026-10-08 00:48:42 +00:00
parent 44a0f4df3f
commit 6a4e6867b6
1 changed files with 2 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
docs/plans/release-0.3.22.md
View file

@ -180,3 +180,5 @@ The 0.3.23 node's heights move Devnet 3's digest to ba75bf6f, and the one-box-at
**dfbd1e10 fully gated:** the fast-time SUMMARY PASS cross-0324-dfbd1e10 at 01:12:57 BST (the shipped binaries; 12 of 12 program ids equal to the CPU verifier's, the stale node refused 95 of 95, the restart across the v5 boundary resynced in 36.2 s with its own mining held, four sinks equal), on top of every box gate green at 01:01:52; the kit's fingerprint 82b19cbde8557ea5 equal on four platforms (Metal, Apple OpenCL, CUDA, and the RX 9070 XT at 01:16:14 BST on PC 1, the v4 control PASS; Intel held with PC 2); class-v5 091a0758 (text arm and page rows, ids unchanged) landed at 01:40 as a post-freeze commit and is 0.3.25's with 8ca66afa; 0.3.24 pairs with 1c420786 as published. **dfbd1e10 fully gated:** the fast-time SUMMARY PASS cross-0324-dfbd1e10 at 01:12:57 BST (the shipped binaries; 12 of 12 program ids equal to the CPU verifier's, the stale node refused 95 of 95, the restart across the v5 boundary resynced in 36.2 s with its own mining held, four sinks equal), on top of every box gate green at 01:01:52; the kit's fingerprint 82b19cbde8557ea5 equal on four platforms (Metal, Apple OpenCL, CUDA, and the RX 9070 XT at 01:16:14 BST on PC 1, the v4 control PASS; Intel held with PC 2); class-v5 091a0758 (text arm and page rows, ids unchanged) landed at 01:40 as a post-freeze commit and is 0.3.25's with 8ca66afa; 0.3.24 pairs with 1c420786 as published.
**The clock line (01:41 BST):** the fleet published no dfbd1e10 move file: build-1's /fleet/move.json still names commit 2720d8d2 with the 22:30 BST minute; no FETCHED count, no named minute, no 0.3.24 DAA reader for the node lane, no answer on build-1's three old-object nodes, no 22:59 or 23:59 BST proven-share line (asks at 01:05, 01:16, 01:41; the fleet lane's last line was its 22:4x report). With the build-server lane dark since 00:17, the two lanes that could move the fleet are silent, and the two that remain cannot (the node lane holds no box keys for the pods; the shipper holds the Mac). So the 02:00 BST minute cannot hold and the 02:53 ceiling (DAA 32,400) stands only if a signed move file lands within minutes and the 34 pullers fetch inside forty. Main's word asked on: (a) waking or replacing the fleet lane tonight (the puller is self-contained: a new signed move file at build-1's /fleet/ naming the node-lane pair and a minute is all the boxes need); (b) otherwise a fourth re-cut of the floor from a morning minute main names, the 0.3.24 Mac entry standing down until then (rule 5: no app alone on b1ba7822); (c) the four words from 01:06 and 01:20 (the hive and the Windows node pair by the shipper; the Windows app chain's owner and hour; take 3's publisher; "PC 2 clear" on the dark lane's ground). The attack-pass interim reads at 01:55 regardless; PC 1's lock-free queue runs (the CA4 unlocked rows from 01:17 BST). The update-return lane's PC 2 S4U proof and PC 1 re-probe stand on take 3, which never ran. **The clock line (01:41 BST):** the fleet published no dfbd1e10 move file: build-1's /fleet/move.json still names commit 2720d8d2 with the 22:30 BST minute; no FETCHED count, no named minute, no 0.3.24 DAA reader for the node lane, no answer on build-1's three old-object nodes, no 22:59 or 23:59 BST proven-share line (asks at 01:05, 01:16, 01:41; the fleet lane's last line was its 22:4x report). With the build-server lane dark since 00:17, the two lanes that could move the fleet are silent, and the two that remain cannot (the node lane holds no box keys for the pods; the shipper holds the Mac). So the 02:00 BST minute cannot hold and the 02:53 ceiling (DAA 32,400) stands only if a signed move file lands within minutes and the 34 pullers fetch inside forty. Main's word asked on: (a) waking or replacing the fleet lane tonight (the puller is self-contained: a new signed move file at build-1's /fleet/ naming the node-lane pair and a minute is all the boxes need); (b) otherwise a fourth re-cut of the floor from a morning minute main names, the 0.3.24 Mac entry standing down until then (rule 5: no app alone on b1ba7822); (c) the four words from 01:06 and 01:20 (the hive and the Windows node pair by the shipper; the Windows app chain's owner and hour; take 3's publisher; "PC 2 clear" on the dark lane's ground). The attack-pass interim reads at 01:55 regardless; PC 1's lock-free queue runs (the CA4 unlocked rows from 01:17 BST). The update-return lane's PC 2 S4U proof and PC 1 re-probe stand on take 3, which never ran.
**Route (A) staged, not placed (01:50 BST):** the dfbd1e10 pair tarball on build-1 and served (fleet/dfbd1e10-node-lane.tgz, 27,495,110 B, sha256 e59ed0e6; igneumd 4870ccf2, igneum-miner aa8c2978); the move file (id mdfbd-1, commit dfbd1e10, want_version igneumd/2.1.0-dfbd1e10, want_digest b1ba7822, both pair slots on that tarball, at_epoch 0) written and signed with the fleet key in the shipper's scratch (r0324/move), the signature verified against the fleet's public key in the puller's namespace; nothing on /fleet/ changed, no box fetched. **The gap that stops (A) as the file alone:** every box's pack gate (box-dn3.sh's PAIR_MINER_SHA16 default list 07246920, c29f33bb, dfdc6883, fb147dd1) lacks aa8c2978 and the puller restarts box-dn3.sh from the box's running environment without reading the move file's miner sha into it, so a move by the file alone restarts every node on b1ba7822 with every miner refused as "not a paired miner" (the 22:31 shape on all 34 boxes at once, chain rate zero until a hand fixes the list); the fix before the minute is one line per box over ssh with the fleet's keys and tooling (on this Mac under ~/igneum-fleet; the Vast proxies limit reach), the fleet lane's work or, on main's word, the shipper's. **Rule for the puller (0.3.25 tooling):** the move file names the pair's miner sha and the puller carries it into PAIR_MINER_SHA16 for the restart, so a pair's miner is paired by the file that moved it, never by a list a hand keeps. **The node lane's three facts (01:5x BST):** a 0.3.24 reader node on build-1 (dfbd1e10's binary, loopback JSON RPC 28690, never mines) dials ten fleet nodes and follows the chain from the move, so the crossing read waits on no fleet reader; the fourth re-cut is one script run from a named minute (about 20 minutes to the pin plus 14 for the fast-time pair); a morning minute later than about 12:50 BST on 8 October puts the floor at or above 79,200 (the difficulty v3 height) and moves the three heights up with it in the same commit.