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

release-0.3.20 plan: the fleet's prover-roll read (12 GB provers paired and unpaid, the claim race, the fix on the node line)

Browse source 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
igneum-labs 2026-10-07 11:36:26 +00:00
parent 8e9da582c8
commit 8c59ebbe9a
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-0.3.20.md
View file

@ -343,3 +343,5 @@ Picked by hash onto ca1742cd, in order: miner-ui-5 810bf5a1 and b322e9fa; ui-ota
**Main (12:3x UK): hold for the isSynced fix.** The 0.3.20 app's execrpc gate and the worker-start rule both key on the node reporting synced; a flapping flag would hold workers back on the machines the cut is meant to make plug-and-play (PC 1's morning in a new form). The node lane carries the fix as a blocker on the node line with a test; the pin waits for it. If it slips past 16:00 UK, main hears the node lane's estimate and decides again. Everything else as set.
**cards_leave_off in (12:33 UK):** the Intel lane's 9e794503 → afb0cfe9 (jobrun.rs: a run job's --cards-off cards restored as OFF with identities and cap kept when the job's bool param cards_leave_off is set, the report line "cards LEFT OFF ... persisted by the app"; jobs.rs kinds doc; publish-jobs.sh --cards-leave-off emitting "cards_leave_off": true beside cards_off). App gate on the box, third pass: 214 + 28 + 8, rc=0, 11:32Z. Nothing else from the Intel lane for the cut. The reliability lane's hashes are app 38a30397 and fork f067f7c1, held until its pod injector finishes (about 45 minutes) and it sends the lines.
**The fleet (12:4x UK), prover roll: a 12 GB prover on the open devnet is paired and unpaid.** p2-4070-1 with the 0.3.17 pair: 7 segments "pair ok", 56 shards accepted, 0 refused, and 0 of 8 claims paid in 25 minutes (7 segment_refused "segment already paid; end to end 203 s", one "does not chain to ... which is proven"; 4 held, 3 held_expired). Its claims are fresh at claim time (margin 432 to 518 DAA, 5 to 7 candidates, rank_by fnv), so a 24 GB box claims the same segment and pays it inside the 4070's 203 s. p1-4070 paid once this morning (2.6 IGN after 176 s). Per tier: a home prover on a 12 GB card earns nothing while a 3090 or 4090 is awake on the same segments; the fix is the claim rule (a settled-depth claim with a per-key reservation, or the candidate hash spread over keys), which the node lane carries on the 0.3.20 line; until then a 4070 prover is a verifier that never pays, not a box fault. The roll continues with p1-3080 and hub-1; p2-4070-1 recorded "pair ok, unpaid (race)". Proving feed 11:35Z: provers_10m 4, 62 shards, lag 248 s. Asked of the node lane: confirm the claim-rule change is on the tip I pin.