Record Muninn trailing parity burst scar

This commit is contained in:
Meta Vens 2026-07-17 01:59:33 +01:00
commit 79781abdbb
2 changed files with 7 additions and 2 deletions

View file

@ -196,3 +196,4 @@
{"ts": "2026-07-16T21:38:23+01:00", "type": "blocker", "status": "pending", "note": "Field Moonlight acceptance cannot run while Raven is offline: Starfire's documented SSH/ICMP/TCP route to 10.77.0.4 times out, and Yggdrasil reports Raven's last WireGuard handshake at 2026-07-15T06:04:12Z versus Starfire's live handshake at 2026-07-16T20:36:32Z. No Raven task or host state was mutated."}
{"ts": "2026-07-17T01:37:22+01:00", "type": "verification", "status": "partial", "note": "Muninn video parity v3 uses 16 Cauchy GF(256) shards; native receiver burst tests 1..16 and Odin 180-test suite pass. Direct Raven LAN at 250 ms recovery deadline ran over one minute with no video assembly expiry or A/V failure; full 10-minute and impairment matrix remain open."}
{"ts": "2026-07-17T01:46:28+01:00", "type": "scar", "status": "fixed", "note": "Per-byte Gaussian elimination for production-sized video FEC can stall the single OBS receive owner and cascade socket loss. Receiver now inverts coefficients once per frame and uses table-driven GF(256); active Release assertions prove every 1..16 consecutive erasure case on 64x1400-byte shards under 50 ms."}
{"ts": "2026-07-17T01:59:33+01:00", "type": "scar", "status": "open", "note": "Direct Raven LAN falsified trailing-parity sufficiency: a 24-datagram sequence gap left a 56-chunk keyframe at 52/56 with parity_present=false because all 16 V3 parity shards followed the data and fell in the same tail burst. Next cut must disperse parity/adapt block layout; deadline increases and repair flooding are not owners."}

View file

@ -97,8 +97,12 @@ accepted_design:
recovery bound. Odin's 180 Muninn tests pass. A per-byte elimination draft
was rejected after one recovery stalled the receiver thread; production
inverts the coefficient matrix once per frame and uses table-driven field
arithmetic. The fresh direct Raven LAN session has no video assembly expiry
or A/V failure in its initial minute at a 250 ms recovery deadline.
arithmetic. Direct Raven LAN evidence then exposed a remaining layout fault:
a 24-datagram tail burst erased four final data chunks plus all sixteen
trailing parity shards, so no parity survived to repair the otherwise small
data loss. Repair retransmission is now capped behind fresh media and video
datagrams use a 50 microsecond spin pacer, but parity dispersion/adaptive
block sizing remains required before the clean ten-minute gate can pass.
Normal frames are forwarded immediately; the recovery deadline is not a
playout buffer. The ten-minute and impaired acceptance gates remain open.
- Muninn now closes receiver congestion pressure into the live NVENC context.