18–24% pipeline frame rejection rate on real May 2026 replays — exceeds shadow deployment stop condition

[reflectionsa]

Summary

In the three working May 2026 replays, 18–24% of pipeline frames are rejected by the frame validator after successful parsing. The parser accepts 0 frames as rejected, so this is a pipeline-level validation failure, not a format issue.

Data

File	Parsed frames	Pipeline processed	Pipeline invalid	Rejection rate
rec_2026-05-01_14-27-52.echoreplay	153,257	29,353	7,097	24.2%
rec_2026-05-01_14-58-25.echoreplay	66,726	11,121	2,079	18.7%
rec_2026-05-01_15-10-00.echoreplay	74,242	12,430	2,930	23.6%

Note: "Parsed frames" = player-frames (one per player per session snapshot). "Pipeline processed/invalid" = per-frame-index counts. The ratio difference is expected (multiple players per frame index).

Shadow deployment stop condition

The shadow deployment guide specifies a stop condition at >20% frame rejection rate. Two of the three files exceed this threshold.

Likely rejection reasons (from internal/pipeline/validator.go)

The validator rejects frames for:

• zero position for player %s — position is [0,0,0]
• NaN/Inf position for player %s
• dt %.4f out of range — DeltaTime outside [minDt*0.5, maxDt*2.5]
• position out of arena bounds — outside arena extents (X: ±12.5m, Y: ±12.5m, Z: ±82m)

The most likely cause is non-active game phases (round_start, score, pre_match, post_match) where players teleport or respawn at [0,0,0], triggering the zero-position check. The pipeline already skips detection during non-active phases but still runs validation, so these frames count as "invalid" rather than "skipped."

However: the shadow guide stop condition is based on the rejection rate being a sign of adapter breakage, not phase transitions. If the 24% is entirely from phase transitions, the stop condition threshold may need to be raised or the counting logic changed to exclude non-active phase frames from the invalid count.

Needed

1. Log the rejection reason per frame (currently the reason is returned as an error string but not counted by category)
2. Determine what fraction of invalid frames are zero-position vs out-of-bounds vs dt-out-of-range
3. If the majority are phase-transition zero-positions, adjust the stop condition or exclude them from the count