← Lighthouse Briefings · Issue shooting · Bad-call spike · 2026-05-17

● ACTIVE INVESTIGATION audio freeze + long connecting-fail iOS + Android 2026-05-17 (Sun) 12 countries REAL DATA

Bad-call rate spike — cross-platform — 2026-05-17

Single Sunday event · all 12 monitored corridors moved · driven by mid-call audio freeze and long connecting-fail (>2s)

Spike date

2026-05-17 (Sun)

Rate (4-Sun avg → 5/17)

9.17% → 10.14%

Delta

+0.97 pp

5-Sunday rank

#1 peak

01 What happened

On Sunday 2026-05-17 the overall bad-call rate climbed to 10.14%, the highest of the last five Sundays. Versus the prior 4 Sundays' average (9.17%), the delta is +0.97pp. 5/15 (Fri) and 5/16 (Sat) both stayed inside their own day-of-week baselines — the anomaly is only Sunday 5/17.

The spike is dominated by audio bad-call (+0.98pp vs 4-Sun avg), with smaller contributions from connection (+0.21pp) and video (+0.26pp). Inside the sub-metrics the signature is clear: mid-call audio freeze (freeze >2% jumped 4.80% → 6.03%, +1.23pp) and long connecting-fail — calls that stayed in connecting state for >2s before failing (never reached connected) went 0.78% → 1.19%, +52% relative. Signaling-layer metrics (no_audio, fast connecting-fail ≤2s) stayed flat — call setup signaling is fine; what fails is media-channel establishment and in-call media transport.

02 Data — bad-call rate (Sunday-on-Sunday)

Bad-call rate · all platforms · global · last 5 Sundays Source: Grafana · Call Dashboard · table ads_botim_voip_result_van_bad_call_1d_di

Sunday is the natural baseline for a Sunday spike — weekday and Saturday traffic shapes differ. 4 prior Sundays sit between 9.05–9.29%; 5/17 jumps to 10.14%.

Sunday rollup · all metrics

Sunday	Total (M)	Bad %	Audio %	Video %	Conn %
2026-04-19	56.55	9.29	3.76	4.17	6.01
2026-04-26	56.16	9.25	3.75	4.15	5.92
2026-05-03	56.51	9.05	3.61	4.08	5.82
2026-05-10	53.47	9.10	3.60	4.06	5.93
4-Sun avg	55.67	9.17	3.68	4.12	5.92
2026-05-17	56.89	10.14	4.66	4.38	6.13
Δ vs 4-Sun avg		+0.97pp	+0.98pp	+0.26pp	+0.21pp

Per-country breakdown · last 5 Sundays (bad-call %)

Country	4/19	4/26	5/03	5/10	5/17	Δ vs 4-Sun avg

All 12 corridors moved on 5/17. Largest deltas: US +2.11pp, EG +1.73pp, DE +1.69pp, GB +1.49pp, JO +1.46pp. PH is essentially flat (-0.03pp) — possible routing/anchor difference worth a separate look.

Audio sub-type · 5/17 vs prior-3 Sun avg

Audio sub-metric	Prev-3 Sun	5/17	Δ
freeze >2%	4.80%	6.03%	+1.23pp
freeze >5%	3.55%	4.60%	+1.05pp
freeze >10%	2.70%	3.64%	+0.94pp
no_audio	0.11%	0.11%	flat

Connection sub-type · 5/17 vs prior-3 Sun avg

Connection sub-metric	Prev-3 Sun	5/17	Δ
connecting slow >2%	2.77%	3.02%	+0.25pp
connecting slow >5%	1.82%	2.02%	+0.20pp
connecting slow >10%	1.22%	1.38%	+0.16pp
connecting fail (total)	0.89%	1.33%	+0.45pp (+51%)
↳ fail ≤2s (fast fail)	0.10%	0.14%	+0.04pp
↳ fail >2s (long connecting-fail)	0.78%	1.19%	+0.41pp (+52%)

03 Observed scope

Platform: iOS + Android both affected — iOS audio +1.22pp, Android audio +0.86pp; iOS total bad-call moved more relatively (7.91% → 9.13%). Not a single-OS regression.
Versions: all versions moved in lockstep — v4.9 (main), v4.8, v4.7, v4.10, v4.0 all show +0.7 to +1.7pp audio. Rules out a client-build regression.
Corridors: all 12 top countries moved on the same day. Largest deltas: US +2.11pp, EG +1.73pp, DE +1.69pp, GB +1.49pp, JO +1.46pp, AE +1.03pp. PH essentially flat (-0.03pp).
Spike shape: sharp single-day event — 5/13 → 5/16 all on DOW baseline, only 5/17 elevated. No gradual ramp.
Symptom mix: real-time media degradation — audio freeze + long connecting-fail (calls stuck >2s in connecting before failing, i.e. media channel never established). Signaling, fast connecting-fail, and no_audio unaffected.
Volume: paired-call volume 57M on baseline — not a volume surge / capacity overflow.

04 Investigation

Status

Active · single-day event · needs causal pin before 5/24 (next Sunday) to confirm whether it repeats

Tracking

Incident doc: incident_bad_call_spike_20260517.md · pipeline partition ads_botim_voip_result_van_bad_call_1d_di / 20260517

Umbrella

[Troubleshoot] Reduce mid-call audio freeze & connection drop — no dedicated umbrella yet; to be opened if 5/24 repeats.

Owners

VoIP Quality (data + triage) · Media infra (relay/SFU telemetry) · Network team (transit & ASN-level diagnostics)

Analysis

Hypotheses and follow-ups (SFU/relay telemetry, cross-border transit, ASN-level RTP throttling, PH routing outlier) are tracked in the incident doc — not in this briefing.