DROP A scorecard.json TO INSPECT ANY RUN · CHOOSE FILE · or append ?scorecard=<relative path>

Dead Air · The Board

BAKED RUN · ADAPTER patter · TIER 2 · 6bee6d2 · CORPUS 629cb08e

FAIL
15 PASSED3 FAILED RUNTIME (STRUCTURAL) 143.79 s / 540.00 s

Metric Radar

PASS RATE PER METRIC · INFO AXES ARE NON-GATING
outcome_success: 17/18 passed (94%)dead_air_latency: 16/18 passed (89%)mishearing_handling: 17/18 passed (94%) · informationalbarge_in_handling: 15/18 passed (83%) · informationaloutcome_successdead_air_latencymishearing_handling · infobarge_in_handling · info

The Call Set

Elapsed/budget columns are (STRUCTURAL) smoke — event ordering under fakes, not measured latency. The dead-air cell is real wall-clock only on the real-Claude (Tier-2) run; on scripted runs it is synthetic and informational. Mishearing and barge-in metrics gate scripted runs; informational where the adapter can't inject. Open a row for its event Tape.

Dead Air · Scripted vs Real

BAKED RUN · fake/T1 → patter/T2 · 2026-07-16T00:28:04.468Z · 6bee6d2 · CORPUS 629cb08e

15 SAME · 3 REGRESSED

Scenarios

ambiguous-intent-clarify Passed Passed SAME
METRICS
outcome_success 1 → 1
dead_air_latency 5 → 5889.097071
mishearing_handling 0 → 0
barge_in_handling 0 → 0
barge-in-yield Passed Passed SAME
METRICS
outcome_success 1 → 1
dead_air_latency 4 → 5685.835153
mishearing_handling 0 → 0
barge_in_handling 1 → 0
book-happy-path Passed Passed SAME
METRICS
outcome_success 1 → 1
dead_air_latency 5 → 5124.242666
mishearing_handling 0 → 0
barge_in_handling 0 → 0
caller-silence-abandoned Passed Passed SAME
METRICS
outcome_success 1 → 1
dead_air_latency 3 → 1710.0616770000015
mishearing_handling 0 → 0
barge_in_handling 0 → 0
caller-silence-reprompt Passed Passed SAME
METRICS
outcome_success 1 → 1
dead_air_latency 4 → 5455.0303429999985
mishearing_handling 0 → 0
barge_in_handling 0 → 0
cancel-confirm-low-confidence Passed Failed REGRESSED
METRICS
outcome_success 1 → 1
dead_air_latency 4 → 10326.772486000002
mishearing_handling 1 → 0
barge_in_handling 0 → 0
cancel-happy-path Passed Passed SAME
METRICS
outcome_success 1 → 1
dead_air_latency 5 → 6987.484561000005
mishearing_handling 0 → 0
barge_in_handling 0 → 0
cancel-unknown-booking Passed Passed SAME
METRICS
outcome_success 1 → 1
dead_air_latency 4 → 3640.1560099999915
mishearing_handling 0 → 0
barge_in_handling 0 → 0
double-barge-in Passed Passed SAME
METRICS
outcome_success 1 → 1
dead_air_latency 4 → 7674.021519000002
mishearing_handling 0 → 0
barge_in_handling 2 → 0
escalation-emergency Passed Passed SAME
METRICS
outcome_success 1 → 1
dead_air_latency 3 → 3996.134818999999
mishearing_handling 0 → 0
barge_in_handling 0 → 0
mid-barge-hangup Passed Passed SAME
METRICS
outcome_success 1 → 1
dead_air_latency 3 → 3594.227574000004
mishearing_handling 0 → 0
barge_in_handling 1 → 0
mishearing-clean-negative Passed Passed SAME
METRICS
outcome_success 1 → 1
dead_air_latency 4 → 4592.837652999995
mishearing_handling 0 → 0
barge_in_handling 0 → 0
mishearing-repair-booking Passed Failed REGRESSED
METRICS
outcome_success 1 → 1
dead_air_latency 4 → 11039.987052000011
mishearing_handling 1 → 1
barge_in_handling 0 → 0
multi-intent-reschedule-and-book Passed Failed REGRESSED
METRICS
outcome_success 1 → 0
dead_air_latency 5 → 2590.4267370000016
mishearing_handling 0 → 0
barge_in_handling 0 → 0
partial-slot-filling Passed Passed SAME
METRICS
outcome_success 1 → 1
dead_air_latency 4 → 4914.890111000001
mishearing_handling 0 → 0
barge_in_handling 0 → 0
reschedule-happy-path Passed Passed SAME
METRICS
outcome_success 1 → 1
dead_air_latency 5 → 5013.4453959999955
mishearing_handling 0 → 0
barge_in_handling 0 → 0
slot-conflict-alternative Passed Passed SAME
METRICS
outcome_success 1 → 1
dead_air_latency 5 → 8119.705276999986
mishearing_handling 0 → 0
barge_in_handling 0 → 0
tool-timeout-escalation Passed Passed SAME
METRICS
outcome_success 1 → 1
dead_air_latency 4 → 2018.8444059999892
mishearing_handling 0 → 0
barge_in_handling 0 → 0

Real-call observability

Planned — Tier 2 / #3

This viewer reads a Tier-1 scorecard: deterministic, scripted runs that validate the harness, domain, and plumbing — not live-call behaviour. The panels below are committed in the design spec (§7) but are not yet implemented, and intentionally show no data.