Round 10 breakdown

scoredleaderSubmission

Round score

0.6531

median speedup minus reliability deduction

Prompts scored

29 / 30

median_e2e_speedup

Below tolerance

answered too little to score

Zeroed total

scored 0.0 without measuring

Workload

Temperature range

0.1 to 1.01 (per prompt)

Request interval

2 ms

Thinking

Disabled

Context limit

262,144 tokens

The same inputs and output limits apply to every engine. Actual concurrency depends on response duration and engine scheduling.

Score calculation

Median speedup

+65.65%

Failed requests

1 / 30 (3.33%)

Deduction

0.1 × 0.0333 = 0.003333

Final score

0.656454 - 0.003333 = 0.653120

The round score is the scoring rule applied to the prompts below. Both engines are compared at the same output token count, so a candidate that stops early cannot buy speed by answering less: a prompt under the tolerance bar scores nothing and shows its gate reason.

Per-prompt breakdown

30 prompts

Prompt	Speedup	Baseline	Candidate	Aligned output	Gate
hf-015	+66.93%	77.368s	25.584s	3945	—
hf-011	+65.93%	78.660s	26.798s	4002	—
hf-016	+64.45%	81.870s	29.103s

Request trace

32 requests

Request	TTFT	Total	Tokens	Slowest gap	Stream
hf-000	242.56ms	34.030s	5120	22.71ms	—
hf-001	258.04ms	30.296s	4577	21.18ms	—
hf-002	254.88ms	33.551s

Per-prompt repetition checks

Each latency-median response is compared with the lowest valid opening-baseline ratio for the same prompt. Drops above the recorded limit fail; absolute checks also apply. These checks detect repetition, not overall writing quality.

Prompt	Candidate char-16	Baseline char-16	Drop	Limit	Result
hf-000	0.9765	0.9362	-0.0403	0.1000	Passed
hf-001	0.9650	0.8729	-0.0920	0.1000	Passed
hf-002	0.9614	0.9291	-0.0324	0.1000	Passed
hf-003	0.9527

Prompt	Repetition	Temperature	Top p
hf-024	1	0.177816	1
hf-025	1	0.394988	1
hf-020	1	0.595151	1
hf-004	1	0.500898	1
hf-021	1	0.142976	1
hf-029	1	0.249246	1
hf-010	1	0.209581	1
hf-009	1	0.674999	1
hf-001	1	0.263132	1
hf-008	1	0.248955	1
hf-022	1	0.584491	1
hf-003	1	0.531719	1
hf-030	1	0.438047	1
hf-015	1	0.54516	1
hf-019	1	0.484014	1
hf-031	1	0.25302	1
hf-013	1	0.504122	1
hf-018	1	0.727228	1
hf-012	1	0.523429	1
hf-005	1	0.368926	1
hf-011	1	0.303068	1
hf-002	1	0.516677	1
hf-007	1	0.445021	1
hf-026	1	0.171189	1
hf-028	1	0.87209	1
hf-014	1	0.60292	1
hf-000	1	0.3273	1
hf-027	1	0.940405	1
hf-016	1	0.681382	1
hf-023	1	0.723001	1
hf-006	1	0.961588	1
hf-017	1	0.991942	1
hf-015	2	0.54516	1
hf-024	2	0.177816	1
hf-001	2	0.263132	1
hf-030	2	0.438047	1
hf-025	2	0.394988	1
hf-031	2	0.25302	1
hf-005	2	0.368926	1
hf-012	2	0.523429	1
hf-029	2	0.249246	1
hf-009	2	0.674999	1
hf-022	2	0.584491	1
hf-016	2	0.681382	1
hf-014	2	0.60292	1
hf-013	2	0.504122	1
hf-020	2	0.595151	1
hf-019	2	0.484014	1
hf-003	2	0.531719	1
hf-008	2	0.248955	1
hf-007	2	0.445021	1
hf-010	2	0.209581	1
hf-021	2	0.142976	1
hf-011	2	0.303068	1
hf-000	2	0.3273	1
hf-002	2	0.516677	1
hf-023	2	0.723001	1
hf-004	2	0.500898	1
hf-026	2	0.171189	1
hf-028	2	0.87209	1
hf-018	2	0.727228	1
hf-017	2	0.991942	1
hf-027	2	0.940405	1
hf-006	2	0.961588	1
hf-024	3	0.177816	1
hf-025	3	0.394988	1
hf-020	3	0.595151	1
hf-004	3	0.500898	1
hf-021	3	0.142976	1
hf-029	3	0.249246	1
hf-010	3	0.209581	1
hf-009	3	0.674999	1
hf-001	3	0.263132	1
hf-008	3	0.248955	1
hf-022	3	0.584491	1
hf-003	3	0.531719	1
hf-030	3	0.438047	1
hf-015	3	0.54516	1
hf-019	3	0.484014	1
hf-031	3	0.25302	1
hf-013	3	0.504122	1
hf-018	3	0.727228	1
hf-012	3	0.523429	1
hf-005	3	0.368926	1
hf-011	3	0.303068	1
hf-002	3	0.516677	1
hf-007	3	0.445021	1
hf-026	3	0.171189	1
hf-028	3	0.87209	1
hf-014	3	0.60292	1
hf-000	3	0.3273	1
hf-027	3	0.940405	1
hf-016	3	0.681382	1
hf-023	3	0.723001	1
hf-006	3	0.961588	1
hf-017	3	0.991942	1

Round 10 breakdown

Workload

Score calculation

Per-prompt breakdown

Request trace

Per-prompt repetition checks

Round 10 breakdown

Round 10 breakdown

Workload

Score calculation

Per-prompt breakdown

Request trace

Per-prompt repetition checks

Replay sampling

Correctness

SLA replay

Engine