{ "scope": "Local screening. Natural output lengths differ; no API sampling seed supplied; Flare default 0. Failed cells excluded from ranking. Cache warmth is observed via hit fraction; filenames are attempted treatment. Same-checkpoint controls do not make incumbent/candidate checkpoint an engine-only comparison.", "rows": [ { "artifact": "capacity/base-ctx10240-c1-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 168.22364171918852, "throughput_tok_s": 115.86355457271537, "time_to_first_output_p50_ms": 813.5912530124187, "ttft_p50_ms": 2265.1859829202294, "e2e_p50_ms": 2620.18905300647, "completion_tokens_mean": 328.6666666666667, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/base-ctx10240-c1-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 167.8454775532386, "throughput_tok_s": 111.05096167488561, "time_to_first_output_p50_ms": 815.8391300821677, "ttft_p50_ms": 2218.7263390515, "e2e_p50_ms": 2482.53126302734, "completion_tokens_mean": 299.5833333333333, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/base-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 60.694628378845316, "throughput_tok_s": 120.4664434031204, "time_to_first_output_p50_ms": 3204.1484899818897, "ttft_p50_ms": 4603.668277035467, "e2e_p50_ms": 5904.855560045689, "completion_tokens_mean": 197.5, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/base-ctx10240-c4-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 67.81512958333084, "throughput_tok_s": 241.4122520733914, "time_to_first_output_p50_ms": 28.50437699817121, "ttft_p50_ms": 2476.1303369887173, "e2e_p50_ms": 3573.331610066816, "completion_tokens_mean": 246.08333333333334, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.9166558329212829, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/base-ctx32768-c1-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 166.08529507788714, "throughput_tok_s": 64.85903357532914, "time_to_first_output_p50_ms": 2463.873633998446, "ttft_p50_ms": 3146.256839041598, "e2e_p50_ms": 3562.471258104779, "completion_tokens_mean": 259.3333333333333, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/base-ctx32768-c1-warm.json.gz", "valid": 11, "attempted": 12, "failed": 1, "eligible": false, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 168.71348721916408, "throughput_tok_s": null, "time_to_first_output_p50_ms": 2454.7802179586142, "ttft_p50_ms": 3579.4091660063714, "e2e_p50_ms": 4008.7096280185506, "completion_tokens_mean": 265.54545454545456, "prompt_tokens_mean": 22922.727272727272, "prefix_cache_hit_avg": 0.00039262343925699334, "request_canary_passed": 11, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/base-ctx32768-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 64.20358483492818, "throughput_tok_s": 65.42544523224456, "time_to_first_output_p50_ms": 9096.083084004931, "ttft_p50_ms": 11090.962184942327, "e2e_p50_ms": 12120.439358055592, "completion_tokens_mean": 221.0, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/base-ctx32768-c4-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 61.148321399487216, "throughput_tok_s": 167.78331567404533, "time_to_first_output_p50_ms": 44.72101200371981, "ttft_p50_ms": 3533.2747279899195, "e2e_p50_ms": 4555.593658005819, "completion_tokens_mean": 251.33333333333334, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.8333914987595494, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-ctx10240-c1-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 168.2084218805494, "throughput_tok_s": 119.1392904973993, "time_to_first_output_p50_ms": 814.0346269356087, "ttft_p50_ms": 2574.0773800062016, "e2e_p50_ms": 2880.2505959756672, "completion_tokens_mean": 348.1666666666667, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-ctx10240-c1-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 167.46854668722278, "throughput_tok_s": 113.9630496628905, "time_to_first_output_p50_ms": 815.8218269236386, "ttft_p50_ms": 2483.3001350052655, "e2e_p50_ms": 2847.4706419510767, "completion_tokens_mean": 302.75, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 82.55807876598494, "throughput_tok_s": 133.33095885314486, "time_to_first_output_p50_ms": 3086.310312966816, "ttft_p50_ms": 4687.278138007969, "e2e_p50_ms": 5423.091597971506, "completion_tokens_mean": 194.91666666666666, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-ctx10240-c4-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 84.0490014680758, "throughput_tok_s": 247.12125876208614, "time_to_first_output_p50_ms": 42.44001000188291, "ttft_p50_ms": 1542.3361139837652, "e2e_p50_ms": 2306.9960449356586, "completion_tokens_mean": 196.75, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.8334852303481749, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-ctx32768-c1-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 178.85009859122258, "throughput_tok_s": 69.01050539318973, "time_to_first_output_p50_ms": 2462.666825973429, "ttft_p50_ms": 3790.170455002226, "e2e_p50_ms": 4139.619276043959, "completion_tokens_mean": 276.8333333333333, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-ctx32768-c1-warm.json.gz", "valid": 11, "attempted": 12, "failed": 1, "eligible": false, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 166.0849444430128, "throughput_tok_s": null, "time_to_first_output_p50_ms": 2454.6180070610717, "ttft_p50_ms": 3850.9530540322885, "e2e_p50_ms": 4288.7270039645955, "completion_tokens_mean": 284.0, "prompt_tokens_mean": 22922.636363636364, "prefix_cache_hit_avg": 0.00039262499625468996, "request_canary_passed": 11, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-ctx32768-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 79.83046058000667, "throughput_tok_s": 74.78788872215377, "time_to_first_output_p50_ms": 9466.572946985252, "ttft_p50_ms": 11408.367299940437, "e2e_p50_ms": 12211.583328898996, "completion_tokens_mean": 244.08333333333334, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-ctx32768-c4-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 69.48684519256686, "throughput_tok_s": 222.47985715815884, "time_to_first_output_p50_ms": 44.92283798754215, "ttft_p50_ms": 1589.7437039529905, "e2e_p50_ms": 2464.6334189455956, "completion_tokens_mean": 205.91666666666666, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.9166848448887881, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed1-ctx10240-c1-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 169.46645423982244, "throughput_tok_s": 98.4767312665712, "time_to_first_output_p50_ms": 810.8437419869006, "ttft_p50_ms": 1427.198842051439, "e2e_p50_ms": 1851.9921200349927, "completion_tokens_mean": 192.0, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed1-ctx10240-c1-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 169.10762459813827, "throughput_tok_s": 95.85937013998269, "time_to_first_output_p50_ms": 813.0746720125899, "ttft_p50_ms": 1397.5187870673835, "e2e_p50_ms": 1759.1756950132549, "completion_tokens_mean": 181.25, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed1-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 76.61771233245003, "throughput_tok_s": 126.34592711731023, "time_to_first_output_p50_ms": 3153.3716870471835, "ttft_p50_ms": 4591.260835062712, "e2e_p50_ms": 5424.38226507511, "completion_tokens_mean": 195.75, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed1-ctx10240-c4-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 81.93370948857726, "throughput_tok_s": 241.63118341458727, "time_to_first_output_p50_ms": 37.94732305686921, "ttft_p50_ms": 1500.490098958835, "e2e_p50_ms": 2173.1913569383323, "completion_tokens_mean": 199.58333333333334, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.8334635162412367, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed1-strict16-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": 71.83760357704192, "throughput_tok_s": 104.36428906114541, "time_to_first_output_p50_ms": 3178.233608021401, "ttft_p50_ms": 4875.5742410430685, "e2e_p50_ms": 5120.6676149740815, "completion_tokens_mean": 140.41666666666666, "prompt_tokens_mean": 7676.833333333333, "prefix_cache_hit_avg": 0.0011723584965342994, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": 1.0 }, { "artifact": "capacity/cap16-fixed2-ctx10240-c1-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 189.2212583372681, "throughput_tok_s": 123.0993812656348, "time_to_first_output_p50_ms": 814.5236679119989, "ttft_p50_ms": 1912.4037050642073, "e2e_p50_ms": 2200.817664968781, "completion_tokens_mean": 283.3333333333333, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed2-ctx10240-c1-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 188.92531212392322, "throughput_tok_s": 125.83011282796454, "time_to_first_output_p50_ms": 817.3108120681718, "ttft_p50_ms": 2120.737898047082, "e2e_p50_ms": 2399.222971056588, "completion_tokens_mean": 302.5, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed2-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 91.98262087536106, "throughput_tok_s": 134.82059537685407, "time_to_first_output_p50_ms": 3281.4152899663895, "ttft_p50_ms": 4376.980005996302, "e2e_p50_ms": 5019.477393012494, "completion_tokens_mean": 213.33333333333334, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed2-ctx10240-c4-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 98.984986790531, "throughput_tok_s": 307.15783671438487, "time_to_first_output_p50_ms": 59.71530103124678, "ttft_p50_ms": 2510.4603990912437, "e2e_p50_ms": 3260.88711200282, "completion_tokens_mean": 283.8333333333333, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.9166341343577279, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed2-host32-ctx10240-c1-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 192.58591777589348, "throughput_tok_s": 127.18409282967, "time_to_first_output_p50_ms": 817.8517880151048, "ttft_p50_ms": 2118.4782220516354, "e2e_p50_ms": 2395.9451620467007, "completion_tokens_mean": 303.3333333333333, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed2-host32-ctx10240-c1-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 191.23846935345452, "throughput_tok_s": 188.223552707566, "time_to_first_output_p50_ms": 39.66020199004561, "ttft_p50_ms": 1121.6442919103429, "e2e_p50_ms": 1410.1892149774358, "completion_tokens_mean": 277.8333333333333, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.9998047129235907, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed2-host32-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 76.52465999252055, "throughput_tok_s": 149.23150473247293, "time_to_first_output_p50_ms": 3251.027219928801, "ttft_p50_ms": 5225.734303006902, "e2e_p50_ms": 6009.018464945257, "completion_tokens_mean": 238.66666666666666, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed2-host32-ctx10240-c4-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 106.60207880717556, "throughput_tok_s": 381.3595378794285, "time_to_first_output_p50_ms": 23.866061004810035, "ttft_p50_ms": 1870.9931860212237, "e2e_p50_ms": 2411.791821010411, "completion_tokens_mean": 258.1666666666667, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 1.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed2-host32-ctx32768-c1-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 192.91742000069817, "throughput_tok_s": 71.10008418725783, "time_to_first_output_p50_ms": 2471.1301090428606, "ttft_p50_ms": 3444.1685599740595, "e2e_p50_ms": 3746.3895700639114, "completion_tokens_mean": 278.8333333333333, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed2-host32-ctx32768-c1-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 193.95855516522667, "throughput_tok_s": 186.7468713340428, "time_to_first_output_p50_ms": 59.23572601750493, "ttft_p50_ms": 1074.8326430330053, "e2e_p50_ms": 1338.9621489914134, "completion_tokens_mean": 278.3333333333333, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.999934563855887, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed2-host32-ctx32768-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 71.75674973151483, "throughput_tok_s": 80.96784434781914, "time_to_first_output_p50_ms": 9413.589234929532, "ttft_p50_ms": 11355.497120996006, "e2e_p50_ms": 13520.642420975491, "completion_tokens_mean": 274.8333333333333, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed2-host32-ctx32768-c4-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 100.31315076542371, "throughput_tok_s": 379.51319745190403, "time_to_first_output_p50_ms": 50.5611120024696, "ttft_p50_ms": 1642.6932900212705, "e2e_p50_ms": 2252.1786539582536, "completion_tokens_mean": 261.75, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.9999854582469294, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed2-host32-strict16-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": 94.60956866029085, "throughput_tok_s": 156.61972349459026, "time_to_first_output_p50_ms": 3121.313238982111, "ttft_p50_ms": 5039.598403032869, "e2e_p50_ms": 5233.552680001594, "completion_tokens_mean": 234.58333333333334, "prompt_tokens_mean": 7676.833333333333, "prefix_cache_hit_avg": 0.0011723584965342994, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": 1.0 }, { "artifact": "capacity/cap16-fixed2-strict16-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": 100.59118894821134, "throughput_tok_s": 145.65044771450968, "time_to_first_output_p50_ms": 3087.208896991797, "ttft_p50_ms": 4708.08144507464, "e2e_p50_ms": 5238.551519927569, "completion_tokens_mean": 213.25, "prompt_tokens_mean": 7676.833333333333, "prefix_cache_hit_avg": 0.0011723584965342994, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": 1.0 }, { "artifact": "capacity/cap16-fixed3-ctx10240-c1-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 193.18716331194014, "throughput_tok_s": 138.45660247475803, "time_to_first_output_p50_ms": 815.6067209783942, "ttft_p50_ms": 2563.674684963189, "e2e_p50_ms": 2867.7306440658867, "completion_tokens_mean": 397.4166666666667, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed3-ctx10240-c1-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 193.44086651719044, "throughput_tok_s": 138.16293200034497, "time_to_first_output_p50_ms": 817.5299540162086, "ttft_p50_ms": 2565.9212049795315, "e2e_p50_ms": 2902.417527977377, "completion_tokens_mean": 396.9166666666667, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed3-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 88.5533558323746, "throughput_tok_s": 182.233124264474, "time_to_first_output_p50_ms": 3188.3144839666784, "ttft_p50_ms": 5633.130050031468, "e2e_p50_ms": 6633.528845966794, "completion_tokens_mean": 325.75, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed3-ctx10240-c4-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 88.32965855765204, "throughput_tok_s": 290.88722382088685, "time_to_first_output_p50_ms": 209.3889380339533, "ttft_p50_ms": 3164.7077900124714, "e2e_p50_ms": 3788.936438038945, "completion_tokens_mean": 298.9166666666667, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.8334418487508368, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed3-ctx32768-c1-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 191.6553115357799, "throughput_tok_s": 77.85273669425291, "time_to_first_output_p50_ms": 2489.904070040211, "ttft_p50_ms": 3661.1929880455136, "e2e_p50_ms": 3966.319374041632, "completion_tokens_mean": 326.6666666666667, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed3-ctx32768-c1-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 196.7653258874724, "throughput_tok_s": 77.68638557558472, "time_to_first_output_p50_ms": 2482.791212038137, "ttft_p50_ms": 3653.290734044276, "e2e_p50_ms": 3957.838093978353, "completion_tokens_mean": 320.3333333333333, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed3-ctx32768-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 71.22452155377755, "throughput_tok_s": 94.00247315769033, "time_to_first_output_p50_ms": 8885.284316027537, "ttft_p50_ms": 13316.476851003245, "e2e_p50_ms": 14417.40833397489, "completion_tokens_mean": 350.9166666666667, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed3-ctx32768-c4-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 82.4246878109153, "throughput_tok_s": 224.16602174404883, "time_to_first_output_p50_ms": 59.51451405417174, "ttft_p50_ms": 4327.554418006912, "e2e_p50_ms": 5124.088173964992, "completion_tokens_mean": 397.3333333333333, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.8333915012969483, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-fixed3-strict16-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": 100.94323809474304, "throughput_tok_s": 144.6126328283347, "time_to_first_output_p50_ms": 3075.298903044313, "ttft_p50_ms": 5275.15656396281, "e2e_p50_ms": 5407.099538948387, "completion_tokens_mean": 205.91666666666666, "prompt_tokens_mean": 7676.833333333333, "prefix_cache_hit_avg": 0.0011723584965342994, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": 1.0 }, { "artifact": "capacity/cap16-host24-ctx10240-c1-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 163.89826025386577, "throughput_tok_s": 109.23968538839229, "time_to_first_output_p50_ms": 821.2969899177551, "ttft_p50_ms": 2145.440137013793, "e2e_p50_ms": 2507.083526928909, "completion_tokens_mean": 286.1666666666667, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host24-ctx10240-c1-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 169.0864067242921, "throughput_tok_s": 168.4593739692857, "time_to_first_output_p50_ms": 38.81679300684482, "ttft_p50_ms": 624.5281789451838, "e2e_p50_ms": 996.9068201025948, "completion_tokens_mean": 204.0, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.9998047129235907, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host24-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 91.3340660952736, "throughput_tok_s": 149.38905116476167, "time_to_first_output_p50_ms": 3127.5117360055447, "ttft_p50_ms": 5127.766535966657, "e2e_p50_ms": 5842.279669013806, "completion_tokens_mean": 229.25, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host24-ctx10240-c4-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 98.6582752399412, "throughput_tok_s": 373.3768976552779, "time_to_first_output_p50_ms": 20.221589016728103, "ttft_p50_ms": 1034.8017059732229, "e2e_p50_ms": 1663.8869390590116, "completion_tokens_mean": 184.0, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 1.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host24-ctx32768-c1-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 169.20883898530528, "throughput_tok_s": 71.9591135702681, "time_to_first_output_p50_ms": 2500.430908985436, "ttft_p50_ms": 3809.8211280303076, "e2e_p50_ms": 4195.194728090428, "completion_tokens_mean": 309.9166666666667, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host24-ctx32768-c1-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 171.12729272076513, "throughput_tok_s": 163.56634974812067, "time_to_first_output_p50_ms": 59.5776989357546, "ttft_p50_ms": 619.9029970448464, "e2e_p50_ms": 1036.9706959463656, "completion_tokens_mean": 185.5, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.999934563855887, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host24-ctx32768-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 86.02632033597908, "throughput_tok_s": 68.16292528974115, "time_to_first_output_p50_ms": 9580.4709920194, "ttft_p50_ms": 10626.519092940725, "e2e_p50_ms": 11554.914314998314, "completion_tokens_mean": 225.08333333333334, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host24-ctx32768-c4-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 96.1226609604673, "throughput_tok_s": 354.8448932867234, "time_to_first_output_p50_ms": 69.64276300277561, "ttft_p50_ms": 2498.566302936524, "e2e_p50_ms": 3277.751458925195, "completion_tokens_mean": 292.4166666666667, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.9999709174454958, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host24-strict128-ctx10240-c4-cold.json.gz", "valid": 0, "attempted": 12, "failed": 12, "eligible": false, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": null, "throughput_tok_s": null, "time_to_first_output_p50_ms": null, "ttft_p50_ms": null, "e2e_p50_ms": null, "completion_tokens_mean": null, "prompt_tokens_mean": null, "prefix_cache_hit_avg": null, "request_canary_passed": 0, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host24-strict16-ctx8192-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 8192, "controlled_output_policy": "strict", "decode_tok_s_p50": 85.37747239137573, "throughput_tok_s": 170.46108325583455, "time_to_first_output_p50_ms": 2514.6668790839612, "ttft_p50_ms": 4264.978516031988, "e2e_p50_ms": 4712.600038037635, "completion_tokens_mean": 224.16666666666666, "prompt_tokens_mean": 6069.25, "prefix_cache_hit_avg": 0.0014828851635309537, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": 1.0 }, { "artifact": "capacity/cap16-host32-ctx10240-c1-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 166.86728150865542, "throughput_tok_s": 110.23085264046732, "time_to_first_output_p50_ms": 821.2028359994292, "ttft_p50_ms": 2016.1585030145943, "e2e_p50_ms": 2343.3520479593426, "completion_tokens_mean": 282.75, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host32-ctx10240-c1-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 169.00125081479345, "throughput_tok_s": 165.45794543374936, "time_to_first_output_p50_ms": 39.11424393299967, "ttft_p50_ms": 615.5667490093037, "e2e_p50_ms": 968.4747159481049, "completion_tokens_mean": 183.91666666666666, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.9998047129235907, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host32-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 78.70115341226813, "throughput_tok_s": 149.21865833717158, "time_to_first_output_p50_ms": 3206.799605046399, "ttft_p50_ms": 4642.0258530415595, "e2e_p50_ms": 5502.520686015487, "completion_tokens_mean": 251.58333333333334, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host32-ctx10240-c4-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 96.70812575904941, "throughput_tok_s": 367.39250617158615, "time_to_first_output_p50_ms": 24.406409007497132, "ttft_p50_ms": 1263.337365933694, "e2e_p50_ms": 1906.656099949032, "completion_tokens_mean": 272.9166666666667, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.999978301436445, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host32-ctx32768-c1-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 166.79242805401518, "throughput_tok_s": 64.66307082985777, "time_to_first_output_p50_ms": 2503.572797984816, "ttft_p50_ms": 3502.702070050873, "e2e_p50_ms": 3892.364798928611, "completion_tokens_mean": 261.75, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host32-ctx32768-c1-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 172.24239106827352, "throughput_tok_s": 163.6874825436999, "time_to_first_output_p50_ms": 59.469912899658084, "ttft_p50_ms": 620.0660601025447, "e2e_p50_ms": 1040.1964880293235, "completion_tokens_mean": 180.0, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.999934563855887, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host32-ctx32768-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 83.78201423599094, "throughput_tok_s": 68.1988806073102, "time_to_first_output_p50_ms": 9765.283280052245, "ttft_p50_ms": 11301.12626904156, "e2e_p50_ms": 12303.761041956022, "completion_tokens_mean": 221.08333333333334, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host32-ctx32768-c4-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 96.77442426015273, "throughput_tok_s": 367.256405608876, "time_to_first_output_p50_ms": 54.1274769930169, "ttft_p50_ms": 1075.4995649913326, "e2e_p50_ms": 1682.9480070155114, "completion_tokens_mean": 201.91666666666666, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.9999854582469294, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-host32-strict16-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": 89.08150181871515, "throughput_tok_s": 134.50336850211585, "time_to_first_output_p50_ms": 3132.125222007744, "ttft_p50_ms": 5105.091236997396, "e2e_p50_ms": 5214.601735933684, "completion_tokens_mean": 188.41666666666666, "prompt_tokens_mean": 7676.833333333333, "prefix_cache_hit_avg": 0.0011723584965342994, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": 1.0 }, { "artifact": "capacity/cap16-p2048-fixed2-host32-ctx10240-c1-cold.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 190.76243177369778, "throughput_tok_s": 116.50608336782796, "time_to_first_output_p50_ms": 928.9650309365243, "ttft_p50_ms": 2091.0068919183686, "e2e_p50_ms": 2369.622456957586, "completion_tokens_mean": 279.5, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-ctx10240-c1-warm.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 190.3437344104787, "throughput_tok_s": 186.22871601862286, "time_to_first_output_p50_ms": 38.82319398690015, "ttft_p50_ms": 1006.6005769185722, "e2e_p50_ms": 1294.2196330986917, "completion_tokens_mean": 250.75, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.9998047129235907, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-ctx10240-c4-cold.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 88.49820001719286, "throughput_tok_s": 153.69197923716115, "time_to_first_output_p50_ms": 3412.9389149602503, "ttft_p50_ms": 5224.02288694866, "e2e_p50_ms": 5794.460078002885, "completion_tokens_mean": 265.25, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-ctx10240-c4-warm.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 106.63977085534681, "throughput_tok_s": 386.0079586511745, "time_to_first_output_p50_ms": 24.28253402467817, "ttft_p50_ms": 1906.7887070123106, "e2e_p50_ms": 2300.370312994346, "completion_tokens_mean": 261.3333333333333, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 1.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-ctx32768-c1-cold.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 190.08809925084412, "throughput_tok_s": 64.12650814864115, "time_to_first_output_p50_ms": 2739.459982025437, "ttft_p50_ms": 3327.665605931543, "e2e_p50_ms": 3736.5146649535745, "completion_tokens_mean": 265.25, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-ctx32768-c1-warm.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 191.1614871675003, "throughput_tok_s": 185.50328387401476, "time_to_first_output_p50_ms": 58.62912395969033, "ttft_p50_ms": 891.2773330230266, "e2e_p50_ms": 1429.4868330471218, "completion_tokens_mean": 302.0, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.999934563855887, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-ctx32768-c4-cold.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 86.58231100841732, "throughput_tok_s": 78.95721937058997, "time_to_first_output_p50_ms": 10846.16882901173, "ttft_p50_ms": 13059.65558392927, "e2e_p50_ms": 14060.89665403124, "completion_tokens_mean": 284.1666666666667, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-ctx32768-c4-warm.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 101.40555260404156, "throughput_tok_s": 363.89662325855136, "time_to_first_output_p50_ms": 46.109026996418834, "ttft_p50_ms": 1866.493320907466, "e2e_p50_ms": 2382.619317038916, "completion_tokens_mean": 277.0833333333333, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.9999927289648954, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-high-ctx10240-c1-cold.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 187.08557986291186, "throughput_tok_s": 64.19990732382976, "time_to_first_output_p50_ms": 964.2346400069073, "ttft_p50_ms": 1093.8638780498877, "e2e_p50_ms": 1433.3713669329882, "completion_tokens_mean": 93.75, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-high-ctx10240-c1-warm.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 184.85651793705907, "throughput_tok_s": 171.88899415202877, "time_to_first_output_p50_ms": 50.76111608650535, "ttft_p50_ms": 181.50198203511536, "e2e_p50_ms": 521.5744690503925, "completion_tokens_mean": 92.83333333333333, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.9998047129235907, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-high-ctx10240-c4-cold.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 68.7490222311799, "throughput_tok_s": 62.18736022593928, "time_to_first_output_p50_ms": 3658.9340890059248, "ttft_p50_ms": 3841.6765680303797, "e2e_p50_ms": 4468.715933035128, "completion_tokens_mean": 78.0, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-high-ctx10240-c4-warm.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 101.43434925487885, "throughput_tok_s": 346.19344594995727, "time_to_first_output_p50_ms": 22.469001007266343, "ttft_p50_ms": 260.1411530049518, "e2e_p50_ms": 820.9098399383947, "completion_tokens_mean": 84.0, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.9999565943961526, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-high-ctx32768-c1-cold.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 192.85981654653702, "throughput_tok_s": 32.80844167210421, "time_to_first_output_p50_ms": 2741.1790849873796, "ttft_p50_ms": 2890.7801379682496, "e2e_p50_ms": 3207.904104026966, "completion_tokens_mean": 108.33333333333333, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-high-ctx32768-c1-warm.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 198.6411418676964, "throughput_tok_s": 177.8472533656514, "time_to_first_output_p50_ms": 58.677271008491516, "ttft_p50_ms": 197.46181508526206, "e2e_p50_ms": 513.6257029371336, "completion_tokens_mean": 107.5, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.999934563855887, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-high-ctx32768-c4-cold.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 89.52292816244329, "throughput_tok_s": 33.29821703363319, "time_to_first_output_p50_ms": 10945.054664975032, "ttft_p50_ms": 11087.665601051413, "e2e_p50_ms": 11968.058387050405, "completion_tokens_mean": 102.33333333333333, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-high-ctx32768-c4-warm.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 101.5575973960762, "throughput_tok_s": 329.89694035830814, "time_to_first_output_p50_ms": 81.04790304787457, "ttft_p50_ms": 434.9897779757157, "e2e_p50_ms": 1006.3897989457473, "completion_tokens_mean": 106.16666666666667, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.9999563763268129, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-high-strict128-ctx10240-c4-cold.json.gz", "valid": 0, "attempted": 12, "failed": 12, "eligible": false, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": null, "throughput_tok_s": null, "time_to_first_output_p50_ms": null, "ttft_p50_ms": null, "e2e_p50_ms": null, "completion_tokens_mean": null, "prompt_tokens_mean": null, "prefix_cache_hit_avg": null, "request_canary_passed": 0, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-high-strict16-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": 105.040218910988, "throughput_tok_s": 69.73123311054957, "time_to_first_output_p50_ms": 3680.9511239407584, "ttft_p50_ms": 3841.0759200342, "e2e_p50_ms": 4229.15110597387, "completion_tokens_mean": 83.66666666666667, "prompt_tokens_mean": 7676.833333333333, "prefix_cache_hit_avg": 0.0011723584965342994, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": 1.0 }, { "artifact": "capacity/cap16-p2048-fixed2-host32-high-strict16-n100-ctx10240-c4-cold.json.gz", "valid": 100, "attempted": 100, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": 53.940814395183764, "throughput_tok_s": 50.157539124740076, "time_to_first_output_p50_ms": 3738.9906990574673, "ttft_p50_ms": 4302.156051038764, "e2e_p50_ms": 4753.194689983502, "completion_tokens_mean": 60.16, "prompt_tokens_mean": 7705.0, "prefix_cache_hit_avg": 0.0011211701278701436, "request_canary_passed": 100, "controlled_output_exact_adherence_rate": 1.0 }, { "artifact": "capacity/cap16-p2048-fixed2-host32-strict128-ctx10240-c4-cold.json.gz", "valid": 0, "attempted": 12, "failed": 12, "eligible": false, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": null, "throughput_tok_s": null, "time_to_first_output_p50_ms": null, "ttft_p50_ms": null, "e2e_p50_ms": null, "completion_tokens_mean": null, "prompt_tokens_mean": null, "prefix_cache_hit_avg": null, "request_canary_passed": 0, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-p2048-fixed2-host32-strict16-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": 90.07446950935514, "throughput_tok_s": 136.56302409108994, "time_to_first_output_p50_ms": 3276.7773140221834, "ttft_p50_ms": 4537.147002993152, "e2e_p50_ms": 4909.578644088469, "completion_tokens_mean": 201.91666666666666, "prompt_tokens_mean": 7676.833333333333, "prefix_cache_hit_avg": 0.0011723584965342994, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": 1.0 }, { "artifact": "capacity/cap16-p2048-fixed2-host32-strict16-n100-ctx10240-c4-cold.json.gz", "valid": 100, "attempted": 100, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": 79.62729667259188, "throughput_tok_s": 123.08725278138257, "time_to_first_output_p50_ms": 4058.191891061142, "ttft_p50_ms": 6659.299143007956, "e2e_p50_ms": 6983.884738059714, "completion_tokens_mean": 220.2, "prompt_tokens_mean": 7705.0, "prefix_cache_hit_avg": 0.0011211701278701436, "request_canary_passed": 100, "controlled_output_exact_adherence_rate": 1.0 }, { "artifact": "capacity/cap16-strict16-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": 84.93112533898211, "throughput_tok_s": 140.12770916272206, "time_to_first_output_p50_ms": 2904.826200916432, "ttft_p50_ms": 4845.8431150065735, "e2e_p50_ms": 4977.093574008904, "completion_tokens_mean": 202.66666666666666, "prompt_tokens_mean": 7676.833333333333, "prefix_cache_hit_avg": 0.0011723584965342994, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": 1.0 }, { "artifact": "capacity/cap16-throughput-ctx10240-c1-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 188.35379288136224, "throughput_tok_s": 106.79791978387642, "time_to_first_output_p50_ms": 812.5570969423279, "ttft_p50_ms": 1434.5189230516553, "e2e_p50_ms": 1806.3284819945693, "completion_tokens_mean": 204.0, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-throughput-ctx10240-c1-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 190.20432486736678, "throughput_tok_s": 112.13868171310102, "time_to_first_output_p50_ms": 815.2275079628453, "ttft_p50_ms": 1424.3687570560724, "e2e_p50_ms": 1767.8119470365345, "completion_tokens_mean": 222.33333333333334, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-throughput-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 90.32648516804329, "throughput_tok_s": 140.1525535231347, "time_to_first_output_p50_ms": 3438.5690030176193, "ttft_p50_ms": 4698.650698061101, "e2e_p50_ms": 5777.190114953555, "completion_tokens_mean": 202.33333333333334, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0011717733111551616, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-throughput-ctx10240-c4-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 50.46625184319009, "throughput_tok_s": 228.2689982789557, "time_to_first_output_p50_ms": 41.39149293769151, "ttft_p50_ms": 1942.0837379293516, "e2e_p50_ms": 3632.8668129863217, "completion_tokens_mean": 204.66666666666666, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.750227847633905, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-throughput-ctx32768-c1-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 187.7286827860364, "throughput_tok_s": 51.04947074706061, "time_to_first_output_p50_ms": 2489.599333028309, "ttft_p50_ms": 3005.828932975419, "e2e_p50_ms": 3331.225630012341, "completion_tokens_mean": 173.75, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-throughput-ctx32768-c1-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 191.188534402587, "throughput_tok_s": 56.17060692337383, "time_to_first_output_p50_ms": 2483.21356496308, "ttft_p50_ms": 3017.3694549594074, "e2e_p50_ms": 3399.159519933164, "completion_tokens_mean": 197.83333333333334, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-throughput-ctx32768-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 50.133852079276046, "throughput_tok_s": 78.56979370839304, "time_to_first_output_p50_ms": 9671.576748020016, "ttft_p50_ms": 11258.671547984704, "e2e_p50_ms": 13307.45966208633, "completion_tokens_mean": 262.8333333333333, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0003926230499169998, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-throughput-ctx32768-c4-warm.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 33.602065265914526, "throughput_tok_s": 148.52158965513203, "time_to_first_output_p50_ms": 431.0618130257353, "ttft_p50_ms": 2737.0667631039396, "e2e_p50_ms": 6535.908814985305, "completion_tokens_mean": 229.83333333333334, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.7500763407936897, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/cap16-throughput-strict16-ctx10240-c4-cold.json.gz", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": 101.21348919921283, "throughput_tok_s": 135.6236764699348, "time_to_first_output_p50_ms": 3108.858431922272, "ttft_p50_ms": 4403.444538009353, "e2e_p50_ms": 4815.520252916031, "completion_tokens_mean": 165.08333333333334, "prompt_tokens_mean": 7676.833333333333, "prefix_cache_hit_avg": 0.0011723584965342994, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": 1.0 }, { "artifact": "capacity/incumbent-ctx10240-c1-cold.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 195.2931828833121, "throughput_tok_s": 119.60780063238772, "time_to_first_output_p50_ms": 877.6218360289931, "ttft_p50_ms": 1871.0420570569113, "e2e_p50_ms": 2146.8527771066874, "completion_tokens_mean": 267.3333333333333, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-ctx10240-c1-warm.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 195.4948668553143, "throughput_tok_s": 192.79085075255628, "time_to_first_output_p50_ms": 36.78748395759612, "ttft_p50_ms": 1183.1074130022898, "e2e_p50_ms": 1569.286294048652, "completion_tokens_mean": 297.0, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 1.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-ctx10240-c4-cold.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 93.56692768056729, "throughput_tok_s": 164.51607899392118, "time_to_first_output_p50_ms": 3922.452778904699, "ttft_p50_ms": 5075.454612961039, "e2e_p50_ms": 5951.165374019183, "completion_tokens_mean": 318.1666666666667, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-ctx10240-c4-warm.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 104.45976748299475, "throughput_tok_s": 375.5832475006309, "time_to_first_output_p50_ms": 101.1842240113765, "ttft_p50_ms": 1705.7775619905442, "e2e_p50_ms": 2383.804040029645, "completion_tokens_mean": 316.5833333333333, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 1.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-ctx32768-c1-cold.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 193.1864346937496, "throughput_tok_s": 70.80228994087706, "time_to_first_output_p50_ms": 2599.7980480315164, "ttft_p50_ms": 3611.277712043375, "e2e_p50_ms": 3952.4721320485696, "completion_tokens_mean": 288.75, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-ctx32768-c1-warm.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 191.57322389184014, "throughput_tok_s": 187.66156182241033, "time_to_first_output_p50_ms": 46.44404898863286, "ttft_p50_ms": 1188.6430280283093, "e2e_p50_ms": 1537.0849709725007, "completion_tokens_mean": 308.25, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 1.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-ctx32768-c4-cold.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 114.01196027143828, "throughput_tok_s": 87.08052472303572, "time_to_first_output_p50_ms": 12685.200973064639, "ttft_p50_ms": 15679.247591993771, "e2e_p50_ms": 16095.354264951311, "completion_tokens_mean": 372.75, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-ctx32768-c4-warm.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 95.9452989769656, "throughput_tok_s": 363.2827476972452, "time_to_first_output_p50_ms": 113.70677605737, "ttft_p50_ms": 2462.217030930333, "e2e_p50_ms": 3227.402388001792, "completion_tokens_mean": 342.5, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 1.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-high-ctx10240-c1-cold.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 187.8292171911925, "throughput_tok_s": 68.22141712540972, "time_to_first_output_p50_ms": 886.5866500418633, "ttft_p50_ms": 1008.6423179600388, "e2e_p50_ms": 1362.842648057267, "completion_tokens_mean": 93.91666666666667, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-high-ctx10240-c1-warm.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 192.78946342437717, "throughput_tok_s": 178.5930908348129, "time_to_first_output_p50_ms": 36.62890603300184, "ttft_p50_ms": 157.1285299723968, "e2e_p50_ms": 504.9757909728214, "completion_tokens_mean": 102.58333333333333, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 1.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-high-ctx10240-c4-cold.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 98.3528555433764, "throughput_tok_s": 69.23369313942291, "time_to_first_output_p50_ms": 4089.004490058869, "ttft_p50_ms": 4261.954981018789, "e2e_p50_ms": 4610.876639955677, "completion_tokens_mean": 90.83333333333333, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 0.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-high-ctx10240-c4-warm.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "observe", "decode_tok_s_p50": 94.70255407306449, "throughput_tok_s": 318.3278124345695, "time_to_first_output_p50_ms": 105.17017298843712, "ttft_p50_ms": 347.97348000574857, "e2e_p50_ms": 918.6739419819787, "completion_tokens_mean": 87.33333333333333, "prompt_tokens_mean": 7680.666666666667, "prefix_cache_hit_avg": 1.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-high-ctx32768-c1-cold.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 189.1888286138499, "throughput_tok_s": 34.06360317007248, "time_to_first_output_p50_ms": 2599.6914400020614, "ttft_p50_ms": 2751.2812359491363, "e2e_p50_ms": 3098.5282219480723, "completion_tokens_mean": 107.5, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-high-ctx32768-c1-warm.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 1, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 189.6751126641199, "throughput_tok_s": 182.77676115887172, "time_to_first_output_p50_ms": 45.24797294288874, "ttft_p50_ms": 278.73152191750705, "e2e_p50_ms": 587.5534159131348, "completion_tokens_mean": 122.83333333333333, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 1.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-high-ctx32768-c4-cold.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 92.5478589150233, "throughput_tok_s": 37.453997251860734, "time_to_first_output_p50_ms": 10967.162353917956, "ttft_p50_ms": 11498.121066950262, "e2e_p50_ms": 11824.70455893781, "completion_tokens_mean": 118.91666666666667, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 0.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-high-ctx32768-c4-warm.json", "valid": 12, "attempted": 12, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 32768, "controlled_output_policy": "observe", "decode_tok_s_p50": 96.90928114216861, "throughput_tok_s": 327.11309401109594, "time_to_first_output_p50_ms": 110.63808796461672, "ttft_p50_ms": 383.3933639107272, "e2e_p50_ms": 1053.514375933446, "completion_tokens_mean": 114.91666666666667, "prompt_tokens_mean": 22922.75, "prefix_cache_hit_avg": 1.0, "request_canary_passed": 12, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-high-strict16-n100-ctx10240-c4-cold.json.gz", "valid": 100, "attempted": 100, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": 122.47274795294936, "throughput_tok_s": 62.46451307823878, "time_to_first_output_p50_ms": 3893.8121099490672, "ttft_p50_ms": 4246.967213926837, "e2e_p50_ms": 4392.035812954418, "completion_tokens_mean": 69.87, "prompt_tokens_mean": 7705.0, "prefix_cache_hit_avg": 0.0, "request_canary_passed": 100, "controlled_output_exact_adherence_rate": 1.0 }, { "artifact": "capacity/incumbent-strict128-ctx10240-c4-cold.json.gz", "valid": 0, "attempted": 12, "failed": 12, "eligible": false, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": null, "throughput_tok_s": null, "time_to_first_output_p50_ms": null, "ttft_p50_ms": null, "e2e_p50_ms": null, "completion_tokens_mean": null, "prompt_tokens_mean": null, "prefix_cache_hit_avg": null, "request_canary_passed": 0, "controlled_output_exact_adherence_rate": null }, { "artifact": "capacity/incumbent-strict16-n100-ctx10240-c4-cold.json.gz", "valid": 100, "attempted": 100, "failed": 0, "eligible": true, "concurrency": 4, "requested_context": 10240, "controlled_output_policy": "strict", "decode_tok_s_p50": 126.44369558226997, "throughput_tok_s": 115.1371395640066, "time_to_first_output_p50_ms": 4761.5796740865335, "ttft_p50_ms": 5711.961172055453, "e2e_p50_ms": 5887.025736970827, "completion_tokens_mean": 175.9, "prompt_tokens_mean": 7705.0, "prefix_cache_hit_avg": 0.0, "request_canary_passed": 100, "controlled_output_exact_adherence_rate": 1.0 } ] }