Report · resources

Rolling chat

Qwen3.8-27B · INT4 variant · NVIDIA NGC vLLM 0.29

Measured results / Rolling chat / Run c4e683141a094897a2b3092b

Measured 2026-10-07 11:12 UTC · Record date 2026-10-07

Failed. 60/105 turns

Answer checks

60/105 turns

Rolling-chat turn checks
All rolling-chat turns
Person · turnFilled inputFirst wordRecent · pinned · oldResultRun
1 · 1
20113
21.54 s
true · true · recalled
Passed
1 · 2
40242
25.69 s
true · true · recalled
Passed
1 · 3
60372
48.54 s
true · true · recalled
Passed
1 · 4
80504
54.80 s
true · true · recalled
Passed
1 · 5
100636
81.66 s
true · true · recalled
Passed
1 · 6
120764
90.72 s
true · true · recalled
Passed
1 · 7
140892
108.69 s
true · true · recalled
Passed
1 · 8
161021
153.80 s
true · true · recalled
Passed
1 · 9
181154
174.22 s
true · true · recalled
Passed
1 · 10
201288
190.20 s
true · true · recalled
Passed
1 · 11
221414
201.22 s
true · true · recalled
Passed
1 · 12
241548
218.35 s
true · true · recalled
Passed
1 · 13
181158
675.79 s
true · true · incorrect
Failed
1 · 14
201294
209.03 s
true · true · incorrect
Failed
1 · 15
221419
219.36 s
true · true · incorrect
Failed
1 · 16
241551
226.73 s
true · true · incorrect
Failed
1 · 17
181159
879.48 s
true · true · incorrect
Failed
1 · 18
201288
161.03 s
true · true · incorrect
Failed
1 · 19
221416
160.06 s
true · true · incorrect
Failed
1 · 20
241544
113.99 s
true · true · incorrect
Failed
1 · 21
181142
775.66 s
true · true · incorrect
Failed
2 · 1
20120
47.93 s
true · true · recalled
Passed
2 · 2
40247
53.98 s
true · true · recalled
Passed
2 · 3
60378
80.99 s
true · true · recalled
Passed
2 · 4
80511
91.75 s
true · true · recalled
Passed
2 · 5
100640
114.39 s
true · true · recalled
Passed
2 · 6
120767
156.65 s
true · true · recalled
Passed
2 · 7
140899
173.26 s
true · true · recalled
Passed
2 · 8
161030
188.32 s
true · true · recalled
Passed
2 · 9
181163
202.21 s
true · true · recalled
Passed
2 · 10
201295
217.22 s
true · true · recalled
Passed
2 · 11
221423
199.50 s
true · true · recalled
Passed
2 · 12
241560
223.94 s
true · true · recalled
Passed
2 · 13
181172
705.09 s
true · true · incorrect
Failed
2 · 14
201305
230.61 s
true · true · incorrect
Failed
2 · 15
221436
201.81 s
true · true · incorrect
Failed
2 · 16
241567
222.46 s
true · true · incorrect
Failed
2 · 17
181179
674.94 s
true · true · incorrect
Failed
2 · 18
201309
157.07 s
true · true · incorrect
Failed
2 · 19
221443
171.15 s
true · true · incorrect
Failed
2 · 20
241573
194.46 s
true · true · incorrect
Failed
2 · 21
181175
585.70 s
true · true · incorrect
Failed
3 · 1
20118
79.51 s
true · true · recalled
Passed
3 · 2
40250
91.76 s
true · true · recalled
Passed
3 · 3
60380
114.35 s
true · true · recalled
Passed
3 · 4
80509
156.67 s
true · true · recalled
Passed
3 · 5
100641
174.42 s
true · true · recalled
Passed
3 · 6
120774
190.17 s
true · true · recalled
Passed
3 · 7
140909
196.07 s
true · true · recalled
Passed
3 · 8
161039
213.35 s
true · true · recalled
Passed
3 · 9
181172
193.97 s
true · true · recalled
Passed
3 · 10
201304
236.96 s
true · true · recalled
Passed
3 · 11
221432
208.11 s
true · true · recalled
Passed
3 · 12
241560
211.37 s
true · true · recalled
Passed
3 · 13
181165
699.41 s
true · true · incorrect
Failed
3 · 14
201296
216.49 s
true · true · incorrect
Failed
3 · 15
221425
178.86 s
true · true · incorrect
Failed
3 · 16
241553
170.98 s
true · true · incorrect
Failed
3 · 17
181151
658.17 s
true · true · incorrect
Failed
3 · 18
201282
195.72 s
true · true · incorrect
Failed
3 · 19
221413
123.06 s
true · true · incorrect
Failed
3 · 20
241546
131.52 s
true · true · incorrect
Failed
3 · 21
181157
483.53 s
true · true · incorrect
Failed
4 · 1
20119
21.97 s
true · true · recalled
Passed
4 · 2
40251
34.33 s
true · true · recalled
Passed
4 · 3
60385
132.56 s
true · true · recalled
Passed
4 · 4
80512
173.90 s
true · true · recalled
Passed
4 · 5
100645
190.15 s
true · true · recalled
Passed
4 · 6
120773
194.92 s
true · true · recalled
Passed
4 · 7
140904
212.43 s
true · true · recalled
Passed
4 · 8
161039
201.03 s
true · true · recalled
Passed
4 · 9
181170
237.34 s
true · true · recalled
Passed
4 · 10
201307
213.38 s
true · true · recalled
Passed
4 · 11
221443
191.56 s
true · true · recalled
Passed
4 · 12
241572
229.25 s
true · true · recalled
Passed
4 · 13
181185
873.99 s
true · true · incorrect
Failed
4 · 14
201318
116.81 s
true · true · incorrect
Failed
4 · 15
221454
109.41 s
true · true · incorrect
Failed
4 · 16
241589
124.11 s
true · true · incorrect
Failed
4 · 17
181195
771.75 s
true · true · incorrect
Failed
4 · 18
201331
90.80 s
true · true · incorrect
Failed
4 · 19
221468
56.39 s
true · true · incorrect
Failed
4 · 20
241604
106.95 s
true · true · incorrect
Failed
4 · 21
181200
398.97 s
true · true · incorrect
Failed
5 · 1
20118
95.45 s
true · true · recalled
Passed
5 · 2
40249
155.09 s
true · true · recalled
Passed
5 · 3
60380
170.42 s
true · true · recalled
Passed
5 · 4
80514
190.21 s
true · true · recalled
Passed
5 · 5
100647
202.23 s
true · true · recalled
Passed
5 · 6
120782
218.11 s
true · true · recalled
Passed
5 · 7
140917
197.37 s
true · true · recalled
Passed
5 · 8
161049
231.76 s
true · true · recalled
Passed
5 · 9
181185
213.42 s
true · true · recalled
Passed
5 · 10
201317
221.38 s
true · true · recalled
Passed
5 · 11
221449
229.28 s
true · true · recalled
Passed
5 · 12
241581
198.52 s
true · true · recalled
Passed
5 · 13
181182
709.10 s
true · true · incorrect
Failed
5 · 14
201320
157.91 s
true · true · incorrect
Failed
5 · 15
221457
158.76 s
true · true · incorrect
Failed
5 · 16
241595
163.12 s
true · true · incorrect
Failed
5 · 17
181189
604.87 s
true · true · incorrect
Failed
5 · 18
201320
127.62 s
true · true · incorrect
Failed
5 · 19
221455
104.38 s
true · true · incorrect
Failed
5 · 20
241584
111.28 s
true · true · incorrect
Failed
5 · 21
181188
414.86 s
true · true · incorrect
Failed

Method

Staggered multi-turn fictional archive, exact engine-tokenized rolling gateway; independent recent, pinned and old probes each turn.

Limits

Lab retrieval and rolling-history evidence only. No summaries. Shared engine cache intervals include overlapping users and are not per-user hits. Gateway path, not authenticated document ingestion or customer acceptance.

What did not work

  • 60/105 requests passed.

What ran

Exact identity

Original publisher model
Qwen3.8-27B
Publisher revision
1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0
Variant · format · precision
INT4 variant · safetensors · INT4
Weights SHA-256
not recorded
Engine · version · feature
NVIDIA NGC vLLM · 0.29.0+5013de39.dev · plain
Hardware
NVIDIA GB10, 128 GB unified memory
Harness / workload
keep-rolling-context-v1 · 1
Workload suite · version
not recorded
Harness SHA-256
not recorded
Fixture SHA-256
not recorded

Workload shape

People in this run
5
Configured window (tokens)
262144
Configured window (tokens)
262144
Output token ceiling
256

Full reproduction pack available to customers and evaluators on request

All recorded numbers

Metric definitions, scope and statistics
All aggregate numeric result fields in this run
MetricValueScopeStatisticSample countBasis / source fieldRun
roll ttft max seconds
879.48 s
per person
max
not recorded
roll_ttft_max_seconds
requests ok
60.00 count
machine total
single
not recorded
requests_ok
roll ttft p50 seconds
675.79 s
per person
median
not recorded
roll_ttft_p50_seconds
normal ttft max seconds
237.34 s
per person
max
not recorded
normal_ttft_max_seconds
normal ttft p50 seconds
173.58 s
per person
median
not recorded
normal_ttft_p50_seconds
requests total
105.00 count
machine total
single
not recorded
requests_total

Data and citation

Skilak Keep evidence, run c4e683141a094897a2b3092b, measured 2026-10-07T11:12:04.658512+00:00.
https://skilak.ai/keep/resources/evidence/r/c4e683141a094897a2b3092b

Ready to scope it?

Start with one workload.

Tell us who uses it, what it can read, and what must stay inside. We will map the build.