T3 Code · Performance over time

Turns completed in the window

All metrics
1 / 120
Turns completed in the window (higher is better)
188
18 Nov−1.1% vs the median of up to 7 previous complete nightlies, within noiseFork perf.9 250, +33.0% vs its upstream base
AgentsRange

When upstream merges a fix, the lines meet.

Upstream nightlyPerformance forkNoise bandRange of attempts green = better, red = worse, beyond the noise band

As agents grow.

Agent turns that finished inside the 300 s measurement window, all agents together. Measured only where the level is not saturated: a turn counts when it starts in the window and ends by its close, and at a saturated level turns outlast the window, so there the value is null with the reason not_measured_saturated and agents served is the measure.

Higher is better. Noise band 22.4% on Linux x64 (n = 6), so a change counts beyond 27.4%.

Unit
n
Better
higher
Noise band
22.4%, threshold 27.4% (n = 6)
Recorded from
17 Oct 2026; before that the series is a gap, never a zero
Platforms
Linux x64
Lane
Every record, inside the agent ladder
Upstream builds
Every upstream build
Source
result.json: metrics.driver.calls["session.turn"].count

Turns completed in the window (higher is better) at 1 to 250 agents, Linux x64

against
Records, 50 agents
DateTagMedianMin to maxValidvs up to 7 previousVerdict
18 Nov 2026v0.1.11-nightly.20261118.4903188188 to 1893−1.1%within noise
17 Nov 2026v0.1.11-nightly.20261117.4851177169 to 1873−7.3%within noise
16 Nov 2026v0.1.10-nightly.20261116.4800191155 to 2003−8.0%within noise
15 Nov 2026v0.1.10-nightly.20261115.4749190187 to 1913−15.6%within noise
14 Nov 2026v0.1.10-nightly.20261114.4709189142 to 1923−16.0%within noise
13 Nov 2026v0.1.10-nightly.20261113.4656140136 to 1623−37.8%regressed beyond noise
12 Nov 2026v0.1.10-nightly.20261112.4606207206 to 2103−8.0%within noise
11 Nov 2026v0.1.10-nightly.20261111.4544220220 to 2313−2.2%within noise
10 Nov 2026v0.1.9-nightly.20261110.4491223222 to 2243−0.9%within noise
9 Nov 2026v0.1.9-nightly.20261109.4443224223 to 2243−0.4%within noise
8 Nov 2026v0.1.8-nightly.20261108.4380226226 to 2313+0.4%within noise
7 Nov 2026v0.1.8-nightly.20261107.4340229227 to 2303+1.8%within noise
6 Nov 2026v0.1.8-nightly.20261106.4288227223 to 2283+0.9%within noise
5 Nov 2026v0.1.8-nightly.20261105.4234226226 to 2273+0.4%within noise, thin reference
4 Nov 2026v0.1.8-nightly.20261104.4192225225 to 22630.0%within noise, thin reference
3 Nov 2026v0.1.7-nightly.20261103.4138225225 to 2263no reference yet
2 Nov 2026v0.1.7-nightly.20261102.4093224221 to 2253no reference yet
1 Nov 2026v0.1.7-nightly.20261101.4048222218 to 2233no reference yet
31 Oct 2026v0.1.7-nightly.20261031.4004224223 to 2243no reference yet
30 Oct 2026v0.1.6-nightly.20261030.3940223223 to 2243no reference yet
29 Oct 2026v0.1.6-nightly.20261029.3892217215 to 2173no reference yet
28 Oct 2026v0.1.4-nightly.20261028.3841216216 to 2173no reference yet
27 Oct 2026v0.1.4-nightly.20261027.3796216215 to 2163no reference yet
26 Oct 2026v0.1.3-nightly.20261026.3740214212 to 2163no reference yet
25 Oct 2026v0.1.3-nightly.20261025.3682218217 to 2213no reference yet
24 Oct 2026v0.1.3-nightly.20261024.3643219216 to 2203no reference yet
23 Oct 2026v0.1.3-nightly.20261023.3592217216 to 2183no reference yet
22 Oct 2026v0.1.2-nightly.20261022.3541221220 to 2213no reference yet
21 Oct 2026v0.1.2-nightly.20261021.3499224223 to 2253no reference yet
20 Oct 2026v0.1.1-nightly.20261020.3438222222 to 2233no reference yet
19 Oct 2026v0.1.1-nightly.20261019.3373211210 to 2143no noise band yet, no verdict
18 Oct 2026v0.1.1-nightly.20261018.3335217215 to 2193no noise band yet, no verdict
17 Oct 2026v0.1.1-nightly.20261017.3272216212 to 2193no noise band yet, no verdict
Ladder as a table
AgentsNightly 18 NovFork perf.9
12525
50188250
100–415
150–608
200–639
250–785
Download this series