Gantry · stats

19 codebases 282 runs archived

409 milestones 2.4k sprints

13k agent sessions

457 plan 2.1k build 108 fix

66.4M tokens out 34k edits

105k turns 180k commands

Execute-agent peak context

2,116 build sessions - exact per-session peaks

The main goal of Gantry is to break work down into chunks small enough for agents to execute without running up large context windows. The key metrics are about context windows, and we focus on the peak number: the largest size the context window reached at any time during the session.

Across 2,116 autonomous execute sessions, the peak context reached stays under the 150k design target in 76% of them - and 89.9% stay under 200k. Only 0.9% of sessions (18 of 2,116) peaked over 300k.

Under 150k target

76%

1,599 of 2,116 sessions

Under 200k

90%

only 213 sessions above

Median peak

109k

across all execute sessions

Over 300k

0.9%

18 sessions

claude execute 861 sessions

0

25

50

75

100

125

150

175

200

225

250

275

300+

150k 200k

72%< 150k

114kmedian

13%> 200k

2.1%> 300k

542kmax

codex execute 1,255 sessions

0

25

50

75

100

125

150

175

200

225

250

275

300+

150k 200k

78%< 150k

106kmedian

8%> 200k

0%> 300k

246kmax

Each histogram is that harness's own execute sessions, bin width 25k, final bar 300k+; bar height is scaled to the busiest bin in that harness.

Supporting roles

the aside - non-execute agents

Execution is the most important type of agent run and the most context-constrained. The bottleneck we are targeting. Here are the other roles and their context peaks.

Per-role median by harness

Role	claude	n	codex	n
execute	114,468	861	106,464	1,255
plan	82,376	206	47,406	251
review	62,831	871	74,498	1,385
fix	58,987	26	57,826	82
replan	54,470	185	40,577	170
milestones	43,867	76	28,590	122

Median over all roles: 76,673. claude peaks higher than codex in every role except review.

Agent runtime

active working time per run and agent runtime by role

How long runs and agents take. The whole-run figure is active time: stopped time is excluded, and parked time is excluded where the journal records it. The harness columns are each harness's median; median, mean and p90 pool both.

Runtime per agent role

Role	claude	codex	median	mean	p90	n
execute	9m 4s	6m 3s	6m 57s	9m 43s	18m 56s	2,113
plan	4m 51s	2m 6s	2m 47s	3m 32s	6m 5s	457
review	1m 44s	2m 4s	1m 55s	2m 30s	4m 17s	2,218
fix	3m 39s	2m 3s	2m 11s	3m 13s	6m 40s	106
replan	2m 16s	1m 13s	1m 45s	2m 24s	4m 17s	229

Sprint runtime median

10m 38s

2,308 sprints - p90 25m 55s

Sprint runtime p25-p75

7m 2s - 16m 26s

execute + test + review per sprint

Active working median

1h 39m

whole-run active time

Active working mean

3h 24m

average across measured runs

Active working p90

6h 47m

90th percentile active time

Active working sample

282

runs with active duration

Stopped human waits are excluded for every active-duration observation. Parked retry waits are excluded for 0 newer runs; 282 older runs have stop-corrected active time only because their journals do not record parked waits.

Decomposition

269 structured builds

Gantry sequences work by modelling it as milestones and sprints, and dispatching agents to implement sprint by sprint, and dispatching agents to do planning on multiple levels. Here is how the milestone-sprint metaphor ends up slicing the work.

Sprints / milestone

4.5

median - p25-p75 3.7-5.3

Milestones / job

3

median of the 144 jobs that split into milestones

Sprints / job

7

median across all 269 builds - max 57

How a plan breaks down

Metric	min	median	mean	p90	max
Milestones / multi-ms job n=144	1	3	2.8	4	8
Sprints / job n=269	1	7	8.8	18	57
Sprints / milestone n=144	1.3	4.5	4.6	6.0	12.0

Milestones per job

269 structured builds, bucketed by milestone count. A flat build is one whose sprints are a single list, with no milestone layer above them.

flat 119

1 ms 6

2 ms 60

3 ms 47

4 ms 19

5 ms 6

6+ ms 4

Sprints per job

269 structured builds, bucketed by total sprint count.

1–2 sp 35

3–5 sp 86

6–10 sp 63

11–15 sp 36

16–25 sp 30

26–40 sp 9

41+ sp 2