Skip to content

Commit 96d4951

Browse files
thymikeethymikee
committed
evidence(ios-snapshot): benchmark results measured at 7c434b5
The four corpora closing the #2188 gate: cold/cold-cold, first-interaction, warm/relaunch with the package-size leg, and the 0/20/80 ms proxy matrix. All legs completed with a clean revision and no failed samples on bench-2188-final (iOS 26.2). README.md states the target, runtime, host-load, app, and harness deviations from the 71fb248 baseline.
1 parent 3837098 commit 96d4951

5 files changed

Lines changed: 19650 additions & 1 deletion

‎README.md‎

Lines changed: 38 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -61,13 +61,50 @@ proof, regression records, and checksums. The immutable tag is
6161
`evidence/ios-snapshot/ae26f7afc0`. Timings are bounded observations under
6262
uncontrolled host load, not a general performance guarantee.
6363

64+
## Convergence final corpus (#2188 gate)
65+
66+
Raw results closing the exact-head evidence sweep that gated #2188 on #2199, measured at
67+
repository commit `7c434b575837e3291c51315bf9bb8b54c8ce7568` on `bench-2188-final`.
68+
The cold and first-interaction legs carry 10 samples per cell, warm/relaunch 20, and the
69+
proxy leg 20 per screen at each of 0, 20, and 80 ms of added RTT. Every leg completed with
70+
`revision.dirty: false` and no failed samples. The package-size leg ran inside
71+
`ios-snapshot-warm-relaunch-local-7c434b575.json`.
72+
73+
Deviations from the baseline corpus above, stated in full:
74+
75+
- Target and runtime: `bench-2188-final` on
76+
`com.apple.CoreSimulator.SimRuntime.iOS-26-2`. The baseline ran on `bench-golden-v2` with
77+
iOS 27.0, whose runtime was not installed on this host at measurement time.
78+
- Host: same machine model, shared with other agent sessions. The 1-minute load average
79+
sampled during the legs was 4.2-31.2 for warm/relaunch, 6.4-30.3 for proxy, and 5.3-96.7
80+
for cold. Timings are bounded observations under uncontrolled host load, not a general
81+
performance guarantee.
82+
- Fixture app: the CI-built `AgentDeviceTester` for this lineage with the measured head's own
83+
JavaScript bundle repacked in. Its trees differ from the baseline app (the catalog screen
84+
exposes 31 nodes here against 35 at the baseline), so response sizes are not comparable
85+
cell for cell.
86+
- Harness: the measured head is one benchmark-only commit ahead of `45e4c594a1824fc8152c5cce41fe1fac50a91bf0`,
87+
which put a bounded anchor wait into the untimed setup admission. No `src/`, `packages/`,
88+
app, or CLI runtime file differs between the two commits.
89+
90+
| File | sha256 |
91+
| --- | --- |
92+
| `ios-snapshot-cold-local-7c434b575.json` | `4663897ee5104569ad54e2ac803c216b284280c1c70fd82a8b2cf7b675d8a8bd` |
93+
| `ios-snapshot-first-interaction-local-7c434b575.json` | `87c686336f5581e3f18111e160cf7b733cd726b41e79ed6d8e5b53e2ab40c3fb` |
94+
| `ios-snapshot-warm-relaunch-local-7c434b575.json` | `d49df3c3c943178f016a2b958449b257c6d46a44c8ffdf8fab75c1634ae1ebce` |
95+
| `ios-snapshot-proxy-7c434b575.json` | `5b5353831851f3a0f60e19d6bfd0cf50db47c3c4db52a283024c10bb17e71573` |
96+
6497
## Fetch into a checkout
6598

6699
```sh
67100
git fetch origin evidence/ios-snapshot
68101
for f in ios-snapshot-cold-local-71fb2483f.json \
69102
ios-snapshot-warm-relaunch-local-71fb2483f.json \
70-
ios-snapshot-proxy-71fb2483f.json; do
103+
ios-snapshot-proxy-71fb2483f.json \
104+
ios-snapshot-cold-local-7c434b575.json \
105+
ios-snapshot-first-interaction-local-7c434b575.json \
106+
ios-snapshot-warm-relaunch-local-7c434b575.json \
107+
ios-snapshot-proxy-7c434b575.json; do
71108
git show FETCH_HEAD:$f > scripts/ios-snapshot-benchmark/evidence/$f
72109
done
73110
shasum -a 256 scripts/ios-snapshot-benchmark/evidence/*.json

0 commit comments

Comments
 (0)