Before 08a463e299 | After 74ef8c4ced. Fresh isolated checkouts and databases; synthetic data only. Every strict pair uses this same revision pair. No product derivation source was instrumented.
What else changed: the complete list per row is below. No unrelated visible differences are accepted in strict comparisons. Live provider demonstrations have independent durations and are explicitly excluded from strict comparisons.
Environment: Headless Chromium 156 Linux via agent-browser, no device emulation, DPR1/font scale100%, W1440×900 dark and N390×844 light, reduced motion, UTC; app glassOpacity100 and side panels closed. Native iPhone17Pro/iOS27.2 on simbox-1, logical402×874, DPR3/fontScale1/light, screenshots402×874@1x, video1206×2622; OS Reduce Motion and Reduce Transparency enabled, America/Los_Angeles. One native simulator; entire status bar must match across each strict pair. Both stills must be identical to a second frame at least500ms later.
Harness scope: R1 deterministic and R2–R6 run real unchanged derivation code and real MessagesTimeline/ThreadFeed consumers with fixed projected items. R7 runs real GuestThread with the original shared synthetic R1 projection and Date frozen before hooks initialize (24h expiry). These labelled harnesses do not prove provider orchestration, guest authentication, transport, or pagination. Real R1 provider demonstrations are retained separately below.
Diff method: max RGB-channel delta >0 (Diff0), or >16/255 (DiffT16). Every changed pixel is saturated pink, unchanged pixels dimmed to22%. Whole surfaces, no alignment, masks, normalization, or crop. Click an image for the original full-resolution PNG.
Unverified additional static states
R1 native recollapsed: Differences outside the duration label; no accepted replacement; R1 native reloaded: Differences outside the duration label; no accepted replacement; R2 native settled-expanded: OS status bar mismatch; no accepted matching-clock replacement; R2 native settled-batch-open: OS status bar mismatch; no accepted matching-clock replacement. Paired recordings remain available for these interactions; rejected stills are not presented as comparable pairs.
Every visible difference: Historical A duration 2m20s → 3.0s. On web, the settled late card moves inside A’s fold, and the Finished notification/answer below reflow upward. On mobile the card was already folded: only the historical duration changes. Initial latest and resumed-live states are exact controls.
web · 1440×900, dark, DPR1, Chromium
Scripted interactions and command-completion timestamps
Times below are relative to record-start completion. Remote native command completion may lag its visual action; the recording itself is authoritative.
before · 08a463e299
4.57s find role button click --name Resume child
7.68s find role button click --name Complete child and notify
10.96s find role button click --name Worked for 2m 20s
14.87s find role button click --name Worked for 2m 20s
17.98s reload
22.48s find role button click --name Toggle main sidebar
22.59s find role button click --name Resume child
22.69s find role button click --name Complete child and notify
25.15s find role button click --name Worked for 2m 20s
28.29s find role button click --name Worked for 2m 20s
after · 74ef8c4ced
4.28s find role button click --name Resume child
7.64s find role button click --name Complete child and notify
11.02s find role button click --name Worked for 3.0s
14.26s find role button click --name Worked for 3.0s
17.38s reload
22.56s find role button click --name Toggle main sidebar
22.68s find role button click --name Resume child
22.75s find role button click --name Complete child and notify
26.07s find role button click --name Worked for 3.0s
29.15s find role button click --name Worked for 3.0s
Scripted interactions and command-completion timestamps
Times below are relative to record-start completion. Remote native command completion may lag its visual action; the recording itself is authoritative.
before · 08a463e299
0.56s scroll top
9.61s click label="Resume child" --settle
10.22s scroll top
18.75s click label="Complete child and notify" --settle
20.09s scroll top
33.02s click label="Worked for 2m 20s" --settle
34.55s scroll top
41.51s click label="Worked for 2m 20s" --settle
42.38s scroll top
55.84s open com.t3tools.t3code.dev --relaunch --launch-url t3code-dev://expo-development-client/?url=http%3A%2F%2F100.101.192.104%3A8084&disableAutoLaunch=1&disableFab=1 --metro-host 100.101.192.104 --metro-port 8084
60.37s open com.t3tools.t3code.dev --launch-url t3code-dev://threads/41a5105a-920d-4084-b81d-4e20ee2f907c/r1-proof
66.89s open com.t3tools.t3code.dev --launch-url t3code-dev://threads/41a5105a-920d-4084-b81d-4e20ee2f907c/r1-proof
69.83s click label="Reset fixture" --settle
71.43s click label="Resume child" --settle
73.00s click label="Complete child and notify" --settle
73.71s scroll top
80.71s click label="Worked for 2m 20s" --settle
81.30s scroll top
88.23s click label="Worked for 2m 20s" --settle
88.91s scroll top
after · 74ef8c4ced
0.60s scroll top
10.16s click label="Resume child" --settle
10.92s scroll top
31.90s click label="Complete child and notify" --settle
32.78s scroll top
39.71s click label="Worked for 3.0s" --settle
40.36s scroll top
53.70s click label="Worked for 3.0s" --settle
54.47s scroll top
64.14s open com.t3tools.t3code.dev --relaunch --launch-url t3code-dev://expo-development-client/?url=http%3A%2F%2F100.101.192.104%3A8085&disableAutoLaunch=1&disableFab=1 --metro-host 100.101.192.104 --metro-port 8085
68.88s open com.t3tools.t3code.dev --launch-url t3code-dev://threads/9c4d05c3-b795-4467-90ec-1a023f1a794b/r1-proof
75.21s open com.t3tools.t3code.dev --launch-url t3code-dev://threads/9c4d05c3-b795-4467-90ec-1a023f1a794b/r1-proof
77.49s click label="Reset fixture" --settle
79.12s click label="Resume child" --settle
80.72s click label="Complete child and notify" --settle
81.26s scroll top
90.48s click label="Worked for 3.0s" --settle
91.06s scroll top
99.60s click label="Worked for 3.0s" --settle
100.71s scroll top
Every visible difference: Historical A duration on web 2m21s → 3.0s, and native 2m20s → 3.0s, while the batch is live or settled. Web keeps the entire batch visible while one child is live, then folds both after settlement. Expanded web and all native states change only the duration label.
web · 1440×900, dark, DPR1, Chromium
Scripted interactions and command-completion timestamps
Times below are relative to record-start completion. Remote native command completion may lag its visual action; the recording itself is authoritative.
before · 08a463e299
4.36s find role button click --name Worked for 2m 21s
7.64s find role button click --name Worked for 2m 21s
8.70s find role button click --name 2 subagents
11.67s find role button click --name 2 subagents
11.77s find role button click --name Settle final child
15.84s find role button click --name Worked for 2m 21s
19.26s find role button click --name 2 subagents
22.65s find role button click --name Open Research the completed task
24.33s back
after · 74ef8c4ced
4.24s find role button click --name Worked for 3.0s
7.43s find role button click --name Worked for 3.0s
8.54s find role button click --name 2 subagents
11.61s find role button click --name 2 subagents
11.74s find role button click --name Settle final child
16.11s find role button click --name Worked for 3.0s
19.20s find role button click --name 2 subagents
22.50s find role button click --name Open Research the completed task
24.17s back
Scripted interactions and command-completion timestamps
Times below are relative to record-start completion. Remote native command completion may lag its visual action; the recording itself is authoritative.
before · 08a463e299
0.57s scroll top
11.68s click label="Worked for 2m 20s" --settle
12.74s scroll top
20.04s click label="Worked for 2m 20s" --settle
40.23s click label="2 subagents, 1 working \u00b7 1 done" --settle
41.94s scroll top
52.18s click label="2 subagents, 1 working \u00b7 1 done" --settle
53.74s click label="Settle final child" --settle
54.24s scroll top
62.19s click label="Worked for 2m 20s" --settle
63.52s scroll top
71.20s click label="2 subagents, 2 done" --settle
71.74s scroll top
78.41s click @e23 --settle
83.02s back --settle
84.98s scroll top
after · 74ef8c4ced
0.97s scroll top
7.41s click label="Worked for 3.0s" --settle
8.51s scroll top
16.82s click label="Worked for 3.0s" --settle
23.33s click label="2 subagents, 1 working \u00b7 1 done" --settle
24.67s scroll top
33.44s click label="2 subagents, 1 working \u00b7 1 done" --settle
34.98s click label="Settle final child" --settle
35.57s scroll top
43.28s click label="Worked for 3.0s" --settle
43.86s scroll top
50.39s click label="2 subagents, 2 done" --settle
51.46s scroll top
58.11s click @e23 --settle
63.47s back --settle
64.93s scroll top
Every visible difference: Worked 2m20s → 20s; Continued 2m10s → 10s. Collapsed late card moves into its Continued fold, moving following content upward. Expanded states differ only in those labels.
web · 1440×900, dark, DPR1, Chromium
Scripted interactions and command-completion timestamps
Times below are relative to record-start completion. Remote native command completion may lag its visual action; the recording itself is authoritative.
before · 08a463e299
4.24s find role button click --name Continued for 2m 10s
7.40s find role button click --name Worked for 2m 20s
10.53s find role button click --name Continued for 2m 10s
11.59s find role button click --name Worked for 2m 20s
after · 74ef8c4ced
4.44s find role button click --name Continued for 10s
7.51s find role button click --name Worked for 20s
10.32s find role button click --name Continued for 10s
11.39s find role button click --name Worked for 20s
Every visible difference: With trailing work before the card, Worked 2m20s → 6.0s. With reversed order it stays 6.0s. Web folds the late card in both orders, and following content reflows. Native card membership is unchanged; reverse-order native controls are exact matches.
web · 1440×900, dark, DPR1, Chromium
Scripted interactions and command-completion timestamps
Times below are relative to record-start completion. Remote native command completion may lag its visual action; the recording itself is authoritative.
before · 08a463e299
4.27s find role button click --name Worked for 2m 20s
7.42s find role button click --name Worked for 2m 20s
8.47s find role button click --name Swap trailing work order
12.63s find role button click --name Worked for 6.0s
15.84s find role button click --name Worked for 6.0s
after · 74ef8c4ced
4.27s find role button click --name Worked for 6.0s
7.34s find role button click --name Worked for 6.0s
8.44s find role button click --name Swap trailing work order
12.33s find role button click --name Worked for 6.0s
15.45s find role button click --name Worked for 6.0s
Scripted interactions and command-completion timestamps
Times below are relative to record-start completion. Remote native command completion may lag its visual action; the recording itself is authoritative.
before · 08a463e299
0.47s scroll top
6.87s click label="Worked for 2m 20s" --settle
7.51s scroll top
15.28s click label="Worked for 2m 20s" --settle
17.13s click label="Swap trailing work order" --settle
18.73s scroll top
26.67s click label="Worked for 6.0s" --settle
27.53s scroll top
34.39s click label="Worked for 6.0s" --settle
after · 74ef8c4ced
1.50s scroll top
7.91s click label="Worked for 6.0s" --settle
9.16s scroll top
16.12s click label="Worked for 6.0s" --settle
17.77s click label="Swap trailing work order" --settle
18.31s scroll top
25.52s click label="Worked for 6.0s" --settle
26.17s scroll top
32.51s click label="Worked for 6.0s" --settle
Every visible difference: No visible change: A remains 3.0s, B remains 11s, ownership/fold positions and null-child card stay unchanged. All before/after controls are exact matches at zero threshold.
web · 1440×900, dark, DPR1, Chromium
Scripted interactions and command-completion timestamps
Times below are relative to record-start completion. Remote native command completion may lag its visual action; the recording itself is authoritative.
before · 08a463e299
4.14s find role button click --name Worked for 3.0s
7.35s find role button click --name Worked for 3.0s
after · 74ef8c4ced
4.54s find role button click --name Worked for 3.0s
8.19s find role button click --name Worked for 3.0s
Scripted interactions and command-completion timestamps
Times below are relative to record-start completion. Remote native command completion may lag its visual action; the recording itself is authoritative.
before · 08a463e299
0.59s scroll top
8.58s click label="Worked for 3.0s" --settle
10.00s scroll top
17.62s click label="Worked for 3.0s" --settle
18.40s scroll top
after · 74ef8c4ced
3.19s scroll top
66.67s scroll top
73.09s click label="Worked for 3.0s" --settle
73.85s scroll top
80.45s click label="Worked for 3.0s" --settle
80.98s scroll top
Every visible difference: Worked 2m20s → 1.0s and Continued 2m10s → 1ms; card remains under its Continued disclosure. No other changes.
web · 1440×900, dark, DPR1, Chromium
Scripted interactions and command-completion timestamps
Times below are relative to record-start completion. Remote native command completion may lag its visual action; the recording itself is authoritative.
before · 08a463e299
4.49s find role button click --name Continued for 2m 10s
7.55s find role button click --name Worked for 2m 20s
10.40s find role button click --name Continued for 2m 10s
11.46s find role button click --name Worked for 2m 20s
after · 74ef8c4ced
4.45s find role button click --name Continued for 1ms
8.18s find role button click --name Worked for 1.0s
11.17s find role button click --name Continued for 1ms
12.26s find role button click --name Worked for 1.0s
Every visible difference: Worked 11m27s → 5.2s; collapsed late card moves inside the fold, producing expected bottom-anchored scroll reflow. Expanded pair differs only in the label. Guest expiry is 24h left on both sides.
web · 390×844, light, DPR1, Chromium
Scripted interactions and command-completion timestamps
Times below are relative to record-start completion. Remote native command completion may lag its visual action; the recording itself is authoritative.
before · 08a463e299
2.01s find role button click --name Worked for 11m 27s
5.92s find role button click --name Worked for 11m 27s
after · 74ef8c4ced
2.06s find role button click --name Worked for 5.2s
5.97s find role button click --name Worked for 5.2s
R1 real-provider demonstrations (separate from deterministic comparisons)
New web runs at08a463e299/74ef8c4ced use Claude Sonnet5.5 with one async Haiku5.5 child, a completed parent, Bash LISTED, then a child resumed after over2minutes and a Finished notification. Independent durations and message timestamps differ; these are demonstrations and are not zero-diff pairs. Each recording includes resume/notification, disclosure and reload. Per-side timestamps below combine driver command completion, synthetic projection message timestamps, and RPC response file time, relative to recording start; subsecond startup uncertainty remains. These are supplementary demonstrations.
before
0.00s Initial settled parent and Bash LISTED visible
27.63s RPC resume same child with again (command complete)
37.68s assistant: PARENT OK
124.91s find ['role', 'button', 'click', '--name', 'Worked for 21s']
125.97s find ['role', 'button', 'click', '--name', 'Worked for 21s']
126.18s reload []
140.22s record ['stop']
The original real native demonstration is retained at its original revisions (before960d5761c1, after956e6f6f82). It supplies provider-flow context; all strict native comparisons above were recaptured at the new requested revisions.
