# Reproduction environment

Web client in Chromium 157 on Linux, light theme, device scale factor 2.
Each commit ran with `T3CODE_DEV_SEED=0 vp run dev` from a separate worktree,
with its own fresh database. No live database was copied.

On both sides, pair with the dev server before capturing. The fresh app creates
its default `server` project and empty `New thread`. Add an empty synthetic
Git directory named `sidebar-demo` through Add project > Local folder > Add.
Run `window.t3dev.replay('todo-list')`, which in this draft state selects the
same default `server` project on both builds. Wait until the replay finishes
with `todo list fixture complete`. Close the thread details panel so the
capture contains no environment name. No paid provider turn is needed.

Run `runEvidence` from resize-evidence-driver.js with the paired tab, side,
commit, and a new timestamped output directory. The driver is unchanged
between sides and uses real browser pointer drag, double-click, resize, and
reload operations. It reads widths without modifying app state or code.
The pointer drag chooses 353.5 CSS px, approximately the requested 350 px.
Wide viewport: 1638 x 900 CSS px. Narrow viewport: 840 x 900 CSS px.
All stills were accepted only after two identical PNG captures at least
500 ms apart. Recordings include normal reload splash screens followed by
fully loaded app states. Black margins during the narrow viewport come from
recording a changing viewport on a fixed canvas; they are not app content.
MP4s preserve the full interaction and add an environment label above it.

Observed widths after widening:

| Scenario | Before | After |
| --- | --- | --- |
| Saved width, narrow reload | 208 px | 353.5 px |
| Reset while narrow | 208 px | 256 px |
| Default width, narrow reload | 208 px | 256 px |

Relative ages and replay duration labels differ because the two real servers
ran separately; project and conversation content are the same synthetic data.

AI-assisted; verified with the tests and tools listed above.
