# Issue filter retry proof

Real iOS app, real server, real WebSocket RPC, synthetic GitHub responses at the server fetch boundary. No product code was changed. Two projects are created through the server API, with synthetic acme/healthy and acme/flaky git remotes. No real GitHub requests or quota are needed.

The fixture returns available-label from healthy. Its description contains the actual per-process IssueCandidates request number. Flaky fails once, then holds each subsequent request for 180 seconds. The driver taps Retry twice before that request completes. The final request ledgers are persisted before assertions.

Run from current main with scenario support and dependencies installed. Place the scenario in scripts/scenarios and the fixture in your chosen scratch directory. Portable copies replace only the original scratch path with ISSUE_FILTER_PROOF_DIR.

```bash
export ISSUE_FILTER_PROOF_DIR="$(mktemp -d)"
# Copy github-fixture-portable.mjs into this directory.
export NODE_OPTIONS="--import=$ISSUE_FILTER_PROOF_DIR/github-fixture-portable.mjs"
export GH_TOKEN=synthetic-fixture-token
export T3CODE_DEV_SEED=0
export T3_SCENARIO_INHERIT_ENV='NODE_OPTIONS,GH_TOKEN,T3CODE_DEV_SEED,ISSUE_FILTER_PROOF_DIR'
vp run scenarios:prove scripts/scenarios/issue-filter-selective-retry.e2e.ts --base 3e1b7c4e93d7f3d035d14296ebf45e373914ae8c --head 3969f93e7ab0d512b4acd44fa4f5d1e47dcce4e1 --upload-command "ghdrop upload"
```

The execution used actionTimeout: 120000 in scripts/e2e.config.ts because a pooled device install exceeded the default 30 seconds. Both final sides ran the same driver. Base: healthy 3, flaky 3. Head: healthy 1, flaky 2. scenarios:prove exited 0.

Videos are unaltered screen recordings trimmed to the same sequence: switch to Issues, open label filter, initial partial failure, Retry, Retry while still running. Evidence PNGs add a clearly labeled ledger panel beside the raw app screenshot; the visible request number in the app comes directly from the synthetic label description.
