# Evidence index — issue [#5817](https://git.cafe/pingdotgg/t3code/issues/5817)

Baseline: [3b95e78](https://git.cafe/pingdotgg/t3code/commit/3b95e7831b11b17627dff16843bc087bfca397ec) (`3b95e7831b11b17627dff16843bc087bfca397ec`). Head: [51cf491](https://git.cafe/pingdotgg/t3code/commit/51cf491624be9461b77412c3e30dbe91a33d93e2) (`51cf491624be9461b77412c3e30dbe91a33d93e2`). No PR at capture.

Environment: Linux x64, Node 24.21.0, Codex CLI 0.163.0; worktree `/home/ubuntu/wt/issue-5817`. Lead ran `vp run dev --no-seed` against isolated `.t3` and the same pre-fix fixture homes. After startup: 2026-10-22 07:19 UTC, server port 14473 / web 6433. No credentials. Lead reports server subsequently stopped and ports confirmed clear; capture agent did not control server/browser.

## Evidence

- [Baseline runtime](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/2652d89c0c97c7fd/before-runtime.txt): three distinct materialization conflicts.
- [Latest sanitized after runtime](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6dd38b9058277c4f/after-runtime-sanitized.txt): real server trace excerpts, all three named provider caches, stopped-server topology and queue sentinels. Supersedes earlier uploads.
- [Red report](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/fc8aa22bd19f92bf/red-tests.txt): summarized baseline result, 2 failed / 8 passed. Earlier test commit b913ce512ebc44d664beef5cff0f7de83ce6fd92; WAL test later narrowed to preserving topology, not reconciliation. Baseline fails before assertions.
- [Current-head green output](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5a71f052c57126b1/green-tests.txt): from `apps/server`, `vp test run src/provider/Drivers/CodexHomeLayout.test.ts --maxWorkers=1`; 1 file, 11/11 passed, no skips reported, duration 5.35 s, start 08:15:42. Command/head attribution supplied by lead; output inspected by capture agent. Typecheck rerun pending; prior pass reported by lead.
- [Audited fixture matrix](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/300af90ea33159b9/issue5817-proof-audit-pass1.md).

## Required runtime fixtures

S1 — `codex-shadow`, `.t3/repro/{shared,shadow}`: baseline `.tmp` conflict; after provider cache identifies installed CLI and expected unauthenticated error, with successful materializer/driver spans. Committed repeated-materialization test creates `.tmp/runtime.lock`, installation_id, queue/goals and future_state_42.sqlite in both homes and asserts retained contents. Actual runtime transcript does not separately enumerate that entire synthetic test file list or prove repeated runtime setup.

S2 — `codex-queue`, `.t3/repro/queue-only/{shared,shadow}`: baseline queue_1.sqlite conflict; after expected unauthenticated result and successful construction spans. Stopped-server inspection shows separate regular 8192-byte queue bases with respective `shared` and `shadow` sentinel rows.

S3 — `codex-sidecars`, `.t3/repro/sidecar-only/{shared,shadow}`: baseline logs_2.sqlite-shm conflict; after expected unauthenticated result and successful construction spans. Stopped-server inspection retains shadow base link to shared base and regular WAL/SHM on each side. Committed test asserts original sidecar bytes and paths after rematerialization. This preserves existing split topology; it does not recover or merge a split WAL.

The excerpted success spans lack instanceId attributes; the cache records identify each fixture. Do not imply an individual span is independently correlated to an instance beyond the lead's captured startup context. Status `error` is expected authentication failure, not authenticated availability.

## Scope and limits

Required visual rows captured: **0/0** (explicit server-only exception; empty visual matrix). Optional screenshot omitted: no render code changed and no new browser work authorized in this capture. No Windows/macOS execution, authenticated turn, adapter session launch, authenticated managed-runtime path, or per-client/remote/relay verification. No claim that SQLite split-WAL state recovered. Historical SHA-256 checks found changed live SQLite files, so runtime full byte identity is unverified; controlled tests supply preservation assertions.

Only CodexHomeLayout.ts and its committed test differ from baseline (221 insertions, 13 deletions); no UI/render code, contracts, routes or styles changed. Web/desktop/mobile consume the unchanged server result; other providers unaffected. Capture agent made no code edits, commits, forge comments, push/merge, or nested agents. Artifacts are outside the worktree and uploaded with the ghdrop skill.

## FRICTION

- tool: some terminal calls yielded only after several minutes; switching to non-login shell helped. No server/browser actions used to work around this.
- brief: current green report arrived during capture; incorporated. Typecheck rerun remains pending with lead.
- runbook: matrix reports missing EVIDENCE.md; no additional unobserved requirements inferred.
- infra: lead reports VM/T3 restart canceled earlier runs and resumed children; no duplicate test/server work started here.
- forge | policy | model | other: none affecting this packaging step.