# Evidence index — issue 20534

Issue: [#20534](https://git.cafe/pingdotgg/t3code/issues/20534); [mirror #20534](https://github.com/pingdotgg/t3code/issues/20534). No PR created by this evidence step.

Scope: `/home/ubuntu/wt/issue-20534`, branch `fix/issue-20534`, base [70e5e384da](https://git.cafe/pingdotgg/t3code/commit/70e5e384da), verified head [a623c6b56b](https://git.cafe/pingdotgg/t3code/commit/a623c6b56b). Worktree clean.

- [Proof audit and empty render matrix](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/23fdf126d09ea535/issue20534-proof-audit-pass1.md): **0/0 capture rows**. Server-only unused SQLite index removal; no render, interaction, animation, or timing path changes. No visual capture or render fixture script is applicable. The committed migration regression supplies state proof.
- [Runtime and red/green transcript](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/17571b7b9ed0571e/issue20534-runtime.txt): supplied red result at [5dedf455d1](https://git.cafe/pingdotgg/t3code/commit/5dedf455d1) fails the index-absence assertion (1 failed, 7 passed); supplied green result at the fixed head reports 33 passing tests, targeted lint, and server typecheck. These execution results were not independently rerun in this evidence-packaging step.

## Independent validation

Read the complete four-file diff (52 added lines), runtime transcript, canonical before/after logs, proof audit, and synthetic benchmark source. Production changes only register migration 68 and execute `DROP INDEX IF EXISTS idx_orch_events_correlation_id`; the other changes are regression coverage and preview migration expectations. No client, contract, provider, render, or interaction code changed.

Opened `/tmp/issue20534-canonical/userdata/statev2.sqlite` using SQLite URI `mode=ro`. Fresh query confirms migration 68 `DropOrchestrationEventCorrelationIndex`, no correlation index, eight remaining event indexes (matching the before log except for the removed index), and four events with four non-null correlation values. The before log records canonical migration 67, zero events, and a 4096-byte correlation index. Historical value preservation is asserted by the committed seeded regression, not by this initially empty runtime fixture.

Environment: Linux x86_64, kernel 6.1.102. Supplied lead runtime used Node 24.21.0, node:sqlite, isolated `vp run dev --home-dir /tmp/issue20534-canonical`, and a paired Chromium web client at 1280×800 to create a thread without starting a provider turn. The lead reports its owned dev server stopped. This packaging step launched no browser, server, or simulator, and changed no repository or live data. It only read the isolated database and wrote this temporary public evidence index.

## Limits

Large populated canonical migration timing was not measured. Populated copied `.t3` data contains fork migrations 68–70 and is reproduction context, not canonical upgrade proof. Synthetic benchmark results are noisy: no percentage speedup, end-to-end latency improvement, reporter database size reproduction, or immediate filesystem shrink is established. Native clients and remote/relay/tunnel modes were not independently exercised by this packaging step. Server scope is supported by migration/state evidence; broader performance claims remain unverified. No extra fixture scripts were committed because there are no affected render/capture rows.

## FRICTION

category|what happened|avoidance
---|---|---
fixture provenance|Copied populated state carries fork migrations 68–70.|Use canonical 67-to-68 runtime evidence and the seeded regression for historical-value preservation.
evidence provenance|Red/green and earlier browser execution are supplied transcript results.|Label them as supplied; independently query canonical SQLite read-only without restarting services.

RESULT: Evidence packaged; 0/0 visual rows, no render changes, no forge writes, pushes, or commits.