No blocking proof findings for [#25696](https://git.cafe/pingdotgg/t3code/pull/25696).

Reviewed after [20d22421d5](https://git.cafe/pingdotgg/t3code/commit/20d22421d54cce014bfacdeb9d86de83a00a1e4b) against before [d73fd1279a](https://git.cafe/pingdotgg/t3code/commit/d73fd1279a9cbf3ef4677b88b4ad82f0a55583c7). The entire diff is seven membership expressions and one regression, with no other source changes.

Read the evidence index and all 32 linked artifacts. Every acceptance transcript identifies the proper pinned source. Repeated migration blocks and repeated JSON objects were read through lossless deduplication; no result fields were excluded. Both full parity dumps were parsed, checked byte for byte, and their complete distinct object values and row memberships inspected. Every acceptance failure is an expected baseline SQL failure or the deliberate typed-corruption probe. Discarded fixture failures remain explicitly superseded and are not counted as acceptance evidence.

| Row | Status | Verified before/after evidence |
| --- | --- | --- |
| R1 | SUFFICIENT | 32,766 idle threads; service reads and authenticated HTTP/WS succeed on both sources. Exact ID counts, uniqueness and digests agree. |
| R2 | SUFFICIENT | 32,767 idle threads; baseline causal SQL failure, head full delivery. Regression transcript: baseline 1 failed / 4 passed / 42 filtered; head 47 passed, exit 0. |
| R3 | SUFFICIENT | Baseline both HTTP requests return 500; head returns 200 and 32,767 unique IDs, with omission flags off/on. Body sizes 46,890,032 and 46,890,114 match index. |
| R4 | SUFFICIENT | Baseline initial WS causal SQL failure. Head initial snapshot at sequence 2; live and one replay at 3; fallback retains all IDs; omitted-marker live update at 4, no synchronized marker. R1 validates the control probe on base. |
| R5 | SUFFICIENT | 1 active + 32,766 archived + deleted sentinel. Only baseline unscoped read fails; head preserves exact split, scoped reads match. |
| E1 | SUFFICIENT | Empty, deleted-only and unmatched project/family/worktree cases return empty snapshots on both. |
| E2 | SUFFICIENT | Explicit empty IDs produce an empty snapshot in complete dumps on both. |
| P0 | SUFFICIENT | 25 thread rows, 13 runs, 347 items, four providers; complete rich shells match, with nonempty fork/count/provider/pending/run-state assertions. |
| P1 | SUFFICIENT | 23 active threads, archived/deleted excluded; complete results match. |
| P2 | SUFFICIENT | One archived thread, active excluded; complete results match. |
| P3 | SUFFICIENT | Project A 22 / project B 1 / missing 0; scoped shells contain no ancestor leakage. |
| P4 | SUFFICIENT | False=23 / true=21; settled timestamp and explicit settled override excluded correctly. |
| P5 | SUFFICIENT | Root and descendant each select the same five members; deleted root retains live child; missing family empty. |
| P6 | SUFFICIENT | Shared exact path selects child/other/quoted ID/root; missing path empty. |
| P7 | SUFFICIENT | One, duplicate, missing, quoted/backslash/non-ASCII and archived IDs; intersecting filters select root, disjoint filters empty. |
| K1 | SUFFICIENT | Single-shell versus snapshot equality for fork/nested/missing/deleted targets/sources and seven run states. Imported history and rolled-back-source controls covered. |
| H1 | SUFFICIENT | 16,383 pairs succeed on both; 16,384 pairs fail only before. File-change plus thread-wait item exercises second-arm blocked-state behavior. Independently rebuilt. |
| H2 | SUFFICIENT | Corrected 32,761 success / 32,762 failure on base; requested 32,765/32,766 failures retained; all four pass on head. Source has five compaction-text bindings and no active-location binding. |
| S1 | SUFFICIENT | 32,767 candidates fail only before; small complete candidate results, scalar/empty/exclusion/provider/pending cases match. |
| X1 | SUFFICIENT | Ordered read/insert/decode returns frozen=1, fresh=2; selected corruption remains typed; outside-scope corruption excluded. |
| O1 | OPTIONAL, UNVERIFIED | Distinct-fork-source stress explicitly omitted; not a minimum gate. |
| N1 | N/A | All production count/provider/pending-reader call sites supply lists; dormant undefined branches need no runtime fixture. |
| N2 | N/A | No UI or transport branch changed; no visual/native/render-performance claim. |

Rebuilt the entire uploaded byte-identical service-harness.mjs on both sources using /home/ubuntu/.local/share/vite-plus/js_runtime/node/26.12.1/bin/node. Both runs exited 0 on fs-20, Linux 6.1.102, SQLite 3.54.0, MAX_VARIABLE_NUMBER=32766. Exact command arrays are in each run's exit.json. The driver hash is 688bafdc26910378aea0c0c0f54c7930881f3b8e3d6a1629b1cadc0b2459b1c1, matching both delivered transcripts.

Rebuilt captures:

- [Before transcript](rebuild-node26-before-20261027T161015/transcript.log), [raw delta](rebuild-node26-before-20261027T161015/raw-transcript.diff), [nonmetadata delta](rebuild-node26-before-20261027T161015/nonmetadata-transcript.diff).
- [After transcript](rebuild-node26-after-20261027T161015/transcript.log), [raw delta](rebuild-node26-after-20261027T161015/raw-transcript.diff), [nonmetadata delta](rebuild-node26-after-20261027T161015/nonmetadata-transcript.diff).
- [Comparison](rebuild-comparison.json): zero nonmetadata delta on either side. Raw deltas contain only scratch invocation paths and migration log timestamps; even fiber IDs match. No assertions/results/error causes were filtered. Both rebuilt parity dumps exactly match the delivered 441,095-byte dumps, SHA-256 efa2b0efbcb1e901bee55fc7f434878b759f8ec10a070f679a6f7751dc9b6906.

All four supplied diffs (service-results.diff, app-r1.diff, app-r2.diff, regression.diff) exactly reproduce from their supplied inputs. The empty parity diff is independently confirmed. The what-else-changed list is consistent: service parity has no differences; R1 changes are invocation/process/port/fixture metadata and byte lengths; R2 changes are the intended SQL failure-to-success transition. The uploaded head test file is byte-identical to the pinned source, with the stated hash. The nonfatal-startup, 47-passed and corrected H2 claims have direct supporting artifacts.

Credential scan and content inspection of all uploads found no credential values, pairing URLs, tickets, or non-synthetic payloads. The one wsTicket match is variable construction in app-driver.mjs, not a ticket value. The app driver creates its own fresh home/database; the service driver uses memory persistence. Uploaded fixtures and parity data are synthetic. The index's 102 tracked PIDs recompute from final app transcripts plus the discarded initial-launch cleanup; none remains in /proc.

No real dev server, browser or provider was started during this audit. Both worktrees remain clean at the pinned SHAs, neither has .t3, and 31 GB remains. See audit-environment-cleanup.json and audit-download-hashes.json. No source edits, pushes, PR/forge writes, or comments. Audit completed within the 60-minute budget; all mandatory rows covered, O1 intentionally not rebuilt.

FRICTION:
- tool | Python urllib downloads returned HTTP 403; curl successfully fetched every artifact | use the supported curl path for these upload URLs.
- tool | The cwd-sensitive node shim selected Node 24.22.1 / SQLite 3.53.4 for the first replay; those output directories are retained as discarded audit runs. Both sources were rerun successfully with the exact delivered Node 26 executable | pin the absolute executable before spawning from apps/server.

VERDICT: SUFFICIENT