# Issue 25172 — server-only evidence index

Visual matrix: EMPTY, 0/0 captures required/completed. No figures required. [Prior surface/proof audit](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/22182801bbd9e6a2/issue-25172-proof-pass1.md).

## Revisions

- Requested comparison base: `5fee09a2c109fa973294c1d5fc3729bfd1a7f093`.
- Actual before runtime: `0ad399c6c5fc521f26e42cfcd50ec056390e4870` (different from requested base).
- Reviewed implementation / after runtime: `15b7291404409de78da6a7f6c46f1c72bf4dfa87`.
- Final head: `57a7bd0a8dda42029eab327ecdac478dce334cae`.

Local git comparison confirms PullRequestSyncReactor.ts and PullRequestSyncReactor.test.ts are byte-identical between reviewed implementation and final head; those are the only changed files relative to requested base. Final checkout HEAD matches requested final head and status is clean.

## Before / after

| Measurement | Before | After |
|---|---|---|
| Fixture | 31 stopped projections: 30 synthetic + development bootstrap | 31 synthetic stopped projections |
| Shared workload | One shared open GitCafe PR, active settlement override | Same reported shared PR/project and override |
| Completed sweep 1: retained groups Maps after GC | 1 | 0 |
| Completed sweep 2: retained groups Maps after GC | 2 | 0 |
| Branded-link regression | Expected 0, actual 30 (rerun red; linked output) | 49/49 reactor tests green on final head (recorded suite output) |

Counter rerun independently against open-before-1, open-before-2, after-1 and after-2 local snapshots reproduced 1, 2, 0, 0. Provenance rerun on open-before-2 found both groups-owning contexts containing scope, nowMs, nowIso, groups, linkedThisSweep, stacksReadThisSweep, persistence, syncEntry, syncGroup, skips and readGroup, matching reactor sweep locals. Two fixed sweep telemetry records report Success with original tracing span names preserved.

## Safe evidence

- [Red regression command and failure](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/00d7a5b7c57bc049/red-test.txt)

- [Runtime transcript](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/1611d7266ea54e16/runtime.txt)
- [Final test output](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/659d1bb5e20ab840/final-tests.txt)
- [Fixed sweep records](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/be4d0a2295e3f6db/fixed-sweeps.json)
- [Validated counts](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/93ea7c37d8af8944/validated-counts.json)
- [Counter script](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/bcf672282b1e4e51/count-groups.py)
- [Provenance script](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/42e34ad4fa0a628b/provenance.py)
- [Fixture script](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c0fe5939fe97d054/seed.py)
- [Provenance output](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/bab95e6eab81be61/validated-provenance.txt)

## Unverified limitations / FRICTION

- Lead reran the final regression against the baseline implementation: expected 0 / actual 30, exit 1, with exact invocation and output linked below. Production file was restored byte-for-byte afterward; git status clean. Final green output records its invocation and 49/49. check:changed success is reported in runtime.txt; all scoped stages passed.
- Actual baseline revision differs from requested base; no claim of a runtime capture at requested base. After runtime was captured at implementation revision, with byte-identical server files at final head.
- Migration-history mismatch prevented real-data auto-seeding. Synthetic fixtures differ in composition as disclosed above; neither reproduces the reporter's full 18,736-link workload or establishes a matching GB/hour growth rate.
- Generic groups-Map counting plus matching context locals supports attribution; this is not a complete retaining-path audit. Raw heaps remain local and unchanged because they may contain credentials.
- Server lifecycle, fixture contents during capture, sync work and full-GC behavior are supplied lead evidence; this evidence pass launched no servers/browser and reran no tests. Sweep Success supports completion but no independent persisted projection-output inspection was performed here.
- Initial web fetch of prior audit failed; read-only curl succeeded. No upload/setup blocker. Only safe text, JSON and scripts were uploaded using ghdrop; no raw heaps, credentials or source databases.
- Repository unchanged; no push, forge write, merge, watch or server/browser launch.