fix(results): harden remote sync dogfood

[christso]

Summary

Remote result sync is now safer under adversarial production-readiness scenarios: offline/auth-style sync failures keep reporting cached remote runs, and /api/runs no longer returns duplicate local plus remote:: rows for the same synced run.

The durable av-fis dogfood report, throwaway-remote evidence, Dashboard screenshots, and production-readiness verdict are stored outside this public repository:

• Private repo: EntityProcess/agentv-private
• Private branch: dogfood/av-fis-remote-sync-evidence
• Private commit: 5d81c8d docs(dogfood): add av-fis remote sync evidence
• Private path: dogfood/av-fis/2026-06-08-remote-sync/

This public PR contains only product/test code changes.

Dogfood Coverage

Covered scenarios include:

• Remote tag add/edit/clear overlays, sync persistence, dirty-state behavior, and clean-clone verification.
• Local/remote discovery, local delete after sync, empty remote, missing metadata, corrupt index.jsonl, branch/default-branch mismatch, offline/missing remote, and multiple project isolation.
• Local combine/delete lifecycle, remote combine/delete rejection, safe dirty metadata push, unsafe dirty blocks, behind/ahead/diverged/conflicted states, interrupted retry simulation, and concurrent sync guarding.
• Dashboard/browser paths plus CLI-adjacent paths: eval auto-export, agentv results combine, agentv results delete, and confirmation that no direct agentv results remote sync/status command currently exists.

No production remote was touched. All remote dogfood used throwaway /tmp file remotes and temporary clones.

Fixes

• syncRemoteResults() now computes run_count from the cached results clone in its error path, so a failed fetch no longer makes available cached remote runs appear empty.
• listMergedResultFiles() now dedupes merged local/remote run rows by canonical raw_filename, preferring the local copy before sorting and pagination.
• Added focused serve.test.ts regressions for both behaviors.

Follow-Ups

• av-fis.1: Dashboard sync button can remain Syncing... after concurrent remote sync; data path is safe, presentation can be misleading until reload.
• av-xqm: Dashboard remote sync status can show stale last_error after a later conflict state.
• av-fis.2: Decide whether AgentV needs a first-class CLI contract for remote results sync/status.
• av-fis.3: Add committed regression coverage for interrupted sync retry; dogfood passed a simulated .git/index.lock retry, but coverage should be made durable before documenting retry guarantees.

Verification

• bun test apps/cli/test/commands/results/serve.test.ts
  • 83 pass, 0 fail, 271 expect() calls
• Pre-cleanup code verification, before private evidence relocation and with the same product fix shape:
  • bun test packages/core/test/evaluation/results-repo.test.ts apps/cli/test/commands/results/remote-metadata.test.ts apps/cli/test/commands/results/delete.test.ts apps/cli/test/commands/results/combine.test.ts apps/cli/test/commands/results/serve.test.ts apps/dashboard/src/lib/run-dedupe.test.ts apps/dashboard/src/lib/project-sync-status.test.ts apps/dashboard/src/lib/run-list-actions.test.ts
  • 121 pass, 0 fail, 381 expect() calls
  • bun run build passed.
• git show --check HEAD passed.
• Public diff hygiene: PR diff against current origin/main includes only apps/cli/src/commands/results/remote.ts and apps/cli/test/commands/results/serve.test.ts; no dogfood report/evidence files remain in the public branch.
• Private evidence hygiene: dashboard-setup.env contains only throwaway /tmp fixture paths, and the private evidence tree was scanned for common secret/token patterns before commit.

Readiness verdict from the private report: the remote result data path is production-ready for controlled rollout after these fixes. The full Dashboard production experience still needs follow-up/waiver for av-fis.1 and av-xqm, and av-fis.3 should add durable retry regression coverage before documenting retry guarantees.

---

[cloudflare-workers-and-pages]

Deploying agentv with    Cloudflare Pages

Latest commit:	c81c89c
Status:	✅  Deploy successful!
Preview URL:	https://ee0e9fdb.agentv.pages.dev
Branch Preview URL:	https://dogfood-av-fis-remote-sync.agentv.pages.dev

View logs