Repository navigation
mcpx testing: one deterministic-hypervisor (Antithesis) experiment on the pkgs/mcpx container image #2641
Description
Filed from #1963 (approach C, step 4) during its close-out — queue task 3259, 2026-10-30.
Summary
Run one deterministic-hypervisor experiment: execute the existing pkgs/mcpx container image under Antithesis (or an equivalent deterministic hypervisor) and record whether end-to-end determinism — including the deno/bun/node runtime subprocesses that no in-process seam can make deterministic — is achievable and worth pursuing.
Evidence (from #1963's record)
- mcpx testing: deterministic execution as a long-term aspiration (retrofit vs Rust rewrite) #1963's blockers table: runtime subprocesses and real MCP servers are "not feasible" for in-process determinism; a deterministic hypervisor running unmodified containers is the only realistic end-to-end path (approach C's endgame, explicitly "a prerequisite that is now met, not a plan").
- The container image exists:
pkgs/mcpx/Containerfile(debian bookworm-slim runtime, bun, ~413 MB;docs/container.md), added by Phase 5: tag event bodies must travel with the event #182.
Prerequisites / blockers (named, not hidden)
- The container image should have a CI job first — tracked by mcpx testing: environments -- nix flake, container, local, remote -- as CI matrix entries #1983 (environments as CI matrix entries, incl. container). Running the experiment before the image is CI-built risks experimenting on a stale image.
- Antithesis is a commercial external service: account access and cost need Drew's decision before any run. This issue records the experiment and its prerequisites; it does not authorize signing up or spending.
Acceptance criteria
- A go/no-go note is recorded on this issue once both prerequisites are resolved (CI-built image available; access/cost decision made).
- If go: one experiment run is executed, and a short report is posted here — what was run, whether execution was deterministic across repeated runs from one seed, what broke the determinism if anything, and a recommendation (adopt / revisit / drop) for mcpx testing: deterministic execution as a long-term aspiration (retrofit vs Rust rewrite) #1963's approach-C endgame.
- If no-go: the reason is recorded here and the issue is closed as dropped, so the question is not re-litigated ad hoc.
- No determinism implementation work in
pkgs/mcpxitself under this issue.
Parent record: #1963 (approach C). Related: #1983, #1896 (fuzzing tracker).
Metadata
Metadata
Assignees
Labels
Package/app: mcpx (pkgs/ or apps/ in this repo)Test coverage, test infrastructure, or test-only fixesNice to have; do when convenientQueue or user escalation stateHas a queued task in the Supabase agent queueFiled by the new-issue-creator jobQueue or user escalation stateNot yet triaged; needs a first lookQueue packet kind: verify, reproduce, and route this itemQueue or user escalation state
Activity
dezren39 commentedon Nov 3, 2026
agent-note:v1
state: escalated (blocked on a spend/access decision — not on engineering)
Triage verdict at main @
4c75cc808(2026-11-03): still live, one prerequisite left, and it is not mine to resolve.Prerequisite 1 — CI-built container image — met. #1983 closed
status:done, and.github/workflows/mcpx-tier2-containers.ymlnow doesdocker build -f Containerfile -t mcpx-tier2-image .before running the tier-2 legs; that workflow's last three runs are green (latest 2026-11-02 21:40 UTC). The experiment would not be running against a hand-built local image.Prerequisite 2 — Antithesis account + cost — open, and explicitly a Drew decision. The issue body already says the tracker "does not authorize signing up or spending". That is the only thing standing between this issue and a run.
My read (worker's opinion, not a decision): recommend decline, or defer indefinitely. Antithesis buys a one-off go/no-go report, not a capability we can keep — it cannot become standing CI without a contract, and the tier-2 container legs plus the in-process seams from #1963 already cover the determinism we actually need. Spending should need a concrete flake the existing seams cannot reach; none is named here. If the answer is decline, the issue closes as dropped per its own acceptance criteria ("so the question is not re-litigated ad hoc").
Escalated for: yes / no on a commercial Antithesis engagement (account + budget). Nothing else is blocked.
Queue: task 7971 (triage,
repo-health-code-pilot),repo-worker-code, lane02cafe76-eafd-4d0b-aaa6-2b7a737d68dc(OpenChamber scheduled lane). Blockedescalated-to-user.Found via: triage packet for #2641 (coverage pass 2026-11-03)
developing-today commentedon Nov 3, 2026
agent-note:v1
state: triaged (coverage pass)
Coverage pass 2026-11-03: reviewed in the fleet-wide per-issue pass (drift / overlap / already-handled / stale / blockers / ordering). Verdict: still live; queued in repo-health-code-pilot as triage at priority 30 (priority/low). If this is already handled at current main, the triage closes it with evidence; if it needs splitting, the triage files the split.
Found via: coverage pass 2026-11-03