[EPIC][P0] Production Loop to Runtime execution, routing, concurrency and evidence

[wesleysimplicio]

Mission

Finish the production integration between Simplicio Loop and Simplicio Runtime without changing the portable nature of Loop.

Loop must continue to operate standalone. When the runtime-backed profile is explicitly selected, Runtime becomes the single governed executor for effects, processes, validation evidence and delivery receipts. The coordinator/LLM remains responsible for semantic reasoning; Runtime never becomes the cognitive owner.

Current verified gaps

• RuntimeBridge exists and is tested, but is not the main runner's production effect path.
• RuntimeBridge defaults to simplicio-runtime while the published binary is simplicio.
• One global lock serializes unrelated workspaces.
• simplicio.execution-route/v1 exists but is not a required runner stage.
• The full installed Loop -> Runtime -> Mapper -> Dev CLI -> watcher -> HBP chain is not release-proven.
• Runtime/Loop documentation and capability lists drift from compiled behavior.
• Internal JSON migration is already tracked separately and must not be duplicated here.

Child issues

• [P0][Runtime Bridge] Resolve the canonical simplicio binary and verify MCP capabilities before effects #692 — resolve canonical Runtime binary and negotiate capabilities
• [P0][Concurrency] Replace the global RuntimeBridge lock with bounded per-workspace sessions #691 — bounded per-workspace Runtime sessions and concurrency
• [P0][Routing] Integrate worker-agent-hybrid execution decisions into the production runner #694 — integrate worker/agent/hybrid routing into the runner
• [P0][Integration] Make RuntimeBridge the production effect executor for Runtime-backed Loop runs #695 — make RuntimeBridge the production executor for runtime-backed runs
• [P0][E2E] Prove installed Loop to Runtime MCP execution through Mapper, Dev CLI, watcher and HBP #693 — installed-package E2E and release gate

Related existing programs

Runtime:

• wesleysimplicio/simplicio-runtime#2983 — warm persistent integration and self-exec removal
• wesleysimplicio/simplicio-runtime#3043 — authoritative CLI/MCP capability manifest
• wesleysimplicio/simplicio-runtime#3134 — coordinator/runtime ownership
• wesleysimplicio/simplicio-runtime#3218 — EffectTransaction/v1
• wesleysimplicio/simplicio-runtime#3459 — loop-execution/v1 consumer
• wesleysimplicio/simplicio-runtime#3570 — fail-closed receipt durability
• wesleysimplicio/simplicio-runtime#3571 — compiled policy as documentation source

Loop:

• [Binary formats] Migrate internal JSON state to HBP, HBI and TOML #644/[Quality] Enforce no-internal-JSON and prove binary-format migration E2E #645 — internal JSON migration and enforcement; this epic consumes their output and does not duplicate it.

Delivery order

Phase 0 — baseline and contracts

1. Freeze installed release versions and source SHAs.
2. Capture process trace, p50/p95 latency, tokens, RSS and current direct-effect paths.
3. Define standalone versus runtime-backed profile behavior.
4. Freeze minimum Runtime capability and EffectTransaction versions.
5. Add failure-state vocabulary: unavailable, incompatible, denied, uncertain, evidence-incomplete, cancelled and verified.

Phase 1 — reliable bridge

6. Implement [P0][Runtime Bridge] Resolve the canonical simplicio binary and verify MCP capabilities before effects #692 binary identity/capability negotiation.
7. Implement [P0][Concurrency] Replace the global RuntimeBridge lock with bounded per-workspace sessions #691 per-workspace lifecycle, queues and locks.
8. Prove reconnect, cancellation, bounded output and zero orphan processes.
9. Make cold CLI fallback explicit and observable; never silent.

Phase 2 — routing and executor integration

10. Implement [P0][Routing] Integrate worker-agent-hybrid execution decisions into the production runner #694 before any production work item is dispatched.
11. Implement [P0][Integration] Make RuntimeBridge the production effect executor for Runtime-backed Loop runs #695 as the only runtime-backed effect adapter.
12. Carry run/turn/item/attempt/lease/fence/idempotency/deadline IDs end to end.
13. Reject direct filesystem/process bypass in runtime-backed mode.
14. Preserve standalone Loop behavior.

Phase 3 — evidence and recovery

15. Correlate Mapper, Dev CLI, Runtime, watcher and Loop receipts.
16. Require HBP durability for VERIFIED mutation delivery.
17. Reconcile uncertain effects instead of replaying.
18. Ensure corrupt, stale or incomplete evidence never becomes success.
19. Verify loop-execution/v1 through Runtime.

Phase 4 — installed E2E and rollout

20. Implement [P0][E2E] Prove installed Loop to Runtime MCP execution through Mapper, Dev CLI, watcher and HBP #693 outside source checkouts.
21. Test minimum/latest package compatibility.
22. Test two independent coordinator hosts.
23. Run cold/warm and one/many-workspace benchmarks.
24. Publish raw evidence and negative results.
25. Enable runtime-backed profile only after release gate passes.

Global acceptance criteria

• Loop works standalone with no Runtime installed.
• Runtime-backed mode resolves the published simplicio binary automatically.
• Every runtime-backed effect passes through one persistent Runtime session.
• No silent direct mutation/process fallback exists.
• Independent workspaces run concurrently within bounds; conflicting writes are fenced.
• Every work item has an execution-route receipt.
• Deterministic work can complete with zero LLM calls.
• Agent escalation is evidence-driven and coordinator-neutral.
• Runtime verifies the final Loop contract as VERIFIED only with complete HBP evidence.
• Crash/reconnect/cancellation cannot duplicate effects.
• Installed Linux/macOS/Windows paths are proven or explicitly reported unavailable.
• p50/p95 latency, tokens, process count, CPU/RSS and evidence bytes are measured from raw data.
• Missing metrics are null with machine-readable reason.
• No paid GitHub Actions is required.
• Documentation, CLI status and compiled capabilities agree.

Definition of done

This epic cannot close because child code exists or unit tests pass in isolation. It closes only after #693 proves the installed default and failure lanes, all child issues are re-fetched as closed, and the final Runtime verification receipt is VERIFIED.

[wesleysimplicio]

🔁 Simplicio Loop — status da execução

Campo	Valor
Estado	CLAIMED
Run / attempt	codex-696-20260723 / attempt-1
Lease / fence	/ fence-696-1

Objetivo e escopo

[EPIC][P0] bounded cross-cutting production Loop-to-Runtime execution/routing/concurrency/evidence slice; preserve standalone Loop; do not close #696 or child issues

[wesleysimplicio]

Progress receipt: PR #712 merged as �8f520c0. Local evidence: py_compile, git diff --check, and 3 cross-cutting integration evidence tests passed. Issue remains open for full production-shaped E2E and resource/latency metrics.

[wesleysimplicio]

CLAIMED

• worker: Codex Work Mode
• run: codex-drain-20260723
• attempt: 1
• base: main @ f142860
• scope: continue the evidence-gated repository drain after merged PRs feat(scheduler): fair admission and safe inference coalescing (#676) #698–test(scheduler): add issue 676 admission evidence #728; fix current regression-gate failures, complete the remaining installed/runtime/concurrency evidence, and close children only after live re-query plus passing local evidence
• guardrails: no paid GitHub Actions; standalone behavior preserved; environment-unavailable AF_UNIX lanes reported separately from genuine regressions