← All agent-stack verdicts

Durable workflow orchestrators

Adopt and wrap Confidence Low Group Orchestration Evidence as of 2026-09-23

Written by VerdictsOrchestration · Reviewed by control-plane pane 2 on 2026-09-23 (separate AI agent sessions; review record)

Bracketed tags are evidence tiers from RULEBOOK.md §1: [Verified] means checked directly, [CI-observed] read from a project’s CI pages, [Maintainer claim] asserted by the project, [External] a dated third-party source, [Inference] our judgment. A gold path:line link opens the cited line. NODUS is the four-ring verdict scale (Invest, Pilot, Explore, Monitor) used for the FrankenSuite assessments; the agent-stack verdicts use their own scale (Adopt, Adopt and wrap, Build clean-room, Watch). TRL is technology readiness level, scored 1 to 9. CI is continuous integration: automated checks that run on every push.

Bottom line

Inference, low confidence: run long or crash-prone agent work on an existing durable-execution engine (Temporal, Cadence or Conductor), and do not write your own event log and replay engine. Wrap it, because none of these engines checks your workflow code for determinism when you compile it: determinism is only caught when a recorded run is replayed, so your test suite must replay recorded histories, including one deliberately nondeterministic history that must fail, under a virtual clock. Confidence is low because the evidence pack calls itself thin and ran none of these suites.

Adopt, do not rebuild

License checked on 2026-09-23; non-permissive licenses are named in the text.

  • temporalio/temporal and its official SDKs MIT: the durable-execution server that records each workflow's event history and replays it after a crash, with five official SDKs; its SDKs ship in-process test environments and a replayer, so rebuilding it would recreate both the engine and its test tooling. [Verified] (ecosystem/pickup/_evidence/workflow-orchestrators.md:7 durable execution server (Go) with 5 official SDKs; defines workflow-as-code, event histories, replay; ecosystem/pickup/_evidence/workflow-orchestrators.md:18 Official Python SDK; contains temporalio/worker/_replayer.py)
  • cadence-workflow/cadence Apache-2.0: the sibling history-replay engine (the old uber/cadence path now 404s), with simulation suites for its history, matching and replication services. [Verified] (ecosystem/pickup/_evidence/workflow-orchestrators.md:9 Uber-originated durable execution engine (Temporal's sibling lineage))
  • conductor-oss/conductor Apache-2.0: an orchestration engine that ships user-facing test-harness modules, so adopters get a test workflow and not only documentation. [Verified] (ecosystem/pickup/_evidence/workflow-orchestrators.md:8 Netflix-originated orchestration engine; highest-star repo in this set; ships test-harness/test-util modules for users)

Copy these practices

  • Replay recorded histories, with a negative control: replay canonical recorded histories on every change, and keep one deliberately nondeterministic history that must fail replay with a named error, so a determinism checker that stopped working is caught. Starter kit: Acceptance shape: positive observable + planted negative + no-claim line (provisional). [Verified] (ecosystem/pickup/_evidence/workflow-orchestrators.md:28 Replay tests against recorded event histories (determinism checking; a deliberately nondeterministic fixture exists))
  • Replay stored agent runs: the Temporal Python SDK replays stored histories of OpenAI Agents, LangGraph, Google ADK and deepagents workflows; copy that shape for any agent that runs inside a workflow. Starter kit: none. [Verified] (ecosystem/pickup/_evidence/workflow-orchestrators.md:29 Agent-workflow replay tests (LLM-agent orchestration frameworks replayed from stored histories))
  • Virtual clock in tests: workflow tests run in-process under a mocked clock that skips ahead, never on real timers. Starter kit: none. [Verified] (ecosystem/pickup/_evidence/workflow-orchestrators.md:27 In-process workflow test environment with a mocked/virtual clock (time skipping; no real timers))
  • Blocking workflow code is a test failure: detect a workflow coroutine that never yields and fail the test, with the timeout plumbed through the suite. Starter kit: none. [Verified] (ecosystem/pickup/_evidence/workflow-orchestrators.md:30 Deadlock detection for workflow coroutines (workflow code must yield; blocks are failures))
  • Test the journal on its own: unit-test the event-history or journal tables as the durability primitive, before engine-level tests. Starter kit: none. [Verified] (ecosystem/pickup/_evidence/workflow-orchestrators.md:33 Journal/event-history tables unit-tested as the durability primitive)
  • Track flaky tests in the open: shard the heavy suites and publish a flaky-test report instead of retrying until green. Starter kit: none. [Verified] (ecosystem/pickup/_evidence/workflow-orchestrators.md:36 Heavy, sharded CI with flaky-test tracking)

Build only if

  • No evidenced constraint rules out the incumbents for durable execution. The one evidenced gap is a checker, not an engine: no surveyed repository had a static (compile-time) determinism analyzer, so a builder who needs one would build it beside an adopted engine. [Inference] (ecosystem/pickup/_evidence/workflow-orchestrators.md:42 the determinism story is runtime replay + deadlock detection, not compile-time checking)

Where FrankenSuite touches this

  • asupersync is an async runtime, not a durable-execution engine, but it ships a deterministic lab runtime with virtual time, deterministic scheduling and trace replay; its packet verified the code exists and did not execute replay; TRL 6, Pilot. [Verified] (packets/asupersync-assessment.md:81 Deterministic lab runtime: virtual time, deterministic scheduling, trace replay | demonstrated (existence; replay not executed); packets/asupersync-assessment.md:202 Technology readiness | TRL 6 | Ships as 0.5.0 on crates.io; packets/asupersync-assessment.md:68 (NODUS: Pilot — see §4.9.))
  • frankenterm runs transactional multi-pane missions for agent fleets with prepare, commit and compensate steps and an idempotency ledger, a saga-style pattern rather than a replayable event history; TRL 7, Explore. [Verified] (packets/frankenterm-assessment.md:143 transactional missions (prepare/commit/compensate, idempotency ledger, kill switches, failure injection); packets/frankenterm-assessment.md:178 Ring: Explore. The ring rules are decisive)
  • franken_engine captures program, policy snapshot, evidence stream and randomness transcript so a security decision can be replayed; that is incident replay inside a JS runtime, not workflow orchestration; TRL 4 to 5, Explore. [Verified] (packets/franken_engine-assessment.md:120 Replay captures IR3 program, policy snapshot, model snapshot, evidence stream, and randomness transcript; packets/franken_engine-assessment.md:197 Technology readiness | TRL 4–5 | Lab-validated components)

What we cannot say

Revisit when

  • An engine ships a static determinism checker for workflow code: the wrap shrinks from "replay everything" to "run the checker plus a smaller replay set".
  • A replay or time-skipping test framework is found and read in restatedev/restate, which would let it join the adopt list.
  • A replay suite is observed failing or passing on live CI for an agent-workflow history, which would move the practice claims past file existence.

Sources

Starting a build here?

Begin from the pickup companion for Durable workflow orchestrators: it carries the charter seed, the oracle candidates and how to verify them, and the first claims to register, so a new project starts from evidence instead of from scratch.

Help fix it

Each opens a GitHub issue form with this page’s address filled in. You need a GitHub account.

Made by Joshua Nowak (ZestStream) with AI coding agents. Not affiliated with or funded by Jeffrey Emanuel; Joshua uses his public tools and is a paying subscriber to jeffreys-skills.md.

Source on GitHub MIT licensed