> For the complete documentation index, see [llms.txt](https://docs.concurrence.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.concurrence.com/design/provable-releases.md).

# Provable Releases

Why Concurrence releases agent changes through pinned versions, simulations, reviewed evidence, and run records, and how to prove a release yourself.

## The Problem

An agent's behavior depends on its instructions, its Context Graph, its tools, the models behind it, and the workspace data it reads. A change to any of them can fix one case and break another. Without a fixed record of what ran and evidence from before and after a change, a team can't tell whether a release improved the workflow or only moved the failures.

## Our Design

A release in Concurrence is a reviewed change to what a service resolves, backed by evidence at each stage.

1. **Pin what runs.** A service's version sets select agent and Context Graph versions, dynamic behavior versions, and model preferences. Every Platform service has a `release` set. A pinned component changes only when its stored version changes. An unpinned one resolves to the latest version.
2. **Test the candidate.** Simulations run synthetic callers through the production reasoning path, with tools in simulation mode returning a `dry_run` result, so writes never run for real. A launch can run chosen reads for real or return configured results. Saved cases group into suites and benchmarks for repeatable regression runs. Voice simulation and test calls cover audio behavior.
3. **Decide with evidence.** Your team sets the acceptance criteria and reviews suite results, transcripts, tool calls, and failure reasons. Promotion copies the reviewed candidate to `release`.
4. **Record what ran.** Each run reports `session_provider` and a `manifest` with the agent and Context Graph versions, the version set, the models, the build, and hashes of the prompt templates its first session started with.
5. **Measure after release.** Production evals can score completed conversations against the workspace's active definitions on a best-effort basis. Trace analysis and the voice judge add per-call analysis where available.

## What It Guarantees

* A pinned `release` changes only through an explicit version-set update.
* A run's `manifest` comes from what its first session started with, so later configuration changes don't rewrite it.
* Writes in simulations don't reach an integration, a world write, or a platform function.

## What It Doesn't Cover

* **No automatic gate or rollback.** Simulation results and drift analysis don't block a promotion or roll back a release. Promotion and rollback are deliberate actions by your team.
* **Not a replay.** Reruns are comparison evidence. The agent, workspace state, external tools, and model providers can still vary.
* **Not a safety certification.** Simulation and playground evidence don't certify clinical safety or reproduce production.
* **Partial evidence.** Production evals, trace analysis, recordings, and the voice judge can be missing for a call. Count what wasn't evaluated separately from passes.
* **External effects stay.** Rolling back configuration doesn't undo external writes that already completed.

## Verify It Yourself

1. Create a named version set on a test service, pin explicit agent and Context Graph versions, and read the set back.
2. Run a saved suite against it, and record the pass counts, error sessions, and failing cases.
3. Copy the reviewed candidate to `release`, keeping the previous configuration for rollback.
4. Start a new test run and read its `manifest`. Confirm the versions match the ones you promoted.
5. After real traffic, read production eval results for the release window and compare them with the window before it.

Follow [Build and Prove](https://docs.concurrence.com/developer-guide/guides/build-and-prove) for the procedure, and see [Deployment Model](https://docs.concurrence.com/platform-overview/deployment-model.md), [Testing Overview](https://docs.concurrence.com/testing/testing.md), and [What Ran a Run](https://docs.concurrence.com/developer-guide/platform-api/conversations/runs#what-ran-a-run).


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.concurrence.com/design/provable-releases.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
