Production ATCR
Completed reconstructed production tasks ÷ eligible reconstructed production tasks.
Answers what agents actually completed in your live product.Agents are becoming software's primary users.
GrowthOS reconstructs how agents discover, choose, and use your product in production, then runs the same real-world tasks across any agent and model.
Trusted by teams at
Engine 01 · Production
Reconstruct the task your existing analytics split apart.
One agent objective can cross docs, authentication, an API, an SDK, and a backend, including through parallel sessions and delegated work. GrowthOS joins the evidence into a journey graph without inventing certainty.
Every connection carries its confidence, supporting evidence, and unresolved alternatives.
How reconstruction worksIllustrative reconstruction
TASK · SDK-001
Install the SDK and send the first valid event.
completed · 00:00
Discover the SDK
The agent finds the SDK overview through documentation search and opens the installation guide.
1 / 6
Engine 02 · Controlled
Test the same real-world task across every agent and model.
Turn a production journey or qualified workflow into a managed Eval. Hold the success criterion steady while you compare agents, models, interfaces, documentation, and product treatments.
Completion is only one answer. Inspect friction, latency, tokens, retries, failure modes, and the trajectory behind the result.
Explore GrowthOS EvalsIllustrative reconstruction
| Agent | Model | Treatment | Stage | Completion | Friction | Latency | Tokens | Retries | Failure mode | Evidence |
|---|---|---|---|---|---|---|---|---|---|---|
| Agent A | Model 1 | Baseline | configure | Blocked | 4/5 | 41s | 8,650 | 2 | Auth scope unresolved | |
| Agent A | Model 1 | Docs treatment | complete | Completed | 1/5 | 29s | 6,120 | 0 | - | |
| Agent B | Model 2 | Interface treatment | complete | Completed | 2/5 | 33s | 7,010 | 1 | - | |
| Agent C | Model 3 | Docs treatment | complete | Completed | 1/5 | 31s | 6,550 | 0 | - |
How each metric is defined
Completion uses the task criterion. Friction counts avoidable recovery work. Latency, tokens, and retries describe the controlled run. Failure mode names the observed blocker.
One evidence system
Production tells you what to test. Evals tell you what to change.
The two engines share the same task definition and evidence model, creating a loop from real behavior to controlled improvement and back to production verification.
The outcome metric
Measure completed tasks, not agent traffic.
Agent Task Completion Rate (ATCR) gives production behavior an outcome unit. It belongs beside, but never collapses into, controlled Eval pass rate or reconstruction confidence.
Eval pass rate
Controlled runs that meet a fixed success criterion ÷ eligible controlled runs.
Answers what completed under comparable Eval conditions.Evidence confidence
The strength of the evidence connecting activity into a task journey.
Qualifies the reconstruction; it is not a performance score.The full agent experience
A task succeeds only when every layer holds.
Use one connected evidence view to see whether agents can discover, access, use, and transact with your product, not four disconnected scorecards.
ACTIVE LAYER · DISCOVERY
Can the agent find the right path?
Trace how an agent discovers product pages, documentation, schemas, and integration surfaces, and whether the chosen route matches the task.
- 01Search and referral path
- 02Resource selection
- 03Task-fit signal
A lower-commitment starting point
Request the Agent Traffic Blindspot Report.
See why sessions and pageviews miss agent demand, how to define production task completion, and which evidence makes agent behavior actionable.
Request the reportIllustrative report spread
01 · Invisible demand
02 · What to measure
ATCR
Questions about the evidence
Start with one task
Show us one task agents should be able to complete.
We’ll scope the surfaces, evidence, completion criterion, and the clearest first reconstruction.
Not ready for a reconstruction? Request the Agent Traffic Blindspot Report.