Definition
Evidence from sustained real operation used to judge whether a system keeps producing acceptable outcomes after launch. It includes user results, defects, incidents, interventions, cost, latency, maintenance, recovery, and behavior as the environment changes.
Dark Factory Dev uses production truth to separate a convincing pre-release result from durable operational performance. A benchmark can show capability on a defined set. A test can show that a known condition holds in a controlled environment. Production reveals demand shifts, integration behavior, rare states, adversarial inputs, organizational workarounds, and slow failure modes that the test environment did not contain.
Production is not automatically ground truth. Telemetry can be incomplete, user feedback can be biased, and a successful status code can hide a wrong real-world result. Useful production truth connects traces and side effects to the outcome the system was meant to produce, then records uncertainty and missing observation rather than treating silence as success.
Distinguish it from nearby terms
Build truth asks whether a candidate completed its local task. Evaluation evidence asks whether it meets a defined test before promotion. Production truth asks what happened after real use over time. All three matter. Production evidence should strengthen and revise tests, not excuse the absence of pre-release verification.
Check your understanding
An agent reports success, its test passes, and the deployment returns HTTP 200. Customers still receive the wrong refund amounts. Which signal is production truth? The customer and ledger outcomes are. Connect them back to the run, binding, decision, and side effect so the failure can become a regression case.