reported authored records; rights and execution are audited separately
Technical evaluation work, made inspectable.
Applied Work Labs is the public analysis layer for how we define, validate, evaluate, review, repair, and release technical AI work. It distinguishes program intent from completed evidence at every step.
the proposed selection set, not a completed release count
agent tasks and rule-governed operational workflows
reproducible support before any public capability claim
Two programs. One controlled quality system.
Project A is designed to demonstrate managed production and validation of technical agent work. Project B is designed to demonstrate diagnosis, repair, and maintenance inside an Applied-owned operating business.
Project AValidation in progress
Applied Validation
A managed technical delivery program for independently developed Terminal-Bench‑style engineering tasks.
- Purpose
- Demonstrate task production, validation, review, and release control.
- Unit
- One versioned task package and its evidence record.
- Evidence now
- Program definition, validation controls, and reported candidate-supply figures.
- Next gate
- Task-level rights review, clean replay, and independent acceptance.
- Boundary
- Independently developed; no official Terminal-Bench affiliation.
Project BMethodology preview
Buildalytic evaluation repair
A planned internal operating case study using Buildalytic, Applied’s SaaS business for certified payroll and public works labor compliance.
- Purpose
- Demonstrate diagnosis, correction, regression, and release control.
- Unit
- One synthetic rule, record, decision, and evidence chain.
- Evidence now
- Public workflow context, analysis method, and case-release criteria.
- Next gate
- Authorized case selection, reproduction, regression, and independent review.
- Boundary
- Authorized synthetic records only; no customer or worker data.
How to read the lab.
Each tab answers a different operational question. Together they form the release record.
Applied Validation
Defines Project A, its Terminal-Bench-style scope, qualification controls, and release states.
Evidence
Specifies the artifacts, attempt ledger, review record, and version history required for acceptance.
Evaluation
Separates measurement validity from behavior, localizes observed failures, and defines when downstream training claims are supportable.
Repair
Defines Project B: a planned Buildalytic case study for evaluation repair and maintenance.
State before scale
Candidate, selected, validated, independently reviewed, and released are separate states.
Cause before score
A result is interpreted only after the task, environment, evaluator, resources, and infrastructure survive review.
Evidence before claim
Every public statement must resolve to a versioned artifact, reproducible check, and accountable decision.