Every TargetSpace Labs engagement runs the same controls as the open benchmark — sealed forecasts, R1/R2 baselines, calibration and permutation gates, evidence ablation — and produces a TargetSpace Report. The question is always the same: does the system show target-specific predictive skill, beyond routine and retrieval, at a stated level of confidence.
Our highest-value engagement: an end-to-end prospective evaluation of your system on your users or study cohort. We design the evaluation, seal the forecasts, run the baselines and gates, and deliver the full report with a readout for your team.
Teams whose product is the personal-model claim: personal agents and memory systems, wearables and ambient AI, enterprise copilots, and research labs studying personal and organizational world models.
Several weeks of evaluation and consent design, followed by prospective evaluation windows of 30–90 days. Scope, cohort, and infrastructure requirements are settled during scoping.
A one-time or recurring evaluation of a specific personalization, memory, or modeling claim — scoped to one product surface and delivered as a standardized report.
Each audit tests a single stated claim on a single product surface, so the result is attributable and the finding is unambiguous.
The claimed capability is compared against the same system with the capability disabled — the difference is the measured contribution of the claim.
Findings arrive in the same TargetSpace Report format as every engagement: calibrated prediction results, baselines, gates, and limitation notes.
A readiness assessment against target-specificity, calibration, and evidence-minimization thresholds — the path toward a future certification mark.
Certify is a readiness program today. No certification marks have been issued; Certify assesses readiness against published thresholds.
Skill must depend on the correct target: re-scored against the wrong target, it must collapse.
Stated probabilities must match observed frequencies across the evaluation window.
The system must demonstrate which observations earn their cost — and what can be removed without losing validated skill.
A self-serve evaluation platform and API for running sealed prospective evaluations continuously. Pilot participants shape its design — join the pilot to influence what ships.
The decision-grade deliverable every engagement produces: lift over R1 and R2, calibration, wrong-target specificity, evidence ablation, and minimum sufficient observation. View the sample report — synthetic and clearly labeled.
Apply → scoping call → agreement → evaluation → report. Scope, cohort, and timelines are fixed before any forecast is sealed.
Evaluations can run federated or on-premises so raw observation data stays inside your environment. See Security for details.
A mutual NDA is available before scoping. Evaluation results are yours; nothing is published without your written consent.
We are accepting a limited number of companies and research labs into the founding pilot cohort. Applications are reviewed manually.