TA-14 ACA · FOUNDING PILOT KIT

A test protocol built to survive a negative result.

A credible data-center pilot cannot begin with an assumed savings claim. It begins with a frozen workload, explicit admissibility rules, complete computational accounting, and an agreement that failure remains part of the record.

PILOT RECORD

What must be frozen before the comparison.

01Workload identity and version

02Purpose and bounded objective

03Required output / quality threshold

04Baseline architecture and configuration

05ACA gate version and policy

06Available evidence and authority

07Resource envelope

08Measurement boundary and telemetry

09Material-change rules

10Stop / safety conditions

MEASUREMENT MATRIX

Count the work before claiming the consequence.

01Baseline total work02ACA gate work03Admitted downstream work04Shifted / retry work05Net downstream work avoided06Latency07Required quality / task success08False rejection09False admission10Model / accelerator time11Tokens / model calls12Retrieval / ranking / embeddings13Tool and agent calls14Network / storage activity15IT energy where measured16Facility / thermal effect where measured17Water where attributable18Emissions / infrastructure where attributable

Net computational benefit = baseline total work − (ACA gate work + admitted downstream work + shifted/retry work)

MATERIAL CHANGE

Standing at T1 is not standing at T2.

IdentityEvidenceAuthorityScopePurposeResourcePolicyOutcome

If a material change occurs, prior computational standing must not silently authorize the changed proposal. Revalidation begins from the changed reality.

ADVERSARIAL TEST SET

Try to break the gate before celebrating it.

A1Request fragmentation

A2Stale cached ALLOW

A3Hidden pre-commit compute

A4HOLD denial-of-service

A5Escalation laundering

A6Binding substitution

A7Branch explosion

A8Metric gaming

A9Evidence spoofing

A10Outcome self-certification

MINIMUM PUBLISHABLE RESULT

SUPPORTED · NEUTRAL · NOT SUPPORTED · INCONCLUSIVE

A pilot result should preserve the workload definition, baseline, gate version, determinations, changed-context events, accounting, quality effects, environmental attribution level, uncertainties, failures and raw evidence references. The outcome does not create automatic authority for the next run.

Ready to put a real workload against it?

Use the interactive lab to model the accounting, then contact TA-14 to establish a bounded founding pilot.