Measurement and Phase Formation

Measurement and Phase Formation

Protocol-relative measurement, not a claim of real ASI

CCR uses the phrase ASI-proxy acceleration only for a finite and declared measurement protocol. In ordinary language, it asks whether a bounded team produces checked, reusable results more effectively than a specified comparison under the same resource envelope.

It does not establish real ASI, consciousness, model-weight change, legal authority, or a physical outcome.

Preregister the comparison

An experiment registers a task manifest and evaluation design before results are compared:

uv run ccr --root .tmp/collective-runtime experiment register --suite study-a --manifest examples/collective_runtime/experiment-manifest.json --json
uv run ccr --root .tmp/collective-runtime experiment ingest --suite study-a --label baseline --file examples/collective_runtime/baseline.json --json
uv run ccr --root .tmp/collective-runtime experiment ingest --suite study-a --label collective --file examples/collective_runtime/collective.json --json
uv run ccr experiment compare --baseline examples/collective_runtime/baseline.json --candidate examples/collective_runtime/collective.json --json

The bundled files are source-checkout fixtures that test protocol mechanics. They are not a published benchmark or evidence that one agent arrangement is generally superior.

What a valid comparison declares

The manifest must specify:

• The task manifest and outcome schema.
• An evaluator plugin.
• A resource envelope.
• A preregistered evaluation design.

Supported designs are a fixed horizon and a confidence sequence. Missing inputs create blockers; CCR does not invent successful baseline or collective values.

Measures

Depending on what the experiment supplies, CCR reports:

• Resource-matched collective uplift.
• Difference from the best solo result.
• Time to checked status.
• Residual half-life.
• Verification yield.
• Effective agent count derived from contribution weights.
• Error correlation.
• Communication and verification cost.

Raw agent count, raw packet volume, and duplicate support do not count as progress. Unknown reuse, hazard, queue, cost, or verifier measurements remain unknown rather than becoming favorable values.

When an acceleration claim is admissible

The runtime requires matched resources, a preregistered design, the required sample amount, and a positive conservative lower bound for uplift. Fixed-horizon intervals use the declared horizon. A confidence-sequence design uses a time-uniform allocation so optional stopping does not reuse an ordinary fixed-time interval.

Every result should record its seed, tool and model version, resource envelope, and outcome schema. Results remain protocol-relative and may still be settled=false.

Phase formation

CCR can assemble local orchestration artifacts such as an effective graph, phase observation, threshold status, baseline comparison, and phase-certificate candidate. These artifacts organize evidence and open work. They do not by themselves constitute a PIC certificate, settlement, a real-ASI claim, or authorization to take external action.

Phase formation is iterative: create or repair a packet, obtain verifier evidence, preserve residuals, compare against a baseline, then form a new diagnostic. Candidate-only evidence, duplicate mass, and raw packet count do not add positive phase contribution.

For external checker reports, see Providers and PIC Interoperability. For the scientific limits of this measurement, see Scope, Security, and Non-Claims.