Chapter 6.1 opens part 6, the reference part, after part 5, the task library, and precedes chapter 6.2, The campaign boundaries and their pooling rules.

Admission gate is the check that decides whether a cell may start. See the admission gate. source: Appendix E, Protocol.md

Agent-coder is the alias for the local coding model. See the agent-coder. source: Appendix E, Protocol.md

Arm is a software version used in the comparison. See the arms and the experiment. source: Appendix E, Protocol.md

Boundary is a change after which cells cannot be pooled with earlier cells. See the campaign boundaries. source: Appendix E, Protocol.md

Cell is a measured task run on an arm and a seed. See the measurement of one cell. source: Appendix E, Protocol.md

Claude Code is a command-line coding agent made by Anthropic. See the container isolation entry. source: Appendix E, Protocol.md

Closeout is the program chain that records the end of a batch. See the closeout chain. source: Appendix E, Protocol.md

Cone is a calculator that identifies affected relationships. See the measuring instrument. source: Appendix E, Protocol.md

Container is a sealed execution environment for the agent and its workspace. See the container isolation entry. source: Appendix E, Protocol.md

DGX Spark (the campaign’s local inference machine) is used for local model inference. See the model-serving routes. source: Appendix E, Protocol.md

Dose is the amount of documentation assigned to a build. See the documentation doses. source: Appendix E, Protocol.md

Disposition is the recorded outcome of a cell. See the scoring of a cell. source: Appendix E, Protocol.md

Evidenceline is the compliance platform used as the experimental product. See the Evidenceline product. source: Appendix E, Protocol.md

Fabric is the model-serving controller that assigns leases. See the model-serving routes. source: Appendix E, Protocol.md

Falsifier is a task intended to disprove a claim. See the falsifier tasks. source: Appendix E, Protocol.md

Freeze is a recorded lock on the contents and hashes of a build. See the construction and freezing entry. source: Appendix E, Protocol.md

Gate is a required check with an evidence record before a step proceeds. See the construction and freezing entry. source: Appendix E, Protocol.md

HASP (the build and instrumentation method) is a name used for the documented codebase. See the construction and freezing entry. source: Appendix E, Protocol.md

Harness is the machinery that constructs builds and runs measured tasks. See the harness test suite. source: Appendix E, Protocol.md

Hidden check is a check run by the host and kept from the measured agent. It is used to score work. See the purpose of the hidden tier. source: Appendix E, Protocol.md

K, the repetition count behind a comparison, is described at the claim and its falsification conditions. source: Appendix E, Protocol.md

Lane is an isolated model-serving process used by a batch. See the closeout chain. source: Appendix E, Protocol.md

Lease, a reserved slot in the serving controller, is assigned to a cell. See the model-serving routes. source: Appendix E, Protocol.md

Manager agent is a planned role that directs a worker agent. See the manager and worker agent topology. source: Appendix E, Protocol.md

Paid route is a hosted model path used for comparison. See the measurement of one cell. source: Appendix E, Protocol.md

Parity repair is a repair that restores a control build to the specification. See the control build and its history. source: Appendix E, Protocol.md

Payload is the measured documentation carried by a build. See the documentation doses. source: Appendix E, Protocol.md

Probe is a hidden check that scores required behaviour. See the purpose of the hidden tier. source: Appendix E, Protocol.md

Run set is a batch of cells recorded together. See the batch ledger. source: Appendix E, Protocol.md

Rung is a position on the documentation ladder. See the documentation doses. source: Appendix E, Protocol.md

SAP (statistical analysis plan) fixes the analysis rules. See the pre-registered analysis plan. source: Appendix E, Protocol.md

Seed is a value that fixes the task selection and run state. See the measurement of one cell. source: Appendix E, Protocol.md

Shakedown is testing of the measurement instrument before confirmatory work. See the findings of the shakedown batches. source: Appendix E, Protocol.md

Source line is a citation to the repository location for a claim. See the documentation experiment. source: Appendix E, Protocol.md

Substrate is the surface on which a task operates. See task applicability by arm. source: Appendix E, Protocol.md

Task is a unit of maintenance work performed by the agent. See the construction of a maintenance task. source: Appendix E, Protocol.md

Tier is a level of the documentation treatment or checking system. See the purpose of the hidden tier. source: Appendix E, Protocol.md

Token is a unit of text read or written by a language model and a measure of work cost. See the token ledger. source: Appendix E, Protocol.md

Variant is a frozen build tree used by the experiment. See the differences between the variant trees. source: Appendix E, Protocol.md

Verification is a check that an instrument or build matches its plan. It confirms that the work follows its specification. See the witness and verification records. source: Appendix E, Protocol.md

Version is a defined state of the software under test. See the maintenance prompt and its versions. source: Appendix E, Protocol.md

Visibility is the extent to which the agent can know about hidden checks. See the claim and its falsification conditions. source: Appendix E, Protocol.md

Vocabulary is the collection of terms used by the campaign. See the project manifest. source: Appendix E, Protocol.md

Waiver is a recorded exception to a rule. See isolation by the container. source: Appendix E, Protocol.md

Witness is an implementation used to show that a check can pass. See the witness and verification records. source: Appendix E, Protocol.md

Worker agent is a planned role that performs maintenance work. See the manager and worker agent topology. source: Appendix E, Protocol.md

Previous: The Experiment section | Next: The repository map
source: Appendix E, Protocol.md