Convention. Assays for transfer, remix and construction [ftip-00OU]
✍️source

    A research campaign should expose separate outcomes before assigning a hull grade. The semantic discovery records supply the trace, and this convention supplies the corresponding fresh-task assays. Every arm fixes the task law, model versions, source corpus, reveal schedule, checker, interaction permissions and resource accounting before final evaluation.

    1. Retrieval and associative application. Reveal a target whose relevant source concept is in the permitted endowment. Measure whether the system selects and applies it, while charging search and interpretation.
    2. Transfer. Withhold the target construction and present a fresh problem or representation. Require a transfer map, a preserved relation and a checkable target result. New wording with the same mathematical structure is a limited transfer test; a new composition regime is a separate extrapolation assay.
    3. Remix. Hold out the joint composition of admitted source concepts. Require an explicit interface and a result that depends on the new dependency, then compare against each source concept alone and against irrelevant material of similar size.
    4. Construction. Hold out the target object or principle and expose a verified obstruction. Require a portable difficulty problem, a positive construction, an essential-resource account and at least one consequence beyond the motivating instance.
    5. Consequence and verification. Reveal fresh instances and adversarial counterexamples after the construction is frozen. Check certificates, proof obligations, failure cases and provenance with an evaluator independent of the proposing transcript.
    6. Acquisition. Remove the contributor, search aid and temporary context, freeze the retained artifact, and test a new recipient or fresh recipient instances under a common deployment budget. Report library, procedural, parameter and combined retention separately.

    The campaign includes controlled rediscovery, withheld research and open research as distinct task regimes. A controlled case tests whether the system can reconstruct a known concept; a withheld case tests development without the published construction; an open case adds a historical-novelty question. A success in one regime does not establish success in another. Initial difficulty is measured under the declared admission procedure in Initial difficulty, discovery and retained capability, and costs and outcomes follow Costs, measurements and outcomes.

    Report the profile fields and the evidence status independently. Do not collapse retrieval, transfer, remix, construction, consequence, verification and acquisition into one headline score. A failed lower-level assay limits the interpretation of a higher-level claim; a successful assisted result without retention remains an assisted-deployment result. These controls make the harness auditable while leaving the capability definitions in Conceptual transfer and constructive invention.