Convention. Semantic records for conceptual discovery [ftip-00OT]

✍️sourceAGENTDRAFTED

A research trace is evidence-bearing when each important transition names its input, operation, output, cost and checking status. The following records are semantic slots; a human, model, tool or mixed process may fill them. They make the route from failure to reusable capability inspectable without treating a particular harness as the source of the ability.

1. Obstruction. Record the attempted method, the instance or family on which it failed, the smallest verified reason for failure, and the evidence that the failure is relevant to the target rather than an implementation error.
2. Difficulty problem. Reformulate the obstruction as a portable question. State what structure is missing, which boundary is held fixed, what would count as a solution, and which information is withheld from the research process. A vague request for a better idea is not a difficulty problem.
3. Direction. Record a proposed route, its source domain or representation, its predicted obstruction or gain, its resource estimate, and the experiment or proof that could reject it. Diverse or opposite directions are useful only when their differences are explicit.
4. Transfer map. Identify the source relations, the target objects, the map or analogy between them, the relations preserved, the adaptation needed, and the target-side certificate. This record separates conceptual transfer from a shared word or an untested analogy.
5. Construction. State the new object, representation, question or principle; identify the hole it fills; list its essential dependencies; and record the construction, checking work and cost. If the construction uses a contributor, archive or temporary context, record that causal input separately.
6. Consequence. State a result that follows from the construction and extends beyond the motivating instance. Include a fresh test, counterexample search or proof obligation that could invalidate the claimed consequence.

A complete trace may branch and may contain rejected records. Rejection is part of the evidence: the trace preserves the failed direction, the reason for rejection and the work charged to reach that decision. A successful transfer does not imply a successful construction, and a construction does not imply a proved consequence. Independent verification and frozen-recipient acquisition remain later records in the research-capability profile.

The record schema also separates ability from facilitation. The transcript names which information was available to the model, which operation was performed by an external setup, and which artifact remained after the setup was removed. Search, retrieval, communication, interpretation, checking and failed attempts are charged to the declared campaign. A record that omits these causal boundaries can document an assisted result but cannot identify a learned or retained capability.

These records support the hull-grade convention and the prospective controls in A prospective comparison of research campaigns. They are an evidence interface, not a benchmark score and not a claim that any current system fills every slot.