AITP Research Mode Classifier

Environment gate

• This skill runs inside an AITP session. Confirm AITP routing has already claimed the task.
• If the topic already has a recorded research mode in classification_contract.jsonl, skip to the "Verify or override" section.

Classification task

Determine which research mode best describes the current topic from these four options:

Mode	When to choose
formal_derivation	The topic centers on mathematical proof, formal derivation, theorem verification, operator algebra, bootstrap arguments, consistency conditions, or any argument that must close by logical deduction rather than numerical evidence.
toy_model	The topic studies a simplified or lattice model (Ising, Heisenberg, spin chain, TFIM, SU(2) lattice, MPS/DMRG) to extract qualitative or semi-quantitative physics. The model is analytically or numerically tractable at small system sizes.
first_principles	The topic uses ab initio methods (DFT, GW, QSGW, BSE, QMC, Hartree-Fock) or computational chemistry / condensed matter codes (VASP, Quantum ESPRESSO, ABINIT, LibRPA) to compute observables from electronic structure. Convergence, basis sets, and benchmark comparison are central concerns.
exploratory_general	The topic does not clearly fit the above categories. This includes literature surveys, open-ended idea exploration, physics discussion without a bounded computational or formal target, and topics still being scoped.

Reasoning priority

1. Explicit declaration: If the user, topic contract, or source material explicitly names a research mode (e.g., "this is a first-principles calculation" or "we study the Heisenberg spin chain"), use that mode.
2. Research question semantics: Read the research question and its target claims. What kind of answer would close the question? A proof → formal_derivation. A numerical result from a lattice model → toy_model. A converged ab initio observable → first_principles. An open-ended exploration → exploratory_general.
3. Contextual signals: Look at the surrounding materials — cited papers, named methods, mentioned codes, symbols used. These support but should not override the above.
4. Default: If no clear signal, choose exploratory_general.

Recording the classification

After reasoning, call the MCP tool:

aitp_record_classification(
    topic_slug=<current topic>,
    classification_type="research_mode",
    value=<one of: exploratory_general, first_principles, toy_model, formal_derivation>,
    rationale=<1-3 sentence explanation of why this mode was chosen>,
    signals_used=<list of key signals that informed the decision>,
    source="ai_reasoning"
)

Verify or override

If a prior classification exists, check whether the current human request or topic evolution has changed the research character. If so, record a new classification (the contract is append-only). If the prior classification is still valid, do nothing.

Hard rules

• Do not use substring or keyword matching. Read the full context and reason semantically.
• A topic can have multiple classifications over time; the latest one is active.
• Never skip recording. The classification must be durable before execution continues.