FPF Reference

    G.9:8 — Archetypal grounding (informative; SoTA‑oriented)

    Preface node heading:g-9-8-archetypal-grounding-informative-sota-oriented:81012

    What this page is

    This is generated FPF reference text from the specification preface or supporting sections. It helps interpret FPF; it is not FPF Reference product documentation.

    Methodology

    Use it to understand how the specification wants to be read, then return to a route, pattern, or work packet for active work. Cite generated IDs only when the wording changes the task decision.

    Content

    Show‑A — Multi‑tradition parity for decision systems (post‑2015 practice). ParityPlan pins a rolling evidence window and comparator refs; ParityReport publishes a selected-set outcome plus the evidence trace. Family labels such as preference-learning comparators, causal decision pipelines, offline-RL evaluation pipelines, and robust BO-style selectors remain illustrative until a G.2 SoTA pack or named current source pins the exact family being compared; the parity report still must preserve the selected set or partial order rather than collapse everything into a single scalar.

    Show‑B — QD parity (MAP‑Elites lineage → CMA‑ME / DQD / QDax JMLR 2024, with QDHF or QDAIF refs only when a feedback-guided QD claim is live). ParityPlan pins descriptor/distance definitions and archive insertion policy editions. ParityReport includes archive outcomes and telemetry deltas needed for refresh, without silently converting illumination summaries into dominance.

    Show‑C — Open‑ended parity (POET as lineage; current generator-family claims require a named G.2 SoTA pack or exact current source). ParityPlan pins transfer rule editions and exploration policy refs. ParityReport publishes selected-set outcomes plus transfer‑keyed traces (PathSlice), enabling refresh reruns when any pinned policy changes.

    Show-D — Causal method rung parity. A team compares an observational predictor, an intervention optimizer, and a counterfactual policy strategy under one "best causal method" headline. G.9 first runs CausalRungParityScreen: if rungs, support bases, estimands, or outcome windows differ, the screen returns degraded parity or abstain before a full record is fabricated. When full parity remains plausible, G.9 requires CausalMethodRungParityRecord: each method declares targetCausalUseClaimKind, target CausalityLadderRung, estimandRef, interventional-action basis, CausalEvidenceSupportBasis, relevant C.28 support record and verdict when consumed, follow-up window, outcome measure, source population and target population basis, and estimation-validity basis. If those fields differ, the parity report names declaredCausalityLadderBridgeOrLossRef, transportability or estimation refs where available, and degraded parity or abstain result. The admissible output may be a selected set by comparable rung, not one scalar winner.

    Last Updated: 2026-06-08 — upstream FPF commit 093d30e8 (github.com/ailev/FPF)