research-document FE-BND-DECISION-001
Decision
Decision
Status and classification
Stage A: incomplete. Stage B: blocked-before-stage-b. Boundary classification: inconclusive by the preregistered rule. This is a gate result, not evidence for H1, H0, or H2.
No protocol deviation occurred. The planned stop rules were applied: material comparator coverage is insufficient; a sealed identity key cannot be access-isolated in this repository; recognition has not been administered; and the eligible human/external panel is unavailable.
Mechanism-level mapping status
| Candidate | Strongest plausible comparator(s), curator screen only | Status |
|---|---|---|
| A01 separate identity/function | B01 + B03 | unreviewed; prediction incomplete |
| A02 extract-redesign-retest | B01 + B02 | unreviewed; likely combination mapping |
| A03 finite compositional procedure representation | B04 + B01 | unreviewed; strongest surviving question |
| A04 reasoning/execution separation | B04 + B02 | unreviewed; interface underspecified |
| A05 typed observation/interpretation/alternatives | B06 + B05 | indeterminate; B06 coverage insufficient |
| A06 prospective verification/reassessment triggers | B02 + B06 | indeterminate; B06 coverage insufficient |
| A07 IDs/freeze/lineage/supersession | B05 + B03 | unreviewed; current implementation contradicts aspiration |
| A08 integrated workflow | B01 + B02 + B03 + B05 | H2 audit fails; no non-additive interaction specified |
These are curator interpretations, not reviewer judgments or adjudicator decisions.
Evidence and falsification
Direct observation: all cards are under 500 words and use one schema; every A-card names a comparator combination capable of supplying much of its operation. Strongest evidence against distinguishability is the broad coverage of method tailoring, lifecycle engineering, traceability, process modeling, and provenance. Strongest evidence for continued testing is A03’s typed representation of epistemic reasoning and A04’s proposed invariant interface between reasoning and coordination; neither yet has a complete unique prediction.
H1 was challenged by generic relabeling, combination mappings, domain-specialization checks, lower-complexity baselines, and prediction-collapse checks. H0 was challenged by seeking missing epistemic typing/composition fields and divergent tasks. Neither challenge can be adjudicated with the frozen evidence.
Outcomes, sensitivities, and validity
Reviewer agreement, recognition accuracy, full-subsumption rate, indeterminate rate, and unique-prediction count are not estimable. Sensitivity analyses are not runnable. Reporting zeros would be false.
Main threats: curator roles were produced in one agent context; standards detail is partly paywalled; identity cannot be concealed with repository permissions; no recognition participants or qualified human reviewer; and several predictions are curator inference. These threats bias toward apparent coherence and must not be relaxed.
Evidence that would change this result
Obtain licensed section-level text for B01–B04 and verified primary/authoritative sources for B06; have separate curators revise cards; store the sealed key in an access-controlled location; pass recognition with two new participants; recruit three eligible independent reviewers including a qualified human and provider-family cap; then lock responses and run the frozen analysis.
Smallest justified next experiment
This is not another mapping run. It is a Stage A repair v1.1 limited to source completion, true role separation, key isolation, and recognition pretest. Only after it passes should the existing eight-by-six mapping packet be issued. If A03 survives, preregister a focused graph-defect/round-trip task against a complexity-matched process-model baseline.
No canonical theory, discipline status, hypothesis confidence, or downstream registry was changed. Downstream records affected are this package only; FEH-001 remains open.