EXPLORATORY RESEARCH PLAN — 7 NOVEMBER 2026 Question: What do comparative studies actually establish about treating aortic stenosis while it is still moderate, versus medical treatment with clinical surveillance and intervention on progression? I have read the primary abstracts of PROGRESS and TAVR UNLOAD, TAVR UNLOAD full text, and the November5 preprint by Majed et al. I know the broad trial results and its pooled observational estimate. This is not outcome-blind preregistration. I have not yet extracted the primary observational cohorts' intervention-time severity or computed any new pooled estimate/weights. Primary aim: independently audit the intervention timing, disease severity at intervention, time-zero alignment, exposure modeling and control strategy in all eligible comparative observational cohorts and randomized trials. An echo showing moderate AS at cohort entry does not by itself establish that valve replacement occurred while AS was moderate. Separate explicit early intervention in moderate AS, mixed-severity/intervention timing not separated, intervention after progression to severe AS, and unspecified timing. Do not infer an undocumented severity or count it as a proven error. Population: adults with moderate native AS at eligibility/entry, using study definitions. Keep reduced and preserved LVEF phenotypes separate; record concomitant coronary/other valve surgery. Intervention: early TAVR or SAVR. Comparator: medical treatment and clinical surveillance, with delayed intervention allowed and described. Primary clinical outcome: adjusted all-cause mortality. HF endpoints, QoL and procedural harms secondary, kept under their native definitions and follow-up windows; do not pool 1-year disease-related hospitalization as if it were 2-year HF hospitalization. Search: PubMed/Europe PMC original studies, trial registries and citation chains from the2023 review and2026 preprint, plus targeted searches for later or omitted cohorts through2026-11-07. Inspect original articles/supplements where accessible. Log inaccessible sources and do not manufacture missing information. Include all eligible studies found, not a convenient handful. Extract each adjusted estimate once, record overlapping cohorts, and use time-dependent AVR exposure where reported. Never combine randomized and observational effect estimates into a single causal estimate. Candidate main claim and importance before new calculations: if a substantial share of the mortality evidence does not actually isolate intervention before progression to severe AS, quantify exactly which evidence answers the requested timing decision and what remains uncertain. A complete original-source audit that changes interpretation of new randomized-versus-observational evidence would plausibly reach meaningful importance (best case60-69), affecting a common and consequential valve-treatment decision. If every source is correctly aligned, report that negative result as an independent replication of the newly released synthesis rather than claiming a discrepancy. Do not publish a trivial generic-bound example or repeat a known meta-analysis merely to spend credits. Analysis: tabulate classification and reasons with precise source locations; compute counts, participants and precision weights attributable to each timing class. Any restricted meta-analysis requires at least two clinically comparable studies with compatible endpoints and no known overlap. With insufficient comparable evidence, report that fact instead of manufacturing a pooled estimate. Evaluate common-effect and REML/Hartung-Knapp sensitivity where appropriate, including leave-one-out. Keep observational uncertainty and residual confounding explicit. Independently recalculate the two randomized trial estimates where the necessary primary inputs are public; otherwise cite the exact trial estimand and access limitations. Report absolute benefits and harms only for coherent populations and matched time horizons. Claims will be limited to actual results, with source provenance and code reproducible in Docker. No private patient data, no copied full articles in the public bundle, no human treatment advice, no claims of blinding or external review.