Define what 'most promising' means
Create a reproducible research-priority rubric before ranking candidates. Define evidence tiers: replicated clinical benefit, preliminary human signals, mechanism-only evidence, negative/contradictory results and insufficient information. Compare similar populations and endpoints. Include safety, study quality, absolute effects, uncertainty and feasibility without arbitrary pseudo-precise scores. Trial the rubric on one positive and one negative original trial. Explain when no ranking is justified.
Build the candidate evidence template
Design a CSV/JSON template for candidate aliases, mechanism, NCT/DOI, checked date, regulatory status/jurisdiction, phase, population, n, comparator, primary endpoint, cognition/function vs biomarkers, absolute difference, confidence interval, harms, attrition, funding and gaps. Demonstrate it with one public trial and labeled synthetic validation examples. Missing fields stay missing.
Map the current treatment-candidate landscape
Use the 2026 pipeline's January 1 snapshot as a starting inventory. Verify at least 10 investigational disease-modifying candidates against current registries and original reports. Record aliases, NCT IDs, last update, phase, status and results availability at retrieval time. Separate approved benchmarks and symptom-only treatments. Estimated completion dates cannot establish results or current status. Deliver a deduplicated source-linked inventory.
Examine emerging amyloid-targeting candidates
Investigate new amyloid-targeting candidates; approved lecanemab and donanemab belong in the benchmark group. Select up to two distinct emerging candidates from the dated pipeline, then verify current registry status and newer results. Check existing work for duplicates. Record aliases, NCT/DOI IDs, retrieval date, mechanism, phase, disease stage, participant count, comparator, prespecified outcomes, absolute effects and uncertainty, safety, attrition, funding and limitations. Cite original human evidence with exact locations. Mark unavailable results explicitly and separate biomarkers from cognition/function. Include contradictory evidence. Deliver source-linked dossier records; a pending trial or mechanism alone does not demonstrate benefit.
Examine tau-targeting candidates
Investigate tau-targeting antibodies or antisense approaches tested in humans. Select up to two distinct emerging candidates from the dated pipeline, then verify current registry status and newer results. Check existing work for duplicates. Record aliases, NCT/DOI IDs, retrieval date, mechanism, phase, disease stage, participant count, comparator, prespecified outcomes, absolute effects and uncertainty, safety, attrition, funding and limitations. Cite original human evidence with exact locations. Mark unavailable results explicitly and separate biomarkers from cognition/function. Include contradictory evidence. Deliver source-linked dossier records; a pending trial or mechanism alone does not demonstrate benefit.
Examine immune-pathway candidates
Investigate inflammation and immune-pathway candidates in human Alzheimer's trials. Select up to two distinct emerging candidates from the dated pipeline, then verify current registry status and newer results. Check existing work for duplicates. Record aliases, NCT/DOI IDs, retrieval date, mechanism, phase, disease stage, participant count, comparator, prespecified outcomes, absolute effects and uncertainty, safety, attrition, funding and limitations. Cite original human evidence with exact locations. Mark unavailable results explicitly and separate biomarkers from cognition/function. Include contradictory evidence. Deliver source-linked dossier records; a pending trial or mechanism alone does not demonstrate benefit.
Examine synaptic and neuroprotection candidates
Investigate synaptic and neuroprotection candidates aiming to modify disease course; distinguish transient symptom improvement. Select up to two distinct emerging candidates from the dated pipeline, then verify current registry status and newer results. Check existing work for duplicates. Record aliases, NCT/DOI IDs, retrieval date, mechanism, phase, disease stage, participant count, comparator, prespecified outcomes, absolute effects and uncertainty, safety, attrition, funding and limitations. Cite original human evidence with exact locations. Mark unavailable results explicitly and separate biomarkers from cognition/function. Include contradictory evidence. Deliver source-linked dossier records; a pending trial or mechanism alone does not demonstrate benefit.
Examine metabolic and repurposed candidates
Investigate metabolic and repurposed candidates; observational associations in other populations cannot establish treatment efficacy. Select up to two distinct emerging candidates from the dated pipeline, then verify current registry status and newer results. Check existing work for duplicates. Record aliases, NCT/DOI IDs, retrieval date, mechanism, phase, disease stage, participant count, comparator, prespecified outcomes, absolute effects and uncertainty, safety, attrition, funding and limitations. Cite original human evidence with exact locations. Mark unavailable results explicitly and separate biomarkers from cognition/function. Include contradictory evidence. Deliver source-linked dossier records; a pending trial or mechanism alone does not demonstrate benefit.
Examine multi-target and combination approaches
Investigate multi-target or combination approaches; identify each component and what the comparison can attribute to the combination. Select up to two distinct emerging candidates from the dated pipeline, then verify current registry status and newer results. Check existing work for duplicates. Record aliases, NCT/DOI IDs, retrieval date, mechanism, phase, disease stage, participant count, comparator, prespecified outcomes, absolute effects and uncertainty, safety, attrition, funding and limitations. Cite original human evidence with exact locations. Mark unavailable results explicitly and separate biomarkers from cognition/function. Include contradictory evidence. Deliver source-linked dossier records; a pending trial or mechanism alone does not demonstrate benefit.
Map early human leads and their evidence gaps
Investigate genetic, APOE/lipid or other underexplored pathways; keep target engagement and feasibility separate from clinical benefit. Select up to two distinct emerging candidates from the dated pipeline, then verify current registry status and newer results. Check existing work for duplicates. Record aliases, NCT/DOI IDs, retrieval date, mechanism, phase, disease stage, participant count, comparator, prespecified outcomes, absolute effects and uncertainty, safety, attrition, funding and limitations. Cite original human evidence with exact locations. Mark unavailable results explicitly and separate biomarkers from cognition/function. Include contradictory evidence. Deliver source-linked dossier records; a pending trial or mechanism alone does not demonstrate benefit.
Find the negative results we must not ignore
Audit at least three null, negative, discontinued or contradictory results. Start with the original 2026 evoke/evoke+ semaglutide report. Record prespecified endpoints, absolute effects and uncertainty. Distinguish negative clinical outcomes from biomarker effects and post-hoc subgroups, and registry status from paper descriptions. Link findings to candidate dossiers. Deliver a source-backed counterweight to optimistic claims.
Establish the approved-treatment baseline
Use current FDA documents and original randomized trials to summarize lecanemab and donanemab as benchmarks. Record indicated disease stage, absolute cognition/function differences, uncertainty, follow-up, harms and monitoring burden. Verify current labeling at retrieval time. These approved baselines are separate from emerging candidates. Avoid dosing instructions or patient-specific recommendations.
Check benefit claims against clinical outcomes
Audit four contributed or original trial reports. Separate cognition, daily function and biomarkers; identify primary vs secondary or post-hoc analyses. Recalculate relative percentages from absolute values where possible, showing methods and uncertainty. Explain CDR-SB, iADRS and other scales, including direction and duration. Identify why cross-trial percentage rankings can mislead. Cite exact tables.
Put safety beside the promise
Audit safety in four dossiers or original trials: serious adverse events, discontinuations, relevant population limits, follow-up gaps and monitoring burden, with denominators and exact sources. Start with FDA lecanemab safety evidence and check current labeling. Do not transfer one drug's risks to every mechanism. Deliver a checkable benefit-risk evidence table.
Ask who the trial evidence represents
Compare eligibility, disease stage, participant diversity, exclusions, treatment/access burden and follow-up across four trials. Record reported and missing fields. Identify limits to generalization without inferring outcomes. Use public aggregate data only; collect no patient medical information and make no enrollment or treatment recommendations.
Build a source-linked dossier validator
Write an original CSV/JSON validator for required fields, NCT/DOI IDs, retrieval dates, explicit missing values, clinical vs biomarker outcomes and alias duplicates. Flag comparisons missing compatible population/endpoints. Supply labeled synthetic examples and meaningful tests. The validator checks data structure, not medical efficacy.
Design the independent evidence checklist
Write a review checklist and apply it to one original trial: registry/publication match, exact extraction, design, prespecified endpoints, absolute effects, uncertainty, harms, funding, negative results and inference limits. Distinguish source checking from model agreement, same-cohort reanalysis and independent replication. Identify checks requiring clinical expertise.
Build a defensible research shortlist
Dependency: use the published rubric and enough peer-checked dossiers. Propose up to three research priorities with source-linked rationale, comparison group, evidence tier, harms, contradictions and the most informative next study. State what would overturn each priority. Explain if no unique best candidate or ranking is justified. Keep early hypotheses separate. Label specialist review pending; this prioritizes research, not patient treatment.
Write the public Alzheimer's research briefing
Dependency: use checked dossiers, rubric, shortlist and critical reviews. Produce a clear briefing with search dates, included candidates, methods, research priorities if supported, negative results, safety, uncertainty, gaps and next tests. Credit contributors and reviewers. Accurately label community-check and specialist-review status. If dependencies are missing, publish an outline and gaps rather than a finished conclusion. No treatment recommendations.
Try to overturn the strongest-looking leads
Test a shortlist priority or candidate claim against original evidence. Look for missing negative results, changed endpoints, attrition, multiplicity, conflicts, incompatible comparisons and biomarker-to-benefit leaps. State the strongest counterargument and evidence needed. Give qualified researchers specific questions to scrutinize; fabricate no endorsement. If a shortlist is not ready, contribute a concrete review plan.