Methodology v2-review
How Cancel Death reads the evidence
Cancel Death continuously collects longevity research, explains what it reports, and separates new information from findings that have been checked. AI helps organize the work; people make scientific assessment decisions.
1Find
Research, trials, preprints, news and reader submissions enter the collection.
2Connect
Duplicates are merged and sources are linked to studies, claims and interventions.
3Explain
AI proposes structured facts and a plain-language summary with source passages.
4Check
Scientific assessments wait for a person to verify methods, results and overlap.
5Publish
Approved findings update the evidence record. News remains clearly marked context.
New evidence and accepted corrections return to the record without erasing what was previously published.
Three labels to know
Collected
The source is in the library. It may still be awaiting analysis or review.
AI-assisted explanation
A readable first pass grounded in the source. It is not a completed evidence assessment.
Reviewed assessment
A person checked the source, scope and proposed evidence change before approval.
News and public-health coverage can help explain a topic, but it does not independently change an evidence assessment. A new paper can also leave the evidence unchanged.
Details and limitations
Open the part you want to inspect
What counts as an evidence change
A source is collected, its reported facts are extracted, and an assessment is proposed. Scientific assessments require a human decision. A published story may still be awaiting review, and a trial-registry milestone without results remains unassessed.
Study design, species, sample, duration, endpoint, bias, replication and effect size can inform a signed experimental movement. Small signals can round to zero. Human evidence stands on its own, safety acts as a final ceiling, and unknown safety is disclosed rather than treated as a finding.
Coverage versus evidence strength
Not yet assessed means a baseline review has not been completed. Partial coverage means some evidence has been assessed but the historical record is incomplete. Neither label means an intervention lacks scientific support.
Headline scores are withheld until a reviewer records a sourced baseline, scope and safety assessment. Scores describe recorded assessments under an experimental method; they are not probabilities of benefit, predicted years of life, or medical recommendations.
AI, sources and human review
AI proposes extraction and interpretation. Its reported confidence is not a measured accuracy rate. Primary-source results and verified passages are required before scoring; protocols, news coverage and missing results cannot establish efficacy.
Reviewers check the source material, population, outcomes and overlap with earlier studies. Inputs, source versions and decisions are recorded so an accepted correction can supersede an earlier assessment without silently rewriting history.
What the collection monitors
Scheduled queries monitor PubMed, ClinicalTrials.gov and preprint sources, with Crossref used for metadata enrichment. Readers can also submit sources. Coverage is selective rather than exhaustive, and regulatory coverage currently depends on submitted sources.
The coverage page shows collection freshness and the current review backlog.
How the cumulative longevity graph works
The homepage graph combines selected human all-cause mortality syntheses. Its reference is the SSA 2023 US period life table. Selected hazard ratios modify age-specific hazards, and integrating the resulting survival curve produces an illustrative remaining-life-expectancy scenario.
The lines compare the population reference, randomized evidence, selected human evidence, and more favorable source bounds. Shading reflects individual source bounds; it is not a joint 95% confidence interval. Source tags beside stories show which research inputs can enter the graph.
Inputs come from different populations and have not been tested together as a treatment package. Overlap and transfer controls are sensitivity assumptions. Confounding, correlation, nonfatal harms, contraindications, interactions, adherence, access and future advances are not fully modeled. The result is a research comparison, not a personal forecast or treatment recommendation.
For auditability, each compatible input identifies its source, study population, effect estimate, interval and overlap group. Cumulative risk ratios and odds ratios are excluded; corrections or withdrawn review status suspend an input.
Corrections and version history
Use the correction link on an evidence page. An accepted correction updates the underlying record, retains its audit history, supersedes affected assessments and informs followers. Method changes and historical backfills are distinguished from newly reported science.