Set the standard of evidence for longevity claims

In LON-1.1 you defined aging as progressive functional decline and rising mortality risk. The rest of this track is about intervening in that process, and the field is loud, well-funded, and full of people who want to sell you something. So before we touch a single intervention, we install the filter that everything downstream runs through. This lesson is not about a mechanism. It is about the standard of proof you will hold every claim to, including the exciting ones, especially the exciting ones.

Here is the one-line version to carry with you. For any "this reverses aging" claim, demand three things before you believe it: the organism it was shown in, whether there was a proper control group, and whether it moved a real outcome or just a marker. Most headlines fail at least one. Let us earn each of those three demands.

The ladder of evidence

Evidence in biology is not flat. A result carries different weight depending on where it was produced, and the levels form a ladder from "cheap and fast and weak" to "slow and expensive and strong." From the bottom:

• Cell culture (in vitro). Cells in a dish. Fast, cheap, and mechanistically clean, because you control everything. It is also the furthest from a living body: no bloodstream, no immune system, no other tissues, often an immortalized or cancer-derived cell line behaving unlike a real cell. A compound that rejuvenates cells in a dish has cleared the lowest bar there is.
• Short-lived model organisms. The worm C. elegans (lifespan about two to three weeks), the fruit fly, then the mouse (about two to three years). Short lives make lifespan experiments actually finishable, which is why almost all longevity biology is discovered here. But the worm is not a small human. Its body cells are postmitotic in the adult (they have stopped dividing, unlike your constantly-renewing skin, gut, and blood). It lacks most of your organs, and it shares only deep, ancient machinery with you. Mice are mammals and much closer, yet still routinely mislead (more on that below).
• Human observational studies. Real people, followed over time, correlating a habit or a molecule with an outcome. This is where you finally get human relevance, but you pay for it: nobody was randomized. The people who take vitamin D or eat more olive oil differ from those who do not in a hundred other ways, and any of those differences could be the real cause.
• Human randomized controlled trials (RCTs). People are randomly assigned to intervention or control. Randomization is the whole point: it breaks the link between the treatment and every confounder at once, so a difference in outcome can be pinned on the treatment. This is the top of the ladder, and it is rare in longevity because a true lifespan RCT in humans would take decades.

Recall the mutation types from S7.1 and the hallmarks you will meet in LON-2: those are usually discovered at the bottom of this ladder, in cells and worms and mice. Discovery there is legitimate. Believing a human claim from there is not.

The ladder is a model, not a law. A well-run mouse study can beat a small, sloppy, or poorly-blinded human trial. The word "RCT" is not a magic pass, and "only in mice" is not an automatic dismissal. The ladder tells you how much doubt to carry by default, and where the burden of proof sits, not which single study wins.

Healthspan is not lifespan

Two different targets hide under the word "longevity," and confusing them muddies almost every discussion.

• Lifespan is years alive. Time from birth to death. One number.
• Healthspan is years of good function. The stretch of life spent free of the functional decline and disease that LON-1.1 defined as aging itself.

They are related but not the same, and an intervention can move one without the other. Keeping a frail, sick animal alive longer extends lifespan while shrinking the fraction of life that was worth living. Most people, asked carefully, want healthspan first. So when a study reports "lived longer," ask "lived better, or just longer." The second is easy to measure and the first is what you actually care about.

Surrogate markers are not hard outcomes

This is the trap that catches trained scientists, and it maps perfectly onto something you already know.

A hard outcome (also called a clinical endpoint) is a thing that matters in itself: died, or did not. Developed cancer, or did not. Years lived free of disease. A surrogate marker (a biomarker endpoint) is a measurement you hope stands in for that outcome because it is faster and cheaper to read: a cholesterol number, an inflammation panel, the reading off an epigenetic clock (a DNA-methylation estimate of biological age, examined in depth in LON-3). The surrogate is a proxy for the outcome, not the outcome.

Here is your analogy. A surrogate marker is a proxy metric, and a hard outcome is the real KPI. You have watched a team optimize click-through rate until it went up and revenue did not budge, because click-through was only ever a proxy for the thing they cared about. This is Goodhart's law: once a measure becomes the target, it stops being a good measure. Biology has its own graveyard of exactly this. Drugs have raised HDL "good" cholesterol (a surrogate) while increasing deaths (torcetrapib). A trial suppressed heart arrhythmias (a surrogate) while the antiarrhythmic drugs raised mortality (the CAST trial). The marker moved the right way and the humans died anyway.

Why mice lie, and why humans are worse to study

Even a clean, positive mouse result translates to humans far less often than intuition suggests. Mouse and human physiology, metabolism, and drug handling diverge, lab mice live in a sanitized cage under conditions nothing like a human life, and many published mouse "wins" do not even replicate in another mouse lab. The serious answer to that last problem is the US National Institute on Aging's Interventions Testing Program, which tests compounds in genetically varied mice at three independent sites with randomization and blinding, so a result has to survive replication before anyone believes it. Very few compounds pass. Rapamycin is one that did. That is the bar a mouse claim should clear, not a single hopeful paper.

And the human observational tier has its own signature failure. Because nobody was randomized, confounding routinely produces effects that later reverse under an RCT.

Reading a claim like an engineer

Turn the three demands into a checklist you run on every headline. Read this panel as a mental gate, not runnable code: it encodes the standard, it does not compute truth.

# Gate a longevity claim on the three demands. This filters HYPE.
# It is a checklist, not a p-value, and passing it is necessary, not sufficient.

EVIDENCE_TIER = {
    "cell_culture": 1,         # a dish of cells, no organism
    "worm_or_fly": 2,          # very short lived, very different biology
    "mouse": 3,                # a mammal, but translation to humans is weak
    "human_observational": 4,  # real people, but confounded, not randomized
    "human_rct": 5,            # a randomized control group, the top rung
}

def triage(claim):
    tier = EVIDENCE_TIER.get(claim["organism"], 0)
    reasons = []
    if tier <= 3:
        reasons.append("shown only in cells or a short-lived animal, may not translate")
    if not claim["has_control_group"]:
        reasons.append("no control group, effect could be placebo or regression to the mean")
    if claim["outcome"] == "surrogate":
        reasons.append("a marker moved, not a life lengthened or a disease avoided")
    return {"believe_yet": len(reasons) == 0, "why_not": reasons}

worm_headline = {
    "organism": "worm_or_fly",
    "has_control_group": True,
    "outcome": "surrogate",
}
print(triage(worm_headline))
# believe_yet is False: a short-lived animal AND only a surrogate marker moved

One more filter the code does not show, because it lives in the statistics. Watch for relative risk with no absolute risk (a "50 percent reduction" of a rare event can be trivial in real terms), for tiny sample sizes, for outcomes that were not the study's pre-registered goal (fishing many endpoints until one turns up significant), and for the funder. The field carries genuine noise and real financial conflicts of interest: supplement sellers, biotech founders, and longevity clinics all have reasons to round up. None of that makes a claim false. It sets how hard you make the claim work.

Where this leaves you

You now hold the filter the rest of the track runs through. For any longevity claim, you ask three questions in order: what organism, was there a control group, and was the outcome real or a moved marker. You keep lifespan and healthspan separate, you treat a surrogate as a hypothesis rather than a result, and you remember that mouse wins translate rarely and observational correlations reverse routinely. Hold this standard evenly. Apply it to the claims you dislike and, harder, to the ones you are rooting for. That even hand is what turns you from a longevity enthusiast into someone who can actually tell which of these interventions is worth your one real life.