The discipline of what we do not publish is itself part of our methodology. The Institute maintains a clear separation between the public corpus — methodology, killed signals, structural research — and privileged work that lives in client engagements, sponsored research, and the trading record. The lines below explain where those separations fall, and why.

Why we do not publish per-signal verdicts publicly

The verdict matrix — which signals we judge CONFIRMED, which we hold at SUGGESTIVE or CONTESTED, which we treat as actively tradable — is the operational output of the research programme. It is also what we sell to institutional clients and what informs the trading book. Publishing it would give away both at once.

This is not paywalling for its own sake. The public surface of the Observatory carries the methodology in full, the killed list in full, and live aggregate statistics — corpus size, verdict distribution, kill rate. A reader who wants to evaluate whether the framework is rigorous has everything they need on the public side. What they cannot see, by design, is which specific signals we currently believe in and how strongly. That information is the work product.

Why we do not name commercial sponsors of cited trials

Many of our cluster analyses reference active clinical trials. The sponsors of those trials are matters of public record on ClinicalTrials.gov and in regulatory filings. Even so, we do not surface sponsor identities on Institute pages alongside our convergence findings.

The reason is operational, not legal. Trial readouts and FDA decisions move markets. A page that pairs an Institute-confirmed mechanism with a named pivotal-trial sponsor reads, to a sophisticated reader, as a directional bet — whether or not we intend it that way. We surface the science: the mechanism, the dosimetry, the endpoint structure, the cross-disciplinary support. Sponsor identification belongs in privileged research delivered under engagement, where the audience has accepted the framing.

Why we do not surface holdout-set statistics

Out-of-sample performance — how a signal behaves on data the model never saw during construction — is the single most informative number for distinguishing a working model from a fit-to-history one. We compute holdout-set statistics on every battery-validated signal. We use them internally to gate promotion from CONFIRMED to higher tiers.

We do not publish them. Holdout performance, paired with a public verdict, is sufficient to reconstruct what we believe and how aggressively. What we do publish is the kill rate — the fraction of tested hypotheses that failed our gates. That number is the honest aggregate signal of whether the methodology discriminates. Per-signal holdout numbers stay inside the system.

Why we do not maintain “what survives” lists

A killed-signal page is pedagogically useful. It shows what we tested, what failed, and why — which is exactly the kind of disclosure that distinguishes a research programme from a marketing exercise. We publish ours in full.

The mirror page — a “signals that survived our gates” list — looks symmetric but is not. A surviving-signals list is a model of commercial direction: every entry is a thesis the Institute is willing to defend, and the collection as a whole describes the shape of the trading book. We publish the killed list because it costs us nothing and informs the reader. We do not publish the surviving list because it would cost us the work.

Why our findings have not been externally peer-reviewed

The most important caveat on this site is also the easiest to overlook, so we state it here in its own right. None of the findings published by the Institute — including those carrying our strongest verdicts — have been subjected to adversarial external peer review or independent replication by unaffiliated teams. Our validation is rigorous and internal: a statistical battery, a devil’s advocate protocol, and an epistemic rubric, all run inside one research programme.

That is a real gap, not a formality. External replication by parties with no stake in the outcome is the step that moves a finding from “survived our stress-testing” to “part of the scientific consensus.” We have not taken that step, and a reader should weight every verdict accordingly. The full discussion of what this means for verdict interpretation is on the Methodology page.

What we do publish

Methodology in full, including the eight-step validation framework, the canonical verdict tiers, and the conditions under which a signal is downgraded or killed. Killed signals with mechanism-level explanations. Structural research papers on long-cycle phenomena. Cross-disciplinary synthesis where convergence is unambiguous and the commercial surface is thin. Live aggregate statistics on the corpus.

The discipline of the public corpus is itself the credibility signal. A research programme that publishes its kill list, its failures, and its methodological caveats — without leaking the verdict matrix or the holdout statistics — is making a claim that the Institute stands behind both halves of the separation. Readers who want to engage on the private side can reach the Institute through proprietary research. Readers who want the public corpus in full can begin at the Observatory.