Methodology & Sourcing

How the work is built, graded, and sourced — and where its limits are.

AtomProphet is anonymous by design, so the work has to stand on its method, not a résumé. This page is the audit trail: how every causal link in the Cascade Graph is graded, where the numbers come from, and an honest account of what the analysis does and does not claim.

The Cascade Graph in one paragraph

The Cascade Graph has 477 nodes, 729 directed mechanism edges, and 46 directed cycles. Nodes cover drivers, chokepoints, jurisdictions, geographies, substitutes and terminal sets. Each arrow states a proposed mechanism; 60 do not currently carry a direct citation URL. Of the cycles, 36 have an undetermined reinforcing/balancing sign. A plausible mechanism, a live link and a profitable trade are three different claims.

How every link is graded

An edge carries its previous basis grade when its evidence is attached. A missing direct citation is now marked citation missing, and unsupported ticker links are thematic link unverified, not recast as measured facts. A reachable page alone does not prove that the stated causal relationship follows.

Edges also carry a relationship type — drives demand for, strains, is the bottleneck for, constrains supply of, controls production of, is an input to, enables, shifts the geography of, transmits price to, substitutes for, accrues value to, exposes risk to, and the loop-closing amplifies — each with a defined direction of effect. Open any node's evidence file in the graph and you can read each link, its grade, its magnitude, and its time horizon.

Where the data comes from

The research draws on the IEA, World Nuclear Association, S&P Global, peer-reviewed science, official records, issuer filings and sponsor fund pages. The October 2026 audit found cross-issuer citation reuse in inherited ticker bundles and inaccessible node sources; newly reviewed issuer links are separated from unreviewed identity candidates. An HTTP-accessible source is not automatically claim-supporting. Historical article-price charts are not current valuations.

Seven previously metricless driver nodes now carry dated, primary-source figures with scope caveats. Four substitute/backstop nodes were also rechecked against IEA, IRENA and peer-reviewed sources: sodium-ion is not universally nickel-free; LFP still needs lithium; recycling is not a guaranteed commodity price cap; and alternative-protein resource savings here are modeled, not observed. Two unsupported robotics numbers were withdrawn from displayed evidence; other broken node citations were replaced only when the original claim matched an exact source. A second reader examined 253 formerly uncited mechanisms, narrowing or neutralizing wrong-endpoint claims; 60 links still lack direct citations. Older node metrics were not all independently rechecked.

Where issuer evidence contradicts the narrative, the data wins: for example Mosaic's sulfur input costs undermined an inherited sulfur-beneficiary label. Unsourced edges remain visible only with explicit missing-citation or unverified labels pending rework.

How tickers are selected — the screen we actually run

As of 2026-10-02, the graph contains 94 terminal sets with no identified issuer-evidence exception, 186 sets with gaps or contradicted instrument links, and 4 no-clean-listed-play dispositions. Across 690 instrument mappings, the reconciled issuer/fund review reports 27 supported, 325 conditional, 153 gaps, and 185 contradicted exposure claims. A separate second reader examined 598 formerly single-pass mappings; 92 other mappings retain their earlier scope. A retrieved quotation is a provenance check, not proof of economic materiality. Structural connectivity, issuer identity and a tradeable security do not prove value or personal suitability.

A terminal set carries up to three instruments. Some carry only one because one documented exposure is more honest than three weak proxies. Four legacy terminal sets are explicitly marked no clean listed play. Forces, jurisdictions, and geographies are not forced to carry direct tickers when the causal path properly runs through a downstream chokepoint.

What this is not: a valuation, price target, return forecast, full credit/solvency rating, entry-timing signal, portfolio plan, suitability judgment or buy list. Investment suitability has not been assessed for any security. A real physical shortage does not mean the equity is cheap or profitable. “Reviewed” means only the node-specific issuer evidence was checked to the stated scope; current price attractiveness was not.

Size & liquidity reference

Below is an 18-ticker historical sample generated 2026-06-26, not a live or exhaustive graph-wide market screen. For each graph instrument use its separately dated card and current market/issuer information. This table does not screen solvency or valuation.

Who is AtomProphet?

The author writes anonymously to keep the focus on the data, not the byline. But the background shapes the method, so it is worth being plain about it. AtomProphet is written by a former computational mathematician turned applied scientist and engineer, who has spent a career building large-scale machine learning and anomaly-detection systems—the kind of fraud and bot-detection work that means living in petabytes of messy data and learning, the hard way, how easy it is to fool yourself.

That work comes down to one stubborn habit: pulling a faint, real signal out of overwhelming noise, and staying honest about the difference between the two. For years now, quietly and as a serious second pursuit, the author has turned that same discipline toward portfolio construction and macro research. The Cascade Graph is nothing more exotic than that—an anomaly-detection mindset, patiently pointed at the physical economy.

The standing rule on honesty

The goal is to show real numbers with their source, date and scope, and label projections or inferences separately. Some inherited graph claims have not yet been independently rechecked; unsupported numbers are withdrawn from display, while still-auditable hypotheses carry explicit gaps. Historical returns need their drawdowns beside them; neither a cited constraint nor a market-data check is proof of a sound investment. Counter-arguments and failure modes belong in view, not buried.

Common questions