How these analyses are built

Throughout the site, four kinds of statement are deliberately distinct:

• Direct observations — counts pulled straight from a source (e.g. a paper appeared in Nature; a trial is recruiting).
• Calculated metrics — defined arithmetic on those counts (the diffusion score, growth CAGRs, interventional %).
• AI-assisted classifications — categories assigned by a language model (a paper’s application area or research theme).
• Expert interpretation — the written conclusions and directional calls. These are judgment, not measurement.

Where a number is an estimate or a directional signal rather than a precise measurement, the surrounding text says so.

Question. How far has each technology diffused through the research community, and how much novelty premium does it still command?

Source & cutoff. Papers from PubMed via the NCBI E-utilities API, by publication year. Data current as of June 2026; the 2026 year is partial and annualized or flagged where shown.

Inclusion. Each technology is a fixed PubMed query (e.g. NGS = “next-generation sequencing” OR “high-throughput sequencing”). High-plex multi-omics deliberately requires a named method (CITE-seq, 10x Multiome, Olink + sequencing, spatial multi-omics, proteogenomics) rather than the loose term. Full queries are in each index’s methodology note.

Definitions. The diffusion score = top-tier papers ÷ total papers × 1,000 for a year. Top-3 = Nature, Science, Cell. Tier 1+2adds the leading specialist journals. The novelty premium is that score; it is aleading indicator that falls before adoption peaks. Adoption phase (Rogers: Innovators → Early Adopters → Early Majority → Late Majority → Laggards) is a separate,concurrent axis — a field can be Early Majority by adoption while its score is near the floor.

Phase calculation. A forward-only state machine combines two signals — publication-growth momentum (median of 1-, 3-, and 5-year CAGR) and the novelty premium (normalized to each technology’s own peak). Early phases are volume-gated; the chasm crosses when the premium falls below 30% of peak or a field exceeds 750 papers/year; late phases trigger on growth deceleration relative to the ambient growth of the literature (~4%/yr). Thresholds were calibrated against expert ground truth, then held fixed across all technologies. The Adoption Index is the single source of truth for phase; pages and notes derive their phase labels from it.

AI-assisted classification. Application categories and research themes are assigned by Anthropic’s Claude (model: claude-haiku-4.5, via the Batch API) from each paper’s title and abstract, into a fixed, tech-specific taxonomy. Country attribution is keyword matching on the first-author affiliation string, not a model. Aggregate distributions were reviewed for face validity; individual classifications were not hand-checked.

Limitations. This measures scholarly diffusion, not clinical or commercial value — a low or falling score is silent on whether a technology matters in the clinic. Keyword queries miss papers that don’t name a method and may catch tangential ones. Term commoditization (e.g. “multi-omics”) inflates volume without implying novelty. Some years hit PubMed’s result cap (flagged on those pages). The atlas/method and clinical/translational classification boundaries are where the model is softest, so per-category splits are directional; the macro shape is robust.

Update frequency. Refreshed when the underlying PubMed pulls are re-run (roughly quarterly); the data-cutoff label updates with them.

Source. Live queries to the ClinicalTrials.gov API (v2) at page-render time, cached and refreshed daily. The counts you see are current; the written conclusions describe the data at the last fetch and can shift as trials update.

Status & denominators. Unless stated otherwise, counts are recruiting trials only. Every percentage states its denominator (e.g. interventional % = interventional ÷ that group’s recruiting trials). The MRD tracker’s per-cancer counts use broad keyword searches that can overlap (a trial spanning two cancers is counted in each), so the headline total is the sum across tracked types, not a de-duplicated global figure.

Curated vs live. Spotlight trials, decision-type breakdowns, and milestone lists are hand-classified from registry entries and are point-in-time; they sit alongside the live counts and are labeled as such.

Limitations. Registry data is self-reported and uneven. “Interventional study type” does not guarantee the assay drives a treatment decision. Keyword searches over- and under-count. These trackers measure registered clinical activity, not real-world use or reimbursement.

Source. Public-market prices via Yahoo Finance, refreshed hourly. Baskets are equal-weighted averages of per-ticker returns, indexed to a common start; the denominator is the count of tickers returning data. Era bands and event annotations are editorial.

Limitations. Equal-weighting and basket membership are editorial choices; returns are price-only (not total-return) and not investment advice.

These analyses are living instruments and will contain errors. If you spot one — a number that doesn’t reconcile, a misclassified paper, a stale claim — please flag it and it will be fixed and noted.

Methodology last updated June 12, 2026.