How we compute every reading.

• Decision support, not recommendations. Readings describe measurable market conditions — regime, breadth, base rates, probability, historical context, market structure — and what historically followed them. No page says buy or sell, names a price target, or picks stocks. Studies that found no edge are published with that conclusion stated, not shelved.
• One database, one universe. Every market tool is computed from the same internal database of daily OHLCV bars for roughly 5,500 stored symbols, sixteen years deep. Each indicator applies documented eligibility rules; a typical breadth session contains roughly 4,750 US common stocks. That consistent base means breadth tools, signals, and heatmaps agree on what "the market" is.
• Reproducible market calculations. Each market indicator is computed by code, not by hand, on a fixed schedule. Its formula — periods, thresholds, and smoothing — is documented on the tool's own page. Proprietary strategies follow the separate disclosure standard below.
• Deviations disclosed. Where our implementation departs from a classic textbook specification (universe size, integer-priced symbols, threshold choices), we say so and explain how it changes the reading. Different inputs produce different historical counts; we show our inputs.
• Dated to the close. Every reading is stamped with the market session it was computed from. If you quote this site, you know exactly which close you are quoting.
• Freshness-gated. The pipeline refuses to compute on incomplete data, and pages flag any reading that has fallen behind. We would rather show a staleness notice than a confidently wrong number.
• Honest base rates. Our signal studies report the full sample — average, median, hit rate, best, and worst — never a cherry-picked statistic. A small sample is labeled as a small sample.

Strategies are held to a second standard because publishing a complete formula would disclose the product itself. The public record is designed to let a reader evaluate the evidence and limitations without making the model reconstructable.

• Point-in-time signals. A simulation can use only information available by that session's close. Any resulting portfolio change is priced at the following market open, matching the timing shown to subscribers.
• Separate development and evaluation. Strategy choices are made on an earlier research window and then tested on later history that was not used to fit those choices. The parameters are not quietly re-tuned to rescue the later result.
• Costs before publication. Published performance uses the cost-adjusted record at the displayed $100,000 account size where the execution model is calibrated. A stronger frictionless backtest cannot substitute for a failed post-cost result.
• A fixed publication gate. Every strategy must clear the same post-cost return and drawdown standard on both the full record and the untouched holdout. The publisher refuses the catalog if any included strategy lacks a calibrated cost record or fails that gate.
• Failures stay out of the product. A compelling story, isolated winning period, or strong gross return does not override the gate. Passing it is still not evidence that future results will resemble the backtest.

Public strategy pages explain the economic idea, broad classes of inputs, portfolio behavior, execution timing, testing assumptions, and principal risks. Exact formulas, weights, lookback windows, ranking cutoffs, filters, entry and exit rules, development experiments, and implementation code remain proprietary. The systematic strategies are no longer published here; see the public record of material product and research updates.

• Equity prices & internals — our own daily OHLCV database.
• Economic indicators — the Federal Reserve's FRED service, with the source series id listed on each indicator page.
• Fund holdings — 13F filings from SEC EDGAR, with CUSIP-to-ticker resolution.
• Options — end-of-day options-chain snapshots.

Each tool page carries its own methodology block — see, for example, the Hindenburg Omen (where we explain the McClellan confirmation and clustering rules) or the signal studies (where we explain the cooldown and forward-return windows). For the bigger picture, read how it works.