How AURYN reaches a view
Deterministic calculations, separate evidence and explicit limitations. Current screening version: evidence-lenses-v1.
Updated October 23, 2026 · research-release-20261023
Three separate questions
Business evidence, valuation and timing are shown separately. Short-term weakness does not by itself invalidate a long-term business case. Published analyst opinions remain distinct from assumption-driven sensitivities.
Eight evidence lenses
Valuation, profitability, growth, financial health, cash quality, owner returns, moat and ownership/expectations organize the evidence. Numeric scores use published absolute anchor rules in lib/decision-research.ts. They are screening scores, not validated stock-return predictors. Missing metrics do not become zero. A composite requires at least four scored lenses.
Style and coverage
Balanced, Value, Growth, Income and Quality change the blend of available category scores, not the underlying facts or category calculations. Peer percentiles, own-history percentiles, model agreement and probability-like confidence are omitted unless their supporting data and evaluation exist. An evidence count is not a probability of success.
Period and currency discipline
Financial comparisons use compatible period scope, reporting basis and currency. Flow metrics use comparable annual or complete trailing periods. ROE, ROA and accruals require comparable beginning/end balance data. Ratios with non-positive denominators are withheld rather than set to zero.
Cash-flow sensitivity
The interactive tool starts from a reconciled OCF-less-capex proxy per current share. Growth fades to terminal growth across ten years. It discounts each cash flow plus the terminal perpetuity. It is not a full enterprise DCF or a claim that reported FCF is distributable equity cash flow. Defaults are assumptions; no calculated WACC is implied.
Reverse and implied calculations
Bisection solves for initial growth or a discount rate only when the target lies inside a valid bracket. The implied discount rate is not a prediction of realized return. Banks, insurers, crypto and unreconciled foreign listings are excluded from the generic sensitivity model.
Methodology changelog
October 18, 2026: evidence-lenses-v1 introduced disclosed absolute anchors and five style blends. October 19, 2026: source-status details, observed-view comparison, financial crossing alerts, cash-runway and working-capital context were added. No historical scores are recalculated and presented as observations made at the time.
Worked examples
If revenue is 100 and operating profit is 20 in the same currency and period, operating margin is 20%. If operating cash flow is 10 and capex is 30, cash after capex is −20 and FCF margin is −20%. A negative margin can exceed 100% in magnitude when spending exceeds revenue. These are arithmetic illustrations, not company data.
Flags and review levels
Flags apply disclosed deterministic thresholds to available data, including thin interest coverage, cash use, high stock pay, dilution and liquidity. Review levels are on-screen prompts, not automatic trading rules or proof of business failure. They do not certify that untested risks are absent.
Timing and DCA
Daily technicals use completed dated observations. The monthly accumulation policy requires consecutive month-end closes and distinguishes confirmed from forming readings. Technicals help describe entry conditions; they do not establish business value or a moat.
Cycle-relative monthly policy
Monthly-v4 uses the mean of the previous 12 completed monthly closes and their sample standard deviation, with a spread floor of 10% of the mean. Eligibility retains 24 consecutive prior months. The 2×/3×/4× pullback gates require 10%/25%/40% below that mean and 0.5/1.35/1.5 spread units. Upper extension reviews retain the v3 robust-trend pause guard; a price at least three times the cycle mean and two spreads above it also triggers a review. A severe cycle review requires three spreads and a 50% period rise. Monthly parameters were calibrated against a small observed reference sample, not learned from a disclosed private formula. Weekly v1 is separate. US daily bars settle after 5 PM New York; crypto uses completed UTC days. Prior versions remain in source for reproducibility. Matching selected signals is not evidence of better returns.
Historical policy comparisons
Per-company simulations use the same 100-unit monthly deposits, carry unspent cash without interest, prohibit borrowing, and execute only after the prior month signal at the next month first observed open. The final chronological 30% is shown separately with accounts restarted at zero. Costs are configurable. Month-end drawdown does not measure intramonth losses. Historical prices can be revised; dividends, taxes, delisted-universe coverage and point-in-time fundamentals are not modeled. Synthetic tests establish calculation mechanics, not investing performance.
Evidence that needs more work
Company descriptions do not prove competitive advantage. Management guidance hit rates need actual management guidance, not analyst consensus. Institutional changes need dated holdings history. Historical valuation ranks need price and financial snapshots that were actually available at each date. These values are not fabricated from a current snapshot.
Research and commercial review
A point-in-time archive, delayed outcome evaluation and independent model validation are needed before publishing performance advantages. Data licensing and the product’s actual advice, personalization and execution features require review before commercialization. A disclaimer does not itself create a publisher exemption.