Methodology

PCI — Pythia Company Index

Composite company-quality index: a Greenblatt-style rank aggregation of the PAS, PGS and PVS percentile ranks, published as a percentile of the ranked universe with a #rank.

Each of PAS, PGS and PVS is turned into a percentile across the ranked universe, those percentiles are averaged, and the average is itself percentile-ranked across every ranked company — so the published 0–100 reads as "this share of the ranked universe scores below this company". A company needs both PAS and PGS to be ranked at all. PVS joins where it exists but never gates, because its inputs backfill over a multi-day rotation and gating on it would empty the list. The honest caveat: PVS is only defined for about three-quarters of ranked companies — it is structurally undefined for banks and insurers, and for issuers with a negative long-run margin — so its percentile is measured against that narrower group, which differs systematically from the rest of the universe. Whether that helps or hurts a given company is not measurable, because the excluded companies have no PVS to compare against; what we can do is stop the difference leaking into the headline. Averaging the legs and then re-ranking the average puts every company back on one scale regardless of how many legs it had, and the size of the PVS group is disclosed on every company so the narrower base is never hidden.

Cites: Pythia methodology; ranking process per Greenblatt (2005) rank aggregation. Published 0–100. Per company.

PAS — Pythia Academic Score

Empirical-finance sanity check: Profitability 25% + Value 25% + Quality 25% + Momentum 15% + Growth 10%, sector-neutral z-scored within (GICS × country), displayed 0–100.

Five pillars from the empirical finance literature, weighted Profitability 25%, Value 25%, Quality 25%, Momentum 15%, Growth 10%. Every underlying metric is z-scored WITHIN its own (GICS sector × country) group before it counts, so a software margin is judged against software and a utility against utilities — the score answers “good for its kind”, not “good in the abstract”. Pillar scores are the normal-CDF of the mean of their available z-scores, and the composite renormalises over whichever pillars resolved, so one missing input dilutes nothing. Since 2026-08-30 the PUBLISHED score and pillars are the percentile rank of those figures across every scored company — a 92 reads “better than 92% of scored companies” — because the raw composite was compressed into 11–82 and its top words were unreachable; ordering is unchanged, so PCI does not move. Two further pillars — capital return and stability — are computed and stored but carry ZERO weight in the headline while their evidence is validated; they are labelled Labs wherever shown. An issuer whose sector is unclassified gets no score rather than a comparison against a group it does not belong to.

Cites: Novy-Marx 2013; Fama-French 5-factor; Piotroski 2000; Altman 1968; Beneish 1999; Jegadeesh-Titman 1993. Published 0–100. Per company.

PGS — Pythia Guru Score

Investment-philosophy thesis check: % of the 12 voting Gurus-NEW hurdles passed (13 measured — the Graham net-net vote is a badge that does not score; Buffett, Graham, Greenblatt, Lynch, Marks, Pabrai, Drucker, Thorndike, CAPEX Efficiency), displayed 0–100.

Thirteen pass/fail hurdles taken verbatim from named investors — Buffett, Graham, Lynch, Greenblatt, Marks, Pabrai, Drucker, Thorndike — each at the threshold that investor actually published; twelve of them vote and the Graham net-net test is shown as a badge. The score is the share of voting hurdles the company PASSES, out of those it can be judged on. A hurdle whose input is missing is excluded from the denominator rather than counted as a failure: absent data is not a bad result. Below eight judgeable hurdles nothing is published, because a percentage of three tests is not a verdict. The Mary Buffett relative-value pair (relative value, initial rate of return) is deliberately EXCLUDED from the score per the founder spec — the two are algebraically the same test and live on the Valuation › RV to Bonds page instead.

Cites: Gurus-NEW workbook; per-hurdle citations in guru-score/hurdles.ts. Published 0–100. Per company.

PVS — Pythia Valuation Score

Reverse-DCF reality check: the revenue growth, EBITDA margin, and expected return the current price implies, scored against academic base rates. Higher = cheaper expectations. Fills per company as the phase94 cash-flow history backfills.

Runs the published DCF backwards. Rather than asking what the company is worth, it takes today’s price as given and solves for what the market must be assuming: the revenue growth rate, the steady-state EBITDA margin, and the return implied by paying this price. Each implied assumption is scored against how often companies have historically delivered it, weighted 40/30/30 toward growth. A high raw reading means the price implies assumptions history says are commonly met; a low one means it requires an outcome few companies achieve. Since 2026-08-30 the PUBLISHED score is that reading’s percentile within the company’s own sector × market-cap-quintile cohort (falling back to the sector, then the universe, and saying which), because the raw reading called 71% of the universe overvalued and averaged 29 in the largest cap quintile against 43 in the smallest — a flat discount rate reads every large quality name as demanding. The engine is unchanged; the score now answers “expensive for its kind”. It refuses rather than guesses: banks and insurers are structurally excluded because an EBITDA-and-net-debt model does not describe them, and an issuer with a negative long-run margin, fewer than five adjacent fiscal years, or no price gets no score and states which. About three-quarters of ranked companies carry one.

Cites: Rappaport & Mauboussin, Expectations Investing (2001); Mauboussin & Callahan, The Base Rate Book (2016); Fama & French (2000) margin mean-reversion. Published 0–100. Per company.

PBS — Pythia Behaviour Score

How the market is treating the company: momentum, price behaviour, news tone and performance against its own sector-and-country cohort. Refreshed after each session’s prices; deliberately NOT a component of PCI.

A behaviour reading, not a quality or value one — higher means the market is treating the company more favourably and more calmly. Four areas from the founder’s specification: market momentum, price behaviour, market sentiment, and performance relative to its cohort. Each price measure is ranked WITHIN the company’s own sector and country rather than against a fixed threshold, because there is no published answer to “is 34% volatility high?” except “relative to what” — so a biotech is not marked down for out-moving a utility. News tone is the one absolute scale, because the classifier already emits a bounded −1 to +1 where zero means neutral. Beta is shown and deliberately NOT scored: orienting it would require claiming a direction, and neither “lower is better” nor “closer to 1 is better” survives as a statement about behaviour. A missing input leaves the denominator instead of scoring 50, so a company measured on one area is scored on that area and told you so; with no area measurable at all the score is absent rather than a midpoint. It does not enter PCI — PCI remains PVS, PAS and PGS.

Cites: Jegadeesh & Titman (1993) momentum; Asness, Moskowitz & Pedersen (2013) skip-month convention; Baker, Bradley & Wurgler (2011) volatility; Tetlock (2007) news tone. Published 0–100. Per company.

PMI — Pythia Macro Index

External-environment supportiveness: Macro + Market + Sector scores (rates, inflation, GDP, valuations, sector phase). Market-level, refreshed daily at 10:30 UTC.

A market-level reading, not a company one: the average of an economy score, a market score and a sector score, each built from named public series with a published band mapping the raw value onto 0–100. Every input carries its own status. Live inputs are measured and counted; inputs we cannot source are carried at ZERO weight rather than filled with an estimate — the score never borrows credibility from a number we do not have. The count of live inputs is stated alongside, because an average of six measured inputs and an average of two are not equally strong claims.

Cites: Pythia methodology v1; FRED + FMP inputs with published Buffett-indicator note. Published 0–100. Market-level.

PI — Pythia Insight

Synthesis of PCI (company), PMI (environment) and PB (behaviour): the equal-weight average of the live legs, plus the written interpretation of that average. Requires at least two legs. Narrative is on-demand, cached by inputs-hash.

The number is the equal-weight average of PCI, PMI and PB — the identity PI = PCI + PMI + PB, computed at render so it cannot drift from the three legs. A missing leg leaves the denominator; fewer than two legs refuse rather than invent a midpoint. The written interpretation reads those same three and explains where they agree, where they conflict, and what would have to change. It cannot introduce a fact that is not in the payloads, and it regenerates only when the underlying numbers change.

Cites: Pythia methodology; generated via the AI gateway, grounded in score payloads only. Narrative, grounded strictly in the score payloads. Per company.

Conviction

How strongly does the evidence support the view?

Each Investment-View dimension and the overall conviction read on one five-word scale. A dimension takes the word for its own 0–100 (the company index, an Academic pillar, the published Valuation score) or for its hurdle pass share; the Insights leaf takes the word for how convergent, broad, stable and directional the evidence is across the lenses, with a valuation disagreement carried as the stated constraint rather than as lost conviction.

Scale: Very Low · Low · Moderate · High · Very High.

Reads: Evidence convergence across the business lenses, the principal contradiction, evidence breadth, stability and direction since the last recorded snapshot.

Valuation View

What kind of valuation case is this?

The valuation view names the case the methods make together: how far the median of the resolved methods sits from the price, whether the methods agree, what kind of business Lynch’s categories say it is, and whether the Academic quality pillars back a quality qualifier. A word that needs evidence the record does not carry is never printed; cyclical and asset-value cases are disclosed as not yet assessable.

Scale: Deep Value · Quality at a Reasonable Price · Fairly Valued Compounder · Premium Quality · Expensive Growth · Potentially Overvalued · Speculative Valuation · Cyclical Value · Asset Value Opportunity · Turnaround Value.

Reads: The median of the resolved valuation methods against the live price, the board’s agreement, the Lynch category and the PAS quality pillars.

Earnings Environment

Are macro conditions helping or hurting this company’s earnings outlook?

The macro index is the same for every company in a sector; the earnings environment is how it reaches this one. Each live macro channel — rates, inflation, growth, credit, market valuation, risk appetite and the sector — is weighted by the company’s own sensitivity to it, and the weighted reading names the environment and its primary driver. Channels Pythia cannot source for a company (currency, commodities, consumer spending, the capex cycle) are disclosed as not assessed rather than guessed.

Scale: Strong Headwind · Headwind · Neutral · Tailwind · Strong Tailwind.

Reads: The live PMI components by channel (rates · inflation · growth · credit · market valuation · risk appetite · sector), each weighted by this company’s sensitivity (multiple, growth, leverage, cover, beta, cyclicality, sector premium).

Behavior Profile

How is the market behaving around this company?

Each raw behavior leg is standardized on its cohort — momentum, realized volatility, maximum drawdown, news tone and return versus the cohort — the four perspectives are scored from them, and the behavior profile is the pattern the four perspectives and their directions make together. A missing direction is never read as uncertainty; fewer than three live perspectives gives no profile.

Scale: Euphoria · Momentum Driven · Healthy Participation · Constructive Accumulation · Quiet Accumulation · Balanced · Uncertain · Declining Conviction · Capitulation.

Reads: The four behavior perspectives (momentum, price behavior, market sentiment, relative performance), each read on its own scale, and their direction since the last recorded snapshot.

Pythia Insights

How attractive is the investment?

The Insights word reads the numeric PI — the mean of the company index, the macro index and the behavior score — on a six-word scale from Unfavorable to Exceptional.

Scale: Unfavorable · Weak · Neutral · Favorable · Strong · Exceptional.

Reads: The numeric PI — the mean of PCI, PMI and PB.

Risk / Reward

Is the potential reward worth the risk?

Six legs vote: the reward (the median of methods against the price), how demanding the price already is, business quality, financial resilience, the macro backdrop and price behavior. The word is the balance of favorable and unfavorable votes over the legs that could be assessed; without a reward leg nothing is said.

Scale: Highly Unfavorable · Unfavorable · Balanced · Favorable · Highly Favorable.

Reads: The implied upside (median of methods vs price), the published PVS, the PAS Profitability and Quality & Safety pillars with the Altman zone, PMI and PB price behavior.

Market Temperature

How favorable is the investment environment?

Market temperature reads how hot the market is: valuation, risk appetite, sentiment, the macro backdrop, and — for a company — the expectations embedded in its own price. Cold reads as the favorable end because a cold market is the cheaper one.

Scale: Cold · Cool · Neutral · Warm · Hot · Overheated.

Reads: The PMS valuation components, the VIX and market-sentiment composites, PES, and optionally this company’s published PVS.