Sigma Score
The platform’s canonical score summarizes multiple dimensions of company and market data. The authenticated workspace provides factor contributions, input availability and weighting controls. A score is an analytical summary, not a promise of return.
Historical Fair Value Corridor
The historical valuation view compares market price with model-derived fair value and a P25–P75 corridor. Its valuation assumptions and available history matter. This corridor is not a statistical confidence interval or a guarantee that price will stay inside it.
Relative Map
Quality and relative value are separate 0–100 axes. Their combination places companies in Premium quality, Relative opportunity, Weak relative setup or Value with trade-offs. Missing or non-comparable inputs should not imply a favorable classification.
Analyst view
External analyst opinions are separate from Sigma Score and the valuation models. Ratings, coverage and target prices reflect their source and observation date; they are not independent proof of value.
Portfolio
The personal portfolio tracks transactions, positions, cash and performance. Deposits and withdrawals are treated separately from investment returns. Available historical data and currency conversion affect comparisons.
Model Portfolio
Model actions are subject to eligibility, evidence, allocation and risk rules. BUY, ADD, TRIM and SELL describe changes in model positions. The official live record is separate from illustrative examples and rehearsal runs.
Research
Sigma Estimates and Earnings Memos
What our estimate is
The Sigma Estimate is a statistical estimate: it starts from analyst consensus and adjusts it by how far the company has typically beaten consensus over its last eight quarters. Version 2.1, now in validation beside version 2.0 and used for published estimates only after it passes its backtest gate, adds: recent quarters weigh more (a quarter one year back counts half as much), the adjustment is pulled back toward consensus when recent beats have narrowed, and it never exceeds the company’s own 75th-percentile beat. Where the company has published guidance for the quarter in its SEC earnings release (Form 8-K, Exhibit 99.1), each estimate states whether it is within, above or below that range, and how often the company has exceeded the top of its own range. The range is reported, not used as a cap: in our backtest companies finished above the top of their own range in about three quarters of guided quarters, and capping made the estimate less accurate.
How the stance is set
The memo stance (Constructive, Neutral or Cautious) comes from a published rule with five inputs, each scoring +1, 0 or −1. The rule does not use our own estimate, so a company that usually beats consensus does not score automatically.
- Company guidance (next quarter, midpoint) vs. consensus — +1 at ≥ 1% · −1 at ≤ -1% (SEC 8-K Exhibit 99.1)
- Consensus revision over 90 days (next quarter revenue) — +1 at ≥ 2% · −1 at ≤ -2% (SigmaThesis consensus snapshots)
- Last quarter: reported revenue vs. consensus — +1 at ≥ 2% · −1 at < 0% (company report)
- Sigma Score (0–100) — +1 at ≥ 70 · −1 at ≤ 40 (screener)
- Sigma Value factor (0–100, higher = cheaper) — +1 at ≥ 60 · −1 at ≤ 30 (screener)
- Total +2 or more is Constructive, −1 or less is Cautious, Neutral in between. Conviction is High at a total of ±3 or more, Medium at ±2, otherwise Low; with fewer than four available inputs it is at most Medium.
What this does not capture
The stance is a quarter-level signal, not a valuation and not advice to buy, sell or hold. It does not capture product and pricing cycles, contract timing or advertising-market swings, nor the items below.
- Macro and currency effects.
- One-off items such as restructuring, legal or tax charges.
- Changes in management or strategy.
Our record
Every estimate is locked before the report and settled against the reported figure, which is checked against the company’s SEC filings. Live results, including the quarters where consensus was closer, are on the track record.
In development: fundamental estimates built from industry drivers in public SEC data (first: remaining performance obligations for software companies). They run in shadow beside the statistical estimate and replace it for a company only after at least four live quarters with a smaller error.