What is the gross-profitability threshold FrontierLab uses?

Gross profit ÷ total assets (GP/A), pass at ≥ 0.33, caution at ≥ 0.20, judged on its absolute level — from Novy-Marx (2013).

Do published stock-screening edges decay?

Yes. McLean–Pontiff (2016) find anomalies are ~26% weaker out-of-sample and ~58% weaker after publication on average. FrontierLab treats every signal as decaying and validates forward.

Why is the ETF diversification verdict sometimes only 'caution'?

Top-10 concentration is not available on the current data tier, so when it can't be measured the diversification verdict is capped at caution rather than showing a confident pass on holdings count alone.

A research filter, not a predictor — not financial advice. Grades rank growth stocks on evidence-based quality signals from end-of-day data; they are not buy/sell calls or price forecasts. Always do your own research.

Evidence

This is the backbone the rest of the site links to. Every number the grader uses is here with its source and its caveat. Two honest limits frame all of it: published edges decay once widely known — about 26% weaker out-of-sample and ~58% weaker after publication on average (McLean–Pontiff, 2016) — and the base rate is humbling: most individual stocks underperform Treasury bills over their life; index returns come from a small minority of extreme winners (Bessembinder, 2018). FrontierLab is a filter that trims the obviously weak and obviously overpriced, not a path to beating the market. A grade is not advice.

1. Quality signals (the grade)

1.1 Gross profitability — GP/A = gross profit ÷ total assets. Pass ≥ 0.33, caution ≥ 0.20. Novy-Marx, "The Other Side of Value: The Gross Profitability Premium" (JFE, 2013). Among the most robust quality signals; judged on its absolute level (sector-ranking it would contradict the evidence).
1.2 Rule of 40 — revenue growth % + profit margin %. Pass ≥ 40, caution ≥ 30. A software-economics heuristic (Brad Feld / SaaS practice); applied to software-like models only, not capital-heavy sectors.
1.3 Revenue growth & trajectory. Pass ≥ 20%/yr, caution ≥ 10%. Plus whether gross margin is improving year-over-year. Durable top-line growth, not a one-off.
1.4 Net share issuance (dilution). Heavy share issuers systematically underperform; net repurchasers outperform (Fama–French, "Dissecting Anomalies", 2008; Pontiff–Woodgate, 2008). Persistent dilution caps the grade; buybacks are a positive.
1.5 Cash runway & burn (early-stage). Runway pass ≥ 18 months, caution ≥ 9; burn multiple pass ≤ 1.5×, caution ≤ 2.5×. Can a pre-profit company fund itself before a dilutive raise, and how efficiently does it burn.
1.6 Distress gate. Campbell, Hilscher & Szilagyi, "In Search of Distress Risk" (J. Finance, 2008): distressed firms earn low returns despite their risk. A distress signal disqualifies a top grade regardless of growth.
1.7 Valuation context — PEG ≤ 1 undemanding, ≤ 2 fair, ≤ 3 rich; P/S fallback ≤ 5 / ≤ 15 / ≤ 25. How much greatness the price already assumes (the "glamour" trap: Lakonishok, Shleifer & Vishny, "Contrarian Investment, Extrapolation, and Risk", 1994). Context, never a price target.

2. Gross margin — sector-calibrated absolute bands (pass / caution, %)

An absolute level per sector, so a thin-margin industry isn't failed against a SaaS bar it could never clear (within-cohort percentiles manufacture false precision in tiny cohorts). Default where unspecified: 60 / 40.

Cloud / SaaS 72 / 55 · Cybersecurity 72 / 50 · AI & chips 60 / 40 · Health / MedTech 60 / 38
Semiconductors 50 / 35 · Fintech 45 / 25 · Robotics 45 / 28 · Space 45 / 25 · Consumer 38 / 22
Energy 35 / 22 · Industrials 32 / 22 · AgriTech 28 / 18 · EV 25 / 15 · Materials 25 / 15

3. ETF rubric (a separate character read, never the single-stock grade)

Cost — expense ratio ≤ 20 bps = core, > 50 bps = fee drag. Bogle / Sharpe's arithmetic of active management; SPIVA scorecards.
Size & liquidity — AUM ≥ $1B (caution ≥ $100M); ≥ $10M traded/day (caution ≥ $1M). Sub-$100M funds carry closure/tracking risk (Fed FEDS 2020-097). Both inputs are shown so you see which drove the verdict.
Diversification — holdings ≥ 50 (caution ≥ 25); top-10 concentration ≤ 30% when available. ~25–30 names removes most unsystematic risk (Statman; Alexeev–Tapon, 2013). Top-10 concentration is not in the current data tier, so the diversification verdict is capped at "caution" when it can't be measured — an absent input never reads as good news.
Structure & maturity — age ≥ 3y; actively-managed funds capped at "caution." Active outperformance does not persist (SPIVA). Leveraged/inverse funds are hard-excluded from a buy-and-hold grade (daily-reset decay; FINRA/SEC).

4. What we deliberately do NOT do

No price targets or "fair value." DCFs and analyst targets run systematically optimistic; we don't publish them.
No RSI-as-a-sell-signal — momentum is shown as context, never a timing call.
Insider selling is not bearish — it's usually routine diversification (Lakonishok–Lee, 2001). Only cluster buying is a positive note.
No fake precision — every track-record figure carries its sample size (N) and window; small samples read "too early to tell."
End-of-day data only — no intraday, no real-time, no stop-loss simulation; cached data is never shown as live.
Grades frozen at entry in the paper-trade journal, so the record can't be rewritten after the fact.

5. Honest limits

These per-flag figures are observational — a flag firing is not a randomized assignment, so any edge can reflect what kind of names trip a flag, not the flag itself. Descriptive, not causal.

The forward track record measures ordering among names still covered at each horizon's end within the curated universe — not a survivorship-clean whole-market backtest; dropped names count as attrition, not hidden, and extreme single-session moves are treated as split/data artifacts and excluded. Read every figure with its sample size.

How a grade is computed → · The forward track record → · Full disclaimer →