← Quantery Blog

Strict gates vs score gates in a stock screen

September 22, 2026 · 8 min readthesis-buildingscreeningmethod

A strict gate admits a company only when every scored criterion clears its minimum. A score gate admits it when the criterion scores add up to a chosen total, so strength in one area can offset weakness in another. Use a strict gate for conditions your thesis says are individually necessary. Use a score gate when the thesis is about a combination of imperfect evidence.

That choice is part of the claim. It decides whether a cash-rich company can compensate for mediocre profitability, whether a wonderful business can compensate for a stretched valuation, and whether one missing input can be washed out by points elsewhere. Don't leave the decision until after you see which companies pass.

What does a strict gate require?

In Quantery, a strict gate requires every criterion to score at least one point. A zero anywhere rejects the company. The criteria can still have graded bands, but the weakest one controls admission.

That structure fits a thesis built from necessary conditions. Consider a net-current-asset-value screen. The discount is the reason for the screen, short-term solvency protects the liquidation floor, cash burn tells you whether that floor is melting, and dilution tells you whether management is transferring it away. If the company isn't trading below the floor, a superb current ratio doesn't rescue the claim. If it is burning through the floor, a deeper discount may only mean the arithmetic is aging quickly.

This is a conjunctive claim: discount and solvency and controlled burn and controlled dilution. The bundled Graham Net-Net template expresses that with mode: strict. The gate prevents a company with one glaring failure from collecting enough points elsewhere to pass.

Strict doesn't mean every criterion is equally important. A criterion can award a stronger score for a deeper discount or a better current ratio, and those points can still order the companies that pass. The gate answers admission. The total score answers rank among admitted names.

The weakness of a strict gate is the cliff. A company scoring zero on one criterion and strongly everywhere else is out, while a small move across that criterion's lower boundary puts it in. Measurement noise can therefore change membership abruptly. If the economic story never treated that boundary as a dealbreaker, strict mode may be making a stronger claim than you intended.

What does a score gate permit?

A score gate sums the criterion scores and admits a company when the total reaches min_score. It represents compensating evidence. A strong result in one family can make up for a weak result in another.

Suppose a quality-and-value thesis scores operating returns, cash backing, balance-sheet strength, and valuation. There are plausible companies that deserve examination despite one middling dimension. A debt-free business with durable cash generation may have only average recent growth. A high-return business may offer an acceptable valuation without clearing the strongest yield band. If the claim is that several clues together identify an interesting case, forcing every clue to pass can discard the very trade-offs the score was meant to describe.

Compensation has a price. A total of four can hide many shapes. It might be two strong criteria and two failures, or every criterion sitting in the middle. Those aren't the same company. A score gate collapses the pattern unless you inspect the criterion scores behind the total.

It can also let a dominant family do too much work. If profitability appears in several overlapping criteria, a company can pile up points from the same underlying fact and offset weak cash backing or leverage. Before using a total-score threshold, remove duplicate votes from the thesis. Compensation only makes sense across evidence that can disagree for real economic reasons.

Missing data needs separate treatment. Quantery marks a name partial when a feature listed under quality.required is null, and partial names fail the gate regardless of their points. Use that block for inputs without which the thesis can't be evaluated. Don't rely on a score gate to handle an absent filing value; otherwise other criteria may compensate for something you never measured.

Which rules belong in the gate?

Write each rule as a sentence before choosing a gate mode. Use “must” only when you'd reject the thesis even if every other observation looked good.

A necessary-condition sentence sounds like this: the balance-sheet discount must exist. Or: trailing cash flow must be positive. If your claim falls apart when that sentence is false, make it impossible for points elsewhere to rescue the company. A strict gate is one way. A required quality field or a narrowly defined thesis can do the same job.

A supporting-evidence sentence sounds different: higher cash conversion strengthens the case. Or: lower leverage makes the case safer. These facts can contribute without being mandatory. Put them in scored bands and let the total express the accumulated support.

The distinction is economic, not cosmetic. A valuation rule may be mandatory in a deep-value thesis and supporting in a quality thesis. Leverage may be a dealbreaker when refinancing risk is the subject, but only a graded caution when screening mature cash generators. There isn't one correct gate for a ratio in isolation.

Keep the lower bands defensible. In strict mode, the one-point boundary is the real gate. In score mode, every point helps a company reach the total. A throwaway “acceptable” band isn't harmless in either design. It changes membership.

How do you write both designs?

The shape below is illustrative, and the bundled templates are the reference for exact fields. The Quantery thesis documentation covers the full DSL.

# Four distinct claims: illustrative
params:
  roe_ok: 0.10
  conversion_ok: 0.70
  leverage_max: 2.0
  yield_ok: 0.03

features:
  ni_ttm: ttm(net_income)
  fcf_ttm: ttm(free_cash_flow)
  equity: latest(total_equity)
  debt: latest(total_debt)
  cash: latest(cash)
  ebitda_ttm: ttm(ebitda)
  roe: if(equity > 0, ni_ttm / equity, null)
  conversion: if(ni_ttm > 0, fcf_ttm / ni_ttm, null)
  leverage: if(ebitda_ttm > 0, (debt - cash) / ebitda_ttm, null)
  fcf_yield: if(market_cap > 0, fcf_ttm / market_cap, null)

criteria:
  quality:
    rules:
      - { when: "roe >= $roe_ok", score: 1 }
      - { else: 0 }
  cash_backing:
    rules:
      - { when: "conversion >= $conversion_ok", score: 1 }
      - { else: 0 }
  balance_sheet:
    rules:
      - { when: "leverage <= $leverage_max", score: 1 }
      - { else: 0 }
  valuation:
    rules:
      - { when: "fcf_yield >= $yield_ok", score: 1 }
      - { else: 0 }

gate:
  mode: strict

With mode: strict, every criterion must score at least one. To test a compensating-evidence claim, keep every feature, rule, universe setting, and threshold fixed. Change only the gate:

gate:
  mode: score
  min_score: 3

Now one failed criterion can be offset by the other three. That's a materially different thesis, even though none of the accounting has changed. Give it a new version, write down why one failure is acceptable, and inspect which failure the admitted companies tend to carry.

You can also protect one true dealbreaker while using a score gate for the rest. Put the indispensable input under quality.required, narrow the universe, or separate the screen into a mandatory condition and supporting criteria. The important part is that the code match the sentence you wrote before the run.

How should you compare gate designs?

Run the strict and score-gated versions over identical dates with the same universe, benchmark, reporting lag, and rebalance cadence. These backtests are hypothetical and exclude costs. You're comparing structures, so changing another setting at the same time ruins the comparison.

Start with membership, before returns. Which companies appear only under the score gate? For each one, record the failed criterion. If nearly every added name fails cash backing, the score gate isn't broadly forgiving; it is making a specific decision that cash backing can be offset. Say that out loud. Does it still sound like the thesis?

Then count failure patterns across rebalance dates. A minimum total may repeatedly admit one shape because the easiest criteria move together. That's another route by which a score can turn into a hidden sector bet. Break out the added names by sector and size before attributing any performance difference to better gate design.

Read the return comparison last. Ask whether the direction holds across subperiods and small threshold changes, following the one-dial-at-a-time robustness method. If the score gate wins only because one cohort with the same failed criterion had an exceptional period, you've learned about that cohort. You haven't established that compensation is generally superior.

Also compare turnover. A score gate near its minimum can admit and remove companies whenever one criterion gains or loses a point. A strict gate can be twitchy at each criterion's lower boundary. Either design may generate membership churn, and the backtest excludes the trading costs that churn would create.

The gate should expose the claim

A strict gate says every named condition is necessary. A score gate says the evidence can compensate. Neither is inherently more rigorous. Rigor comes from choosing the structure before seeing the output, handling missing data outside the point total, and checking the criterion patterns hidden by the final verdict.

Start with the business sentence. Mark the words that mean must. Keep those conditions non-negotiable. Let genuinely independent supporting evidence add up, then run the alternate design as a challenge. Don't make it a replacement chosen after the curve appears.

Quantery keeps the gate beside the features and scoring rules because this decision shouldn't be buried. Change it, preserve both versions, and compare what crossed the boundary. The rules are yours. The trade-offs should be visible too.

Want to try this on your own rules? Quantery is free for 14 days: the full app, no card required.

← All articles