Part XIII · Discipline · Chapter 13B

The conviction score

When a setup deserves capital

Where we are

Chapter 13A measured how much of a paper Sharpe ratio survives into a retail account. This chapter builds the test that every candidate trade must pass before it gets capital. The test is a conviction score that multiplies five independent checks, so one failed check sinks the whole score. It ends with the Real Value score, the companion test for single stocks.

What you need first

The retail haircut (Chapter 13A).
The four sources of edge (Chapter 1G), and the five categories and evidence tiers of Chapter 10A.
Expected shortfall (Chapter 1B), tail correlation (Chapter 1F) and the stress scenarios S1 to S3 (Chapter 1D).
The composite signal (Chapter 11G), Weinstein's stages (Chapter 11E), and gamma exposure with its flip level and walls (Chapters 8A and 8C).
Contango and backwardation in the VIX term structure (Chapter 5E).

1 · What conviction should mean

"High conviction" should not mean that you believe in a trade strongly. It should mean that every independent layer of evidence converges: the evidence is robust, the mechanism can be stated, the regime is supportive, execution is feasible and the size is survivable. If any single layer fails, conviction collapses, however strong the signal feels.

In plain terms

Before a trade gets money, ask five questions: is the evidence good, do you know why it works, does today's market suit it, can you trade it, and can you survive it going wrong? Score each answer from 0 to 1 and multiply the five scores together. A trade that fails one question badly scores near zero, whatever its other answers.

Four conditions first

The Edge Book came to its score through a reviewer's challenge. The reviewer of its second version pointed out that it never defined "high conviction" formally. Without a definition, a framework is just a collection of edges with no clear threshold for action. The first answer was a four-part test: trade only if all four conditions hold.

\begin{aligned} &\hat{\mu}_{\text{net}} > 0, \quad \text{ES}_{95} < L_{\text{tol}}, \\ &\Pr(\text{margin breach}) < \epsilon, \\ &\rho_{\text{tail}} \text{ acceptable} \end{aligned}

(13B.1)

Here $\hat{\mu}_{\text{net}}$ is the estimated expected return after costs, and $\text{ES}_{95}$ is the expected shortfall at 95%, the average loss in the worst 5% of outcomes (Chapter 1B). $L_{\text{tol}}$ is the largest tail loss you can tolerate, and $\epsilon$ is a small probability you choose. The last symbol, $\rho_{\text{tail}}$ , is the tail correlation of the trade with the positions you already hold (Chapter 1F). In plain English, expected value must be positive after costs, the tail loss survivable, the probability of a margin failure low, and correlation with your existing positions in stress acceptable.

Chapter 1F wrote a threshold version of the same idea, in which the net return must beat penalties on tail risk and on execution cost.

Worked example: Four conditions on a put-writing sleeve

Step 1: the sleeve. Suppose a $500,000 portfolio puts $100,000 into cash-secured monthly XSP puts, with a hypothetical positive net expected return, so the first condition holds. Chapter 1D put the PUT index's monthly $\text{ES}_{95}$ at roughly −8%, which here is $8,000, or 1.6% of the portfolio.

Step 2: the next two conditions. If your tolerance $L_{\text{tol}}$ is 5% of the portfolio, $25,000, the tail loss passes. The puts are fully secured by cash, so no margin call can breach the account, and the third condition passes too.

Step 3: the fourth. In a crash the sleeve falls with the stocks you already hold, since its tail beta is about 1.0 (Chapter 10A). Whether that is acceptable depends on how much equity the portfolio carries, which no formula answers for you.

These four conditions are necessary but not jointly sufficient: strong evidence and good mechanism design also matter. So the Edge Book extends them to a score over five domains.

2 · The conviction score

For any candidate move $i$ at time $t$ , the conviction score, which the Dealer Mechanics Lab calls the C-score, is

C_i(t) = E_i\, M_i\, R_i(t)\, X_i(t)\, S_i

(13B.2)

Each of its five factors is a score from 0 to 1:

Factor	Domain	What it asks
$E_i$	Evidence quality	How well supported is the edge in academic and practitioner work, and has it been replicated?
$M_i$	Mechanism clarity	Can you say why the edge exists, and what would make it disappear?
$R_i(t)$	Regime alignment	Is today's market regime supportive of this specific move?
$X_i(t)$	Execution feasibility	Can you actually trade it, given its liquidity, spread, fill quality and operational complexity?
$S_i$	Sizing survivability	Can you size it so that S2 is operational and S3 recoverable, with kill criteria written down?

The two factors marked $(t)$ move with the market, so the same move can score differently next week. Evidence, mechanism and sizing belong to the move itself and to the size you choose. The Edge Book sets three thresholds:

Score	Conviction	Action
$C \ge 0.5$	High	Allocate, at full intended size
$0.3 \le C < 0.5$	Medium	Take a small allocation, or test the sizing
$C < 0.3$	Insufficient	Do not trade

The thresholds are heuristic but disciplined. The point is not the exact number but the multiplicative structure, which forces every domain to clear the bar.

Why multiply

The multiplicative form is the load-bearing assumption, since a single factor at 0 makes the whole product 0 and one weak factor drags down every strong one. Three examples show it, each with its other factors at 1:

Stellar evidence with unsurvivable sizing, $E = 0.95$ and $S = 0.10$ , gives $C = 0.095$ , however good the paper looks.
Strong evidence in a poor regime, $E = 0.9$ and $R = 0.2$ , gives 0.18, which is not high conviction.
A great mechanism with unsurvivable sizing, $M = 0.9$ and $S = 0.1$ , gives 0.09.

Each domain is a necessary condition. In operational reality you cannot compensate for one broken dimension with extra strength in another, and the product encodes exactly that.

Worked example: Averaging hides the broken factor

Step 1: one weak factor. Take four factors at 0.9 and a fifth at 0.2, whose average, $(4 \times 0.9 + 0.2)/5 = 0.76$ , looks strong.

Step 2: the product. Their product is $0.9^4 \times 0.2 = 0.656 \times 0.2 = 0.131$ , below the floor of 0.3: do not trade.

Step 3: no weak factor. Five factors at 0.9 multiply to $0.9^5 = 0.59$ , comfortably high. The average hides the one factor that the product exposes.

How good must each factor be? If all five sat at one level $g$ , their product would be $g^5$ . So the level that a score $C$ demands of the factors, on average, is its fifth root:

g = C^{1/5}

(13B.3)

This $g$ is the geometric mean of the five factors. High conviction, $C = 0.5$ , needs $g = 0.5^{1/5} = 0.87$ , and the floor of 0.3 needs $g = 0.79$ . A single factor at 0.6 therefore demands that the other four average $(0.5/0.6)^{1/4} = 0.955$ for high conviction, which leaves little room for any other weakness.

3 · The five factors, scored

The Dealer Mechanics Lab gives a scoring guide for each factor as it applies to swing trades. The Edge Book's guide, written for strategies in a portfolio, differs in places. The table gives the Lab's bands, with the Edge Book's differences in the notes after it.

Factor	Top band	Middle band	Low band
E, evidence	Tier 1, peer-reviewed, multi-decade and replicated: 0.8 to 1.0	Tier 2, strong empirical results that are sensitive to the sample: 0.5 to 0.8	Tier 3, thin, recent or decayed: 0.0 to 0.4
M, mechanism	A clear mechanism with academic support: 0.7 to 1.0	A partial story about the mechanism: 0.4 to 0.7	"I do not know why this works": 0.0 to 0.3
R(t), regime, for long swings	Composite of +40 or more, Stage 2 confirmed, positive GEX, contango: 0.8 to 1.0	Composite of +10 to +40, Stage 2 turning to Stage 3, neutral GEX: 0.5 to 0.7	Composite below 0, Stage 4, negative GEX, backwardation: 0.0 to 0.3
X(t), execution	SPX, SPY, QQQ or the Mag-7 with standard order types: 0.85 to 1.0	A multi-leg index option with manageable spreads: 0.6 to 0.8	A thin single-stock option with wide spreads: 0.2 to 0.4, or infrastructure you do not have: 0.0 to 0.2
S, sizing	Sized for survivability, kill rules committed in writing, drawdown plan documented: 0.85 to 1.0	Sized "comfortably", without explicit stress testing: 0.5 to 0.7	Sized aggressively, without kill criteria: 0.1 to 0.3

Evidence asks how well supported the setup is in academic and practitioner work, and whether it has been replicated across datasets and periods. Its bands follow the evidence tiers of Chapter 10A. The Lab's Tier 1 examples are Jegadeesh-Titman momentum, Connors RSI(2) mean reversion in trending markets, and Ni-Pearson-Poteshman pinning near options expiration (OPEX). That pinning paper studied optionable single stocks from 1996 to 2002 and found at least 16.5 basis points of return distortion per expiration date. It did not study the index: for S&P 500 futures the evidence is Golez and Jackwerth (2012). Chapter 11H grades Connors's specific RSI(2) rule lower than this list does. It calls the rule practitioner work, published in trade books rather than peer-reviewed journals, that rests on an academic short-horizon reversal effect.

Mechanism asks whether you can articulate why the setup has a positive expected value. In the Edge Book's version, naming the source among the four of Chapter 1G earns 0.7 to 1.0: risk premium, behavioral, institutional or microstructure. Evidence and mechanism are different questions: evidence asks whether the effect has held up in data, while mechanism asks why it should hold in the future and what would stop it.

Regime asks whether the current market suits this specific setup, and it changes with time. For long swing trades the Lab reads it from the composite signal of Chapter 11G, a score from −100 to +100. Three other readings join it: Weinstein's stages (Chapter 11E), the sign of gamma exposure (GEX, Chapter 8A) and the term structure. Its bands skip composites between 0 and +10, a zone that Chapter 11G calls neutral and in which it takes only the highest-conviction setups, at half size. The Edge Book's examples are for strategies rather than setups. For short volatility a backwardated VIX curve scores low, as does the middle of a drought for trend. The even week of the FOMC cycle scores high for equity (Chapter 10G).

Execution asks whether you can actually trade the setup in a retail account, with good liquidity, a low spread and clean operations. The Edge Book scores a liquid ETF at 0.9 to 1.0 and a multi-leg index option at 0.5 to 0.7, a little below the Lab's 0.6 to 0.8. Both score a thin single-stock option at 0.2 to 0.4.

Sizing asks whether you can size the trade so that the S2 stress scenario (SPX −20%, VIX up 25 points) is operational and S3 (SPX −35%, VIX up 50) is recoverable, with explicit kill criteria. Operational means the account keeps working through S2, with no margin breach and no forced sale. Recoverable means the S3 loss is one the account can repair (Chapter 1E).

4 · The score as a worksheet

The figure lays out the five factors with their scoring guides, one slider each, from 0 to 1 in steps of 0.05. The readout gives $C$ and a verdict: at 0.5 or more it prints "HIGH CONVICTION" and says to take the trade at full intended size. From 0.3 it prints "MEDIUM CONVICTION" and sizes the trade at 60% to 80% of the intended size. Below 0.3 it prints "INSUFFICIENT CONVICTION" and says to pass or reduce existing positions.

When $C$ falls below 0.5, the figure highlights the binding constraint, the lowest factor, which is what is killing conviction. The starting scores are the calm-regime Trend Pullback of the next example, with $C = 0.527$ . Set $R$ to 0.20, $X$ to 0.65 and $S$ to 0.40 to move it into the stress regime.

Factor and scoring guide	Score
E · Evidence Tier 1 peer-reviewed (0.8-1.0); Tier 2 (0.5-0.8); Tier 3 thin (0.0-0.4)	0.85
M · Mechanism Clear + academic (0.7-1.0); partial (0.4-0.7); 'don't know why' (0.0-0.3)	0.90
R · Regime (time-dependent) Composite ≥ +40 (0.8-1.0); +10-40 (0.5-0.7); < 0 for longs (0.0-0.3)	0.85
X · eXecution SPX/SPY/QQQ (0.85-1.0); multi-leg index (0.6-0.8); thin single name (0.2-0.4)	0.90
S · Sizing & survivability Kill rules pre-committed (0.85-1.0); 'comfortable' (0.5-0.7); aggressive (0.1-0.3)	0.90

conviction C = 0.527

HIGH CONVICTION. Take at full intended size.

One weak factor caps the whole score. Score a swing setup on the five factors of the conviction score, C = E × M × R × X × S, each from 0 to 1 by the guide under its name. C of 0.5 or more is high conviction, 0.3 or more is enough for some size, and below 0.3 the answer is to pass; when C falls below 0.5, the highlighted row is the binding constraint, the lowest factor. The starting scores are the calm-regime Trend Pullback of the worked example below (C = 0.527): set R to 0.20, X to 0.65 and S to 0.40 for its stress regime.

The stress settings give $C = 0.040$ , with the regime factor highlighted. Now restore $R$ to 0.85 and $X$ to 0.90, leaving $S$ at 0.40: $C$ recovers only to 0.234, still a pass. A single weak factor is enough to keep a trade out.

5 · One setup in two regimes

Worked example: A Trend Pullback on SPX in calm and in stress

Step 1: the setup. Take the canonical RSI(2) pullback in Stage 2, the Trend Pullback of Chapter 11H. Hold the setup, its evidence and its mechanism fixed, and change only the regime.

Step 2: a calm regime. SPX is above its rising 200-day simple moving average (SMA), the VIX is at 14 and the composite is +50. GEX is positive, above the gamma flip, with the term structure in contango. The scores are $E = 0.85$ , $M = 0.90$ , $R = 0.85$ , $X = 0.90$ and $S = 0.90$ , so $C = 0.85 \times 0.90 \times 0.85 \times 0.90 \times 0.90 = 0.527$ . High conviction: take the trade at full intended size.

Step 3: a stress regime. SPX is in the transition from Stage 3 to Stage 4, the VIX is at 28, the term structure is backwardated and the composite is −30. GEX is negative. A Consumer Price Index (CPI) report has just shocked the market. Evidence and mechanism are unchanged, but $R = 0.20$ , $X = 0.65$ and $S = 0.40$ , so $C = 0.85 \times 0.90 \times 0.20 \times 0.65 \times 0.40 = 0.040$ . Insufficient conviction: do not enter, and reduce existing positions.

The same setup, with identical evidence and mechanism, went from 0.527 to 0.040 as the composite fell from +50 to −30. This is what operational discipline looks like: selectivity is mostly not trading.

6 · The same numbers for a put-writing sleeve

The Edge Book scores the VRP harvest, at-the-money (ATM) monthly cash-secured XSP puts at 15% of the portfolio, with exactly the factor values above. Its reasons differ, since this is a strategy held for years rather than a trade held for days, and it adds a third scenario.

Worked example: The VRP harvest in three settings

Step 1: VIX at 14, contango, a normal regime. The evidence is Tier 1, from Bondarenko, from Carr and Wu, and from Jurek and Stafford: 0.85. The mechanism is a risk premium with a clear theoretical foundation: 0.90. The VIX is low to normal, the curve is in contango and no event is pending: 0.85. XSP is liquid, with tight spreads, and easy to roll: 0.90. At 15% of the portfolio the size is well within survivable bounds, while holding the puts cash-secured removes margin risk: 0.90. The product, $C = 0.53$ , is high conviction, so allocate.

Step 2: VIX at 28, backwardated for 5 days, after a CPI shock. With evidence and mechanism unchanged, backwardation would trigger the strategy's kill criterion, so the regime scores 0.20. Spreads are wider in volatility but still tradeable: 0.65. Scaling down is mandatory in this regime, and the full 15% is unsafe: 0.40. So $C = 0.040$ , insufficient conviction: enter no new positions, and reduce existing ones.

Step 3: VIX at 18, a normal regime, but 50% of the portfolio. Evidence, mechanism, regime and execution score 0.85, 0.90, 0.85 and 0.90 again. But 50% of the portfolio's notional in short puts means that a gap to a VIX of 30 with SPX down 20% likely force-liquidates, so sizing scores 0.20. The product falls to $C = 0.85 \times 0.90 \times 0.85 \times 0.90 \times 0.20 = 0.117$ , which is insufficient conviction.

In Step 2 the same strategy, with the same evidence and mechanism, loses its regime alignment and its sizing survivability. Its score drops by an order of magnitude, which is what discipline looks like. In Step 3 the evidence and the regime do not matter, because the size is wrong, and the framework correctly catches it.

The warning about forced liquidation in Step 3 assumes puts written on margin. Fully cash-secured puts cannot be force-liquidated by a margin call (Chapter 1E). A 20% fall still costs them close to 20% of their notional, less the premium collected. At a 50% allocation, that is nearly 10% of the portfolio.

7 · Scoring a popular strategy: GEX-level scalping

Consider the rule "buy SPX at the put wall, sell at the call wall", run on dealer-positioning data from vendors such as SqueezeMetrics or SpotGamma.

Worked example: GEX-level scalping

Step 1: evidence. The academic literature on dealer gamma (Garleanu, Pedersen and Poteshman) supports directional effects. Specific rules for trading levels, though, have not validated since 2022. A Cboe study of 2025 found that hedging of 0DTE options by market makers (MMs) is about 0.2% of liquidity, so evidence scores 0.25. That study measures same-day hedging only, so using it against walls built mostly from longer-dated open interest extends the measurement.

Step 2: mechanism. The general story of gamma hedging is plausible. The specific mechanism, that dealers must hedge at exactly these strikes, is not robustly supported. It depends on assumptions about who is long and who is short at which strikes: 0.30.

Step 3: regime. Dealers are sometimes net long gamma and sometimes net short, which public GEX models do not reliably distinguish: 0.40.

Step 4: execution. Although SPX is liquid, the strategy needs specific intraday entries near a wall, which makes it complex to execute: 0.50.

Step 5: sizing. Where to put the stop is unclear, since there is no natural exit if the wall does not hold. The crash risk is asymmetric as well: if the wall fails, you are caught long as the index falls through the put strikes, so sizing scores 0.35.

Step 6: the score. The five factors multiply to $C = 0.25 \times 0.30 \times 0.40 \times 0.50 \times 0.35 = 0.005$ , effectively zero conviction.

The score is not a judgment of the people who trade the idea: the framework is correctly identifying that a popular strategy fails even basic conviction screening. Most of its appeal is narrative, and the multiplicative structure exposes that.

Both traditions of this book reject scalping level by level (Chapters 10A and 13F). They differ on how much the sign of dealer gamma should count. The Dealer Mechanics Lab reads the sign into the regime factor above and the walls into the Real Value score below. The Edge Book gives dealer positioning only low confidence, as a minor input to its regime filters (Chapter 10H).

8 · Scoring a slow allocation: trend through DBMF

Worked example: A 15% allocation to DBMF

Step 1: the factors. The evidence is Tier 1, from Hurst, Ooi and Pedersen's 140-year record and from Moskowitz, Ooi and Pedersen: 0.90. There are several complementary mechanisms, under-reaction, compensation paid by hedgers and liquidity provision, and their decomposition is well understood: 0.85. In 2023 and 2024 trend was mixed, in a drought between strong years, so the regime scores 0.55. It is mid-cycle rather than a peak, at which the Edge Book would not allocate at all. DBMF is a single ETF, easy to buy: 0.95. A 15% allocation is well within survivability. Trend also has positive crash skew, with rare upside in stress: 0.90.

Step 2: the score. The product is $C = 0.90 \times 0.85 \times 0.55 \times 0.95 \times 0.90 = 0.36$ , which is medium conviction.

The regime score is the binding constraint. The structural case for the allocation is strong, with high evidence, mechanism, execution and sizing. The timing of the entry is not. The Edge Book's operational implication is to build the position over time, by dollar-cost averaging, rather than in one purchase, and to scale up if a drawdown creates a better entry point. This kind of nuance is exactly what the conviction score is meant to surface.

Chapter 10D, from another chapter of the Edge Book, advises against timing trend's entries at all: buy and hold across its cycles. The two pieces of advice meet in staging, which spreads the entry without waiting for a signal. Scaling up after a drawdown is a timing decision that Chapter 10D would not make.

9 · Five ways a trade fails the score

Both sources name the same five recurring patterns that look like edge but score low. Each has an example from swing trading and one from a portfolio, a factor that collapses, and a rule.

Strong evidence, hostile regime. A trend pullback in a Stage 4 market, or a VRP harvest in backwardation, has high evidence and mechanism but a low regime score. Respect the regime filter even when the evidence behind the setup is strong.
Right idea, oversized. A single swing trade holding 25% of the portfolio, or put-writing holding 50%, scores high on every factor except sizing. Edge cannot compensate for an unsurvivable size, because sizing is multiplicative, not additive.
Beautiful backtest, no mechanism. Optimized parameters that give a great paper Sharpe ratio, with no articulable reason to work in the future, score medium to high on evidence and low on mechanism. A backtest without a mechanism is a warning sign of data mining, which produces high Sharpe ratios in any sufficiently large sample (Chapter 1D).
Real edge, wrong infrastructure. A cross-sectional dispersion edge that needs daily multi-leg, delta-hedged positions on 50 names scores high on evidence and low on execution. So does an option-mispricing edge that needs a portfolio of 200 names, rebalanced and hedged daily. An edge that needs infrastructure you lack is not an edge for you.
Vivid narrative, thin evidence. GEX scalping has a compelling story about dealer positioning and no backtest evidence behind it, so its mechanism feels high while its evidence is low. A compelling narrative is not evidence quality.

The framework catches all five systematically, because its multiplicative structure forces every domain to score on its own.

Key idea

High-conviction swing trading is mostly not trading, because most observations do not survive interpretation and most interpretations do not survive the conviction score. The trades that remain have cleared all five gates, so the Dealer Mechanics Lab concludes that they should be larger and more durable than the trades you would otherwise make.

10 · The trade-admission checklist

The score is a number that should rest on a written checklist, which every candidate move passes through:

Box	The check
1. Source identified	Which of the four is it: risk premium, behavioral, institutional or microstructure?
2. Category assigned	Foundation premium, factor, overlay, filter, or rejected (Chapter 10A)?
3. Evidence tier	T1, T2 or T3, with the primary source cited
4. Mechanism stated	Why does this edge exist, and what would make it disappear?
5. Academic Sharpe ratio	What does the literature suggest?
6. Retail haircut applied	What is your realistic user-account Sharpe ratio (Chapter 13A)?
7. Regime check	Is the current regime supportive, and what would invalidate it?
8. Tail loss calculated	The $\text{ES}_{95}$ , and the profit and loss (P&L) in S2 and in S3
9. Margin behavior	How far does the margin requirement expand under stress?
10. Position size	Survivable in S2 (operational) and recoverable in S3, with no irrecoverable loss?
11. Kill criteria	Specific, measurable and committed in advance
12. Correlation to the existing book	Does the trade duplicate exposures you already hold?
13. Conviction score	Is $C = E \times M \times R \times X \times S \ge 0.5$ ?

If you cannot tick all 13 boxes, the trade is not ready for admission. The Edge Book insists that this is not bureaucracy but the discipline that prevents 80% of bad trades.

The checklist also beats intuition, which links strategies to vivid narratives rather than to the quality of their edge. It forces you to state each domain separately before you commit capital. Stories die under separate analysis. Real edges survive it.

Familiar ground

The checklist works like a trial's eligibility criteria, which are conjunctive. A patient who meets eleven inclusion criteria and fails the twelfth is not enrolled, however strongly they meet the other eleven. The conviction score is a graded version of the same rule. A factor near zero acts like a failed criterion, while a middling factor shrinks the product rather than excluding the trade outright.

11 · The decision tree

The same analysis can be walked as seven questions, in order:

Step	The question	What settles it
1	What is this?	Its source, whether risk premium, behavioral, institutional or microstructure, and its category, from foundation premium to rejected
2	Is the evidence robust?	Peer-reviewed, multi-decade and replicated is T1; strong but sensitive to the sample is T2; thin or decayed is T3, and likely not a trade
3	What is the mechanism?	Why it works, what would make it stop, and whether anything has changed that affects it
4	Does it survive the haircut?	Whether the realistic Sharpe ratio after the academic-to-account haircut is still positive, and materially better than simply holding SPX
5	What is the worst-case path?	The stress P&L for falls of 10%, 20% and 35% in SPX, the margin expansion, and the risk of forced liquidation
6	What is the kill rule?	A specific trigger, a pre-committed action, and the criteria for revalidating
7	What is the conviction score?	At 0.5 or more, allocate; from 0.3 to 0.5, trade small or test; below 0.3, do not trade

This is not seven separate analyses but one analysis, structured to surface the issues that informal evaluation misses.

12 · What the score is not

A promise. High-conviction trades still lose, since the score measures expected risk-adjusted return after frictions, not certainty.
A reason to size big. Even at $C = 0.7$ , the size comes from the tail loss and the portfolio context. Conviction tells you whether to trade, not how much.
A substitute for a backtest. The score is a thinking tool, beside which backtests, stress tests and live paper trading keep their value.
A static number. Regime and execution shift over time, so the score updates as conditions change.

What the score is: a forcing function for honest evaluation across every domain before you commit capital. It works because its multiplicative structure mirrors reality: you cannot compensate for unsurvivable sizing with great evidence, or for poor execution with a strong mechanism.

Where the traditions differ on size

The two sources disagree about the second point. The Edge Book keeps conviction as a gate that decides whether a trade happens. The tail loss and the loss budget of Chapter 13C decide its size, though a medium score does call for a small or test allocation. The Dealer Mechanics Lab lets conviction scale the size. It takes high-conviction trades at full intended size and medium ones at a reduced size, 60% to 80% in the figure. Its worked examples in Chapter 13D follow that rule. Both size from an explicit loss budget, so they agree on the ceiling: conviction never takes a trade above the size the loss budget allows. They differ only on whether a medium score should shrink it below that.

Common mistake: "my intuition is strong, so I do not need to score it"

Strong intuition is your unconscious mind applying patterns from past experience. Useful as it is, it is also exactly the channel through which behavioral biases enter. The score forces a conscious evaluation that surfaces what your intuition skipped. If your intuition agrees with the score, you have cross-validated. If they disagree, one of them is wrong. Only the score leaves a written audit trail you can review.

13 · Single stocks: the Real Value score

The conviction score is calibrated for index and broad-market swing positions, where regime, execution, evidence and sizing dominate. A single stock, particularly with an options overlay, has layers that the conviction score does not capture cleanly. For single stocks the Dealer Mechanics Lab uses a companion, the Real Value score:

\text{RV} = F\,T\,O\,G\,V\,X\,S

(13B.4)

With each of its seven factors running from 0 to 1, a trade is taken at $\text{RV} \ge 0.4$ . That is slightly lower than the conviction score's 0.5, because more factors, all at most 1, multiply a uniformly decent position down faster. By the reasoning of (13B.3), the level each of seven factors must average is $\text{RV}^{1/7}$ , and $0.4^{1/7} = 0.88$ . That is almost exactly the 0.87 that high conviction asks of five factors, so the two thresholds set the same bar.

The first five factors are new, while the last two are the conviction score's own.

F, fundamentals, asks whether revisions to revenue and to earnings per share (EPS) trend the right way and the margin trend is constructive. It also asks whether analysts' estimates are converging as a consensus forms, guidance revisions are positive (negative for a short), and the sector or industry has a tailwind.
T, technical structure, asks for Stage 2 confirmed and a proper consolidation rather than noise. It wants good relative strength against the benchmark, volume drying up on the pullback, and a clean pivot or reclaim.
O, option flow signature, looks for net premium bought at the ask in the direction of the thesis, in opening transactions rather than closing ones. Rising open interest (OI) the next day confirms that the flow opened new positions. It also wants trades initiated on the thesis's side, flow repeated across sessions rather than a one-off, and a consistent change in skew and implied volatility.
G, GEX path support, asks whether dealer positioning supports the move. Is the price below a call wall that becomes a magnet on the upside, or above a put wall that makes a floor? And is the gamma flip near enough to make the regime unstable?
V, vol-value, asks whether the option is priced reasonably against realized volatility, event risk and its own history of implied volatility. It also asks whether the IV percentile is attractive for the direction, and whether the expected move is plausible against the average true range (ATR) and the announced catalysts.

Factor	0.9 to 1.0	0.5 to 0.7	0.2 to 0.4	0
F	Rising estimates, narrowing dispersion, positive guidance, large institutional buying	Stable estimates, mixed analyst views, a neutral sector	Falling estimates, widening dispersion, mixed or negative guidance	A clear negative break in the fundamentals
T	A textbook Stage 2 pullback: tight consolidation, strong relative strength, a clean pivot	Stage 2, but an imperfect base or mediocre relative strength	Choppy structure, weak relative strength, an ambiguous pivot	Stage 4, or distribution in Stage 3
O	Large net premium bought at the ask in the supporting direction, opening confirmed, repeated over 3 or more sessions	Modest aligned flow in a single session, opening status unclear	Mixed flow, possibly closing transactions, hard to interpret	Flow opposite to the thesis
G	Clear path support: positive GEX above the flip on a long thesis, the call wall well above the target, the put wall well below the entry	A neutral path: flat GEX, no walls nearby	An obstructed path: the flip in the way, or a call wall directly overhead on a long thesis	GEX actively opposes the move, such as negative GEX under a long mean-reversion thesis
V	Cheap against realized volatility and its history, for a buyer; rich against realized, for a seller	Fairly priced, no edge to either side	Mispriced against the direction: rich for a buyer, cheap for a seller	Structural arbitrage against the trade, such as buying a long call into an IV percentile above 90

Flow alone is never a signal. O is only a confirmation factor, which counts once F or T already scores 0.6 or more.

The execution factor X asks for a tight bid-ask spread, enough liquidity at the intended size and a favorable time of day, away from the known fragility windows. Those windows are the last 30 minutes, the first 5 minutes after the open, and the lunch hours for less liquid names. The sizing factor S asks that the position survive a stop blown through by 3 times its distance, with its correlation to the existing book accounted for. The book's total vega and delta must stay within portfolio limits too; Chapter 13H sets the limits on vega.

After computing RV, classify it:

RV	Reading	Action
0.5 or more	A real signal: fundamentals, technicals, flow and the volatility surface all agree	Take it, at full size (Chapter 13C)
0.3 to 0.5	Borderline: most factors aligned, one or two weak	Take it at half size, or skip it for the next setup
0.15 to 0.3	Mixed: enough green to tempt, enough red to be dangerous	Skip it: a common loss-maker for retail traders
Below 0.15	Avoid	Mostly noise: do not trade

The table's borderline band starts at 0.3, below the threshold of 0.4. Between the two, skipping satisfies both rules.

Worked example: A single-stock score, and a bad vehicle

Step 1: a strong candidate. Suppose a stock scores 0.9 on fundamentals, technical structure, option flow and GEX path, 0.8 on vol-value, 0.95 on execution and 0.9 on sizing. By (13B.4), $\text{RV} = 0.9^5 \times 0.8 \times 0.95 = 0.449$ . That clears 0.4 but falls in the borderline band, so take it at half size.

Step 2: the same thesis through an overpriced option. If the option is rich against the direction, vol-value falls to 0.25 and $\text{RV} = 0.9^5 \times 0.25 \times 0.95 = 0.140$ , in the avoid band. The thesis has not changed, only its expression, which is the second of the pathologies below.

Four single-stock pathologies show up as a low RV at a glance:

Flow-only noise. Massive call buying at the ask, but no fundamental story (F below 0.4) and no technical setup (T below 0.4). It drives chasing trades that fade at once, and RV is typically below 0.1.
Good thesis, bad vehicle. F and T are both above 0.7, but V is below 0.3 because the option is priced unreasonably. Though the thesis is real, you are paying too much for its expression. It often resolves to trading the shares instead of the options.
Good setup, bad path. F, T and O are all above 0.6, but G is below 0.3, with a call wall directly overhead or the gamma flip in the way. The directional thesis is solid, but dealer mechanics block its path. Wait for the obstacle to clear, or pick a different strike or vehicle.
Stretched, stretching further. Each factor looks fine alone. The stock, though, has already moved 30% in the prior month, so the story is in the price. Here RV correctly downgrades the trade, because F should ask whether the story is already reflected in the price, not just whether it is positive.

The two scores are not redundant. Use the conviction score for index swings and the Real Value score for single-stock swings. They share the execution and sizing factors and differ in the layers that matter for each kind of decision.

14 · In practice

In practice

Score every candidate in writing before you enter, and record the five factor values in your journal, not just the product.
Name the binding constraint and ask what would fix it. A low R says wait, a low S says trade smaller and a low X says change the vehicle. A low E or M says it may not be a trade at all.
Rescore open positions when the regime moves. A position that scored 0.53 in calm can score 0.040 a week later, which says reduce it, not hold and hope.
Use the same scoring guide every time, the bands above or your own written version of them, so that your scores stay comparable.

When it fails

Optimism compounds. Each factor is a judgment, and small biases multiply. Scoring five factors at 0.95 where 0.90 was honest inflates $C$ by $(0.95/0.90)^5 = 1.31$ , about a third, which is enough to turn 0.45 into 0.59 and a medium trade into a high one.
The score ignores the rest of the book. Two high-conviction trades on correlated names are one large bet. The checklist's correlation box catches that, not the score (Chapter 13D, Example 10).
A score goes stale. Regime and execution move, so a number computed last week is not today's.

Decision rule

Score every trade on all five factors before entry, in writing, and name the binding constraint.
At $C \ge 0.5$ , take the trade at the size the loss budget of Chapter 13C allows.
From 0.3 to 0.5, take a small or test size, such as the figure's 60% to 80% of the intended size.
Below 0.3, do not trade, and reduce any open position whose score has fallen there.
For single stocks, use the Real Value score: skip below 0.4, take half size from 0.4 to 0.5, and full size above.

15 · Where this fits

Chapters 1B to 1F gave you the primitives, Chapter 1G the sources of edge, Chapter 1D the reasons the Sharpe ratio misleads, and Chapter 13A realistic retail expectations. This chapter ties them into a decision rule: the conviction score that decides whether a candidate move becomes a position. Chapter 10A's taxonomy then decides how to implement a move that passes, as a foundation premium, factor, overlay or filter. Chapters 10B to 10I supply the edges one by one. Chapter 13E walks the full conviction analysis through each major strategy. The conviction framework is the spine of the book: every other chapter either defines the inputs to the score or applies it to specific strategies.

What to carry forward

High conviction means that every independent layer converges: evidence, mechanism, regime, execution and a survivable size.
The conviction score multiplies five factors from 0 to 1, $C = E\,M\,R\,X\,S$ . High conviction starts at 0.5 and trading stops below 0.3, so the factors must average about 0.87 for a high score.
One weak factor sinks the product. The same setup scores 0.527 in calm and 0.040 in stress. The same strategy scores 0.117 at an unsurvivable size.
The score says whether to trade, while the loss budget of Chapter 13C caps how much. The two traditions differ on whether a medium score should shrink the size further.
For single stocks, the Real Value score multiplies seven factors and takes trades at 0.4 or more.

16 · Check yourself

Question 1

State the conviction score. Why is it multiplicative rather than additive?

Answer

$C = E \times M \times R \times X \times S$ , each factor from 0 to 1. Each domain is a necessary condition. A product lets a single failed domain sink the score, which an average cannot. Four factors at 0.9 and one at 0.2 average 0.76 but multiply to 0.131, below the floor of 0.3.

Question 2

A candidate scores $E = 0.6$ , $M = 0.7$ , $R = 0.9$ , $X = 0.8$ and $S = 0.9$ . What is $C$ , and what binds?

Answer

$C = 0.6 \times 0.7 \times 0.9 \times 0.8 \times 0.9 = 0.272$ , below 0.3, so do not trade. Evidence, at 0.6, is the binding constraint. A Tier 2 edge with only a partial mechanism cannot reach high conviction even in a good regime.

Question 3

Why does "strong intuition" not satisfy any of the five domains?

Answer

Intuition is pattern-matching from past experience, running through exactly the channel by which behavioral biases enter. It attaches to vivid narratives rather than to evidence quality, so it can make a mechanism feel strong while the evidence is thin. Each domain must be scored consciously, separately and in writing.

Question 4

What is the difference between evidence, $E$ , and mechanism, $M$ ?

Answer

Evidence asks whether the effect has held up in data: peer-reviewed, replicated, over decades. Mechanism asks why it should keep holding: who pays for it, and what would make it stop. A data-mined backtest can have evidence of a kind and no mechanism, while a compelling story can have a mechanism and no evidence. Both score low.

Question 5

In the VRP harvest example, how does $R(t)$ change between the calm scenario and the stress scenario, and what follows?

Answer

It falls from 0.85 to 0.20, because a backwardated VIX curve would trigger the strategy's kill criterion. With execution and sizing also lower, $C$ drops from 0.53 to 0.040, an order of magnitude. The operational implication is to enter no new positions and to reduce existing ones.

Question 6

Why is the Real Value score's threshold 0.4 rather than 0.5?

Answer

With seven factors instead of five, a product of numbers below 1 multiplies down faster. The level each factor must average is the seventh root of the threshold, $0.4^{1/7} = 0.88$ , about the same bar as $0.5^{1/5} = 0.87$ for the conviction score.