Diagnostic Index: Symptom → Cause → Fix

This is the global troubleshooting index for the whole Atlas - the aggregated “clinic” that collects the failure modes spread across every topic’s §5 Failure Modes & Practice page, the pillar hubs’ §4, and the home-page matrix, deduplicated and organised by symptom.

How to use it. Find your observed symptom in the left column (scan the area heading that matches where the problem bites: backtest, execution, risk, portfolio, derivatives, ML, or systems). The root cause names the first-principles violation underneath, the remedy says what to change, and the canonical page takes you to the full treatment - math, code, and the complete failure-mode catalogue.

What this supplements. This index is a pointer, not a replacement. Each canonical page’s §5 Failure Modes & Practice remains the authoritative, full-depth treatment of every failure mode here - with the derivations, runnable code, and numbered checklist. Use this page to route yourself fast; use the §5 to actually fix it.

Research & Backtesting

Observed symptom	First-principles root cause	Remedy	Canonical page
Backtest shows 3.5 Sharpe, instantly loses money live	Backtest overfitting & selection bias: the reported Sharpe is a maximum over $N$ trials, not a sample statistic	Report $N$ , $\overset{ρ}{^}$ , $\hat{N}$ ; gate on DSR ≥ 0.95, PBO ≤ 0.5	Backtesting Hygiene
High-turnover strategy’s Sharpe collapses to ~0 live	Transaction-cost amnesia: drag $\propto τ c / σ_{ann}$ ignored; quadratic impact omitted	Model proportional spread + size-dependent (Almgren) impact in the backtest	Backtesting Hygiene
Repeatedly “touching” the holdout keeps improving results	Repeated holdout: the OOS set spent as a selection device converts ~5% false positives into expected ones	One vaulted use of the OOS; any post-hoc change resets it	Backtesting Hygiene
Strategy overweights today’s winners/constituents	Survivorship bias: universe filtered to names that survived, dropping the losers	Point-in-time universe, include delisted names and terminal returns	Backtesting Hygiene
Equity curve unnaturally smooth; alpha too stable	Look-ahead via vintages: fundamentals/index membership/vendor prices known only later fed in early	Strict point-in-time, as-of-known-at-timestamp joins	Backtesting Hygiene
Treated 1,000 correlated variants as 1,000 independent bets	Effective- $N$ blindness: effective trials $\hat{N} \approx \overset{ρ}{^} (M - 1) + 1$ far below raw $M$	Report $M$ , $\overset{ρ}{^}$ , $\hat{N}$ ; correct the selection threshold	Backtesting Hygiene
Sharpe quoted for a negatively-skewed, fat-tailed stream (carry, short vol)	Non-normality ignored: true standard error of a non-normal Sharpe far larger	Use DSR’s denominator (skew/kurtosis correction); quote ES not VaR	Backtesting Hygiene
Alpha is statistically real but too small to matter, or decayed by deployment	Capacity & alpha-decay blindness: net alpha concave in AUM, premium erodes post-publication	Build impact into the backtest, cap participation, haircut published alphas	Factor Investing & Timing
Sharpe computed on illiquid mark-to-model or smoothed returns	Metric gaming: understated variance inflates the ratio	Use audited, mark-to-market returns; reject smoothed series	Backtesting Hygiene
Pairs spread widens and stops reverting (divergence trap)	Cointegration structural break: linear relation $z_{t} = P_{A} - β P_{B}$ broke via merger/index/dividend shift	Rolling/recursive cointegration + CUSUM on residual; hard statistical stop	Statistical Arbitrage & Pairs
Pairs pair chosen after the fact; great backtest, dead live	Data snooping: pair selected ex-post is an extreme order statistic	OOS formation/trading split, deflated Sharpe, purged CV	Statistical Arbitrage & Pairs
Spread-trading costs eat two legs every rebalance	Transaction costs & turnover: turnover scales with spread re-entry; gross ≠ net	Cost-aware thresholds, longer half-life screen, portfolio netting	Statistical Arbitrage & Pairs
Short leg unborrowable exactly when divergence largest	Short borrow & asymmetric costs: one-sided, hidden risk	Screen borrow availability; budget short-side cost	Statistical Arbitrage & Pairs
Correlated unwinds across same-side books (Aug 2007 quant quake)	Crowding: many funds run the same residual-reversion book	Monitor crowding/capacity; de-size correlated sleeves	Statistical Arbitrage & Pairs
Stale closes / non-synchronous legs fabricate convergence on paper	Non-synchronous pricing: bid-ask bounce invents mean-reversion that cannot be captured live	Use midpoint quotes, synchronous timestamps, lag-adjusted estimates	Statistical Arbitrage & Pairs
Event study attributes a same-window shock wholly to the event	Event-window contamination: overlapping announcements confound attribution	Short event window + screen out firms with other news; no statistical fix	Event Studies
Small, thin-stock event t-stats overstated	Non-synchronous trading biases OLS beta	Scholes–Williams/Dimson lag-lead corrections, or restrict to liquid firms	Event Studies
Event t-stat inflated ~1.7× from correlated firms	Cross-sectional dependence/clustering ignored	Portfolio-time-series SE (short), calendar-time portfolio alpha (long)	Event Studies
Event significance overstated when event-period vol rises	Event-period variance increase not in estimation-window SE	Event/non-event variance ratio; nonparametric rank tests	Event Studies
Long-horizon abnormal returns unreliable	Long-horizon joint-test problem & low power: risk-adjustment error compounds	Short horizons; report expected-return-model sensitivity	Event Studies
Momentum portfolio crashes in bear-market rebounds	Momentum crashes (negative skew): option-like short-loser leg; clustered, partly forecastable	Size/hedge; do not trust variance-swap hedging (Daniel–Moskowitz)	Momentum
Momentum bleeds slowly over 12–18 months (whipsaw)	Whipsaw in a mean-reverting regime: buys tops, shorts bottoms	Regime filter; cost-aware entry; accept lower frequency	Momentum
”Hedged momentum” looks great in backtest only	Hedged-momentum illusion: hedge used ex-post/forward-looking betas	Use implementable ex-ante hedge; verify no look-ahead in hedge	Momentum
Momentum short leg expensive/illiquid in a squeeze	Transaction costs & short-leg capacity	Cost-aware shorting, monitor squeeze risk	Momentum
Factor’s modeled beta-risk “low” but realized tail huge	Crowding & liquidity black holes: everyone holds the same long-short	Treat crowding as a risk input; de-size before the unwind	Factor Investing & Timing
Published factor premium already shrunk by the time you deploy	Post-publication decay: ~35% haircut, larger for easy-to-arbitrage	Haircut backtested alphas; re-estimate on post-publication data	Factor Investing & Timing
In-sample-optimal timing weights collapse out-of-sample	Factor timing high-variance, low-IC bet: timing breadth ≈12	Treat timing as a small ±0.25–0.5× tilt; purged CV	Factor Investing & Timing
Overlapping style factors make regression slopes uninterpretable	Multicollinearity in style factors: nearly singular design	Orthogonalize/standardize exposures; ridge/lasso; drop redundant factors	Fundamental Multi-Factor Models
”Diversification” across sleeves is illusory	The diversification illusion: residual correlation bounded by factor structure; few effective eigenvalues	Orthogonalize (Gram–Schmidt/PCA), count eigenvalues not sleeves	Fundamental Multi-Factor Models
Vol model’s persistence too high; forecasts unresponsive	GARCH persistence overstated by breaks: structural break masquerading as memory	Test for breaks (CUSUM), model the break not the memory	GARCH & Volatility Modeling
Symmetric GARCH wrong on downside moves	Missing asymmetry: leverage effect in equity returns	Engle–Ng sign-bias test; add asymmetry/GJR term	GARCH & Volatility Modeling
Vol forecast “accurate” but VaR systematically breached	Backtest the quantile, not the variance: wrong tail model	Kupiec/Christoffersen tests on the VaR itself	GARCH & Volatility Modeling
Gaussian-shock GARCH understates tails	Fat tails in shock distribution	Use standardized Student- $t$ /GED shocks	GARCH & Volatility Modeling
Kalman-filtered beta drifts quietly wrong	Divergence = over-confidence, not instability: process noise collapsed to zero	Floor on $Q$ ; monitor standardized innovation variance ratio	Signal Processing & Kalman
Rolling-OLS beta lags the true beta by half a window	Fixed windows leak, recursive filters lag by construction	Choose and justify the filter’s lag via $Q$ ; prefer Kalman with explicit lag	Signal Processing & Kalman
Kalman filter numerically blows up / PSD drifts negative	Numerical fragility: ill-conditioned covariance, symmetricity drift	Joseph form / square-root filter; solve not invert; keep $P$ standardized	Signal Processing & Kalman
Beta jumps at index reconstitution, filter reacts too slowly	Regime breaks are not random walks: jump ≠ continuous drift	Model the jump (Markov-switching/jump-diffusion state)	Signal Processing & Kalman
Dynamic-beta strategy’s edge eaten by induced turnover	Filter-induced turnover consumes alpha	Pass filtered-alpha strategy through DSR/purged-CV battery	Signal Processing & Kalman

Machine Learning & AltData

Observed symptom	First-principles root cause	Remedy	Canonical page
ML model 95% in-sample accuracy, ~0% live	Leaky cross-validation: features scaled/screened on full data before splitting	Preprocessing and feature selection inside each training fold only	Financial ML Pitfalls
Today’s index constituents backtested to 2005	Point-in-time (look-ahead) bias: data used before it was available	Strict PIT datasets; knowable-at timestamps	Alt-Data Pipelines
Universe filtered to today’s S&P 500 drops the names that prove the model wrong	Survivorship bias	Backtest on point-in-time constituent lists	Financial ML Pitfalls
Best config’s in-sample Sharpe inflated by tuning	Selection/multiple-testing bias: winner chosen on noise	Deflated Sharpe over the number of trials	Financial ML Pitfalls
Purged-CV accuracy rises as fold count grows	Look-ahead leakage survives in the split (the leak diagnostic)	Find and close the leak; rerun the fold-count diagnostic	Purged CV & Backtest Hygiene
Single backtest Sharpe quoted as the verdict	Path overfitting: a lone Sharpe is the most volatile estimate	Report a CPCV Sharpe distribution (mean ± sd), never one number	Purged CV & Backtest Hygiene
Honest split starves the model of training data	Sample loss from purging at long horizons	Confirm per-fold training count ≫ parameter budget; gather more data	Purged CV & Backtest Hygiene
Deep net overfits; train error 0, test error high	Data hunger / overfitting: effective samples finance rarely supplies	Weight decay, dropout, early stopping, smaller models first	Deep Learning for Sequences
LSTM forecast degrades silently over months	Non-stationarity: model is a snapshot of a moving process	Walk-forward/purged validation, rolling refits	Deep Learning for Sequences
Sequence model “peeks” at the future	Look-ahead via missing causal mask / overlapping labels	Causal masking; purge + embargo; audit recurrence/attention	Deep Learning for Sequences
Sequential inference too slow for microsecond decisions	Latency/cost: RNN slow per step, full attention $O (T^{2})$	TCN or small model; compute enters alpha	Deep Learning for Sequences
Tree model with unbounded depth overfits low-SNR data	Too many boosting rounds: test error has a shallow minimum in low SNR	Early stopping; depth 2–4, few hundred trees, chosen OOS	Tree & Boosting Methods
MDI-based feature selection picks high-cardinality noise	MDI is in-sample and biased	Use MDA/permutation with clustered/orthogonalised features	Tree & Boosting Methods
Correlated factors cancel in permutation importance	Substitution effects in MDA	Permute correlated groups together; orthogonalize first	Tree & Boosting Methods
Features screened on all data then CV’d	Feature selection before fold split = leakage	Screen inside each fold on training data only	Tree & Boosting Methods
Standard K-fold hyperparameter search double-overfits overlapping labels	Unpurged hyperparameter search	Purged/embargoed CV; proper loss (log-loss) not accuracy	Tree & Boosting Methods
HMM/GMM regime labels swap between runs	Label switching: likelihood invariant to state permutation	Order-constrain states at fit; never hard-code semantics onto unconstrained indices	GMM)
Likelihood keeps climbing as regimes are added; states clone	Overfitting the number of regimes	Choose by BIC / out-of-sample likelihood, never max-likelihood	GMM)
Regime labels smooth away the future	Look-ahead from smoothed/Viterbi-decoded labels	Label online with the filtered posterior; purge/embargo backtests	GMM)
Hard regime switch flips a coefficient on one wrong label	Hard switching vs label error	Use the soft posterior (mixture-of-experts weighting)	GMM)
Gaussian-emission HMM manufactures a phantom “spike” regime	Gaussian-emission misfit on fat tails/jumps	$t$ -distributed/heavier emissions; stress-check vs no-regime baseline	GMM)
RL agent over-trades / exploits gross-P&L or Sharpe reward	Reward hacking: gross reward or Sharpe-only reward is gameable	Reward net-of-cost P&L with explicit risk/turnover penalty	Reinforcement Learning for Trading
RL policy exploits instant fills at mid / benign impact	Sim-to-real gap: agent exploits simulator artifacts	Conservative impact; validate simulator P&L vs real fills	Reinforcement Learning for Trading
RL policy’s guarantees expire as regimes shift	Non-stationarity: learned policy claims a fixed DGP	Regime-conditional policies, short replay windows, constant monitoring	Reinforcement Learning for Trading
RL result flips with the random seed	Reproducibility/evaluation noise: seed-sensitive deep RL	Report distributions over seeds; never one run vs baseline	Reinforcement Learning for Trading
NLP sentiment feature counts “liability/tax/gross” as negative	Dictionary misfit: generic lexicons are finance-blind	Use Loughran–McDonald / finance-native lists	Financial NLP & Transcripts
Many-n-gram NLP model collapses out-of-sample	Overfitting with many n-grams on sparse features	Screen inside folds, regularize, cap n-gram order	Financial NLP & Transcripts
Sentiment feature built from restated/leaked text	Look-ahead via restated/leaked text	Timestamp by public release; features only from text knowable at as-of	Financial NLP & Transcripts
Alt-data vendor rewrites history on re-run	Backfill/reinstatement time-travel: storing only latest value	Keep multiple knowledge-dated versions; forbid backwards edits	Alt-Data Pipelines
Vendor attrition looks like a demand collapse overnight	Panel/coverage drift read as demand	Model share or normalize by panel size	Alt-Data Pipelines
Best dataset’s t-stat inflated by the search	Overfitting the evaluation: winner selected on noise	Deflate for number of datasets tried (deflated Sharpe/PBO)	Alt-Data Pipelines
Web-scraped/leaked data exposes fund to MNPI liability	Compliance/MNPI: non-public information embedded in feed	Use public or properly-licensed primary sources only	Alt-Data Pipelines

Execution & Microstructure

Observed symptom	First-principles root cause	Remedy	Canonical page
Execution backtest books “trade printed” as “I filled”	Optimistic fill accounting: ignores queue position & cumulative outflow	Model cumulative outflow (trades + cancels) against queue position; record quantity	Execution Backtesting & Simulation
Fill-model loss grows with assumed latency	Latency omission / stale-quote pick-off	Model latency explicitly; cancel-ahead logic	Execution Backtesting & Simulation
Whole-day flow used for a short-lived quote inflates fill rate	Look-ahead in replay: non-causal ordering	Strict causal ordering; events affect decisions at/after timestamp	Execution Backtesting & Simulation
Size-blind backtest omits self-impact	Ignored self-impact: large order moves its own price	Calibrate Almgren impact; charge per leg	Execution Backtesting & Simulation
Correct fill model still overstates edge	Ignored adverse selection on fills	Price the conditional drift into fill P&L	Execution Backtesting & Simulation
Queue ahead vanishes simultaneously when a sweep arrives	Phantom-liquidity cascades: correlated cancellations	State/event-dependent cancel intensity	Execution Backtesting & Simulation
Impact/fill params calibrated and evaluated on same data	Calibration/benchmark drift: in-sample optimism	Out-of-sample calibration; fixed pre-declared benchmark	Execution Backtesting & Simulation
Almgren–Chriss schedule overpays and mis-times the trade	Linear-impact delusion: impact is concave (square-root law)	Power-law impact; size-dependent $η$	Optimal Execution & Almgren–Chriss
Execution trajectory wrong because risk aversion guessed	Risk-aversion mis-estimation: $λ$ is a mandate, not a data parameter	Set by mandate; oversizing over-trades	Optimal Execution & Almgren–Chriss
Trajectory wrong in the same direction the market is stressed	Non-stationarity of impact & vol: both shift on stress days	Re-solve intraday; treat parameters as state variables	Optimal Execution & Almgren–Chriss
Deterministic execution schedule front-run / predatory	Adversarial/predictable execution: open book	Randomize timing, vary venue and participation, hide the shape	Optimal Execution & Almgren–Chriss
Order left unfinished because liquidity under-resourced	Ignored opportunity cost of the unfilled tail (Perold)	Price the trades the market will not let you make	Optimal Execution & Almgren–Chriss
Static execution schedule suboptimal in trending names	Serial correlation/drift break time-homogeneity	Use the genuinely dynamic optimum; occasionally optimal to not sell	Optimal Execution & Almgren–Chriss
VWAP report looks great but is pure selection	VWAP is gameable: low measured cost can be a zero-sum transfer	Randomize schedules; use real-time volume/POV; adaptive IS	POV)
Large orders auto-flatter VWAP regardless of handling	VWAP order-size dependence: you become the benchmark	Rank on ungameable benchmarks (arrival) for large orders	POV)
Passive participants get picked off by informed flow	Fast laggard = adverse selection (Glosten–Milgrom)	Let algorithm react to price/variance; cut adverse selection	POV)
Router trades through the protected quote	Acting on a snapshot: NBBO is a past observation	Atomic re-check; sweep best protected prices in strict order	Smart Order Routing & Fragmentation
Venue cheap on screen, expensive in reality	Underpricing latency: pick-off loss an explicit term	Put latency in the effective price	Smart Order Routing & Fragmentation
Fee-blind SOR silently misroutes at scale	Optimizing the displayed price: fee misroute is silent and uniform	Rank on cum-fee price	Smart Order Routing & Fragmentation
Resting order rarely filled, or filled only when wrong	Queue-position risk: expected fill set by size ahead, not price	Model cumulative outflow; account for position resets on reprice	Queue Position & Fill Probability
Stale passive quotes filled by faster takers at wrong moment	Latency pick-off	Cancel-ahead / quote-fade logic; genuine latency reduction	Queue Position & Fill Probability
Frequent reprice destroys fill quality	Refresh cost: cancel-and-reinsert resets queue position	Account for expected position resets in reprice decision	Queue Position & Fill Probability
FOK order can’t fill; market order too large for level	Order-type mismatch is structural, not operational	Size child to visible level; choose type from partial-fill tolerance	Market Microstructure & Order Types
Clock skew ≥ margin ⇒ random order timing	Clock skew beyond the edge: violates simultaneous clocks	PTP grandmaster + GPS discipline <100 ns; holdover audit	Colocation & Clock Sync
Latency budget sized on mean, fails ~10× too often	Mean-vs-tail budget: p99.9 ≫ mean	Size tick-to-trade budgets to p99/p99.9 tail	Colocation & Clock Sync
One-way latency estimated from RTT/2 badly biased	Asymmetric route bias	Measure one-way with synchronized clocks, never RTT/2	Colocation & Clock Sync
FPGA runaway sprays orders at clock rate	Catastrophic malfunction: no hot-fix in hardware	Out-of-band kill switches; order-rate interlocks in fabric; heartbeats off-path	Hardware Acceleration & FPGA

Market Making & Liquidity

Observed symptom	First-principles root cause	Remedy	Canonical page
Maker fills 100 consecutive buys right before price crashes	Toxic/adverse-selection flow: fills are informative, quote is an unpriced option	Detect toxicity (VPIN/PIN), widen/pull quotes; price conditional drift	Adverse Selection (Glosten–Milgrom)
Spread too thin; one-sided loss from informed fraction	Underestimated informed fraction: spread below toxicity ceiling	Re-estimate $π$ (toxicity); widen spread	Adverse Selection (Glosten–Milgrom)
Flat quotes through announcements/reopens	Spread not widened for rising uncertainty	Widen/thin in step with belief uncertainty	Adverse Selection (Glosten–Milgrom)
Volatile-but-uninformed stock quoted like informed one	Equating volatility with information	Measure permanent impact, not bounce	Adverse Selection (Glosten–Milgrom)
Maker quotes stale spreads into a changed market	Belief lag: parameters never updated	Update toxicity/beliefs with experience	Adverse Selection (Glosten–Milgrom)
Avellaneda–Stoikov quotes get picked off by informed flow	Adverse selection invisible to AS: spread covers inventory only	Layer VPIN/Lee–Ready toxicity detection; widen/pull	Avellaneda–Stoikov
Hard-coded quoting params misquote when regime changes	Parameter fragility: $κ, γ$ non-stationary	Refit intraday; shrink toward a prior	Avellaneda–Stoikov
Dealer lets inventory grow unbounded in a trend	No inventory cap: skew only discourages, not limits	Add hard inventory limit/penalty	Avellaneda–Stoikov
Skewed quotes cross the mid for large size	Quote crossing / marketable quotes	Clamp and cap before publishing	Avellaneda–Stoikov
Random-walk inventory breaches in a trend with near-certainty	Uncontrolled inventory = guaranteed breach	Skew strong enough the cap is rarely binding	Inventory Management & Quote Skewing
Forced liquidation realizes paper losses at worst moment	Breach under a trend converts MTM into realized loss	Pre-empt: flatten/hedge before breach	Inventory Management & Quote Skewing
Overnight gap bypasses intraday quote skewing	Overnight/gap risk: market closed, no quoting protection	Flatten/hedge before the close	Inventory Management & Quote Skewing
Rebate-masked maker desk was losing all along	Rebate masks, does not cure: subsidy moves break-even toxicity	Re-measure P&L net of rebate tier; monitor edge erosion	Market Maker Economics & Rebates
Risk system monitors vol but not order-flow imbalance	Inventory blowups are flow-composition, not price, events	Monitor order-flow imbalance (OFI)	Market Maker Economics & Rebates
Reserve sized against average adverse selection	Adverse selection clusters: informed flow autocorrelated	Size capital against the burst, not the mean	Market Maker Economics & Rebates
LOB reconstruction state silently wrong	Feed desynchronization: dropped/duplicated/out-of-sequence messages	Sequence-number gap detection; periodic snapshot resync	Limit Order Book Mechanics
Depth metrics fooled by spoofing/fleeting liquidity	Manipulative/fleeting liquidity: depth is not belief	Use intent-inference (cancellation patterns), size-weighted signals	Limit Order Book Mechanics
Roll model returns NaN spread	Positive autocovariance: Roll’s sign assumption fails	Check autocovariance sign; use quote-based estimators	Spread Decomposition & Roll Model
Roll underestimates the spread	Serial-correlated order flow: institutional splitting clusters flow	Treat plain Roll as a lower bound; use decomposition with direction data	Spread Decomposition & Roll Model
Impact regression inflates persistent impact	Bid-ask bounce contaminates impact estimate	Use midpoints or generalized-Roll separation	Market Impact & Depth
OLS impact coefficient not causal	Endogeneity: prices cause flow as well as flow causes prices	IV or event-based designs	Market Impact & Depth
Linear impact fit to square-root data misprices large orders	Functional-form misspecification	Power-law/square-root impact; avoid arbitrary kernels	Market Impact & Depth
Amihud, OFI, metaorder impact conflated	Category errors across impact families (units mismatch)	Carry units; never equate per-dollar with per-share	Market Impact & Depth
Full-day average misprices trades at open/close	Ignoring intraday seasonality (U-shape depth/spread)	Schedule-aware impact/participation	Market Impact & Depth
Calm-regime impact estimate under-charges in stress	Non-stationarity of impact with regime	Re-estimate in/for stressed regimes	Market Impact & Depth
Illiquidity treated as diversifiable	Liquidity covaries across assets and is a priced factor	Model liquidity as a priced factor, not diversifiable	Liquidity Risk & Asset Pricing
OTC desk quotes constant liquidity into a fair-weather model	Liquidity is state-dependent, collapses in stress	Stress the intensities, not just prices	Dealer Banks & OTC
Fire-sale spiral worsens every dealer’s loss	Fire sales feed back: same thin buyers, prices fall	Clearing/capital buffers; recognize the super-additive spiral	Dealer Banks & OTC
CCP failure becomes systemic	Clearing moves risk, does not delete it	Diversify CCP exposure; monitor concentrated residual risk	Dealer Banks & OTC
All makers pull quotes on the same VPIN signal	Self-fulfilling kill-switch: shared alarm evaporates liquidity	Decouple thresholds; distributional position not fixed cutoff	Toxic Order Flow & VPIN
VPIN reads volatility, not information	Conflating toxicity with volatility	Read VPIN as a flow-imbalance gauge, not an oracle	Toxic Order Flow & VPIN

Derivatives & Pricing

Observed symptom	First-principles root cause	Remedy	Canonical page
Delta-hedged book bleeds cash in fast volatile markets	Discrete hedging error & jump risk: continuous rebalancing $d t > 0$ ; unhedged gamma $\frac{1}{2} Γ (Δ S)^{2}$	Hedge trade-off: control gamma vs costs; accept jump risk is unhedgeable	Black–Scholes–Merton
Black–Scholes misprices by strike/maturity	Constant-vol delusion: one $σ$ can’t fit a smile	Move to stochastic/local vol (Heston, SABR, Dupire)	Black–Scholes–Merton
Hedge cannot protect a discontinuous gap	Jumps & fat tails: market incomplete, price not unique	Price under a jump/regime model; size jump limits	Black–Scholes–Merton
American option priced with European closed form	European formula on American put understates early-exercise premium	Use binomial/tree or finite-difference American engine	No-Arbitrage & Binomial
Tree gives meaningless prices (bracket violated)	Unguarded $u < e^{r Δ t} < d$ bracket in low-vol/long-maturity	Assert the bracket; raise on violation	No-Arbitrage & Binomial
Real-measure discounting used in a risk-neutral tree	Wrong discounting measure: $r \neq = μ$	Discount with $r$ (risk-neutral) for pricing	No-Arbitrage & Binomial
Single favorite tree price hides oscillation error	Single- $n$ reporting: error oscillation	Report an $n$ -study or averaged value	No-Arbitrage & Binomial
Long-dated option priced with BAW approximation	BAW error grows with maturity	Use BS-2002/tree for long maturities; benchmark	American Options & Optimal Stopping
Bermudan priced as American overstates value	Continuous value overstates by discretization gap	Use the discrete exercise schedule for structured notes	American Options & Optimal Stopping
”Exercise when deep ITM” heuristic low-biases price	Fixed exercise rule is suboptimal	Use LSM/tree; exercise decision is a solved optimum	American Options & Optimal Stopping
Delta-hedging an American short near the boundary blows up	Gamma discontinuity at the early-exercise boundary	Finer rebalancing near boundary; proper American delta	American Options & Optimal Stopping
Finite-difference prices oscillate at payoff kinks	Ringing without instability: CN symbol negative at high frequency	Rannacher start-up, Richardson-extrapolated implicit Euler, smooth payoff	Numerical Methods (Pricing)
Explicit scheme values meaningless above CFL bound	CFL violation: stability is a hard constraint	Implicit/CN, or exponential fitting	Numerical Methods (Pricing)
MC error stubbornly present at any path count	Discretization bias masquerading as noise (Euler weak order 1)	Log transform / weak-order-2 / Richardson; Brownian interpolation for extrema	Numerical Methods (Pricing)
Truncation bound too small biases the answer invisibly	Truncation error is a bias, not mesh error	Set bound from payoff support; verify by widening	Numerical Methods (Pricing)
Importance-sampled MC variance worse than plain	Badly tilted IS: likelihood-ratio variance inflates	Monitor weight second moment; stratify instead of bigger tilt	Numerical Methods (Pricing)
“The model says 10.42” hides all other errors	Reporting one number with no error budget	Report price + MC SE + mesh/boundary/scheme terms	Numerical Methods (Pricing)
Barrier priced with continuous monitoring closed form	Continuous-form assumption: misses discrete-monitoring hits	BGK/adjust for discrete monitoring	Exotic & Path-Dependent Options
Binary/barrier deltas via pathwise differentiation = 0	Pathwise Greeks on step payoffs are invalid	Likelihood-ratio or finite differences with CRN	Exotic & Path-Dependent Options
Quanto/spread price unstable to correlation	Correlation instability: $ρ$ drifts, less observable than vol	Re-run price across a $ρ$ -band; sensitivity not point estimate	Exotic & Path-Dependent Options
Converged-looking MC on a coarse grid is confidently wrong	MC dimension confusion: path error vs step (monitoring) bias	Push grid; confirm answer changes	Exotic & Path-Dependent Options
LV forward skew flat ⇒ barrier/cliquet/digital mispriced	LV’s flat forward skew dynamics	Use SV/SVJ for forward-skew exposure	Volatility Surfaces & Smiles
Interpolated smile violates no-arbitrage	Interpolation arbitrage in statics	Parametrize with arbitrage-aware objects (SVI + calendar constraints)	Volatility Surfaces & Smiles
Wrong wing extrapolation reprices the whole book	Extrapolation & wing risk	Enforce consistency with liquid VS quotes	Volatility Surfaces & Smiles
Hedging with model delta when implied vol moves with spot	Delta convention risk: BS delta ≠ minimum-variance delta	Use min-variance delta; state the sticky rule	Volatility Surfaces & Smiles
Recalibrating LV surface is itself unpriced P&L	Recalibration risk: LV forward skews/vol-of-vol not lockable	Price the cost of holding model dynamics	Calibration & Market Practice
Local-vol map explodes from raw-quote noise	Ill-posed inversion: differentiation amplifies noise	Smooth/parameterize smile first; regularize; clip local variance	Calibration & Market Practice
Identical fits, different hedges (SABR vs Heston)	Non-identifiability: smile fixes combinations, not parameters	Fix a convention; use term structure; ridge toward a prior	Calibration & Market Practice
Rich model drives today’s residuals to zero, future to nonsense	Overfitting: rich model, sparse data	Cross-validate, penalize, check no-arbitrage	Calibration & Market Practice
Same data/model, different desks, different params	Convention drift: silent reconciliation risk	Lock conventions; report residuals	Calibration & Market Practice
Heston rejected on Feller alone / Euler assumed fine	Confusing Feller condition with scheme positivity	Use Milstein/QE/exact transition; check Feller correctly	SABR)
Full-truncation Euler biases variance down	Truncating instead of fixing	Use Milstein, QE, or exact transition	SABR)
Calibration weights put all error on the wings	Calibrating without vega/bid-ask weights	Weight so a vol-point error costs the same everywhere	SABR)
Two parameter sets, identical RMS, different prices	Trusting the residual: RMS hides parameter stability	Report and monitor parameters and their stability	SABR)
Vol-of-vol carry mis-attributed in P&L	Vol risk premium priced separately, not a forecast	Attribute vol-of-vol carry explicitly	SABR)
One-factor model misprices steepeners/butterflies	Perfect-correlation trap: one Brownian ⇒ all forwards move together	Two-factor or market models with correlation matrix	Interest Rate & Term Structure
Lognormal Black-76 undefined near zero/negative rates	Negative rates break lognormal Black	Bachelier (normal), shifted lognormal, displaced diffusion	Interest Rate & Term Structure
Swaps misvalued post-2008 using one curve	Multi-curve/basis risk: OIS discounting vs LIBOR forwards diverged	Build LFM off curves consistent with the discount curve	Interest Rate & Term Structure
Cross-hedge leaves large residual basis	Basis risk irreducible with cross-hedge	Match contract to horizon; minimize but accept residual	Options Fundamentals & Markets
Hedge illiquid or margin-unfunded	Liquidity & margin constraints ignored	Check executable capacity; fund margin; respect constraints	Options Fundamentals & Markets
Contract-spec errors silently change cash flows	Contract-specification traps: multiplier, delivery options, day-count	Account for wild-card/CTD options; verify multiplier & basis	Options Fundamentals & Markets
Naked short call blows up	Leverage & short-option asymmetry: unbounded loss, margin scaling	Under-margining + leverage is the blow-up mechanism; cap and margin	Options Fundamentals & Markets
Unadjusted splits/dividends wrong by a factor	Dividend/split adjustment: pure specification error	Adjust payoff and hedge ratio for splits/dividends	Options Fundamentals & Markets

Risk Management

Observed symptom	First-principles root cause	Remedy	Canonical page
VaR breaches cluster and exceed promised rate	Fat tails underpriced by normality: delta-normal breaches ~1.56% vs promised 1%	EVT tails, Cornish–Fisher, or move to ES	MC VaR
Historical VaR pinned by a few worst days; jumps when old crash scrolls off	Ghost/window artifact	Filtered historical simulation, volatility weighting	MC VaR
Summed desk VaRs understate firm risk	VaR not subadditive	Move to coherent ES aggregation	MC VaR
VaR never validated against realized breaches	No backtest, no model	Kupiec + Christoffersen backtesting battery	MC VaR
Portfolio optimized on VaR loads hidden catastrophic tails	Tail blindness: VaR misses what sits past the quantile	Constrain ES, not VaR	VaR & Expected Shortfall
Calm window assigns zero probability beyond sample max	Window truncation: calm window is a blindfold	FHS + EVT tail extrapolation	VaR & Expected Shortfall
VaR/ES fall in calm markets, allowing more leverage, spiking after losses	Procyclicality feedback	Stressed VaR/ES (Basel II.5); FRTB liquidity-horizon scaling	VaR & Expected Shortfall
Two desks with identical risk report VaRs a quarter apart	Estimation error: VaR carries sampling noise	Confidence bands; average longer/filtered samples	VaR & Expected Shortfall
Extreme quantile (0.9999) estimated from few thousand losses	Small-sample distant-quantile estimation: extrapolation	Report EVT outputs as a range, not a point	Extreme Value Theory & Fat Tails
EVT on i.i.d. assumption fails in volatility clusters	Dependence/clustering: i.i.d. excesses fail under clustering	GARCH/EWMA filter before EVT; extremal index / point-process	Extreme Value Theory & Fat Tails
Tail index fitted once to 20 years mixes regimes	Regime dependence of tail index	Conditional/rolling EVT; GARCH filtering	Extreme Value Theory & Fat Tails
Threshold choice drives the EVT answer	Threshold no-free-lunch: optimum depends on unobservable second-order term	Pick inside the stability plateau; report sensitivity	Extreme Value Theory & Fat Tails
Margin haircuts spike, forced sales depress prices	Margin/funding spiral (Brunnermeier–Pedersen)	Make haircut a function of volatility & MPOR; stress both	Liquidity Risk & Funding
Catastrophic liquidation during stress	Funding-market coupling ignored: constant haircut deletes the spiral	Stress haircut × MPOR jointly	Liquidity Risk & Funding
On-the-run Treasuries liquid individually, illiquid all together	Liquidity illusion: aggregate market liquidity is a public good	Model aggregate capacity, not sum of individual estimates	Liquidity Risk & Funding
Impact cost from calm sample understates stress cost ~10×	Calm-period calibration: $λ, σ$ procyclical	Use stressed parameters; re-estimate MPOR	Liquidity Risk & Funding
Options book loses on a term never on the risk page	Cross-greek interaction: vanna/volga invisible to delta-gamma-vega report	Report/hedge vanna & volga; stress spot-down-vol-up jointly	Risk Factor Sensitivities
Gamma blows up at expiry; delta flips	Pin risk: gamma → ∞, digital-like at expiry	Close/roll before expiry (operational, not mathematical)	Risk Factor Sensitivities
Netting gammas across strikes loses real scenario info	Aggregation destroys curvature information	Never net a book’s gamma at your peril; keep strike detail	Risk Factor Sensitivities
Stress scenarios chosen to look mild	Scenario selection bias: severe scenarios hardest to sell internally	Build scenario set before seeing results; include reverse-stress	Stress Testing & Scenario Analysis
Stress outputs blended into a “total risk” number	Scenarios have no probability, not aggregable	Keep stress results conditional, separate from VaR/ES	Stress Testing & Scenario Analysis
Diversification vanishes exactly when needed	Correlation breakdown: the tail killer is correlations, not levels	Shock the correlation matrix itself, not just factor levels	Stress Testing & Scenario Analysis
Delta-only stress misses convex losses	Convexity neglect	Full revaluation on option-heavy books	Stress Testing & Scenario Analysis
Merton spread far below market at short maturity	Credit-spread puzzle: default forced at maturity only	Black–Cox first-passage barriers; jump-diffusion	Credit Risk & Merton Model
Merton PD misses sudden default	No jump-to-default: diffusion cannot collapse	Add jump channel; markets price with intensities	Credit Risk & Merton Model
Recovery/LGD treated as constant	Recovery is random, correlated with default	Model stochastic recovery by seniority	Credit Risk & Merton Model
CVA underpriced under correlated default	Independence delusion (WWR): exposure, PD, LGD treated independent	Conditional exposure (Gregory §17.6)	Counterparty Risk & XVA
CDS hedge pays when exposure jumps above hedged level	Wrong-way risk unhedgeable in the tail	Avoid WWR structurally: exclude linked trades from netting; EAD/LGD = 100%	Counterparty Risk & XVA
Zero WWR parameter reported as “no WWR”	Correlation is not independence: zero correlation ≠ independence	Report dependence honestly; it’s a category error	Counterparty Risk & XVA
Delta/CS01 hedge fails to protect the default event	Jump-to-default defeats the Greeks	Explicit JTD limits, not delta	Counterparty Risk & XVA
Collateral assumed to eliminate WWR	Collateral is a timing hedge, not a default hedge	Collateral protects gradual exposure only; size accordingly	Counterparty Risk & XVA
DVA double-counted with FBA	Double-counting DVA	Use one consistent package (DVA or FBA), never a mix	Counterparty Risk & XVA
Gaussian copula underprices the joint tail	Static copula vs dynamic dependence	Regime-condition the copula; mixture/Markov-switching	Copulas & Dependence
Joint tail error compounds beyond marginal errors	Both layers estimated: margins + copula, errors compound	Fit & validate both; prefer semi-parametric margins	Copulas & Dependence
Copula family choice swings the tail answer	Family uncertainty dominates	Report a set of copulas (dependence uncertainty band)	Copulas & Dependence
Tail parameters quoted without an interval	Small-sample tail estimates: bootstrap SE ≈ the estimate	Never quote a tail estimate without an interval	Copulas & Dependence
”Estimated default copula” from discrete data	Discrete credit data under-identifies the copula	Over-claim; CDO correlation is implied, not observed	Copulas & Dependence
Every peek at the validation set improves the number	Validation-set reuse → biased low	Nested/purged CV, true holdout, multiple-testing correction	Model Risk & Validation
Benchmark shares structure/data with the model	Benchmark error: cannot challenge	Benchmark on different data/model class; measure error correlation	Model Risk & Validation
Model silently degrades when regime changes	Drift & non-stationarity: coverage decays silently	Ongoing monitoring, rolling recalibration, stress backtests	Model Risk & Validation
Optimizing “for the backtest” passes tests without improving model	Gaming the validator (Goodhart)	Effective challenge: independent party asks “why did this change?”	Model Risk & Validation
LDA capital dominated by its least-estimable parameter	Data scarcity / tail misestimation (OpRisk)	External databases, scenario analysis, Bayesian pooling, buffers	Operational Risk
Poisson independence wrong in stress	Tail dependence of rare events: clustered arrivals	Negative-binomial/contagion frequency; stress overlay	Operational Risk
Lognormal-vs-Pareto choice swings 99.9% capital ~5.8×	Severity-family selection risk	Report capital range over defensible families, never one number	Operational Risk
Capital falls when losses under-reported	Incentive-corrupted data	Independent, audited loss capture; de-minimis thresholds	Operational Risk
Compliance sells in the worst state	Procyclicality: binding constraint driven by lagging vol	Countercyclical buffers; through-the-cycle calibration	Systemic Risk & Aggregation
”Sum the ES” overstates / “treat as independent” understates	Aggregation fallacy: dependence is unobservable	Disclose the dependence assumption; capitalise the worst copula	Systemic Risk & Aggregation
Capital ratio looks fine but the run kills the bank	Liquidity not in the capital ratio	LCR/NSFR as separate hard constraints	Basel & Regulation
RWA highest exactly when capital can’t be raised	Procyclicality: PDs/vols cycle-dependent	Countercyclical buffer; through-the-cycle calibration	Basel & Regulation

Portfolio Construction

Observed symptom	First-principles root cause	Remedy	Canonical page
Weights swing violently ( $+ 300%$ to $- 200%$ ) on small updates	Inverted covariance noise maximization: $N > T$ , sample $Σ^{- 1}$ amplifies noise eigenvalues	Ledoit–Wolf shrinkage / RMT denoising	Covariance Shrinkage & RMT
Sample covariance singular; pinv papers over it	The $N > T$ cliff: sample covariance not invertible	Shrinkage/factor estimator to regularize the estimate, not the arithmetic	Covariance Shrinkage & RMT
Small-eigenvalue noise dominates weights	Amplification dominates small-sample inference	Accurate small eigenvalues matter most; shrink the low end	Covariance Shrinkage & RMT
Reported risk beats the true optimum in-sample	In-sample evaluation is not a test	Walk-forward/OOS evaluation with trailing-window covariance	Covariance Shrinkage & RMT
MVO bets hardest where evidence is thinnest	Mean-error amplification: mean is least reliable input	Shrink/regularize; Black–Litterman prior	MPT & Mean–Variance
Near-duplicate assets produce garbage weights	Small-eigenvalue explosion from collinearity	Drop/merge near-duplicates; shrink covariance	MPT & Mean–Variance
Frontier curve stable but the point on it is a knife-edge	Portrait-of-frontier-stable illusion: curve stable, weights not	Michaud resampling	MPT & Mean–Variance
Return estimates dominate the answer no matter the covariance	Means > covariances, always (Chopra–Ziemba ~10×)	Fix means (BL/shrinkage) before polishing covariance	MPT & Mean–Variance
Overbetting a strategy ruins growth	Overbetting is asymmetric in danger: past the Kelly optimum flips growth sign	Half-Kelly operating default	Kelly Criterion & Bet Sizing
Kelly computed on Gaussian variance under jumps	Gaussian variance ≠ SD of log growth under fat tails	Estimate exact growth over empirical/heavy-tailed distribution	Kelly Criterion & Bet Sizing
Overbet near $2 f^{*}$ triggers margin calls at the trough	Leverage × drawdown × margin: forced de-lever locks in ruin	De-risk pre-drawdown via fractional Kelly	Kelly Criterion & Bet Sizing
Black–Litterman weights near extremes	Confidence mistaken for conviction: near-zero $τ$ drives naive-MVO extremes	Confidence = reciprocal of estimate variance, not a fervor dial	Black–Litterman
BL still blows up on near-collinear assets	BL never fixes covariance; hides under it	De-risked covariances belong in the BL stack	Black–Litterman
BL confidently wrong in the market’s direction during bubbles/crashes	Equilibrium bias: stability bet that market is mostly right	Overlay factor/tail views; question the prior	Black–Litterman
Backtest trades continuously, alpha only exists without a market	Backtesting at zero cost: removes the no-trade region	Assume cost inside optimizer; apply true cost to realized returns	Constraints & Transaction Costs
Strategy breaks when AUM grows	Underestimating impact for large notional: impact scales with size	Re-estimate impact as a function of your own AUM	Constraints & Transaction Costs
Every fresh alpha estimate looks like a trade	Treating noise in alphas as signal	Shrink alphas AND price the trades	Constraints & Transaction Costs
Multiple constraints interact; binding one invisible	Ignoring constraint prices when combining constraints	Report active set and multipliers, not just weights	Constraints & Transaction Costs
HRP tree flips and large weight blocks move	Clustering instability: tree is estimated	Co-membership/cophenetic diagnostics; average linkage, longer window	Hierarchical Risk Parity
HRP weights noisy from single window	Weight volatility ≫ clustering volatility	Smooth weights across windows	Hierarchical Risk Parity
HRP tree rebuild churns turnover	Turnover from tree churn	Constrain turnover; smooth weights	Hierarchical Risk Parity
HRP underperforms when the premium sits elsewhere	No expected returns ⟹ structurally incomplete	Layer a return view (BL); HRP is a pure risk allocator	Hierarchical Risk Parity
Calm-data covariances understate joint losses in crisis	Correlation crisis (diversification illusion)	Stress covariance; regime conditioning; tail-aware allocation	Multi-Asset & Factor Allocation
Carry’s steady gains hide a heavy left tail	Carry crash: negative skew, mean-variance ignores it	Tail hedge, leverage caps, defensive overlay	Multi-Asset & Factor Allocation
Repeated tactical re-optimisation rewrites strategy with noise	Strategic/tactical confusion	Tracking-error budget keeps the two separate	Multi-Asset & Factor Allocation
Robust optimizer guarantees underperformance	Over-conservatism: oversize uncertainty → shrink to zero	Calibrate uncertainty to a confidence quantile	Robust Optimization
Robust box protects against the wrong adversary	Uncertainty-set misspecification: wrong shape/width	Build from the estimator’s confidence region; test sensitivity to shape	Robust Optimization
Optimizer judged by in-sample Sharpe, ranking reverses OOS	Optimism gap (overfitting)	Report OOS; penalize in-sample-optimal solutions	Robust Optimization
Resampling an overfit center reproduces the overfit	Resampling false comfort	Constrain/shrink before resampling; use to build a distribution	Robust Optimization
Shrinking toward the grand mean erases genuine alpha spread	Shrinkage-target risk	Shrink toward an economically motivated prior (CAPM/BL implied)	Robust Optimization

Dev / Systems

Observed symptom	First-principles root cause	Remedy	Canonical page
Trading engine latency spikes from 1 $μ$ s to 2 ms intermittently	OS syscall / dynamic memory heap lock on hot path	Pre-allocated arenas; no exceptions/allocations/logging in the loop	High-Performance C++
p99.9 destroyed while median unchanged	Hidden allocation → tail explosion (amortised but unbounded)	`reserve`, pools, value types	High-Performance C++
Latency-bound loop from cache misses	Cache misses from locality-blind layout	SoA/flat containers, sequential access	High-Performance C++
Benchmark lies; optimizer deleted/reordered code	Undefined behavior in C++ (signed overflow, strict aliasing, races)	Sanitizers, `std::atomic`, strict-aliasing-safe access	High-Performance C++
Branch misprediction flushes the pipeline	Unpredictable control flow	Branchless/`cmov`; sort/bucket inputs	High-Performance C++
Independent counters slow down ~18× across cores	False sharing: cache coherence line-granular	`alignas(64)` padding of per-thread state	Concurrency & Lockless
Data race silently corrupts shared state	Lost update: no happens-before edge	ThreadSanitizer; treat the GIL as an accident	Concurrency & Lockless
Correct on x86, corrupt on ARM/PowerPC	Memory-ordering bug: missing/relaxed handshake	Review store/load ordering; explicit synchronizes-with chain	Concurrency & Lockless
Kill switch stalls behind an unrelated task	Priority inversion: low-priority holder starves high-priority	Priority inheritance	Concurrency & Lockless
CAS succeeds against reused memory	ABA problem: CAS detects same address, not same object	Tagged/DCAS pointers or generation counters	Concurrency & Lockless
Order transmission delayed by tens–hundreds of ms	TCP Nagle buffering: `TCP_NODELAY` unset	Set `TCP_NODELAY`; tuned low-latency socket stack	FIX Protocol & Connectivity
Venue rejects with `SequenceNumberTooLow`; trading freezes	Sequence-number desynchronization across crash	Journal every message to durable storage before acting	FIX Protocol & Connectivity
Phantom position / double-counted quantity on replay	Duplicate execution on replay: resent `ExecutionReport` processed as new	Dedup on `ExecID(17)`; treat `PossDupFlag` as a warning to check	FIX Protocol & Connectivity
`LeavesQty` diverges from venue; hedge against ghost position	Order-state desync from a dropped report	Assert `151 == 38 − 14`; reconcile on mismatch	FIX Protocol & Connectivity
Quoting into a dead link while OS buffers	Silent session death: no heartbeat enforcement	Enforce `HeartBtInt`, `TestRequest`, auto-disconnect	FIX Protocol & Connectivity
Backtest fills on the generating event	Event-driven look-ahead: fill on the generating event	Assert causal ordering at runtime	Event-Driven Backtesting Engines
Centred rolling window leaks the future	Centred rolling windows: sees both sides of $t$	Audit every data derivation; forbid centred stats	Event-Driven Backtesting Engines
Backtest fills at mid as “conservative”	Mid-price fills: assumes a free option	Fill against the side you would actually cross	Event-Driven Backtesting Engines
Engine fills unlimited size	Infinite capacity: no volume cap	Cap fills by bar volume; state the cap	Event-Driven Backtesting Engines
Backtest can’t be re-run bit-for-bit	Non-reproducible runs: seed/tie-break/data not frozen	Fix seed, tie-break rule, data snapshot	Event-Driven Backtesting Engines
Single flipped bit rides into the model	Silent byte corruption: checksum fails but means/sds intact	Hash at ingest; reject on mismatch	Data Infrastructure & Reproducibility
A `999.99` row survives because the mean looks fine	Mean-mask trap: aggregates average away outliers	Distributional stats (z-scores) and per-cell integrity	Data Infrastructure & Reproducibility
Rerun silently diverges after a dependency/seed change	Config drift / irreproducible reruns	Lockfile; pin deps and seed	Data Infrastructure & Reproducibility
Restated data served to a backtest looks “valid”	Point-in-time leakage: semantic corruption checksums can’t see	Enforce PIT at the store/query layer	Data Infrastructure & Reproducibility
Store can’t represent what was knowable at $t$	Look-ahead in storage: no vintages	Append-only, bitemporal rows	Tick-Level Databases & kdb+
As-of join takes seconds to hours	Unsorted index in the store	Sort by time on ingest	Tick-Level Databases & kdb+
Vendor renames a field and history is reinterpreted	Schema drift: feed spec is a contract that changes	Self-describing formats; schema registry; decode-time validation	Tick-Level Databases & kdb+
Store built from today’s symbols drops the delisted losers	Survivorship in the symbol universe	Point-in-time universe membership	Tick-Level Databases & kdb+
Kernel tuning reverts after a reboot/package update	Silent config drift: tuning not versioned	Lock tuning into boot-time config; diff every deploy; alarm on tail	Low-Latency Linux & Networking
Interrupt storm re-injects variance	Jitter re-injection	Monitor p99/p99.9 continuously; EWMA drift alert	Low-Latency Linux & Networking
Reconciliation “stopped” but nobody alerts	Silent failure: alert on the absence of a clean run	Alert on age-since-last-clean-reconciliation	Production Risk Guards & Kill Switches
Per-name tolerance hides aggregate breakout	Offsetting breaks inside tolerance	Add aggregate exposure check and break-count limit	Production Risk Guards & Kill Switches
Guard enforces a limit against a wrong internal book	Guards vs reconciliation disagreement	Reconcile guards to the external record, not trust internal	Production Risk Guards & Kill Switches
Runbook unreachable in a live incident	No written runbook	Runbook reachable from alert; name first action + authorizer	Production Risk Guards & Kill Switches
Pandas write “didn’t stick”	Hidden copies: fancy/boolean indexing returns a copy	Check `np.shares_memory`; never chain indexing + assignment	Python Quant Stack
`int32` overflow silently; money inexact	dtype traps	Assert dtypes at pipeline entry; upcast accumulators; integer cents	Python Quant Stack
Python threads give ~1× speedup on CPU-bound work	The GIL	Processes for interpreted CPU work; `@njit(nogil=True)`	Python Quant Stack
`apply`/loops reintroduce per-element dispatch overhead	Interpreted dispatch per element	Native reducers, NumPy/numba	Python Quant Stack
Vectorization exhausts RAM on big grids	Production limits: full-array materialization	Stream/polars/Arrow bridges; chunk	Python Quant Stack

Accounting & Finance

Observed symptom	First-principles root cause	Remedy	Canonical page
High earnings, low cash: value screen says cheap but it isn’t	Accrual-gamed earnings: ROE/margins/P/E inflate with accruals	Sloan accrual ratio; Beneish M-score	Accounting Quality & Red Flags
Single-year accrual ratio flags a firm as manipulator	Timing luck mistaken for quality: accruals reverse	Use trend and cumulative measures, not one year	Accounting Quality & Red Flags
Cash-side fabrication invisible to accrual ratios	Structural blindness: fabricated cash revenue leaves accruals untouched	External evidence: channel checks, customer filings, receivables aging	Accounting Quality & Red Flags
Quality screen ranks firms opposite ways across proxies	The proxy determines the answer: “quality” is not one construct	Name the construct before screening; never mix proxies blindly	Accounting Quality & Red Flags
Discretionary-accrual model over-rejects poor performers	Extreme-performance contamination	Add a performance control or restrict the sample	Accounting Quality & Red Flags
Time-series ratios break across an acquisition/standard change	Comparability breaks: ratios assume a stable entity	Mark the break year; rebuild on constant-perimeter basis	Accounting Quality & Red Flags
”Improving” from a catastrophic base scores as excellent	Directional composites reward non-death	Pair every binary signal with a level condition	Accounting Quality & Red Flags
Negative P/E quoted as “cheap”	Negative/tiny denominators: ratio singular point	Skip the broken ratio; switch to EV/Sales, book anchors, Altman Z	Core Financial Ratios
P/E inflates exactly when a cyclical is cheapest	Financials/cyclicals distort the standard set	Mid-cycle normalized earnings; regulatory ratios for banks	Core Financial Ratios
”Current ratio > 2” applied universally	Absolute-threshold reading: ratios are comparative	Compare to sector medians and firm’s own history	Core Financial Ratios
Ratios not comparable across firms	Convention mismatch: averages vs ending, pre/post-tax	Fix one convention and keep it constant	Core Financial Ratios
Leverage death spiral: distress → fire-sale → capital destroyed	Leverage magnifies (MM + distress externality)	Covenant headroom, longer maturities, conservative leverage	Capital Structure & Corporate Finance
Overlevered firm takes excessive risk	Asset substitution: equity captures upside, creditors bear downside	Debt covenants, collateral, short maturities	Capital Structure & Corporate Finance
Deeply levered firm refuses positive-NPV projects	Debt-overhang underinvestment: gains accrue to creditors	Restructure/de-lever before good projects arrive	Capital Structure & Corporate Finance
Distressed equity issuance wipes out existing common	Forced dilution: issue below intrinsic transfers wealth	Raise equity before distress (pecking order)	Capital Structure & Corporate Finance
DCF value dominated by the terminal value	Terminal-value dominance: 70–90% of value is the perpetuity	Test/specific; cap growth; never let perpetuity run away	Equity Valuation
WACC used for equity flows (or mismatched beta/ERP)	Wrong cost of capital: discount rate not matched to cash flow	Match cash flow to discount rate mechanically	Equity Valuation
DCF reverse-engineered to “confirm” the market price	Anchoring to price: psychological confirmation	Estimate intrinsic value independently of price	Equity Valuation
Single-point DCF reported to the cent	Precision illusion: implies accuracy inputs don’t support	Report a range; margin of safety	Equity Valuation
Value trap: cheap because deteriorating	Price reflects expected fundamentals	Require fundamental momentum (F-score, ΔROA, CFO/NI) with cheapness	Fundamental Analysis & Screening
Genuine bargain halves again	Falling knife: prices overshoot	Momentum veto, position sizing, staged entry	Fundamental Analysis & Screening
Screen on a single multiple misses debt/quality/cash	Any one ratio is gameable and incomplete	Pair value with quality and solvency metrics	Fundamental Analysis & Screening
Screen thresholds tuned until the backtest looks beautiful	Overfitting the screen	Keep screens simple, test across decades/regimes, prefer economic signals	Fundamental Analysis & Screening
Averaging down into a trap destroys capital	Concentration and behavioral failure	Written thesis, pre-committed sell rules, diversification	Fundamental Analysis & Screening
Restated fundamentals leak the future into a factor backtest	Look-ahead bias at the sort date	Point-in-time, as-reported data only	Quantitative Fundamental Investing
Value screen’s losers vanish from today’s universe	Survivorship bias: the losers delist	Include delisted firms and terminal returns	Quantitative Fundamental Investing
Many “distinct” factors are the same risk in different clothes	Redundancy across factors	Check correlation; prefer the parsimonious set	Quantitative Fundamental Investing
Value screen looks extraordinary but holds no firm that ever failed	Survivorship bias in screens	Historical universe construction incl. delistings and terminal returns	Data Sources & Corporate Data
Line-item history silently changes when vendor refreshes	Restatement gaps: revisions downside-clustered	Version rows by filing date; flag moved periods	Data Sources & Corporate Data
13-F ownership treated as known today	Look-ahead via 13-F/ownership: legal lag bounds knowledge	Lag by filing date	Data Sources & Corporate Data
Free tier silently samples the universe	API/coverage limits	Bulk endpoints, caching, exponential backoff, identifying UA	Data Sources & Corporate Data
June-FYE firm’s “Q4” read as December	Time-zone and fiscal-calendar ambiguity	Store fiscal period explicitly; stamp next-tradable knowable-at time	Data Sources & Corporate Data
Earnings boosted by accruals without cash	The accruals trap (Sloan): high-accrual firms underperform	Never accept earnings without its cash counterpart	Financial Statements & Accounting
Revenue recognized before the customer can pay	Recognition timing games: channel stuffing, bill-and-hold	Read receivables vs revenue growth; expect later write-offs	Financial Statements & Accounting
Earnings shifted between periods with zero cash effect	Estimate games: bad-debt/depreciation/pension assumptions	Watch for accounting-policy changes	Financial Statements & Accounting
Economic claims never appear as liabilities	Off-balance-sheet operations: leases, guarantees, commitments	Read footnotes and capitalize	Financial Statements & Accounting
Same facts produce different reported earnings across borders	GAAP vs IFRS as a quality variable	Compare on a common basis only	Financial Statements & Accounting

Last note: this index is generated from, and points back to, the canonical §5 Failure Modes & Practice page of each folder - go there for the derivation, the code, and the complete numbered checklist.

Kwant Atlas

Explorer

Diagnostic Index