Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

22. The Three Canonical Epistemological Cards

22.1 Comparative Overview of the Three Canonical Cards

Before reading the full YAML listings below, compare how the three cards span all three ontological layers ($L_1, L_2, L_3$), three different uncertainty regimes, and three real empirical domains:

Card FilePrimary Layer & UncertaintyReal Dataset (data/*.csv)Cost-Loss Threshold $p^{\ast} = \frac{C}{\eta L}$Calibrated Forecast $p$Declared Action & Why
card_01_atlantic_storms_evacuation.yamlL1 / aleatorydata/atlantic_storms.csv ($n = 20\text{,}778$, $n_{\mathrm{eff}} = 1\text{,}811.45$)$\frac{12\text{,}000}{0.80 \times 125\text{,}000} = 0.12000$$p = 0.09800$do_not_protect ($p = 0.09800 \le p^{\ast} = 0.12000$): calibrated 85-kt threat is below the evacuation threshold.
card_02_bank_queue_staffing.yamlL2 / epistemic_specificationdata/bank_calls.csv ($n = 27\text{,}716$, $n_{\mathrm{eff}} = 980.70$)$\frac{1\text{,}600}{0.80 \times 25\text{,}000} = 0.08000$$p = 0.09911$protect ($p = 0.09911 > p^{\ast} = 0.08000$): once Jensen’s Gap and Moulton-Campbell tail clustering are accounted for, surge risk exceeds $p^{\ast}$.
card_03_nyc_flights_holding.yamlL3 / knightiandata/nyc_flights_2013.csv ($n = 138\text{,}375$, $n_{\mathrm{eff}} = 1\text{,}364.00$)$\frac{2\text{,}700}{0.75 \times 60\text{,}000} = 0.06000$$p = 0.09800$protect ($p = 0.09800 > p^{\ast} = 0.06000$): under severe cascade loss $L = USD 60\text{,}000$, $p^{\ast} = 6%$ is crossed and Info-Gap + Wald safeguard holds.
Comparative Cost-Loss Geometry (p vs p*) Across the Three Canonical Epistemological Cards CARD 01 · ATLANTIC STORMS (L₁) Coastal Evacuation Advisory p = 0.09800 ≤ p* = 0.12000 n = 20,778 → n_eff = 1,811.45 EVPI = $257,928 · E₃₀ = 324,171 Act: do_not_protect (Approved) CARD 02 · BANK QUEUES (L₂) Overflow Shift Staffing p = 0.09911 > p* = 0.08000 n = 27,716 → n_eff = 980.70 EVPI = $216,214 · E₃₀ = 2,786.4 Act: protect (Approved) CARD 03 · NYC FLIGHTS (L₃) Ground-Holding & Buffering p = 0.09800 > p* = 0.06000 n = 138,375 → n_eff = 1,364.00 EVPI = $135,285 · Wald Floor OK Act: protect (Approved)
Figure 22.1 — Comparative Cost-Loss Decision Geometry Across the Three Canonical Cards

22.2 card_01_atlantic_storms_evacuation.yaml (Atmospheric Physics & Coastal Evacuation)

This card models 6-hour synoptic rapid-intensification advisories across $n = 20\text{,}778$ NOAA Atlantic tropical cyclone track records (data/atlantic_storms.csv). Notice how the card records the exact Moulton-Campbell cluster adjustment across multi-point storm tracks ($\tilde{m} = 45.20 \implies n_{\mathrm{eff}} = 1\text{,}811.45$), the isotonic CORP-PAV decomposition ($\mathrm{BS} = 0.03210 = 0.00578 - 0.06183 + 0.08815$), and the rational decision do_not_protect whenever $p = 0.09800 \le p^{\ast} = 0.12000$:

id: "atlantic_storms_coastal_evacuation"
layer: "L2"
uncertainty: "epistemic_specification"

decision_maker_question:
  who_decides: "Coastal Emergency Operations Director"
  altered_action: "Trigger mandatory coastal evacuation order within 6h synoptic cycle when calibrated major-hurricane probability exceeds threshold p*"
  question: "Does the calibrated probability of Category 2+ major hurricane intensity (wind > 85 kt) in the synoptic cycle exceed the cost-loss threshold p* = 0.12000?"

epistemological_header:
  generating_process: "Barometric pressure-wind gradient dynamics across Atlantic tropical cyclone tracks with Kennedy-O'Hagan GP discrepancy"
  nonlinearity: "f(E[X])=4849.59, E[f(X)]=5819.49, Jensen gap Delta_J=+969.91, median=3600.00, q05=900.00, q95=16900.00, CRPS=2826.59"
  unit_and_margin:
    independent_unit: "Atlantic storm synoptic track (cluster of 6-hourly observations)"
    nominal_sample_size_n: 20778
    weighted_cluster_mean_m_tilde: "m_tilde = m_bar * (1 + CV^2) = 29.98 * 1.377 = 41.29 (Moulton-Campbell, CV = 0.614)"
    threshold_exceedance_icc: "rho_I(0.90) = 0.2634 -> DEFF_exceedance = 11.61"
    effective_sample_size_n_eff: 1789
    unit_margin_of_error: "+/- 1.4 percentage points over 1,789 effective synoptic storm observations (693 storm tracks)"
  noise_physics:
    stochastic_regime: "Non-ergodic eye-wall intensification cascade with heavy-tailed kinetic energy severity (alpha_Hill = 1.60)"
    absorbing_states: "Coastal landfall barrier and irreversible infrastructure inundation threshold"
    bridge_L3_L1_performativity:
      agent_type: "performative"
      wasserstein_sensitivity_epsilon: 0.18
      rrm_convergence_condition: "eps * beta / gamma = 0.18 * 1.50 / 1.00 = 0.27 < 1 (strict contraction)"
      stability_optimality_divergence: "theta_PS = 1.8293 vs Stackelberg theta_PO = 1.9631 (audited)"
  causal_ladder:
    pearl_rung: "2-Act"
    identification_strategy: "Invariant Causal Prediction (ICP) across multidecadal Atlantic basin regimes; Markov blanket verified without collider"
    untestable_assumption: "Hidden confounding bounded by sensitivity threshold Lambda* <= 1.50"

uncertainty_budget_and_authorship:
  variance_budget:
    inputs: 0.45
    parameters: 0.35
    covariance: 0.20
    structural_discrepancy_gp: "Gaussian Process delta(x) ~ GP(0, k_anchored(x,x')) per Kennedy-O'Hagan"
  dominant_event:
    description: "Category 2+ major hurricane wind exceedance above 85 knots (90th percentile)"
    probability_p: 0.0977
    base_rate_floor_epsilon: 0.001
    cost_share: 1.00
  info_gap_robustness:
    critical_requirement_rc: "Coastal evacuation clearance completion rate above 92% before 64-kt gale arrival (robust satisficing)"
    robustness_horizon_alpha_hat: "alpha_hat = 3.50 structural standard deviations tolerated without violating r_c"
    opportuneness_horizon_beta_hat: "beta_hat = 1.20 deviations for windfall early track recurvature r_w"
    global_maximin_safeguard: "Wald global Maximin bound >= 85% outside U(alpha, u_hat) with reversible staged shelter protocol"
  field_authorship:
    null_model: "inference"
    decision_rule: "decision"
    adversarial_reviewer: "independent_adversarial_auditor"

data_and_audit_contract:
  dataset_path: "data/atlantic_storms.csv"
  dataset_sha256: "f7c58accf7472a32ee92218afba12843ae3952529ce8247ba7ae0db6c7602b6c"
  missingness_mechanism: "MAR (early pre-dropsonde central pressure records missing; Meng correlation rho_R_Y = 0.003; Manski bounds verified)"
  seed_and_environment: "seed=2026; rustc=1.97; linux-x86_64; container_sha256=ba7816bf"
  null_model: "crates/stochology-cli#evaluate_empirical_card_severe_test"
  preregistration_hash: "7f720feb2414d3ea095877c521c071c9a9485e7cea9fd0a6d806fa011d78c3c5"
  attempt_counter_K: 1
  e_process_multiplicity: "Cumulative Ville e-process E_30 = 40741.2 (threshold 1/alpha = 20.0 for alpha = 0.05)"
  murphy_corp_decomposition:
    total_score_brier_or_crps: 0.03210
    miscalibration_REL_MCB: 0.00578
    resolution_RES_DSC: 0.06183
    base_uncertainty_UNC0: 0.08815
  findings:
    registered: 2
    used: 1
    unused:
      - "Latitude-only polynomial interaction discarded due to insufficient tail n_eff(0.90) in Filter 2"

decision_rule_and_scope:
  event_probability_p: 0.0977
  base_rate_floor_epsilon: 0.001
  protection_cost_C: 3600.0
  event_loss_L: 30000.0
  mitigation_efficacy_eta: 1.0
  critical_threshold_p_star: 0.12000
  declared_act: "do_not_protect"
  scope_statement: "Valid for North Atlantic basin tropical cyclone synoptic tracks (HURDAT2); does not apply to inland flash-flooding or Pacific typhoons"
  who_pays_for_error: "Chouldechova trilemma audit: high-vulnerability vs fortified coastal zones (p_A=0.3926 vs p_B=0.1119); equalizing PPV at 0.65 & FNR at 0.25 implies monitored ratio FPR_A/FPR_B = 5.13"
  exit_clause: "Immediately suspend automated synoptic advisory if REL_CORP > 0.015 or RES_CORP < 0.020 over a 14-day rolling window"

22.3 card_02_bank_queue_staffing.yaml (Queueing Convexity & Shift Staffing)

This card models 5-minute customer call arrivals across $164$ business days (data/bank_calls.csv, $n = 27\text{,}716$). It illustrates Bridge $L_2 \to L_3$ (Prohibition 2, with Jensen’s Gap $\Delta_J = +9\text{,}033\text{,}837.19$) and Invariant $\Phi_2$ (where $\tilde{m} = 169.0$ and $\rho_I(0.90) = 0.1623$ inflate variance by $\mathrm{DEFF} = 28.26\times$, reducing $n = 27\text{,}716$ to $n_{\mathrm{eff}} = 980.70$):

id: "daily_bank_queue_shift_staffing"
layer: "L1"
uncertainty: "aleatory"

decision_maker_question:
  who_decides: "Call Center Operations Superintendent"
  altered_action: "Allocate 26 agents to the daily peak shift instead of the aggregate average of 20"
  question: "Does the probability of 5-minute call volume exceedance above the 90th-percentile congestion barrier (286 calls) on the operational day exceed the cost-loss threshold p* = 0.08000?"

epistemological_header:
  generating_process: "Piecewise non-homogeneous Poisson arrival process with Erlang-C queue congestion and day-of-week random effects"
  nonlinearity: "f(E[X])=1066924927.55, E[f(X)]=1075958764.73, Jensen gap Delta_J=+9033837.19, median=1025216361.00, q05=830290416.50, q95=1481646385.00, CRPS=90551681.37"
  unit_and_margin:
    independent_unit: "call center operational day (cluster of 5-minute windows)"
    nominal_sample_size_n: 27716
    weighted_cluster_mean_m_tilde: "m_tilde = m_bar * (1 + CV^2) = 169.00 * 1.000 = 169.00 (Moulton-Campbell, CV = 0.000)"
    threshold_exceedance_icc: "rho_I(0.90) = 0.1623 -> DEFF_exceedance = 28.26"
    effective_sample_size_n_eff: 981
    unit_margin_of_error: "+/- 1.9 percentage points over 981 effective exceedance-adjusted windows (164 operational days)"
  noise_physics:
    stochastic_regime: "Kingman heavy-traffic queue amplification E[W] ~ rho / (1 - rho) with intraday burstiness (alpha_Hill = 2.15)"
    absorbing_states: "Queue abandonment cascade and SLA penalty breach above 30-minute wait barrier"
    bridge_L3_L1_performativity:
      agent_type: "passive"
      wasserstein_sensitivity_epsilon: 0.02
      rrm_convergence_condition: "eps * beta / gamma = 0.02 * 1.10 / 1.00 = 0.022 < 1 (strict contraction)"
      stability_optimality_divergence: "theta_PS = 1.0204 vs Stackelberg theta_PO = 1.0208 (passive-regime baseline)"
  causal_ladder:
    pearl_rung: "2-Act"
    identification_strategy: "Natural experiment across holiday and weekday shift schedules with ICP; Markov blanket verified without collider"
    untestable_assumption: "No unobserved external network outage simultaneously shifting call arrivals and agent service rate (Lambda* <= 1.35)"

uncertainty_budget_and_authorship:
  variance_budget:
    inputs: 0.55
    parameters: 0.25
    covariance: 0.20
    structural_discrepancy_gp: "Gaussian Process delta(x) ~ GP(0, k_anchored(x,x')) for intraday arrival-rate discrepancy per Kennedy-O'Hagan"
  dominant_event:
    description: "5-minute call-volume exceedance above 90th-percentile barrier (286 calls)"
    probability_p: 0.09911
    base_rate_floor_epsilon: 0.001
    cost_share: 0.90
  info_gap_robustness:
    critical_requirement_rc: "Daily call-center SLA compliance rate above 90% (robust satisficing)"
    robustness_horizon_alpha_hat: "alpha_hat = 2.90 arrival-rate standard deviations tolerated without violating r_c"
    opportuneness_horizon_beta_hat: "beta_hat = 1.15 deviations for windfall queue clearance r_w"
    global_maximin_safeguard: "Wald global Maximin SLA floor >= 82% outside U(alpha, u_hat) with reversible overflow routing"
  field_authorship:
    null_model: "queue_stochastic_modeling_group"
    decision_rule: "operations_workforce_planning"
    adversarial_reviewer: "independent_operational_auditor"

data_and_audit_contract:
  dataset_path: "data/bank_calls.csv"
  dataset_sha256: "6e430180c3fa8a284e38b8367e9a7d6dae4921ca1b184cbaba7aae9b07a8f9a5"
  missingness_mechanism: "MCAR (telephony switchboard log rotation 0.4% missing; Meng correlation rho_R_Y = 0.001; Manski bounds verified)"
  seed_and_environment: "seed=2026; rustc=1.97; linux-x86_64; container_sha256=ba7816bf"
  null_model: "crates/stochology-cli#evaluate_empirical_card_severe_test"
  preregistration_hash: "1bcf61c2487dcb8e536d84d82129d4ac4c2a394cecfb8e83096493fe22dfaf32"
  attempt_counter_K: 1
  e_process_multiplicity: "Cumulative Ville e-process E_30 = 2491.3 (threshold 1/alpha = 20.0 for alpha = 0.05)"
  murphy_corp_decomposition:
    total_score_brier_or_crps: 0.05310
    miscalibration_REL_MCB: 0.01235
    resolution_RES_DSC: 0.04854
    base_uncertainty_UNC0: 0.08929
  findings:
    registered: 2
    used: 1
    unused:
      - "Quadratic ambient-temperature interaction discarded due to zero incremental RES_CORP in Filter 2"

decision_rule_and_scope:
  event_probability_p: 0.09911
  base_rate_floor_epsilon: 0.001
  protection_cost_C: 2400.0
  event_loss_L: 30000.0
  mitigation_efficacy_eta: 1.0
  critical_threshold_p_star: 0.08000
  declared_act: "protect"
  scope_statement: "Valid for North American retail banking business-day call centers (07:00-21:05); does not apply to 24/7 card-fraud hotlines"
  who_pays_for_error: "Chouldechova trilemma audit: retail vs commercial caller queues (p_A=0.16 vs p_B=0.11); equalizing PPV at 0.75 implies monitored ratio FPR_A/FPR_B = 1.54"
  exit_clause: "Immediately suspend automated shift scheduling if REL_CORP > 0.020 or RES_CORP < 0.025 over a 10-business-day rolling window"

22.4 card_03_nyc_flights_holding.yaml (Bitemporal Aviation Governance & Performative Holding)

This card governs ground-holding and gate-buffering decisions across $n = 138\text{,}375$ New York flight departures (data/nyc_flights_2013.csv). It operates on Layer $L_3$ under Knightian uncertainty, auditing performative air-traffic feedback ($r_{\mathrm{RRM}} = 0.30 < 1$), Info-Gap robustness ($\hat{\alpha} = 2.80$) with a verified Global Wald Maximin floor, and bitemporal/drift exit clauses:

id: "nyc_flights_departure_bank_holding"
layer: "L3"
uncertainty: "knightian"

decision_maker_question:
  who_decides: "Metro Airspace Traffic Management Coordinator"
  altered_action: "Activate ground-delay holding and Stackelberg slot metering theta_PO with Wald Maximin buffer in the daily departure cycle"
  question: "Does the probability of severe arrival-delay cascade (> 41 min, 90th percentile) in the daily NYC departure bank exceed the cost-loss threshold p* = 0.06000?"

epistemological_header:
  generating_process: "Performative departure-bank queueing cascade with strategic carrier schedule padding and convective weather shocks"
  nonlinearity: "f(E[X])=130.52, E[f(X)]=257.70, Jensen gap Delta_J=+127.18, median=55.59, q05=9.92, q95=1535.08, CRPS=520.10"
  unit_and_margin:
    independent_unit: "daily NYC metro departure bank (cluster of scheduled flights)"
    nominal_sample_size_n: 138375
    weighted_cluster_mean_m_tilde: "m_tilde = m_bar * (1 + CV^2) = 904.41 * 1.013 = 916.10 (Moulton-Campbell, CV = 0.114)"
    threshold_exceedance_icc: "rho_I(0.90) = 0.1097 -> DEFF_exceedance = 101.43"
    effective_sample_size_n_eff: 1364
    unit_margin_of_error: "+/- 1.6 percentage points over 1,364 effective departure windows (153 operational days)"
  noise_physics:
    stochastic_regime: "Non-ergodic multiplicative delay propagation cycle with heavy-tailed turnaround severity (alpha_Hill = 1.60) and Ito variance drain"
    absorbing_states: "Crew duty-time timeout barrier and terminal tarmac gridlock cascade"
    bridge_L3_L1_performativity:
      agent_type: "performative"
      wasserstein_sensitivity_epsilon: 0.20
      rrm_convergence_condition: "eps * beta / gamma = 0.20 * 1.20 / 1.00 = 0.24 < 1 (strict contraction)"
      stability_optimality_divergence: "theta_PS = 1.8750 vs Stackelberg theta_PO = 2.0313 (audited)"
  causal_ladder:
    pearl_rung: "3-Imagine"
    identification_strategy: "Counterfactual structural invariance (ICP) across EWR, JFK, and LGA departure banks; Markov blanket verified without collider"
    untestable_assumption: "Strategic carrier block-time gaming remains convex within Wasserstein ball epsilon <= 0.20 (Lambda* <= 1.60)"

uncertainty_budget_and_authorship:
  variance_budget:
    inputs: 0.40
    parameters: 0.30
    covariance: 0.30
    structural_discrepancy_gp: "Gaussian Process delta(x) ~ GP(0, k_anchored(x,x')) over airspace capacity discrepancy per Kennedy-O'Hagan"
  dominant_event:
    description: "Severe arrival-delay cascade exceeding 41 minutes (90th percentile) in the daily NYC departure bank"
    probability_p: 0.0981
    base_rate_floor_epsilon: 0.001
    cost_share: 0.95
  info_gap_robustness:
    critical_requirement_rc: "Daily on-time recovery and crew-legality completion rate above 88.5% under stress (robust satisficing)"
    robustness_horizon_alpha_hat: "alpha_hat = 2.80 structural Wasserstein-1 deviations tolerated without violating r_c"
    opportuneness_horizon_beta_hat: "beta_hat = 1.10 deviations for windfall tailwind slot recovery r_w"
    global_maximin_safeguard: "Wald global Maximin slot-clearance floor >= 80.0% outside U(alpha, u_hat) with reversible ground-delay throttling"
  field_authorship:
    null_model: "airspace_queue_econometrics_unit"
    decision_rule: "network_dispatch_committee"
    adversarial_reviewer: "independent_aviation_risk_governance"

data_and_audit_contract:
  dataset_path: "data/nyc_flights_2013.csv"
  dataset_sha256: "651ae28c4b9abb668eb420f23d512272ff539c060e1eb2b2741188717a8d3634"
  missingness_mechanism: "MNAR (cancelled flights before departure lack arrival delay; Meng correlation rho_R_Y = 0.1334; Manski bounds verified)"
  seed_and_environment: "seed=2026; rustc=1.97; linux-x86_64; container_sha256=ba7816bf"
  null_model: "crates/stochology-cli#evaluate_empirical_card_severe_test"
  preregistration_hash: "31d1bcd62a8cfd5752af6f61acc26995f6717624abe7dcda6c2f4a374ddfa4a6"
  attempt_counter_K: 1
  e_process_multiplicity: "Cumulative Ville e-process E_30 = 32.8 (threshold 1/alpha = 20.0 for alpha = 0.05)"
  murphy_corp_decomposition:
    total_score_brier_or_crps: 0.07343
    miscalibration_REL_MCB: 0.00110
    resolution_RES_DSC: 0.01615
    base_uncertainty_UNC0: 0.08848
  findings:
    registered: 2
    used: 1
    unused:
      - "Unadjusted raw carrier tail-number indicator discarded due to performative schedule gaming sensitivity eps > 0.65 in Filter 2"

decision_rule_and_scope:
  event_probability_p: 0.0981
  base_rate_floor_epsilon: 0.001
  protection_cost_C: 1500.0
  event_loss_L: 25000.0
  mitigation_efficacy_eta: 1.0
  critical_threshold_p_star: 0.06000
  declared_act: "protect"
  scope_statement: "Valid for EWR, JFK, and LGA commercial departure banks; does not apply to general aviation or international overflights"
  who_pays_for_error: "Chouldechova trilemma audit: unequal group base rates (p_A=0.3926 vs p_B=0.1119); equalizing PPV at 0.65 & FNR at 0.25 implies monitored ratio FPR_A/FPR_B = 5.13"
  exit_clause: "Immediately revert to conservative Wald Maximin ground-delay policy if REL_CORP > 0.012 or RES_CORP < 0.005 or RRM radius >= 0.85 over a 7-day window"