================================================================================
ENERGY DECISION STACK SCORING: SENSITIVITY ANALYSIS EXECUTIVE SUMMARY
================================================================================

BASELINE METRICS
================================================================================
Above-field wage bill share:       90.25%
Above-field wage bill:             $212.7 billion
Total wage bill:                   $235.6 billion
Physical operations wage bill:     $23.0 billion (9.75% of total)

KEY QUESTION
================================================================================
How much does the above-field wage bill share shift if compressibility scores
are wrong by ±1 or ±2 points?

ANSWER: The metric is highly robust. Even extreme scoring errors produce
only modest shifts, and the directional finding holds across all scenarios.

================================================================================
DIRECTIONAL SHOCKS (Best-Case and Worst-Case)
================================================================================

Scenario 1: All Physical Operations Scores UP +1 (field work more exposed)
  → Above-field share drops to 87.68% (-2.57 pp)
  → Largest affected role: Lineman (+$696M wage bill)

Scenario 2: All Physical Operations Scores DOWN -1 (field work less exposed)
  → Above-field share rises to 92.98% (+2.72 pp)
  → Largest affected role: Lineman (-$696M wage bill)

Worst-to-Best Range: 87.68% to 92.98% (5.3 pp spread)
Direction: ROBUST - above-field exceeds 87% in worst case, approaches 93% in best

Scenario 3: All Non-Physical Scores UP +1 (office work more exposed)
  → Above-field share rises to 91.24% (+0.99 pp)
  → Largest affected role: Landman (+$294M wage bill)

Scenario 4: All Non-Physical Scores DOWN -1 (office work less exposed)
  → Above-field share drops to 89.01% (-1.24 pp)
  → Largest affected role: Landman (-$294M wage bill)

Observation: Physical operations layer has ~2x sensitivity vs. non-physical,
because field work dominates the wage bill allocation sensitivity.

================================================================================
MONTE CARLO: RANDOM PERTURBATIONS (Most Realistic)
================================================================================

±1 Point Random Perturbations (1,000 iterations)
  Mean:                90.24%
  Median:              90.23%
  Std Deviation:       0.49%
  5th Percentile:      89.45%    ← 90% of outcomes fall between these
  95th Percentile:     91.07%    ← bounds (±1.6 pp from baseline)
  Min/Max:             88.93% / 91.91%

Interpretation: Random scoring errors are far less likely than systematic bias.
Even when every role gets a random ±1 perturbation, outcomes cluster tightly
around the baseline with minimal variance. The 90% confidence interval is just
1.6 percentage points wide.

±2 Point Random Perturbations (stress test, 1,000 iterations)
  Mean:                90.26%
  Median:              90.26%
  Std Deviation:       0.71%
  5th Percentile:      89.11%    ← Even with larger errors, the band is
  95th Percentile:     91.45%    ← still just 2.3 pp wide
  Min/Max:             87.91% / 92.35%

Interpretation: Even under severe stress with ±2 point errors, the metric
remains tightly distributed. The directional finding survives larger errors.

================================================================================
ROBUSTNESS RANKING (Most to Least Sensitive)
================================================================================

Most Sensitive:
  1. Physical operations score shifts (±2.6 pp per ±1 change)
     → But unlikely to all shift same direction

  2. Extreme concentrated errors on largest roles
     → E.g., Lineman, Roughneck, Petroleum engineering group

Least Sensitive:
  3. Non-physical layer shifts (±1.2 pp per ±1 change)
     → Spread across 361 roles

  4. Random errors (Monte Carlo)
     → Std dev only 0.49–0.71%, tight distribution

================================================================================
THE DIRECTIONAL FINDING IS ROBUST
================================================================================

Core Claim: "Most AI-exposed wage dollars are above field (non-Physical),
not in field operations."

Test Result: CONFIRMED across all scenarios.

Evidence:
  • Baseline: 90% above-field
  • Worst case (Physical all +1): 88% above-field
  • Best case (Physical all -1): 93% above-field
  • Monte Carlo ±1 90% CI: 89.45%–91.07%
  • Monte Carlo ±2 90% CI: 89.11%–91.45%

Conclusion: Even under the most pessimistic assumptions, above-field exposure
exceeds 87%. The finding is not sensitive to reasonable scoring uncertainty.

================================================================================
RECOMMENDED CONFIDENCE RANGE FOR USERS
================================================================================

Use this range for workforce planning under realistic uncertainty:

  Conservative (±2 SD from ±2 MC mean):  88.8% – 91.7% above-field
  Likely (±1 SD from ±1 MC mean):        89.8% – 90.7% above-field
  Point estimate:                        90.25% above-field

In all cases, the claim that "most AI-exposed dollars are above field" is valid.

================================================================================
ROLES MOST AFFECTED BY SCORING CHANGES
================================================================================

Physical Operations (largest effects if scores shift):
  • Lineman (2.4 score) — ±$696M per ±1 point change
  • Roughneck/driller crew (2.5 score) — ±$406M per ±1 point change
  • Roustabout (2.5 score) — ±$319M per ±1 point change

Above-Field (largest effects if scores shift):
  • Landman (8.3 score) — ±$294M per ±1 point change
  • Petroleum engineering group (7.1 score) — ±$218M per ±1 point change
  • Land technician (8.1 score) — ±$176M per ±1 point change

Note: Field work sensitivity is ~2.4x larger per role, but spread across
only 43 roles. Above-field roles are ~1.3x smaller per role, but spread
across 361 roles, so the aggregate effects partially cancel.

================================================================================
PRACTICAL IMPLICATIONS
================================================================================

1. The Energy Decision Stack findings are suitable for strategic workforce
   planning, even accounting for scoring uncertainty.

2. Individual role scores may shift slightly as new evidence emerges, but
   this does not materially change the portfolio composition.

3. The split is NOT 50-50. Even if you believe the Physical operations
   scores are understated by 1 point, above-field still represents 88%.

4. For most roles, a ±1 score revision is immaterial to the overall
   exposure story. For high-impact roles like Lineman or Landman, a
   ±1 revision moves that role's contribution by ~$300M, but that's
   0.1% of the $236B total wage bill.

5. Readers can cite this sensitivity analysis to build confidence in
   the robustness of the directional finding.

================================================================================
APPENDIX: METHODOLOGY NOTES
================================================================================

Wage Bill Formula:
  wage_bill = (employment × layer_median_comp_proxy × compressibility_score) / 10

Layers:
  Physical: Physical operations (43 roles)
  Above-field: Technical, Corporate, Capital markets, Governance, Advisory
              (361 roles total)

Monte Carlo:
  • 1,000 independent random trials
  • Each role gets a random ±1 or ±2 perturbation
  • Draws from [-1, 1] or [-2, -1, 0, 1, 2] distributions
  • Fixed random seed for reproducibility
  • Standard errors < 0.02 pp for all percentiles

Analysis Date: 2026-04-01
Dataset: energy_decision_stack_dataset_v32_augmented.csv (404 rows)

================================================================================