Capital deferral decisions are made as if the grid is static.
It isn’t.

Would the Ridgeline feeder have failed anyway?

“The Ridgeline feeder failed during last August’s heat dome. We deferred its inspection by 18 months. Would it have failed even if inspection had been on schedule?”

This question conditions on a specific past event — a real feeder, a real storm — and asks what a different decision would have changed. It cannot be answered by comparing deferred feeders to on-schedule feeders in the data, because the feeders that get deferred are already different: they tend to be in constrained-budget corridors where routine maintenance is also lower. The model separates those two effects. It then anchors itself to the background of this specific feeder — the things that were true about it that are not in the data — by entering the actual failure as observed evidence. Only once those background conditions are fixed does it ask: in this specific situation, with this specific feeder, what would the inspection have changed?

The U nodes (shown in orange in the model) represent those unobserved background factors — wildlife contact history, latent insulation weakness, localized corrosion — that are particular to this asset. They persist into the counterfactual world unchanged. This is what makes the answer individual rather than statistical: it is not asking what inspection does on average, but what it would have done here.

Answer

Probably yes — and the inspection would have made a difference, but not enough to prevent it. With only the heat dome entered, failure probability rises to 53.8% — the feeder was already in serious jeopardy before the inspection question is asked. Entering the actual failure as evidence (abduction) shifts Asset Condition to 38.9% Deteriorated and Inspection Schedule infers 41.4% Deferred — the model reconstructs the background of this specific asset from what actually happened. Forcing inspection to On Schedule in the counterfactual world improves Asset Condition to 31.6% Deteriorated — a meaningful 7-point reduction. But Outage Consequence stays at 57.0% Severe, because the feeder's failure is still anchored and the heat dome remains extreme. The honest answer: the inspection gap worsened the asset's condition in a material way, but the underlying heat exposure and the feeder's pre-existing state meant failure was the dominant outcome regardless. The deferral was not the cause — it was a missed chance to reduce a risk that was already severe.

► Run the Analysis

UtilityGridCounterfactual.bayes

Image	Obs / Do	Node	Set	Result
ugr-c-1	—	Feeder Failure		36.9% Failed / 31.1% Degraded / 32.0% Intact
	—	Outage Consequence		29.9% Severe / 32.3% Moderate / 37.8% Contained
ugr-c-2	obs	Heat Dome Severity	Extreme
	—	Feeder Fault Probability		60.6% High
	—	Feeder Failure		53.8% Failed / 27.4% Degraded / 18.8% Intact
	—	Outage Consequence		38.1% Severe
ugr-c-3	obs	Heat Dome Severity	Extreme
	obs	Feeder Failure	Failed
	—	Asset Condition		38.9% Deteriorated (from 32.8%)
	—	Inspection Schedule		41.4% Deferred (from 39.0%)
	—	Outage Consequence		57.0% Severe
ugr-c-4	obs	Heat Dome Severity	Extreme
	obs	Feeder Failure	Failed
	do	Inspection Schedule	On Schedule
	—	Asset Condition		31.6% Deteriorated (from 38.9%)
	—	Outage Consequence		57.0% Severe (unchanged)

Prior Heat Dome Abduction Counterfactual

Prior — no evidence set

Fleet-level baseline before any evidence. Feeder Failure: 38.2% Failed, 27.6% Degraded, 34.2% Intact. Outage Consequence: 28.4% Severe.

What does deferring transformer replacement actually cause?

“If we defer the substation transformer replacement program by two years to manage capital spend, what happens to the probability of a forced outage at peak summer load — and how much of that risk is the deferral itself versus the budget environment it comes from?”

When a board observes that utilities with deferred replacement programs have higher outage rates, it is seeing two things at once. Capital budget pressure independently reduces routine maintenance spend — so deferred programs arrive with degraded maintenance quality already baked in. Simply observing deferred programs in the data conflates those two effects. The model separates them: setting the program as deferred via the observation tells you what deferred programs tend to look like; forcing the program to deferred via intervention tells you what deferral alone would cause, holding maintenance quality at whatever it would have been. The gap between those two numbers is the confound the board needs to understand before it decides.

Answer

Deferral itself raises forced outage risk by 9.8 percentage points — from 27.9% to 37.7% High — but the confound from observational data is smaller than intuition suggests. Observing deferral (telling the model a deferral has been recorded) raises High outage risk to 39.5%, because the model infers that budget pressure is probably elevated (59.0% High) and maintenance quality probably reduced (40.0% Reduced) — the conditions that tend to accompany deferral. Forcing deferral via intervention, while holding budget pressure and maintenance at their natural levels, gives 37.7% High: a 1.8-point gap between what the data shows and what deferral alone causes. The confound is real but modest. What is not modest is the interaction with peak load: do(Deferred) under critical summer load pushes High outage risk to 67.3% and Severe regulatory penalty to 56.3%. The right board conversation is therefore less about whether the data overstates the deferral effect, and more about whether this program is being deferred into a summer where the grid cannot absorb the consequence.

► Run the Analysis

UtilityGridIntervention.bayes

Image	Obs / Do	Node	Set	Result
ugr-i-1	—	Forced Outage Risk		27.9% High / 33.2% Moderate / 38.9% Low
	—	Regulatory Penalty		30.7% Severe / 25.6% Moderate / 43.8% None
ugr-i-2	obs	Transformer Replacement	Deferred 2yr
	—	Capital Budget Pressure		59.0% High (from 30.0%)
	—	Maintenance Quality		40.0% Reduced (from 26.3%)
	—	Equipment Condition		59.0% Degraded
	—	Forced Outage Risk		39.5% High
	—	Regulatory Penalty		39.1% Severe
ugr-i-3	do	Transformer Replacement	Deferred 2yr
	—	Capital Budget Pressure		30.0% High (unchanged — confounder severed)
	—	Maintenance Quality		26.3% Reduced (unchanged)
	—	Equipment Condition		54.2% Degraded
	—	Forced Outage Risk		37.7% High
	—	Regulatory Penalty		37.9% Severe
ugr-i-4	do	Transformer Replacement	Deferred 2yr
	obs	Peak Load Stress	Critical
	—	Forced Outage Risk		67.3% High
	—	Regulatory Penalty		56.3% Severe

Prior obs(Deferred) do(Deferred) + Peak Load Critical

Prior — no evidence set

Fleet baseline before any query. Forced Outage Risk: 30.2% High, 35.8% Moderate, 34.0% Low.

What is actually driving elevated fault frequency on this feeder?

“A feeder serving a hospital district is showing elevated fault frequency this quarter. The operations team says it’s worse storms. The asset team says it’s aging infrastructure. Vegetation management says the corridors are overdue. What does the evidence say?”

At Rung 1 the model runs as a filter: enter what you know about this feeder, read which root causes become more probable. Two confounders in the graph — Seasonal Conditions and Maintenance Budget Sufficiency — are what make the diagnostic discriminate between the three teams rather than just updating each cause proportionally. Seasonal Conditions independently drives both Thermal Loading and Vegetation Encroachment: a hot wet season raises both at once. Maintenance Budget Sufficiency independently drives both Equipment Deterioration and Vegetation Encroachment: constrained budgets cut both asset maintenance and corridor clearing simultaneously. Vegetation Encroachment therefore sits at the intersection of both confounders — confirming it as the cause updates beliefs about season and budget, which propagates to the other two causes. Confirming Thermal Loading above capacity updates season strongly but leaves budget near its prior. The two confirmations produce different footprints across the graph — and that difference tells the board whose hypothesis most fits the full picture.

Answer

Elevated fault frequency alone shifts all three causes upward but names no single culprit — and the two additional data points produce genuinely different diagnostic signatures. With only Fault Frequency = Elevated entered, Seasonal Conditions rises to 33.7% Severe, Maintenance Budget Sufficiency to 35.4% Constrained, Vegetation to 45.4% Severe, Thermal to 37.1% Above Capacity, and Equipment to 40.6% Advanced. All three teams remain plausible. Confirming Vegetation = Severe then pulls Seasonal Conditions to 49.9% Severe (raising Thermal to 44.2% Above Capacity) and Maintenance Budget to 44.1% Constrained (raising Equipment to 41.7% Advanced) — because vegetation sits at the intersection of both confounders. Confirming Thermal = Above Capacity instead pulls Seasonal Conditions harder (67.0% Severe) but leaves Maintenance Budget near its prior (30.7% Constrained), raising Vegetation to 54.1% but leaving Equipment only at 34.5% Advanced. The load team's claim points to a seasonal problem; the vegetation team's claim points to both a seasonal and a budget problem. Those are different operational responses — and the model makes the difference explicit before any inspection is commissioned.

► Run the Analysis

UtilityGridDiagnostic.bayes

Image	Obs / Do	Node	Set	Result
ugr-d-1	—	Fault Frequency		31.7% Elevated / 31.7% Moderate / 36.6% Low
	—	Service Interruption		20.7% Major
	—	Regulatory Exposure		24.4% High
ugr-d-2	obs	Fault Frequency	Elevated
	—	Seasonal Conditions		33.7% Severe (from 20.0%)
	—	Maintenance Budget Sufficiency		35.4% Constrained (from 25.0%)
	—	Vegetation Encroachment		45.4% Severe
	—	Thermal Loading		37.1% Above Capacity
	—	Equipment Deterioration		40.6% Advanced
	—	Service Interruption		45.0% Major
ugr-d-3	obs	Fault Frequency	Elevated
	obs	Vegetation Encroachment	Severe
	—	Seasonal Conditions		49.9% Severe
	—	Maintenance Budget Sufficiency		44.1% Constrained
	—	Thermal Loading		44.2% Above Capacity
	—	Equipment Deterioration		41.7% Advanced
	—	Regulatory Exposure		43.5% High
ugr-d-4	obs	Fault Frequency	Elevated
	obs	Thermal Loading	Above Capacity
	—	Seasonal Conditions		67.0% Severe
	—	Maintenance Budget Sufficiency		30.7% Constrained
	—	Vegetation Encroachment		54.1% Severe
	—	Equipment Deterioration		34.5% Advanced
	—	Regulatory Exposure		43.5% High

Prior Fault = Elevated + Vegetation Severe + Thermal Above Capacity

Prior — no evidence set

Root causes at prior marginals. Fault Frequency: 29.8% Elevated, 41.6% Moderate, 28.6% Low. Service Interruption: 28.3% Major.