Dovrane

Storm-restoration decision intelligence

The evidence behind what gets repaired next.

Dovrane combines incomplete storm evidence with feeder, crew, material, access, safety-condition, and critical-load constraints to produce ranked repair-sequence recommendations and a replayable decision record.

See the decision loop Review the validation standard

MVP available · Evidence-bearing recommendations · Human authority preserved

SYNTHETIC EVENT / SYN-042
Evidence
14:21:04 UTC · EVENT TIME

Evidence received

Aerial frame E-184 and field ticket F-77 enter the event record.

01 / 06
PRODUCT BOUNDARY

Read-only advisory intelligence beside OMS, GIS, ADMS, and work management.

  • No autonomous dispatch
  • No switching
  • No grid control

The category gap

A damage map is not a restoration strategy.

Existing systems tell utilities what is happening, where assets are, and how approved work is executed. Dovrane addresses the decision between evidence and execution.

Existing capabilityWhat it providesWhat remains unresolved
ForecastingProbable hazard and outage impactWhich repairs should be sequenced after evidence arrives
Imagery and field assessmentProbable damage evidenceHow uncertainty changes candidate actions
OMS and ADMSOutage and operational stateWhich modeled-feasible alternative should be considered
GISAsset and topology contextHow context interacts with crews, materials, access, and critical loads
Work managementApproved execution workflowWhether the prior decision sequence improved outcomes

The Dovrane decision loop

From incomplete evidence to an accountable recommendation.

Every stage preserves what arrived, what remained uncertain, which constraints bound the choice, and who retained authority.

  1. 01

    Evidence

    Timestamped imagery, field reports, weather, asset context, and read-only system extracts.

  2. 02

    Damage hypotheses

    Probable conditions, conflicting observations, missing evidence, and explicit abstentions.

  3. 03

    Constraint graph

    Topology, crew classes, materials, access, safety conditions, and critical loads.

  4. 04

    Candidate sequences

    Multiple restoration paths generated for comparison under the same event-time state.

  5. 05

    Feasibility gate

    Independent modeled checks withhold candidates that violate known operational constraints.

  6. 06

    Ranked recommendation

    An advisory sequence with estimated outcome, alternatives, and ranking rationale.

  7. 07

    Human disposition

    Accept, modify, defer, or reject—with operator reason preserved in the record.

  8. 08

    Outcome and replay

    Versions, evidence, decisions, and observed outcomes retained for frozen-time evaluation.

One recommendation, fully explained

A recommendation is only as useful as its record.

Dovrane is designed to show evidence, tradeoffs, alternatives, uncertainty, and the constraint trace—not a black-box score.

  • Supporting and conflicting evidence
  • Binding constraints and rejected alternatives
  • Estimated restoration curve and modeled CMI
  • Operator disposition, reason, and version history
Examine the platform method
RECOMMENDATION / RR-0042

Isolate S-17, then replace pole P-204

Modeled feasible
Estimated CMI18,420modeled estimate
Customers restored2,180after action 02
Confidence postureAbstaincrossarm condition
CUSTOMERS INTERRUPTEDEVENT TIME →
BINDING CONSTRAINT

Class-H crew available after 14:36 UTC

ALTERNATIVE 03

Withheld: access road clearance unconfirmed

MODEL 0.3.1 · DATASET SYN-2026-07 · DECISION 14:22:18Z

Designed for disciplined evaluation

Five principles shape every record.

The product thesis is narrow by design: explain the sequence decision, test it fairly, and keep operational authority with the utility.

01

Evidence that admits uncertainty

Missing or conflicting evidence is visible, and abstention is a valid output.

02

Sequences constrained by reality

Topology, crews, materials, access, safety conditions, and critical loads shape candidates.

03

Explanations that show tradeoffs

Each recommendation carries alternatives, binding constraints, and ranking rationale.

04

Replay limited to what was knowable

Evaluation reconstructs the information state at the original decision time.

05

Validation against outcomes

Sequence quality is compared against approved baselines and observed restoration outcomes.

Works beside current systems

Read-only inputs. Human-reviewed output.

Dovrane is designed to consume bounded evidence and system extracts, then return a recommendation and evidence package for review.

OMS / ADMSGIS / topologyWork managementImagery / field reportsWeather / roadsCrews / materials
DOVRANEDecision intelligence
Human-reviewed recommendationEvidence · constraints · alternatives · versions

No control path to dispatch, switching, or operational technology.

Human authority

Advisory by design.

Recommendations are advisory. Human operators retain decision authority.

Failed checks are withheld. Candidates that violate modeled-feasibility rules do not reach the ranked set.

Modeled feasibility is not real-world safety. The utility remains authoritative for clearances, switching, dispatch, and execution.

Validation before claims

Publish the method before publishing a result.

Dovrane evaluates scenario distributions, adverse-tail behavior, modeled-feasibility violations, baselines, and decision traceability so buyers can assess the system behind the result.

Review the validation standard
VALIDATION STANDARDMVP framework
Primary metricEstimated customer-minutes interrupted
Information ruleIdentical evidence at event time
Scenario setsDevelopment · regression · sealed
ReportingDistribution · tail · violations
Decision recordEvidence · alternatives · constraints

Evaluation standard

See how Dovrane earns trust.

Explore the method, evidence discipline, and operating boundaries behind every recommendation.

Contact Dovrane Review trust boundaries