Synthetic strategic validation

P6.7 validates a complete 10-year synthetic strategic case: optimized portfolio, monitoring policy, fixed-cycle BAU, yearly report, and scenario replay.

The committed result is notes/reference/p6-validation.v1.json.

Case

build_synthetic_case creates:

  • three segments in one span with different lengths, service, risk, and uncertainty baselines;

  • ten yearly periods from 2026 to 2035;

  • one mowing option per segment and period;

  • a service target of 0.7 and a risk target of 0.4;

  • annual LiDAR monitoring with a value-of-information selection;

  • a four-year BAU cycle;

  • a 20-row optimized/BAU report;

  • a scenario with canonical portfolio and report digests.

Checks

Code

Check

COVERAGE

Ten optimized and ten BAU yearly rows.

BUDGET

Optimized rows remain within the period budget.

SERVICE

Optimized service reaches the declared target.

MONITORING

Monitoring cost is within budget and at least one policy decision exists.

BAU

The comparison is like-for-like and the BAU plan is feasible.

TOTALS

The report total equals the sum of optimized yearly rows.

REPLAY

The scenario portfolio and report digests replay.

validate_strategic_plan returns each check separately and a single pass/fail. A failed check is never hidden behind the aggregate.

from brushline.strategic.validation import build_synthetic_case

case = build_synthetic_case()
assert case.validation.passed
assert case.portfolio.service_attainment >= 0.7
assert case.report.comparison.like_for_like

Results

On the committed case:

  • optimized service is about 0.707, present treatment cost is 920, and the plan is feasible;

  • BAU service is about 0.413 with a four-year cycle;

  • monitoring costs 200 and selects an annual LiDAR policy;

  • all seven checks pass and the scenario replays exactly.

Limits

  • The case is synthetic; it is not a field or realistic-corridor validation.

  • The option outcomes and contingencies are assumptions, not calibrated forecasts.

  • The validation proves contract and replay consistency, not global optimality.

  • No MIP/CP bound or external solver is used.