Planning Quality

This page answers the hardest evidence question on the site: did Verdify make useful greenhouse decisions, or did it merely create plans?

Planning quality is not a single number. It is the relationship between:

  • what the forecast said would happen
  • what the planner decided to do
  • what the greenhouse actually experienced
  • how much stress, water, and cost remained afterward
  • what lessons were extracted for the next cycle

For the exact parameters the planner can change, see AI-Writable Tunables.

Open the live Planning Quality dashboardGrafana read-only view

Use this for the continuously refreshed panels. The static snapshot below is the crawler-friendly proof layer.

Current scorecard

Static public API snapshot: 2026-05-19 02:06 MDT. Source: evidence snapshot JSON and scorecard API. These values are crawler-friendly receipts; use the live dashboard link above for continuously refreshed panels.

86.8

Today's planner score

84.0%

Both-axis compliance

84.0%

Temperature compliance

100.0%

VPD compliance

0.32h

Stress-axis hours

May 18

Last validated plan

Last validated planiris-20260518-0543

Validated 2026-05-18 20:13 MDT with outcome score 7.

Latest plan statusiris-20260518-2012 · awaiting outcome

Written 2026-05-18 20:15 MDT; age 5.8h at snapshot time.

Stress breakdownheat 0.02h · cold 0.30h · VPD-high 0.00h · VPD-low 0.00h

Total stress-axis hours can exceed wall-clock hours because temperature and VPD are independent stress axes.

Latest lesson#122 · heating · low confidence · validated 1x

Cold wet overcast days with outdoor VPD near 0.1-0.3 kPa validate high moisture thresholds, conservative fog escalation, wide VPD hysteresis, short sealed-mist residence, and bias_cool +4 with gas staging; success should be judged by VPD-low and dp-risk avoidance plus bounded cold stress rather than water use alone.

Did the plan improve the outcome?

What to inspect: score trend, both-axis compliance, and stress-axis hours over the last 30 days.

The compliance panel is percent only: temperature compliance, VPD compliance, and total compliance when both axes are inside the firmware-enforced band.

The stress panel is time only: temperature stress hours, VPD stress hours, and total stress hours. Total stress can exceed 24 hours because temperature and VPD are independent stress axes.

Forecast, band, actual

What to inspect: forecast pressure, the actual-to-forecast firmware compliance band, and the actual greenhouse response across the current planning window.

These panels are the direct decision story. The forecast is the external pressure. The stitched green firmware band shows what the ESP32 considered compliant historically and what the dispatcher projects next. Crop target provenance, cfg readbacks, trigger thresholds, and padding math live in the traceability tables rather than as extra trend lines.

The planned posture maps back to bounded parameters in AI-Writable Tunables.

Compliance and accuracy

What to inspect: waypoint compliance and forecast/plan accuracy grouped by day.

Compliance is waypoint-level: planned setpoints compared against actual hourly extremes. Accuracy groups those checks by day so the pattern is easier to inspect.

Tradeoffs

What to inspect: whether water and cost increases correspond to meaningful stress reduction.

The goal is not simply to minimize cost or water. The goal is to spend them when they buy meaningful stress reduction.

Journal and lessons

What to inspect: plan outcomes, extracted lessons, and whether repeated lessons have become useful operating constraints.

Open full dashboard ↗