Confidence scoring (4-dim weighted model)

Every kW in an Opnor audit carries a confidence score. The score is a single percentage but it's a 4-dimensional weighted model under the hood — and understanding the weighting is the difference between "our data is garbage" and "here are the three assets we need to field-measure."

Why we score, not just average

A common mistake in industrial audits is reporting a single "data quality percentage" per plant. It's correct in spirit but useless in practice — it doesn't tell you which inputs are weakest, where to spend your verification budget, or which ECMs you should hesitate to recommend.

Opnor's confidence model breaks the score into the four input fields that actually drive savings calculations, weights them by their leverage on the energy estimate, and surfaces the result per asset. The audit report can then show you exactly which 3–5 assets need a field walk-through to lift the site-level score above 80%.

The 4-dimensional model

Every asset row contributes four input fields to its annual energy estimate. The confidence score weights them by impact on the result:

Source taxonomy (5 levels)

Each of the four fields takes one of five source-type values, each with its own confidence weight before being multiplied by the dimension weight above.

How the weights aggregate

An asset's confidence score is the dimension-weighted sum of its source-level scores. For an asset with:

• kw_source = NAMEPLATE → 80%
• lf_source = ESTIMATED → 60%
• hours_source = MEASURED → 100%
• fuel_source = N/A (electrical asset) → not counted

The score is:

# Source-level confidence (% from the taxonomy table)
kw_conf      = 80    # NAMEPLATE
lf_conf      = 60    # ESTIMATED
hours_conf   = 100   # MEASURED
fuel_conf    = None  # not applicable — electrical asset

# Dimension weights (electrical: drop fuel, renormalize to 100%)
kw_weight    = 0.444  # was 0.40, now 0.40 / 0.90
lf_weight    = 0.333  # was 0.30, now 0.30 / 0.90
hours_weight = 0.222  # was 0.20, now 0.20 / 0.90

confidence = (
    kw_conf    * kw_weight    +    # 80  × 0.444 = 35.6
    lf_conf    * lf_weight    +    # 60  × 0.333 = 20.0
    hours_conf * hours_weight       # 100 × 0.222 = 22.2
)
# → 78%

For a thermal asset (boiler, kiln, dryer), the fourth dimension stays in and the original 40 / 30 / 20 / 10 weights apply directly.

For purely electrical assets the fuel_source dimension is dropped and the remaining three weights renormalize to 100% (kw 44%, lf 33%, hours 22%). This prevents penalizing electrical assets for not having a thermal field.

Pill thresholds and what they mean

The Asset Hierarchy and Reconciliation Dashboard render every asset's confidence as a coloured pill:

Verification-priority list

On every audit the platform produces a ranked list of the 3–5 assets where spending one day with a clip-on meter would lift the site-level confidence score the most. The ranking is a function of:

• Asset's contribution to total site energy (a 250 kW kiln matters more than a 5 kW office fan)
• Current confidence score (lower → bigger absolute lift)
• Which dimension is weakest (kw_source = ASSUMED is more verifiable than fuel_source = MISSING)
• Cluster effect — verifying one motor in a row of identical assets lifts confidence on all of them

On a typical first-pass audit (no sub-metering, asset list from spreadsheets), the site-level score lands at 60–70%. Verifying the top 5 assets per the priority list typically lifts it to 80–88%, which is enough confidence for ECMs to advance to capital allocation.