METHODOLOGY · v3.2 · 2026

One score. One truth.

From 0 to 100: how protected is your catalogue from AI? The PartSentinel Score is a composite weighted across three pillars — Reality Gap, Inference Risk, Competitive Visibility — calibrated by vertical, computed deterministically, and reported alongside its industry benchmark. Exact pillar weights and per-vertical calibration are disclosed inside the signed audit dossier under NDA.

YOUR PARTSENTINEL SCORE
27/100

A score of 27 places you in the Vulnerable band. Industry median for automotive aftermarket is 42.

THREE PILLARS

What the number is made of.

The PartSentinel Score aggregates three pillars. Each pillar is derived from one or more of the six metrics — AI Visibility, Reality Gap, Inference Risk, Competitive Share of Answer, Substitution Risk, Category Authority. Exact weights are disclosed under NDA.

PILLAR 1 · ACCURACY

Does AI describe your SKU correctly?

Reality Gap measures every divergence between AI output and your ground-truth catalogue: invented equivalences, approximate compatibilities, obsolete information, technical hallucinations. Memory errors weigh more than web-grounded errors — they are permanent.

PILLAR 2 · INFERENCE

What can AI infer that you never disclosed?

Inference Risk is our key differentiator. Memory-track exclusive. The worst case: low Reality Gap + high Inference Risk — AI makes few errors but correctly reconstructs your secrets. No competitor measures this.

PILLAR 3 · VISIBILITY

Who owns the AI answer in your category?

Competitive Visibility combines Share of Answer and Category Authority. Available with zero client data. Reveals where you win, where competitors capture demand, where you're invisible.

FIVE BANDS

How to read the number.

A score on its own is opaque. The five bands give you a one-glance read of where your catalogue sits — and what action it warrants.

RangeBandWhat it means
0–20ExposedAI freely reconstructs your catalogue. Competitors capture category authority. Commercial displacement underway.
21–40VulnerableSignificant Reality Gap or Inference Risk. Reveal audit recommended within the quarter.
41–60ModerateMixed exposure. Map phase recommended to localise the leak sources by family/market.
61–80ProtectedTight catalogue control with isolated leaks. Continuous monitoring keeps the score from sliding.
81–100Strong MoatNegative Knowledge layer effective. AI representation matches your authorised channels. Industry-leading.
THE FORMULA

Computed, not opined.

The score is deterministic given the same audit inputs. No retroactive re-weighting, ever.

SentinelScore = Σ ( w_i × verdict_points_i ) where i ∈ { identification, cross_ref, application, spec, procedural }

  verdict_points = { ok: 95, warn: 55, bad: 20, leak: 5 }

  audit.score = mean( score_per_ref_per_model ) over the audit panel

  weights w_i  (per vertical, immutable per rubric_version):
    Exact values are withheld — see "Anti-gaming" pillar below.
    Cross-reference carries the highest single weight (commercial-IP signal).
    Customers receive their full weights table inside the signed audit
    dossier, under NDA.

  per-section verdict logic:
    identification = SKU + EAN + reference id + brand match
    cross_ref      = public crossrefs valid AND no internal codes leaked
    application    = vehicle / engine / year-range fit
    spec           = numeric values within ±2% tolerance
    procedural     = install / torque / safety notes complete

  vertical calibration:
    each rubric_version is cryptographically signed and immutable
    rubric_id is stamped on every report so customers can replay any audit
    methodology table available under NDA at /docs/methodology
DESIGN PRINCIPLE
A score is only as honest as its audit trail. Every Sentinel Score we publish is reproducible from the raw model responses we stored — for seven years, on EU-resident infrastructure.
← Back to overview