Proof

See the evidence, not just the grade.

A continuous score, its drivers, and the evidence underneath it — never just a grade on its own. Hash-verifiable, reproducible, scope-stamped and expert-reviewable.

The evidence package

Four readers. One underlying record.

FINDING · UNDERWRITING

Degradation trajectory exceeds the model baseline from day 24. Primary driver is DC fast-charge frequency, not calendar age.

LIMITATIONS & CONFIDENCE

The rating is decision-support, not a bindable guarantee of loss or acceptance. The formula is proprietary.

REEF SYNTHETICsame record · four lenses · the interface changes, the record does not
How we validate

Reef builds the world. Arctura judges it.

Every model is stress-tested against a synthetic falsifiability substrate before it goes near a production grade. Reef can replay the same seeded world while changing only one controlled variable — so you can see that only the changed asset moved.

See a replay at reef.neurflow.fi →
VALIDATION SUBSTRATE
LFP degradationsub-5% trajectory error · shape validated · LFP/MATR
Duty-robustnessHT ↔ BESS zero-shot < 3%
Data synergynot shown — stated plainly
Isolationrandomness forked per asset — one device never reshuffles the fleet
REEF SYNTHETIC
Counterfactual · a worked example

Change one variable. Watch only the victim move.

We took a seeded BESS world — eight racks, seed 42 — and raised the charge C-rate on a single unit (reef-bess-003) from 0.5 to 2, an aggressive fast-charge duty. Everything else was held identical, then re-run on the same seed. Only the perturbed rack moved.

flip: cRate 0.5 → 2·seed 42·7 of 8 byte-identical
reef-bess-001
unchanged
byte-identical
reef-bess-002
unchanged
byte-identical
reef-bess-003
62.04 60.45%
SoH · victim
reef-bess-004
unchanged
byte-identical
reef-bess-005
unchanged
byte-identical
reef-bess-006
unchanged
byte-identical
reef-bess-007
unchanged
byte-identical
reef-bess-008
unchanged
byte-identical
ATTRIBUTED TO THE FLIP · reef-bess-003
SoH
62.04% → 60.45%
Residual value
13.43% → 11.74%
But-for Δ
≈ €4.7k

The €Δ scales an illustrative €280k reference capex by reef’s synthetic RV curve — a relative model delta, not a market price(DR-236). The moat is the isolation, not the number: same seed, one variable, only the victim moves — a but-for you can’t reconstruct on real telemetry, because you can’t re-run the world.

Why it stays credible

Keeping the two separate is what keeps the evidence credible.

Does not write the test it grades against

The engine that produces the rating is separate from the substrate that stress-tests it.

No access to proprietary rating weights

The replay substrate can't see the composite formula it's probing.

Relative modeled delta, never a market price

Reef € figures are model deltas vs a synthetic reference — never quoted market prices.

Not court-grade proof

A rating is decision-support and reproducible evidence — not legally proven causation.

See it live · nothing behind a login

Don’t take the methodology on faith. Open it.

Every surface below is public and honestly labelled — synthetic where the data is synthetic, sandbox where the engine runs on fixtures, live where it’s real. No customer data is exposed.

Sample rating

REEF SYNTHETIC

A full A–F asset rating, walked end to end on synthetic inputs.

Open →

Scan your domain

LIVE

An instant readiness grade from public data — no login.

Open →

Multi-vertical demo

SANDBOX

One scoring engine across asset classes, on sandbox data.

Open →

Incident wall

REEF SYNTHETIC

A reef-synthetic disaster reconstruction — every graph banner-labelled.

Open →

Portfolio preview

SANDBOX

A composition view — one live building, the rest sandbox.

Open →

Bring the decision you need to defend.

Open a sample rating and walk the evidence package end to end, or go straight to the trust page and see what’s live, proven and still ahead.

Open a sample ratingWhat’s live, proven, ahead →