What's behind every SOHpro Score

Live catalog

We publish current numbers from public.ev_catalog. No marketing inflation — these are the same rows the scoring engine reads.

Chemistry distribution

Why it matters: chemistry decides how we read ΔV imbalance, how fast a battery should age, and which OCV-SOC curve we compare against. A score computed without chemistry uses fleet-default NMC math, which we surface as lower confidence.

Reference bands published

Every threshold the score uses is a published band with a primary source. The colored zones in your report come from this table — no hidden math. Bands version 4.0.0 (May 2026) recalibrated the LFP knee-region historical cell-spread band to 350 / 550 / 700 mV. Version 3.0.0 introduced the 7-tier letter rating over the composite Score (A+ / A / B+ / B / C+ / C / D) detailed in the section below, plus the customer-facing expectation chip that compares your battery to its physical-age baseline. Version 1.1.0 (April 2026) added five first-visit indicators: round-trip energy efficiency, coulombic efficiency, cell voltage σ, HV insulation margin, and OCV deviation.

Letter rating (A+ → D)

The headline rating in your report is a 7-tier letter scale over the SOHpro Score (a 0–100 composite combining BMS-reported SOH 43%, cell voltage balance 22%, degradation pace 20%, energy efficiency 5%, OCV-SOC consistency 5%, and thermal uniformity 5%). The letter is not raw SOH%; it is the integrated health index.

Lower cuts (70 / 76) reflect the convergence of OEM warranty floors — most manufacturers guarantee ≥70% at 8 yr / 100k mi (Tesla, Ford, VW, Hyundai/Kia, BMW, Mercedes, Polestar, Volvo); the 76 cut separates 'warranty-margin' cars from 'near-floor'. Upper cuts (88 / 93 / 97) reflect the empirical density of the global EV fleet per Geotab 2026 (22,700 vehicles, fleet-avg 2.3%/yr) — most healthy EVs in their first 100k km live above 88, and we want to differentiate 'near-new' (97+) from 'well-cared-used' (88–93) for resale and insurance pricing signal. The 7-tier scheme is product + empirical; no scientific paper claims these exact buckets. The bands_version stamp on every report row preserves the original letter at scan time so future scale revisions cannot re-letter historical scans.

When the letter is rendered

The letter follows the composite Score's confidence — not legacy SOH-only confidence. When it appears, when it hides, and where the SOH number comes from:

• ✓
  It appears when

  The composite Score is non-zero, the scan exposes at least one measurable signal (cell balance, thermal, capacity, BMS SOH, or trajectory), AND at least half of the score's weight actually scored.

• —
  It hides when

  The unscoreable tier (zero signals evaluable, Score 0 by construction), or coverage below half (no measured capacity and no degradation rate). Both the number and the letter give way to a neutral “estimated” state.

• ◐
  Estimated-SOH reports still letter

  Reports routed through estimated-SOH paths (cycle-count, km-age, physical-model) still show a letter when coverage is enough, because cell balance + thermal account for 27% of the composite weight, measured directly.

• ⚖
  Quality gate — at rest, measured-first

  Cell balance (ΔV) and voltage-consistency are scored only when the pack is at rest, and that rest check sizes itself off the battery's measured capacity (the BMS's amp-hours, or the scanner's kWh) — biasing conservative, discarding borderline readings, when only an unverified catalog estimate exists, so an unverified spec can never relax the bar (score engine 1.4.0).

Where the SOH number comes from — measured first

The report always labels which path produced the number, with a “?” explainer next to it (Report Technical Guide §3.1). The legacy 4-tier grade (Excellent / Good / Fair / Attention) is preserved in the database for projection copy but is no longer the headline. Since July 2026 (engine 1.6.0), only the Measured rows of this table may become points in a vehicle's longitudinal SOH history and in our calibration ledger — estimated values are shown on their own report but never feed the trend, the degradation rate, or the model's self-evaluation (Report Technical Guide §3.10, the anticircularity rule). Measured non-BMS points carry an “estimated” disclosure badge next to the trend chart.

Expectation band — Δ vs physical-age baseline

Next to the letter we show how your battery compares to the physical-aging baseline for its age + kilometres + chemistry + climate. This is the existing `sohGradeContextual` Δ ladder (−1 / −3 / −6 percentage-point thresholds, calibrated against the Arrhenius-plus-cycle-life model documented in §3 of the technical guide), surfaced as a 4-state chip. The asymmetric ladder is preserved verbatim from prior bands_version releases; only the top tier splits at +1 pp to introduce a positive-direction signal.

• ↗
  Above expectedΔ ≥ +1 pp

  Battery is healthier than the model predicts for this age + km. Rare and good.

• →
  In line with age−1 ≤ Δ < +1 pp

  Within the model's ±1 pp band — typical aging for this chemistry, climate, and usage.

• ↘
  Below expected−6 ≤ Δ < −1 pp

  Trailing the model. Often early imbalance or above-average thermal stress.

• ⚠
  Far below expectedΔ < −6 pp

  Material divergence from the model. Worth a technician deep-dive at the next visit.

Models in production today

These are the models running every time you generate a Score. Each is chemistry-aware where the science demands it.

• Composite SOHpro Score — weighted combination of BMS SOH (43%), ΔV (22%), degradation pace (20%), thermal (5%), OCV consistency (5%), round-trip efficiency (5%). Components with missing inputs renormalize, never default to mid-band placeholders.
• Chemistry × SOC ΔV thresholds — DELTA_V_STATE_THRESHOLDS_MV keyed by NMC/NCA/LFP/LTO and three SOC zones. LFP plateau (mid-SOC) gets the strictest thresholds because the plateau hides imbalance.
• Contextual SOH grade — sohGradeContextual measures observed SOH against expectedSohAtKmAndAge, which combines cycle fade (km × per-1000-km rate) with calendar fade (K · √year, K per chemistry).
• Cycle-count fallback SOH — sohFromCycleCount uses Sandia/NREL per-FEC loss rates when the BMS doesn't report SOH directly. Refuses to estimate without chemistry; floors at 50% to avoid runaway.
• OCV-SOC consistency with rest gating — refuses to score when the pack is under load (|I|>2 A) or recently cycled (<30 min rest), so polarization isn't charged as imbalance.
• LFP plateau detection — OCV check is automatically suppressed for LFP at 20–80% SOC where the voltage curve is too flat to be informative.
• Projection scenarios — computeProjections returns linear-from-pace; empiricalOptimistProject returns a chemistry-aware dampened scenario (τ NMC=50k / LFP=80k / LTO=120k km).

Country & climate context

Climate adjusts calendar fade through a coarse zone classifier (temperate / warm / tropical). It is best-effort: when zone is unknown the score falls back to the chemistry-only baseline, never blocks. Defaults below; users can override per VIN. Since engine 1.7.0 the same calendar term also carries a thermal-management offset: passively-cooled packs (no active cooling — e.g. Nissan Leaf, Renault Zoe, VW e-Golf) run the physics at +10 °C effective, forced-air at +5 °C, liquid-cooled at baseline. Source for the magnitude: Geotab's 10,000-vehicle fleet study (2015 Leaf, passive: 4.2 %/yr vs 2015 Model S, liquid: 2.3 %/yr) — fleet telemetry, not peer-reviewed; we model the conservative edge of the observed range: https://www.geotab.com/blog/ev-battery-study/ — when the cooling type is unknown, no adjustment is applied.

Range estimation inputs

Translating SOH into kilometres uses four numerical inputs alongside the catalog's WLTP range and usable kWh. Each input has a primary source and is published below — same audit standard as the score reference bands. The technical guide §15 documents the full math; this table is the at-a-glance audit.

Note: cold-climate rows are intentionally not modeled — LATAM-first scope. The table needs cold rows (heater load below 0 °C adds 25–40 % consumption per AAA + Idaho National Laboratory fleet studies) before any cold-market launch. We do not claim the range estimate is accurate to ± any specific number — that would require an internal validation run we have not yet published.

Sources we cite

Deduplicated list of every primary and corroborating source published in our reference bands. Click through — every URL is real.

• Accure — LFP hysteresis effect on OCV curves
• ACS Omega — Heat Generation and Degradation during High-Temperature Aging
• Argonne National Laboratory — Vehicle Technologies Office (battery testing)
• Attia, P. M., Bills, A., Brosa Planella, F., et al. (2022) — Review — 'Knees' in Lithium-Ion Battery Aging Trajectories, J. Electrochem. Soc. 169:060517 (NREL/82801)
• Battery Design — Cell-to-Cell Temperature Gradient
• Battery University BU-1003 — Comparing Battery Round-Trip Efficiency (cumulative DC counters vs plug-side)
• Baumhöfer et al. (2014) — Production-caused variation in capacity aging trend, JPS 247:332
• Beck et al. (2021) — Inhomogeneities and Cell-to-Cell Variations, MDPI Energies 14(11):3276
• Dubarry, M., Devie, A., Liaw, B. Y. (2017) — Identifying battery aging mechanisms in large format Li-ion cells, J. Power Sources 360
• FMVSS No. 305 — Electric-powered vehicles: electrolyte spillage and electrical shock protection
• Geotab 2026 EV Battery Degradation Analysis (22,700 vehicles)
• Geotab 2026 Update — 2.3% Annual Degradation
• ISO 6469-3:2021 — Electrically propelled road vehicles, safety specifications, electrical safety
• Midtronics — Early Signs of Cell Imbalance (field-service heuristic)
• Nature Scientific Reports — Aging of Li-ion above Room Temperature
• Neural Concept — EV Battery Thermal Management (vendor overview; secondary/illustrative, not the threshold source)
• NREL — Battery Energy Storage Testing Procedures (DC-side counter analysis)
• Plett, G. (2015) — Battery Management Systems Vol. II: Equivalent-Circuit Methods
• Preger et al. 2020 (Sandia) — Degradation of Commercial Lithium-Ion Cells as a Function of Chemistry and Cycling Conditions, J. Electrochem. Soc. 167:120532
• Recharged — EV Battery Warranty Comparison (industry warranty aggregation; secondary basis for the 70% 8yr/100k-mi floor)
• Schmid, B., Cabrera, J., et al. (2022) — Knee-point detection in lithium-iron-phosphate batteries via incremental capacity analysis, Energy Storage Materials 50
• Schmid, J., et al. (2025) — Aging study of twenty 18650 graphite/LFP cells, Nature Scientific Data 12
• Schuster et al. (2015) — Lithium-ion cell-to-cell variation during BEV operation, JPS 297:242–251
• ScienceDirect — BTMS Review
• SOHpro composite-Score product calibration
• Tesla Model 3 LFP owner aggregates — InsideEVs (consumer illustration; secondary, not a controlled study)
• The Operation Window of LFP/Graphite Cells Affects their Lifetime (2024), J. Electrochem. Soc. 171
• USABC Electric Vehicle Battery Test Procedures Manual, Rev. 2 (DOE/INL, 1996) — 80% rated-capacity end-of-life definition
• VW Service Action 93P7 — HV Battery Cell Modules (NHTSA)

Our data commitments

• No hallucinated specs. A row enters the catalog with verified=false until two independent sources agree on chemistry, capacity, and pack details. Verified rows show their source list.
• Confidence travels with the data. When chemistry is unknown, the score uses NMC defaults and surfaces the assumption — it does not silently grade as if data were complete.
• Reference bands are locked to published research. We do not move thresholds to flatter results. Every band cites a primary URL you can audit.
• Climate is best-effort. The coarse zone classifier improves the projection when present and is skipped when unknown — it never blocks a report.
• Old reports do not silently re-grade. When we update models, new scans get the new math; existing reports are preserved unless an explicit migration is approved and announced.