Frozen rules. Reproducible evidence. Bounded conclusions.
VolMax Studio Lab builds verification methods and applies them to real technical claims — from energy-system telemetry and battery performance to computational research and formal mathematics. We separate what the evidence demonstrates from what it does not.
We verify the chain from claim to conclusion — and make the stopping point explicit.
Technical and scientific claims are often evaluated by the team that produced them.
In battery and scientific ML, strong reported metrics can fail to transfer when evaluation allows leakage, correlated train/test samples, protocol overlap, or benchmark-specific tuning. In formal mathematics, a machine-checked proof can be valid and still not mean what the abstract claims, if the formal statement doesn't faithfully capture the original claim.
Buyers, investors, insurers, and readers of a paper can't always tell a sound result from a well-presented one by reading the producer's own numbers.
That is a structural conflict of interest.
VolMax is the neutral check. We don't sell a competing model and don't co-author the claim under test, so we have no incentive to flatter a number. Independence is the product.
Two tracks, one methodology.
Research builds the verification tools. Applied work uses them on real claims. Both run under the same frozen-rule discipline.
Verification Research
- P10 Core
Certificate-carrying verification protocol with explicit trust boundaries — frozen milestones, Lean-checked.
- P10 Underdetermination ProfileRelease candidate (v0.1.1)
Release-candidate profile for binding NotDemonstrated(reason=underdetermined) to concrete divergent-world witnesses.
- Representation Lifting S1Preregistered — blind evaluation pending
Preregistered Lean 4 case study on whether a representation-first proof strategy transfers across blind theorem families.
- Formal mathematics (Lean 4)
Machine-checked artifacts that separate derivability from interpretation.
- Scientific reproduction
Independent reconstruction of bounded claims from published computational work and public datasets.
Applied Verification
- Battery & energy-ML claims
SOH/RUL model audits against held-out data and physical law.
- Public market telemetry audits
ERCOT, AEMO, Elexon/BMRS dispatch and capacity claims, reconstructed from primary sources.
- Power-signal & DSP verification
Measured-signal pipelines checked against physics and real sensor resolution.
- Technical due-diligence support
Pre-DD screening for funds, lenders' advisors, and insurers.
Public work, inspectable evidence.
The standard we hold every claim to — including our own — is public and inspectable. These are public, version-pinned, reproducible repositories: the credential that replaces a CV.
Flagship artifacts
P10 Underdetermination Profile
Release candidateA frozen protocol for when two witness worlds are compatible with the same closed evidence and produce different values for the same claim — v0.1.1, Lean-checked witness kernel.
Representation Lifting S1
Preregistered — blind evaluation pendingFormal case study testing whether a representation-first formalization strategy discovers reusable structure on blind theorem families.
Primitive Composition (Square Content, S1)
Machine-checkedLean 4 result on primitive Pythagorean triples: literal integer divisibility, no smuggled parity or positivity assumptions.
ERCOT Contrast — Anole vs. Bat Cave
Applied / telemetryThe same frozen protocol applied to two BESS assets returns two different outcomes — one bounded by SCED limits, one not resolvable from public telemetry alone.
Market Telemetry Audits (ERCOT · NEM)
Anole (ERCOT, 240 MW/480 MWh)
F1 power & F2 energy: Demonstrated at SCED limits. F3 SoC consistency: Inconsistent under the strict 80% rule (55.2% pass, 330 events); rises to 81.8% in an exploratory, non-pre-registered filter on major events (≥10 MWh). F4 SoC field interpretation: Deferred pending ERCOT column definitions.
Bat Cave (ERCOT, 100 MW/100 MWh)
F1 power: Not Demonstrated (Bounded) — peak 72.61 MW under SCED, HSL confirms 100 MW model capacity. F2 energy: Not Demonstrated (Not Verified). F3 SoC consistency: Inconsistent (1.2% pass, 245 events, mean ratio 0.63). F4 SoC field interpretation: Deferred. Each finding is reported separately — not collapsed into one verdict.
AEMO NEM Fleet Dispatch Audit
16 units evaluated; dispatch conformance, generalization gap, and FCAS auditability finding.
Verdicts are reported exactly as the frozen protocol renders them — not compressed into a single headline number.
Battery_Health_Portfolio
NASA PCoE + Severson/Attia, DOI-archived. Honest findings led by their limits: where early prognosis breaks, where impedance is observable vs predictive, capacity-regeneration isolated from true fade. Includes the worked example where the audit caught its own pipeline overclaiming three times. Every number regenerates from reproduce.py.
github.com/VolMax-Studio/Battery_Health_PortfolioOther reproducible projects
Power_Signal_Tools
Test-covered signal library (RMS, THD, DWT, Hilbert). The measurement layer the audits stand on.
Transformer_Health
Hierarchical DGA fault diagnosis, with tested boundaries.
PV_Anomaly_Detection
PV fault detection on real NREL data + injected benchmarks, honestly framed.
Data_Center_Efficiency
The "PUE Loophole" audit: how PSU conversion losses mask real facility savings.
(+ Power Quality, NILM, MCSA, CWRU, Grid Frequency, VPP — domain breadth.)
View all on GitHub →The P10 Verification Method
Every finding is produced by a strict procedure built on four non-negotiables. We apply it to our own work first.
Pre-registered, frozen rules
Publicly timestamped before we look at the data.
Independent evidence
Primary public sources, with a hash captured upon download.
Full reproducibility
Open code and data, DOI-archived for persistence.
Public failure registry
Our own errors are dated and listed, never deleted.
What our evidence does not claim.
This is the differentiator most verification work leaves implicit. We state it directly.
Formal proof ≠ semantic truth
A Lean proof shows a theorem follows from stated axioms. It does not show the formal statement faithfully captures the original claim.
Hash ≠ correctness
A SHA-256 hash fixes which bytes we analyzed. It says nothing about whether the analysis of those bytes is right.
Public telemetry ≠ ground truth
Public market data lets us reconstruct dispatch behavior. It is not a substitute for an asset's own physical instrumentation.
Reproduction ≠ validation of the surrounding theory
Reproducing a published result's numbers confirms the computation. It does not confirm the theory that motivated it.
Applied Verification — Engagements
We don't build your model — we check whether its numbers survive contact with data they haven't seen. Independent audits designed to catch interpretation artifacts.
1 · Battery ML Audit
Independent integrity audit of an SOH/RUL model or a vendor's accuracy claim. We check split integrity (cell-level vs cycle-level), preprocessing leakage, metric honesty (full-set error, no dropped cells), and physical consistency. Deliverable: a reproducible report where every number regenerates from one script.
2 · Independent SOH / RUL Verification
We reproduce a claimed result on a clean, group-aware split and report the gap between the claimed and the verified number — and trace its cause. The honest number is the one that survives a new cell.
3 · Power Signal & DSP Verification
Verification of measured-signal pipelines (power quality, vibration, MCSA, PV) against physics and the sensor's real resolution — built on a test-covered signal-processing core.
4 · Public Telemetry Audit (pre-DD screening)
Independent reconstruction of public market telemetry (ERCOT SCED, AEMO NEMWEB) for a specific asset prior to due diligence: dispatch history, demonstrated vs. nameplate capacity, and SoC reconciliation where semantics allow. Deliverable: a reproducible report with a hash chain and pre-registered rules. Target audience: funds, lenders' advisors, insurers.
What you receive
A reproducible audit report; a plain verdict (where the claim holds, where it breaks, the corrected number); the regeneration script.
What we do NOT guarantee
An accuracy number. We are not selling a model, so we make no "95%" promise — that is the vendor claim we exist to test. A verifier who guarantees a flattering result has the same conflict of interest as the vendor. We do not build the model under test, and we list no client outcomes we don't have — the proof is the public work above.
Ivan Nestorov — founder
25+ years in power electronics and field electrical work, now applying that hardware intuition to independent verification — of energy ML, and increasingly of formal and computational claims. The combination is the point: a pure data scientist doesn't know why a converter loses efficiency at high frequency; a pure hardware engineer doesn't audit a model's train/test split; and neither, on their own, questions whether a formal statement still means what the original claim meant.
VolMax stands at the intersection — between the measurement and the claim.
A measurement is not a conclusion. The chain between them has to be inspectable.
Hardware & embedded R&D
Alongside the verification practice, VolMax runs hardware and edge-ML R&D — embedded signal processing on STM32/ESP32, and a hardware safety-interlock concept (analog-to-logic override, sub-2.5µs actuator-interrupt latency, Serbian IPO priority filed). These inform the measurement side of the audits and are in active development.
Contact
VolMax Studio Lab d.o.o.
Independent verification of technical & scientific claims · Serbia · EU/remote engagements
Email: volmax.core@gmail.com
GitHub: github.com/VolMax-Studio
LinkedIn: linkedin.com/in/ivan-nestorov-274157371
For an enquiry, include the claim type and the relevant dataset, model, or artifact if possible.