For battery R&D & electrochemistry
Clean Arbin, Neware, Maccor and BioLogic exports without flattening the electrochemistry, get a System Passport of the experiment — and compile an executable degradation/dynamics model, verified in free-run, with a Nonlinearity Passport. One platform, prepaid, no seats.
Free · runs in your browser
A screening check that runs entirely in your browser: the file is never uploaded and no account is needed. It maps your columns, checks time integrity, completeness, frozen channels and jumps, detects charge / discharge / rest segments — and plots the raw, unsmoothed dQ/dV, so you can see what noise does to it before any cleaning.
Drag a cycler CSV/TSV here or
Runs locally — your data never leaves the browser. For large files, the first ~32 MB are analyzed as a sample.
Thresholds (screening, not a verdict): gap = sample period > 5× median; frozen = ≥ 50 identical consecutive samples; jump = step > 8× robust scale (MAD) of the channel's increments; rest = |I| ≤ 1% of the file's max |I|; dQ/dV axes are clipped to the 1–99th percentile. Change the column mapping above and re-run if the auto-guess is wrong.
What this screening cannot tell you: whether the noise is separable from the signal, how modelable the system is, or what a cleaned dQ/dV looks like — that is the engine's job, not the browser's. Next steps: price the full cleaning (instant, metadata-only) or try the free tier — no card.
Why it matters
Differential capacity turns charge/discharge plateaus into peaks tied to electrochemical transitions — and numerical differentiation amplifies noise. The field's best practice is explicit: Dubarry & Anseán warn that smoothing can move dQ/dV peaks, and that «smoothing after derivation … should be avoided» (Frontiers in Energy Research, 2022). A cleaner that over-smooths quietly rewrites your electrochemistry.
When there is nothing to remove, the cleaning backs off to the raw signal instead of over-smoothing — RMSE at the input floor on near-clean channels. Abstention is printed, not hidden: an exact 0.0 dB means the engine declined to touch the signal. See the claim policy.
Overlay before/after, confirm the fade curve and the dQ/dV peaks. The workflow — including per-cycler units, format gotchas and step-boundary rules — is in the step-by-step cycler cleaning guide.
Three ways in
Upload a cycler CSV — every column comes back denoised, gaps filled, outliers flagged. Priced by data volume (top-ups from €15), with an instant on-page estimate: your file is read locally, only metadata is sent.
Price a file →Stream cycling channels — current, voltage, temperature — and get cleaned signal, anomaly events with explanations, channel health, and a System Passport: recovered per-channel equations with confidence intervals and an explicit list of what cannot be recovered. From €29/mo.
Explore Filtration →Cycling trajectories → an executable degradation/dynamics model with support and uncertainty — selected in free-run, shipped with a Nonlinearity Passport, manifest and checksums. Early access.
Explore NDC →Deliverables
| Artifact | What it contains |
|---|---|
| Cleaned channels | Per-column denoise, de-spike and gap fill under a do-no-harm criterion — as a clean CSV or over the API. |
| Anomaly timeline | Named events with attribution: outlier, level shift, regime change, frozen channel, degradation — with per-channel health. |
| System Passport | Recovered per-channel equations with confidence intervals, channel verdicts, the unrecoverable list, and a held-out certificate. |
| Model artifact | model.json, Nonlinearity Passport, manifest, checksums and a free-run rollout report — verifiable, hand-off ready. |
Why cycling data fits
Cycling data is synchronized multichannel time series — current, voltage, temperature, protocol state — with the drive recorded alongside the response. No plant hookup is needed: upload what your cycler already exports.
Experiments repeat, so results can be evaluated offline on held-out cycles — and every published number follows the claim policy: executed runs, declared abstention, stated support boundaries.
Models move on — to BMS work, simulation and reports. Artifacts ship with a Passport, manifest and checksums, so the receiving team can verify what they got instead of taking it on faith.
Start
Free tier, no card: a fixed starter volume of processing, the full analytics surfaces, and one demo model compile on a public dataset. Or challenge us: send a cycling case — we run it against open baselines and publish both numbers, win or lose.