Round 2 capacity benchmark: provenance, source conditions and reuse status
Accessed 6 October 2026 (Asia/Shanghai).

1. Original UCL source and why additional EIL codes are not additional cycling cells
Publisher dataset: https://rdr.ucl.ac.uk/articles/dataset/Lithium-ion_Battery_INR18650_MJ1_Data_400_Electrochemical_Cycles_EIL-015_/12159462
Version-specific DOI: https://doi.org/10.5522/04/12159462.v1
Publisher API metadata: https://api.figshare.com/v2/articles/12159462
Exact raw CSV: https://ndownloader.figshare.com/files/23140433
License: CC0, explicitly present in publisher API metadata, https://creativecommons.org/publicdomain/zero/1.0/ . Citation remains appropriate.
Citation: Heenan T, Jnawali A, Kok M, Tranter T, Tan C, Dimitrijevic A, Jervis R, Brett D, Shearing P. Lithium-ion Battery INR18650 MJ1 Data: 400 Electrochemical Cycles (EIL-015). University College London; 2020.
Companion original data paper: https://pmc.ncbi.nlm.nih.gov/articles/PMC7394852/ ; DOI https://doi.org/10.1016/j.dib.2020.106033 . Its source collection DOI is https://doi.org/10.5522/04/c.4994651 . The paper explicitly inventories one electrochemical CSV (EIL-015), four electrode-assembly tomograms (EIL-005,006,007,008), two electrode/particle tomograms (EIL-013,014), and one full-cell tomogram (EIL-016). These imaging files are not additional aging trajectories. The source cannot supply three independent cycling cells in the same EIL series. A broader search did not establish other same-series cycling-cell records, so the prewritten fallback was used.
Conditions: commercial LG Chem INR18650 MJ1, nominal 3.5Ah, NMC811/graphite-silicon as documented by the publisher; chamber24°C; 1.5A CC to4.2V, CV until100mA, 4A CC discharge to2.5V;400cycles. Cell temperature may exceed chamber temperature under load. Analysis uses the reported per-cycle discharge-capacity field. The freshly downloaded CSV has the same SHA-256 as the round1 local CSV, confirming provenance of that file. This UCL cell is a previously inspected reference, not an additional external validation cell.

2. NASA additional real experimental cells
Official repository: https://www.nasa.gov/intelligent-systems-division/discovery-and-systems-health/pcoe/pcoe-data-set-repository/
Official catalog: https://data.nasa.gov/dataset/li-ion-battery-aging-datasets
Publisher-linked download (NASA page section5 Batteries): https://phm-datasets.s3.amazonaws.com/NASA/5.+Battery+Data+Set.zip
Downloaded outer ZIP:209708670bytes, SHA-256 82302a7db4fc1b34e0b6676326610438d43b816bdf11a69d1d012a464ef2f92e.
Internal archive: 5. Battery Data Set/1. BatteryAgingARC-FY08Q4.zip . All four MAT records named by this archive's README are included: B0005,B0006,B0007,B0018. No selection based on model fit is made. Other experimental groups in the outer archive are not claimed to be analyzed.
Recommended repository citation: B. Saha and K. Goebel (2007), Battery Data Set, NASA Prognostics Data Repository, NASA Ames Research Center, Moffett Field, CA.
Reuse status: public download. The current NASA Open Data catalog explicitly states “License not specified”; it does not assign CC0 or another named license. The NASA repository asks users to acknowledge the repository and data donors and supplies data without liability. These statements are preserved, rather than inventing a Creative Commons or public-domain grant. No contact or publication action was taken on the user's behalf.
Raw README defines four physical cells, room-temperature operation, CC1.5A charging to4.2V then CV to20mA, CC2A discharge to respectively2.7,2.5,2.2,2.5V for cells5,6,7,18, and EIS. The ambient-temperature field is24°C for every extracted discharge operation. The rated capacity is2Ah according to the README. Specific electrode chemistry and manufacturer are not identified in these publisher files, so this analysis does not label the NASA cells NMC811 or equate them with the phone battery.
Important target definition: the README describes its Capacity field as capacity (Ah) for discharge to2.7V, even though the operating discharge cutoff differs between cells. We use the publisher-supplied Capacity value unchanged, do not substitute an integral to each cell's deeper cutoff, and make no end-of-life threshold-crossing claim. The dataset's general30%-fade stopping description need not equal every observed final reported capacity.

3. Extraction and provenance chain
Each MAT contains charge/discharge/impedance operations. Only discharge operations contribute a capacity target. The ordinal cycle index counts discharge events in original source order; original operation index, source date vector and ambient-temperature field are retained in each derived CSV. All636discharge capacities in the four named files are finite and positive; none are smoothed or removed. Counts:168each for B0005,B0006,B0007;132for B0018. The split produces126/42per168-cycle cell and99/33for B0018. All raw capacity values are real measurements distributed by the publisher; model curves are fitted predictions, not synthetic observations.
The outer/inner archive and per-file checksum chain is saved in data/download_manifest.json, data/extraction_manifest.json and extraction_audit.json. Every CSV derived from NASA has the same fixed extraction procedure; the raw MAT is retained. Downloaded archive code is never executed; only locally authored Python scripts perform reading and analysis.

4. Interpretation and statistical scope
The four NASA records are distinct physical cells from a common released experiment. They are an external source relative to UCL, but the models are refitted to each cell's own early life. This is replicated within-cell chronological extrapolation, not cross-cell zero-shot or transfer learning. Report all four cells and equal-cell averages, not a row-pooled figure of merit. The protocol was written before evaluating these additional trajectories; it is not a registered preregistration. UCL had already been inspected in round1 and must be labeled accordingly. None of these cylindrical laboratory-cell results validates the smartphone's SOC,TTE,temperature correction, or hardware power model.
