ONE BATTERY CLUSTER · NINE NAMED MEMBERS
Nine RacksA battery through its records.
A number can stay the same while its members change. A recorded operation start need not be the first usable current observation. Explore what becomes visible when we keep the names, clocks and distances.
Each room uses a different set of previously examined records to open one question. The examples are illustrative, not a representative sample or a frequency estimate. See which operations appear in each room ↓
01 / THE BEGINNING
Where does “start” begin?
The operation label and the first usable current record are different coordinates. Neither alone identifies the physical onset.
First usable records, Rack by Rack
Position = seconds after the recorded operation start. Markers show first records only, not continuous support. The leading current zero was excluded by the existing segmentation rule.
Observed
Interpretation
The blank interval does not show that a Rack was idle, disconnected or late to participate. The first usable current record is an observation boundary.
Inspect the native clocks and selection
Both previously retained examples are available here. Source times are used as recorded; their time zone and acquisition-clock synchrony are not established. No new operation has been selected for this exhibit.
02 / SAME SUMMARY
The numbers hold. Do the members?
Switch between two consecutive cache states. These four custom summaries stay exactly equal; some named voltage records do not.
This is our deliberately simple observer, not a reconstruction of the source BMS. Cached records have different timestamps; they are not a simultaneous physical state.
The voltage summaries can be recomputed from the member records served here. Current and reported-SOC medians are precomputed values from the retained example; their per-member inputs are not included in this exhibit.
Keep the member names
Hollow markers retain the other cache state for comparison. The short connector represents a recorded value difference, not a physical trajectory. Axis is rack terminal voltage in volts.
Observed
Interpretation
These four summaries cannot uniquely represent the retained member values. This says nothing about which member information the original BMS kept, or whether a control decision should change.
Inspect every member record and its age
The cache stores the latest available records up to each replay timestamp. All nine members must have current, voltage and SOC records before the four summaries exist. SOC here is the source system’s reported estimate, not independent ground truth. Each example is the first exact-equality instance in its previously selected operation; selection is retrospective.
03 / ORDER & DISTANCE
A change of place has a distance.
Follow named members through retained voltage selections. Then inspect how far apart a pair actually is.
Each Rack’s Ah starts at its own first usable current record. A nominal Ah position is neither capacity nor a shared physical instant. Higher voltage rank does not mean better health or greater burden.
Recorded voltage order
Rank 1 = highest retained rack voltage. Exact stored-value ties share the mean of occupied ranks. Dashed lines guide the eye between selections; the intervening path has not been observed here.
Observed
Interpretation
Order alone discards distance. A tie means equality of stored values, not equality within known uncertainty. Cross-Rack calibration offsets are unresolved; these patterns do not identify material differences.
Inspect the actual Ah, voltage times and current support
I* is the inherited interpolation at the selected voltage time, not necessarily a simultaneous current sample. The selection uses the existing ±2.5 Ah tolerance. The four operations come from a previously examined comparison set; they are not independent repetitions or one transitively matched class. Missing positions remain unfilled.
04 / CURRENT & VOLTAGE
A narrow current band. A moving voltage.
Two selected passages, one charge and one discharge. An existing current-only rule picks each interval; all nine members keep their own current and voltage records inside it.
Why these two? #4 is a previously viewed charge bookmark, located from R1. #291 is the longest retained discharge passage at the 5 A scale, located from R6. They illustrate the question; they do not stand for all operations. This room does not offer the full archive or a 5 A / 10 A scale selector.
A 5 A range is a viewing rule, not a physical constant or a claim of constant forcing. Other conditions are not held equal.
Hover near a point to inspect its source index, recorded time and value.
Native points only. No smoothing, joining, resampling or shared-state reconstruction. Both charts use the same recorded-clock axis; that does not establish simultaneous acquisition.
Observed
Interpretation
A current band does not fix voltage. These records alone cannot separate prior state, electrochemistry, control and measurement contributions. They invite a question, not a diagnosis.
Inspect per-member endpoints and how the passage was chosen
Starting at a current record, the inherited rule extends a passage while max|I| − min|I| ≤ 5 A. It splits at larger ranges, gaps over 10 s, source-index gaps, or sign/zero changes. The charge example is a previously viewed #4 R1 bookmark; the discharge example is the longest retained discharge passage at this scale. Voltage does not choose the boundaries. Each member’s endpoint times can differ.
05 / AFTER THE RECORD
Leave the unrecorded part open.
A marked current decrease occurs at the end of these retained excerpts. What happens afterwards is not supplied by a line drawn across the blank.
The blank twelve seconds are display space, not measured zero current. The previous ten seconds are a viewing crop. Neither is a physical event window.
Observed
Interpretation
We cannot establish recovery, persistence or a shutdown cause from these terminal records. The next useful evidence would preserve the subsequent current, voltage and control records on qualified clocks.
Selection and the unanswered question
These are the existing largest decreases in absolute current among same-sign, no-gap range-decrease boundaries, separately for charge and discharge. That is a restricted record selection, not the largest physical transient. Sign changes, zero transitions and gaps were not candidates. Each selected new passage is a singleton: it has no duration from which to infer a response.
READ THE EXHIBIT WITH ITS BOUNDARIES
The records have a source.
The reading has a scope.
Source & transformation
Qu, Jingbo (2026), Dataset for: A Multi-Agent AI Framework for Explainable Battery System Maintenance, Zenodo, DOI 10.5281/zenodo.20094787. The record lists CC BY 4.0.
LFR selected and arranged the retained excerpts, custom summaries, ranks and current-band passages. This is not a reproduction of the authors’ analysis or an indication of their endorsement. Source material is supplied without warranties under its stated licence.
Related article: Qu et al., A multi-agent AI framework for explainable battery system maintenance. Its scientific conclusions are not evidence for the exhibit’s descriptions.
What is kept, what is unknown
Rack labels, stored values and native timestamps remain attached to each selection. Reported current is negative for charge and positive for discharge. Measurement accuracy, cross-Rack offsets, acquisition synchrony and original control logic are not established here.
Extra stored decimal places do not establish measurement accuracy. SOC summaries retain the source estimate; the temperature channel is not part of this exhibit.
The examples are selected, retrospective excerpts. They are not a complete operating history or an estimate of how frequently a pattern occurs. Descriptions do not become physical findings because they are interactive.
Different rooms, different selections
| Room | Recorded operations | Selection purpose |
|---|---|---|
| 01 · The beginning | #130, #145 | Two retained examples of record-start support. |
| 02 · Same summary | #1, #349, #5, #350 | The first exact-summary match with changed member voltage in each previously selected operation. |
| 03 · Order & distance | #318, #333, #339, #350 | All selections in an existing nominal-Ah comparison set, with missing positions left blank. |
| 04 · Current & voltage | #4, #291 | A viewed charge bookmark and the longest retained discharge passage at the 5 A scale. |
| 05 · After the record | #30, #287 | Largest retained current decreases within a restricted boundary selection. |
These are separate viewing selections, not one matched sample. All nine members remain attached to each displayed case. A broader local research explorer is not included in this public exhibit.
Public inspectability and limits
The exhibition’s selected record payload is delivered to your browser. It supports inspection of the served member records, voltage arithmetic and selections. The current and reported-SOC summary medians are precomputed retained values; their underlying per-member inputs are not served, so those medians cannot be independently recomputed from this exhibit. It is not a public end-to-end replay of the complete LFR analysis. Read the selection and method notes.
Unsupported positions remain blank. All nine members are retained in each displayed case. No raw monthly MAT files or complete internal research archive are served here.