# Nine Racks — selected record exhibition

Source data: Qu, Jingbo (2026), *Dataset for: A Multi-Agent AI Framework for
Explainable Battery System Maintenance*, Zenodo, DOI 10.5281/zenodo.20094787.
Source record: https://zenodo.org/records/20094787
Listed licence: https://creativecommons.org/licenses/by/4.0/

Related article: Qu et al., *A multi-agent AI framework for explainable battery
system maintenance*, DOI 10.1016/j.xcrp.2026.103388. The authors do not endorse
this exhibition. Their conclusions are not used as evidence for its descriptions.
Source material is provided without warranties under its stated licence.

## LFR's changes

LFR selected and arranged record excerpts, constructed a simple latest-record
cache and four custom summaries, retained nominal-Ah voltage selections, and
displayed previously defined current-band passages. The selected payload is
available to the browser in `records.js`. This is not the complete raw archive,
the original BMS, or a public end-to-end replay package for the LFR research.

## Exact selection inventory

- Beginning: retained native-support descriptions of operations #130
  (`202412_Operation40`) and #145 (`202412_Operation55`).
- Same summary: first chronological same-summary / changed-member-voltage cache
  pair in each of the previously selected operations #1, #349, #5 and #350.
  Summary = median voltage, max-minus-min voltage, median signed current,
  median reported SOC, using the latest retained record per member. Equality
  uses unrounded values. Display rounding does not determine equality.
  The public payload includes all member voltage records for both cache states;
  these support recomputing the two voltage summaries and inspecting the member
  differences. The current and reported-SOC medians are precomputed values from
  the retained example. Their per-member inputs are not served here, so those
  two medians cannot be independently recomputed from this exhibit.
- Order: all four operations #318, #333, #339 and #350 and the nominal 75, 100,
  125, 150 Ah positions in the existing display. Two unsupported positions stay
  blank. Each Ah coordinate integrates that Rack's absolute recorded current
  from its own first usable current record. It is not capacity or a common
  physical time. All members, pairs, source indices, actual Ah, native voltage
  times and current-interpolation support spans remain available.
- Current/voltage: the existing #4 R1 charge bookmark and #291 R6 longest
  discharge passage at the 5 A absolute-current range scale. Every retained
  displayed current/voltage point for all nine members inside each fixed
  interval is kept. Native source indices are retained. No smoothing or
  interpolation is applied to these plots.
- End of record: the existing #30 R4 and #287 R5 bookmarks for largest
  same-sign, no-gap range-decrease boundaries. Show ten preceding recorded
  seconds, followed by twelve blank display seconds. They are not the largest
  physical transients, and the blank is not a zero or a recovery observation.
  Each member's last-current age and exact-boundary availability are shown.
  Rack 4 in #287 has no current or voltage record at the selected boundary;
  its last current is 2.880 s earlier. It is retained in the crop but cannot
  support a boundary-endpoint comparison.

The examples are retrospective and illustrative, not a population estimate.
The existing current-band rule splits at gaps over 10 seconds, source-index
gaps and sign/zero changes. It does not identify a physical steady state.
The 5 A scale is a declared viewing choice, not a universal constant.

## Reading limits

All values are observations of records or arithmetic on those records.
Measurement-chain bias, calibration, acquisition synchrony and original BMS
control decisions remain unresolved. Equal stored values do not establish
physical equality. More stored decimal places do not establish more accuracy.
Plots keep native clocks. A cache or a nominal-Ah selection is not a simultaneous
physical state. No complete lifetime or between-operation continuity is claimed.

Observation/measurement layer only: selected retained current and voltage record
descriptions, with reported SOC used only in the custom summary example.
Not SOH, RUL, safety, fault, remaining-life, warranty, underwriting or investment
advice. No physical continuity pattern, material mechanism, BMS omission, control
benefit or intervention basis is established by this page. Material-mechanism
interpretation would require independent physical evidence.

Project-wide boundary below is the observatory's outer limit, not a result
established by this exhibit; this page does not claim physical continuity patterns:

Observation/measurement layer only. Not SOH, not RUL, not safety, not fault, not remaining-life,
not warranty/underwriting/investment advice. It shows response-organisation continuity patterns
visible in the named channels. Material-mechanism interpretation requires independent physical
evidence (EIS / teardown / reference electrode / imaging).

## Presentation edition 0.2

This edition uses the existing LFR website font, navigation, visual tokens and
company footer. The selected record payload is byte-identical to edition 0.1.
No operations, channels, sample values, interval boundaries or calculations
have been added or changed. The exhibition is a guided selection, not the
complete local research explorer. Room 04 deliberately retains only #4 and
#291 at the fixed 5 A scale; other rooms use the separate selections above.