ArraySignal

Demo Portfolio · Synthetic Sample Data

Explore the full demo workspace

Browse existing fleet tools without signing in. This read-only workspace uses bundled fictional examples, never customer records. It does not connect providers, send notifications, run models, upload data, or create work orders.

The 2-site overview, incidents, and report use July 22–August 20, 2026 reference totals. Advanced feature panels are separate illustrative scenarios with their own periods and assumptions—not recalculations of that portfolio, validated accuracy, or promised savings.

Investigation workflows · Synthetic Sample Data

Explore an exported-data analysis, evidence-packet anatomy, and outcome capture. This is a read-only illustration for Cedar Ridge Community Solar, not a processed upload, a customer investigation, or a confirmed field result. The August 20, 2026 scenario is independent of the portfolio’s fixed 30-day totals.

Forensic analysis of an operational export

In an authorized customer workspace, analyze UTF-8 CSV or tab-delimited operational data for a configured site without connecting a live inverter provider. XLS/XLSX files are not supported; export a worksheet as CSV first.

  1. 1. Upload
  2. 2. Map columns
  3. 3. Check quality
  4. 4. Analyze
  5. 5. Detect
  6. 6. Diagnose
  7. 7. Quantify
  8. 8. Prioritize
  9. 9. Review report
Inspect example column mapping and quality boundaries

Illustrative mapping only: no file was parsed and no quality score was computed for this preview.

timestamp_utc → TIMESTAMP
Timezone-aware timestamp; confirm ambiguous local time.
meter_kw → AC_POWER_KW
Site-level AC power in kW, not an inferred sum of wide inverter columns.
poa → POA_IRRADIANCE_WM2
Optional irradiance evidence in W/m²; not sufficient on its own for model-ready weather.
customer_expected → Unmapped
Not substituted for ArraySignal’s expected-production model.
  • Timestamps and AC power or authoritative interval AC energy are required. Cumulative counters are not interval energy.
  • Only deterministic repairs are allowed: identical-row deduplication, timestamp ordering, and supported interval-energy conversion. Conflicting duplicates and ambiguous values remain visible.
  • Missing readings are not filled, set to zero, or counted as invented production loss. Missing model inputs produce an insufficient-data state.
  • Model-ready on-site weather requires GHI, DNI, DHI, ambient temperature, and wind speed. The analysis does not silently substitute clear-sky production.
Inspect the synthetic analysis result

The existing Site Analysis result view below uses bundled simulated production and modeled expectations for Cedar Ridge Community Solar (America/Denver). Its run metadata identifies a reference fixture, not an analysis executed from the example mapping above. No live telemetry, incident queue, or portfolio totals were changed.

Site analysis state

Serious issue

A serious modeled performance issue requires prioritized investigation.

Model confidence

high · 100%

Confidence is a deterministic evidence-quality score, not a calibrated probability.

Deterministic input-quality index

Model Confidence: High

This score describes support for modeled expected production and loss estimates. It is not a statistically calibrated probability.

Quality index

0.858 / 1.000

Positive contributors

  • Validated synthetic reference weather supplies the demo model.
  • Synthetic measured AC power is 99.5% complete.
  • The synthetic reference site configuration is complete.

Limitations

  • Reference mode represents limited synthetic calibration history.

Modeled site-period estimate

Production impact

Estimate available
Modeled expected production
16.4 MWh
Measured actual production
3 MWh
Estimated production shortfall
13.4 MWh
Percentage shortfall
81.6%

This measured-versus-modeled estimate covers the selected period. It is not attributed exclusively to any individual probable cause.

  • This is a site-level measured-versus-modeled estimate and is not attributed exclusively to any individual finding.

Analysis complete

Run ID
d1afa274-10bd-552b-a375-ac0e16458240
Analysis version
arraysignal-reference-analysis-v1
Completed
Aug 21, 12:00 AM
Period
Aug 20, 12:00 AM–Aug 21, 12:00 AM
Model
arraysignal-reference-pv-v1
Detector
arraysignal-reference-rules-v1
Findings
1
Operating condition
underperformance
Estimated curtailed energy
—
Unknown impact
0

production · PROBABLE_OUTAGE

Likely inverter outage

Priority #1 · 91.0/100Critical

Pattern consistent with an inverter outage; inspect inverter availability, alarms, and the affected electrical boundary.

Period
Aug 20, 10:48 AM–Aug 20, 11:48 AM
Persistence
60 min
Confidence
high · 90%
Performance
53.0%
Modeled expected mean
1.38 MW
Measured actual mean
732.7 kW
Estimated lost energy
649.7 kWh
Estimated lost value
$75

Why this is Priority #1

  • Critical severity contributes to operational priority.
  • Deterministic finding confidence is 90%.
  • The site-level production shortfall contributes to priority but is not attributed exclusively to this finding.

The site-level shortfall informs priority but is not assigned solely to this finding.

Decision support · modeled estimate

Decision economics for Priority #1

Investigate remotely

Review remotely within 3 business days

Energy loss rate
649.7 kWh/day
Modeled range: 519.8 kWh–779.7 kWh
Revenue exposure
$75/day
Modeled range: $60–$90
30-day exposure
$2,242
Modeled range: $1,793–$2,690
Priority score
72.0%
Technical reasons
  • • Technical severity is warning.
  • • The underperformance pattern persisted across the analyzed period.
Economic reasons
  • • The modeled loss has an estimated revenue impact.
  • • Investigation cost is not configured.
Recoverable value
$1,513
$897–$2,286
Investigation cost
—
Modeled net value
—
Break-even
Unavailable
Expected persistence
30 days
Modeled range: 14–45 days
Decision confidence
86.0%
Deterministic rule score
Assumptions and uncertainty
  • • Future values are decision-support scenarios, not guaranteed savings.

Detected condition

Likely inverter outage

Detection confidence 90%

Likely classification

Site-Wide Underperformance

Classification confidence

78%

Deterministic score · not a probability

Why ArraySignal thinks this

  • The pattern persisted for 60.0 minutes.
  • Modeled expected output averaged 1382.4 kW.
  • Measured actual output averaged 732.7 kW.
  • The actual-to-expected ratio was 53.0%.

Evidence against certainty

  • The probable cause is not a confirmed field diagnosis.

Missing evidence

  • Inverter-level output, alarm history, curtailment status, and aligned on-site irradiance are unavailable.

Alternative explanation

  • Device Zero Output — 55%

Vendor context

  • The reference adapter supplies no vendor alarm or operating-state confirmation.

Recommended investigation

Compare inverter-level output and alarms with site controls and irradiance before assigning the broad loss to equipment.

Classifier stage24-reference-site-v1

Why ArraySignal thinks this

Probable plant availability issue

Confidence 90%

Deterministic rule confidence, not a calibrated probability or confirmed field diagnosis.

Evidence supporting
  • The pattern persisted for 60.0 minutes.
  • Modeled expected output averaged 1382.4 kW.
  • Measured actual output averaged 732.7 kW.
  • The actual-to-expected ratio was 53.0%.
Evidence against
  • No inverter fault or status code corroboration was evaluated.
  • Only site-level production was evaluated; neighboring inverter behavior is not available.
Competing explanations
  • Inverter or array derating
  • Curtailment or export limiting
  • Soiling, shading, or weather-input bias
  • Meter or sensor bias
Data limitations
  • No material telemetry completeness limitation was detected.

Evidence through Aug 20, 11:48 AM · Rule arraysignal-reference-rules-v1

What should I check next?

Recommended next check

Obtain inverter-level AC output for the affected interval and compare it with available neighboring inverters.

Access
Remote check
Expected information gain
high
Effort
low
Why it matters

This helps distinguish inverter-specific derating from a site-wide irradiance, curtailment, or expected-model issue.

Data needed
  • Timestamped inverter-level AC power
  • Aligned irradiance or expected production

Helps distinguish: Inverter or array derating · Curtailment or export limiting · Soiling, shading, or weather-input bias

Other possible checks

  • Retrieve inverter fault and event codes for the affected interval. (Remote check, low effort)

ArraySignal recommends investigation steps; it does not perform physical tests or confirm field conditions.

Recommended investigation

Review inverter alarms, telemetry completeness, curtailment, weather assumptions, soiling, and maintenance history.

Customer forensic results are kept separate from canonical site telemetry and the live operational incident queue. Selecting and processing a file requires authorized access to that site.

No upload control, file retention, or analysis request is available in this public preview.

Investigation evidence packet anatomy

A customer packet organizes a persisted finding’s production evidence, likely cause, competing explanations, estimated impact, next checks, and limitations. The example below reuses those evidence panels with the same synthetic finding. It is not an issued packet and has no persisted issue, evidence fingerprint, or recorded resolution.

Sample finding: Likely inverter outage

Pattern consistent with an inverter outage; inspect inverter availability, alarms, and the affected electrical boundary. The sample diagnosis remains unconfirmed.

Why ArraySignal thinks this

Probable plant availability issue

Confidence 90%

Deterministic rule confidence, not a calibrated probability or confirmed field diagnosis.

Evidence supporting
  • The pattern persisted for 60.0 minutes.
  • Modeled expected output averaged 1382.4 kW.
  • Measured actual output averaged 732.7 kW.
  • The actual-to-expected ratio was 53.0%.
Evidence against
  • No inverter fault or status code corroboration was evaluated.
  • Only site-level production was evaluated; neighboring inverter behavior is not available.
Competing explanations
  • Inverter or array derating
  • Curtailment or export limiting
  • Soiling, shading, or weather-input bias
  • Meter or sensor bias
Data limitations
  • No material telemetry completeness limitation was detected.

Evidence through Aug 20, 11:48 AM · Rule arraysignal-reference-rules-v1

Decision support · modeled estimate

Decision economics for Priority #1

Investigate remotely

Review remotely within 3 business days

Energy loss rate
649.7 kWh/day
Modeled range: 519.8 kWh–779.7 kWh
Revenue exposure
$75/day
Modeled range: $60–$90
30-day exposure
$2,242
Modeled range: $1,793–$2,690
Priority score
72.0%
Technical reasons
  • • Technical severity is warning.
  • • The underperformance pattern persisted across the analyzed period.
Economic reasons
  • • The modeled loss has an estimated revenue impact.
  • • Investigation cost is not configured.
Recoverable value
$1,513
$897–$2,286
Investigation cost
—
Modeled net value
—
Break-even
Unavailable
Expected persistence
30 days
Modeled range: 14–45 days
Decision confidence
86.0%
Deterministic rule score
Assumptions and uncertainty
  • • Future values are decision-support scenarios, not guaranteed savings.

What should I check next?

Recommended next check

Obtain inverter-level AC output for the affected interval and compare it with available neighboring inverters.

Access
Remote check
Expected information gain
high
Effort
low
Why it matters

This helps distinguish inverter-specific derating from a site-wide irradiance, curtailment, or expected-model issue.

Data needed
  • Timestamped inverter-level AC power
  • Aligned irradiance or expected production

Helps distinguish: Inverter or array derating · Curtailment or export limiting · Soiling, shading, or weather-input bias

Other possible checks

  • Retrieve inverter fault and event codes for the affected interval. (Remote check, low effort)

ArraySignal recommends investigation steps; it does not perform physical tests or confirm field conditions.

Customer provenance, not invented demo history

Real packets use the immutable issue snapshot, source coverage, model and rule versions, and append-only outcome records. Their SHA-256 fingerprint identifies packet content; it is not a digital signature. None is manufactured here.

Identified and anonymized views

Customer anonymized packets omit organization and site names, financial values, and free-text notes. Both views remain authenticated; there is no public sharing token in this demo. Raw source files are not included in a packet.

In a customer workspace, run Site Analysis and open a persisted finding’s evidence-packet link. No PDF is generated or fetched here.

Field outcomes and the learning loop

A detector prediction and a field-confirmed outcome are separate records. Synthetic findings have no customer issue ID, so this preview cannot accept feedback or turn a likely diagnosis into a confirmed cause.

Supported investigation outcomes

These are available choices in the authorized customer workflow, not recorded outcomes for the sample.

  • Confirmed fault
  • Different fault
  • No fault found
  • Data or sensor issue
  • Weather or environmental cause
  • Maintenance performed
  • Monitoring issue
  • Unresolved
Investigation feedback
Record the observed outcome, cause, action, resolution date, and optional recovery in percent or kWh/day. Missing recovery measurements remain unknown, not zero.
Incident resolution
Record the confirmed cause, repair action, optional cost, production-restored status, and resolution time against a persisted incident. The original detector prediction is preserved.
History and interpretation
Investigation feedback is append-only; reports use the latest outcome per issue. A maintenance action or an unresolved result does not by itself confirm the predicted fault.
Analysis usefulness
A customer can also mark an analysis useful, partly useful, or not useful. That rating is not field verification of a physical cause.

0 confirmed field labels in this demo

No diagnostic precision, false-positive rate, measured recovery, or calibration performance is claimed. Authorized owner/admin reporting uses real labels when available; outcome capture does not train or deploy a machine-learning model.

No operator-confirmed resolution has been recorded for this sample. Physical inspection and equipment work require appropriately qualified personnel.

No outcome form, artificial field history, model promotion, or customer write is exposed.

Explore this site’s evidenceInspect the empty diagnostic-performance report

Ready to use your own data? Your customer workspace preserves your existing session and applies your organization role and plan. Uploads, saved changes, verified routes, schedules, and connections require an authorized account.

Open your customer workspace