ACP Consultants

ACP scientific tools · Version 1.0 · 17 November 2026

Use the tools.
Keep the evidence.

Instructions, worked examples, sources and limits for both tools.

Before you start

Work from copies. Keep original recordings, detector settings, field notes and review decisions. These browser tools describe supplied data; they do not approve a survey or provide an automated species identification.

Privacy and saving

Inputs stay in this browser. There is no upload, automatic saving, editing telemetry or background account storage. An exported CSV project contains the supplied CSV text and may therefore contain sensitive site references. Share it only with an appropriate recipient.

Compatibility and ownership

Use a current desktop browser for larger files. Phone-width controls are provided, but chart detail may need horizontal scrolling. Preserve ACP credit on original chart templates and keep the recording creator’s credit separate. These tools support your review; they do not replace professional ecological judgement.

T11 / CSV explorer

From records to an auditable rate

  1. Choose the data grain. Each row represents one project/site/station/night/taxon or group, with a consistent pass definition. A recorder deployment, sound file, event, pass and individual bat are different units. Aggregate raw detections through a reviewed protocol before using this explorer.
  2. Open the CSV. Choose comma, semicolon or tab before importing. Headers must be unique. Decimal numbers use a full stop; thousands separators inside numeric values are rejected. Quoted notes may contain commas, quotes and newlines.
  3. Map columns. The six required fields are site, night, planned effort, valid effort, reviewed passes and status. Additional project/station IDs, review state, UTC-offset timestamps, downtime and notes strengthen the audit. Header names need not match if you use the mapping controls.
  4. Declare the units. Select hours or minutes for all three effort columns. Valid effort is usable recorder operation in the counting interval, after downtime. It is not the total duration of triggered WAV files. Do not combine effort measures with different definitions.
  5. Keep clocks explicit. Optional start/end timestamps use 2026-07-01T20:00:00+01:00 or Z. Both are required if either is given. Offsets resolve elapsed time across midnight and daylight-saving transitions. An offset is not an IANA time-zone name; check it against your actual recorder metadata. Unknown -00:00, leap seconds and invalid calendar dates are rejected.
  6. Add an expected schedule if available. Supply site_id,station_id,night, one planned station-night per row. If station is omitted, site_id is used. Use the same site and station identifiers as the observations and a schedule with matching project scope. Missing rows are listed separately, never filled with zeros. With no schedule, the tool cannot know which nights should exist.
  7. Run the audit. Resolve errors in the source, then reopen it. Duplicate keys block rates. Check warnings for unreviewed counts, missing schedule rows and inconsistent night labels. Export the audit to preserve what was found.
  8. Confirm comparability. The declaration confirms the included counts have been reviewed against matching effort and a consistent counting protocol. It is your judgement, not independent software validation. Counts marked unreviewed or uncertain remain excluded from rates.
  9. Choose one group and summarise. Select one taxon/group. Pooled rate divides accepted passes by accepted recorder-hours. Equal-site or equal-station mean first computes each group’s rate and gives each group equal weight. These can differ because their weighting differs.
  10. Save the handover. Export the review JSON, audit, summary and chart. The JSON keeps the original CSV, schedule, mappings, units, declaration and source fingerprint. Reopen it to confirm the result. Keep the original source file as well.

An effort adjustment does not remove differences in equipment, settings, position, weather, habitat or detectability. These descriptive outputs do not fit an occupancy model, estimate abundance, establish absence or measure population trends.

CSV fields and decision rules

FieldMeaning / rule
site_id / station_id / project_idKeep IDs stable. Station defaults to site when absent; project uses your supplied declaration. Multiple recorders require a clear aggregation design.
nightA real YYYY-MM-DD date, normally the local calendar date at the start of the night. The tool warns when a supplied start timestamp uses a different date.
planned_hours / valid_hoursNames follow the existing toolkit. The units control may convert minutes. Planned > 0; 0 ≤ valid ≤ planned; at most 48 hours per station-night.
reviewed_passesA non-negative whole count. Zero means a genuine reviewed zero within positive valid effort. Missing counts stay blank.
statusincluded, excluded or not_recorded. not_recorded requires zero effort and a blank count. Add reasons for exclusions.
taxon / review_statusOne taxon/group per summary. Review states: reviewed, unreviewed or uncertain. Absent review columns use your explicit dataset declaration.
row_idOptional unique record ID. Repeated IDs are reported, even if other fields differ.
start_time / end_timeExplicit-offset timestamps, seconds required, optional milliseconds. Planned effort cannot exceed the elapsed interval. Scheduled gaps need explanation.
downtime_hours / noteOptional downtime must equal planned minus valid effort within one second, using the chosen units. Notes preserve reasons and context.

T12 / WAV workbench

Measure the selected signal reproducibly

  1. Check the source. Open original-rate WAV, not a slowed listening copy or zero-crossing output. The workbench accepts little-endian RIFF PCM16/24/32 and IEEE float32, mono or stereo, 8–768 kHz. Compatible extensible headers are accepted only when valid bits equal container bits. Compressed WAV, RF64, RIFX, more than two channels and ambiguous layouts are rejected.
  2. Retain ownership and evidence. Enter creator, source, permission/licence and transformations. Link project, station, event and recording IDs to your field records. The bundled real excerpt carries its external creator’s attribution and a CC BY 4.0 link.
  3. Choose the channel. Stereo channels are not mixed automatically. Results refer to the selected channel. The file fingerprint covers the complete supplied file, including its header.
  4. Select the interval. Enter source-relative seconds, up to two seconds at a time within a source no longer than five minutes and 64 MB. Bounds are rounded to the nearest sample and the actual bounds are exported. A manual region is not automatically the duration of a call.
  5. Set the frequency band. The upper frequency cannot exceed half the sample rate (Nyquist). The tool does not reconstruct frequencies removed by recording hardware, filtering or conversion.
  6. Choose FFT and hop. A periodic Hann window is applied to complete frames with no padding or detrending. Window duration = N / sample rate; bin spacing = sample rate / N; hop sets the time between frames. Overlap adds sampling positions, not independent calls or extra resolving power.
  7. Inspect all three views. The waveform shows minimum and maximum sample values in display bins. The spectrogram colour is relative to the strongest displayed time-frequency bin. The mean spectrum averages unnormalised squared FFT magnitudes, then references its full-spectrum maximum. These relative scales have different references and are not dB SPL or calibrated PSD.
  8. Read the quality metrics. RMS, mean, absolute peak and rail-sample fraction use all samples in the selection. Integer rails or float values at/outside ±1 flag possible saturation; a rail sample alone is not proof of analogue clipping. Silence has no peak frequency.
  9. Record a selection. Add a label and limitations, then add the current result to the measurement log. Repeat with a different interval if useful. Each entry retains exact settings, actual sample bounds and the source fingerprint.
  10. Export and reopen. Save the JSON, log CSV, metrics, spectrum and chart. The JSON does not contain the audio. Open the identical source WAV before restoring settings; a different file is rejected. Keep audio and JSON together. Check that each export saved successfully.

There is no automatic call boundary detector, species classifier, noise removal, sound-pressure calibration or abundance estimate. Signal shape depends on recording geometry, hardware and context. Inspect a full sequence with suitable expertise before ecological interpretation.

Worked examples with known checks

CSV: the existing teaching dataset

Eight invented sites and 24 site-nights. Accepted: 21 rows, 147 recorder-hours and 464 reviewed passes. Pooled rate: 3.156463 passes/hour. Equal-site mean: 3.112338. Two accepted nights have zero passes; two failed nights have no count. One row is excluded.

The extended example adds synthetic station IDs, offset timestamps and downtime. Its numerical result matches the earlier R/Python toolkit. The quality example deliberately adds duplicate keys, bad effort and an ambiguous clock, and removes one expected row. It should block rates and identify the absent schedule row.

Reviewed CSV · Quality example · Expected schedule

WAV: tone, sweeps and an attributed excerpt

The 256 kHz synthetic tone has a 45 kHz frequency and approximately 0.35355 digital RMS. With N=1024, it lies exactly on a 250 Hz-spaced bin. Four synthetic descending sweeps illustrate why time and frequency views are complementary. Neither is a recording of a bat.

For the real 1.1-second excerpt, N=512, hop=64 and a 38–55 kHz band reproduce the earlier scripts: peak bin 47.5 kHz, RMS 0.006406942, 4,393 complete frames. This is a calculation check, not independent confirmation of the source species label.

Known tone · Synthetic sweeps · Recording credit

Sources, methods and rights

Research refreshed 17 November 2026. The existing ACP scripts and their expected outputs provide a reproducibility reference; the browser implementation is also tested against independent known counts, a known tone and a direct DFT calculation.

External recording attribution

Recording © Uploads from the Somerset Bat Groups events and members. Source observation 243795320; CC BY 4.0. ACP reuses the existing 1.1-second original-rate excerpt, with metadata removed and no denoising or amplitude normalisation. Community label: Pipistrellus pipistrellus, not independently confirmed by ACP. No endorsement is implied. The archived licence check and file hash are retained; the source page was not retrievable in this refresh.

Original ACP chart templates and synthetic examples are distinct from this external recording. Retain source, licence and modification notices with any shared audio-derived chart. Project files can contain your own confidential evidence; the tool does not grant permission to redistribute it.

Review checklist

CSV

  • Can you map your headers without changing the source?
  • Are duplicates, failed nights and missing schedule rows clear?
  • Do counts and denominators match your reviewed protocol?
  • Does reopening JSON restore the same result?

WAV

  • Do the file rate and selected channel match your recorder?
  • Are axes, colour references and measurements readable?
  • Do source attribution and settings survive export?
  • Does a saved record reject a different source?

The supplied examples and desktop and phone-width layouts have been checked. Representative field-data trials and physical phone checks remain separate requirements for your workflow. No server-side R/Python runner is included.