./skraak import segments --db ./db/skraak.duckdb --dataset abc123 --location loc456 --cluster clust789 --folder /path/to/data --mapping mapping.json

./skraak import segments --db ./db/skraak.duckdb --dataset abc123 --location loc456 --cluster clust789 --folder /path/to/data --mapping mapping.json # requires mapping.json

# Call analysis (extract from ML predictions, review/classify)
./skraak calls from-preds --csv predictions.csv                    # Extract calls, write .data files

# Create .data files 
./skraak calls from-preds --csv predictions.csv                    # Extract calls from OPSO, write .data files

./skraak calls show-images --file recording.wav.data               # Display spectrograms
./skraak calls classify --folder ./data                            # Interactive classification (reviewer + bindings from ~/.skraak/config.json)
./skraak calls classify --folder ./data --filter opensoundscape-kiwi-1.0

# Summarise .data files

# Display spectrograms
./skraak calls show-images --file recording.wav.data               

# TUI for manual classification (reviewer + bindings from ~/.skraak/config.json)
./skraak calls classify --folder ./data                            # Interactive classification 
./skraak calls classify --folder ./data --filter opensoundscape-kiwi-1.0

# Agentic call analysis

# Export OpenSoundScape clip_labels-format CSV from .data files

# .data files to OPSO multihot csv (requires mapping.json)

./skraak time                            # Current time as JSON

# Get current time
./skraak time                            # Current time as JSON

**`isnight`** — Night detection for bioacoustic recordings. Determines if a WAV file was recorded at night (between sunset and sunrise) at the given GPS coordinates. The recording timestamp is read from the WAV file metadata, not from the filename — this works reliably because bioacoustic recorders (AudioMoth, BAR-LT, Song Meter, etc.) embed an accurate timestamp in the WAV header at the time of recording. AudioMoth comments are parsed automatically including the embedded UTC offset. For non-AudioMoth files without a recognized filename pattern, the timestamp falls back to the file modification time. Use `--brief` for batch/agent use to return only `file_path` and `solar_night`.

Secondary bindings for a, eurbla, are accessed by shift-a, a/c/s

**Secondary bindings for a, eurbla, are accessed by shift-a, a/c/s**

6. **Export OpenSoundScape clip_labels-format CSV:**

6. **Export .data files to OpenSoundScape multihot CSV:**

Reproduces OpenSoundScape's `BoxedAnnotations.clip_labels()` output
exactly — same row layout, byte-identical CSVs — but in Go, fast, and
without round-tripping through Raven `selections.txt`.
**Algorithm.** For every `.data` file, generate fixed-duration clip
windows from `[0, Duration]` using OPSO's `generate_clip_times_df`
(supports `--final-clip` of `full | remainder | extend | none`). Every
window is emitted as a row; for each output class column, the value is
`True` when at least one cert-100 annotation of that class overlaps the
window by ≥ `--min-label-overlap` seconds, else `False`. Gaps just emit
all-`False` rows.
Only certainty=100 labels participate. `mapping.json` (from the
`/data-mapping` skill) translates `.data` species strings to canonical
class names. Two sentinels with distinct semantics:

- **`"__NEGATIVE__"`** — clip IS emitted, **all class columns False**.
  Overrides any positive labels in the same clip's union. Use for
  confirmed-negative training examples (e.g. `Noise`, `Not`, rain, wind,
  silence, chainsaw, helicopter).
- **`"__IGNORE__"`** — the segment is dropped from output. Any
  segment whose species maps to `__IGNORE__` triggers the drop, regardless
  of filter. Use for files whose annotation set is incomplete: emitting any 
  clip from them as confirmed-False would poison the training set with 
  possibly-wrong negatives.

- **`"__NEGATIVE__"`** — segmens IS emitted, **all class columns False**.
- **`"__IGNORE__"`** — the segment is not in the dataset. 

Override order within a clip: `__NEGATIVE__` beats real classes. (File-level
`__IGNORE__` is checked before any clip is generated.)

```
{
  "Kiwi":       {"species": "Kiwi"},
  "Geese": {"species": "__NEGATIVE__"},
  "Not":        {"species": "__NEGATIVE__"},
  "Don't Know":        {"species": "__IGNORE__"}
}
```

If `--output` exists, the run **appends**. Column-set mismatch with the
existing header → hard error. Duplicate `(file, start_time, end_time)`
row (within the run, or vs existing rows) → hard error on first
occurrence. Any `.data` parse error, missing `Duration`, or species
missing from `mapping.json` aborts before any row is written.

If `--output` file already exists, the run **appends** to the file. 

## Segments Import

## Import .data files to database

use claude skill to guide user through creation of species calltype mapping to db

use agent skill to create mapping

```bash
# All Makefile
make test
# Lines of code
make count
# Go unit tests
make unit 
# Shell script integration tests (make sure db/test.duckdb available and FK's applied)
make shell 
```