Skip to content

schemacast.output

A cast on disk.

File Holds Same bytes on replay
mentions.jsonl one graflo RecordGraph per episode: what each text names yes
graph.json the schema and what the records' keys identify (graflo GraFloOutput) yes
staged.jsonl records not identified, with why, and the relations they hold yes
groundings.jsonl one grounding per record, relation and property yes
episodes.jsonl one result per episode yes
quarantine.jsonl what was left out, and unmapped items yes
manifest.json versions, settings, counts and usage no: usage differs
write.json what was sent to the database, when one was given —

A pick (save_picks) writes picks.jsonl (one pick per staged record), episodes.jsonl and manifest.json to its own directory.

With a database, staged.jsonl holds what is still staged after the lookups there. manifest.json names the output_format_version. Version 6: a cast is a record graph, quotes and spans in evidence; unmapped is outside it. A record of a known entity (ref k1…) has evidence {"known": role, "label": label} and no quote or span, and no grounding.

Attributes

QUARANTINE_FILE = 'quarantine.jsonl' module-attribute

STAGED_FILE = 'staged.jsonl' module-attribute

Classes

Functions:

save_cast(result, out_dir, *, write_report=None, resolution=None)

Write result under out_dir (created if needed); returns the directory.

resolution, from a database write, replaces the offline one in staged.jsonl.

Source code in schemacast/output.py 
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
def save_cast(
    result: CastResult,
    out_dir: str | Path,
    *,
    write_report: BaseModel | None = None,
    resolution: Resolution | None = None,
) -> Path:
    """Write *result* under *out_dir* (created if needed); returns the directory.

    *resolution*, from a database write, replaces the offline one in
    ``staged.jsonl``.
    """
    out = Path(out_dir)
    out.mkdir(parents=True, exist_ok=True)
    (out / "mentions.jsonl").write_text(_jsonl(result.graphs), encoding="utf-8")
    (out / "graph.json").write_text(
        _json(result.output.to_dict()) + "\n", encoding="utf-8"
    )
    (out / STAGED_FILE).write_text(
        staged_lines(resolution or result.resolution), encoding="utf-8"
    )
    (out / "groundings.jsonl").write_text(_jsonl(result.groundings), encoding="utf-8")
    (out / "episodes.jsonl").write_text(_jsonl(result.episodes), encoding="utf-8")
    (out / "manifest.json").write_text(
        _json(result.manifest.model_dump(mode="json")) + "\n", encoding="utf-8"
    )
    if write_report is not None:
        (out / "write.json").write_text(
            _json(write_report.model_dump(mode="json")) + "\n", encoding="utf-8"
        )
    return out

save_picks(result, out_dir)

Write a pick run under out_dir (created if needed); returns the directory.

Source code in schemacast/output.py 
 98
 99
100
101
102
103
104
105
106
107
def save_picks(result: PickResult, out_dir: str | Path) -> Path:
    """Write a pick run under *out_dir* (created if needed); returns the directory."""
    out = Path(out_dir)
    out.mkdir(parents=True, exist_ok=True)
    (out / "picks.jsonl").write_text(_jsonl(result.picks), encoding="utf-8")
    (out / "episodes.jsonl").write_text(_jsonl(result.episodes), encoding="utf-8")
    (out / "manifest.json").write_text(
        _json(result.manifest.model_dump(mode="json")) + "\n", encoding="utf-8"
    )
    return out

staged_lines(resolution)

Staged records (a record key), then held relations (a relation key), one JSON object per line.

Source code in schemacast/output.py 
58
59
60
61
62
def staged_lines(resolution: Resolution) -> str:
    """Staged records (a ``record`` key), then held relations (a ``relation``
    key), one JSON object per line."""
    staged = resolution.staged
    return _jsonl([*staged.records, *staged.relations])