format-adapter-jsonl

The identity adapter. Emit canonical training example records (@agentic/code/frameworks/training-complete/schemas/example-record.yaml) as one-record-per-line JSONL with schema validation. This is the reference serialization for canonical form and the source format consumed by the Parquet adapter.

When to Use

• Shipping canonical records to tooling that speaks AIWG-native training JSONL
• Acting as the upstream input for format-adapter-parquet
• Producing a lossless intermediate that can be fanned out to any other adapter later
• Diagnostics: dump any record set with zero field loss for inspection

Parameters

• <input-glob> (required) — glob of canonical records
• --output <path> (optional) — default: .aiwg/training/exports/canonical-<timestamp>.jsonl
• --validate-round-trip (optional) — reload output and verify byte-level (or structural) equivalence

Format Spec

One canonical JSON object per line — no transformation, no field dropping:

json
{"id": "550e8400-...", "task_type": "instruction_following", "input": {"system": "...", "user": "..."}, "output": {"assistant": "..."}, "metadata": {"quality_grade": "HIGH", "license": "CC-BY-4.0", "provenance_id": "prov-...", "created_at": "2026-04-15T01:00:00Z"}}

Operation

1. Load canonical records — parse each input file; validate against example-record.yaml.
2. Transform — identity pass; no field mapping.
3. Validate target — re-serialize and confirm every record satisfies example-record.yaml validation rules.
4. Round-trip check (if --validate-round-trip) — reparse output and assert structural equality with input.
5. Write output + log — emit JSONL, no sidecar required, append format-convert event with adapter: jsonl and lossless: true.

Round-Trip Invariants

All fields round-trip. Zero loss. This adapter is the benchmark against which other adapters' invariant preservation is measured.

Sidecar Metadata

Not required — JSONL preserves the full canonical record. A sidecar is emitted only when callers explicitly request per-line indexing (<output>.index.yaml mapping line numbers to record IDs for random access).

Acceptance Criteria

• Every input record produces exactly one output line; no rejections except for schema-invalid inputs.
• --validate-round-trip succeeds with 100% structural equality.
• All round_trip_invariants from example-record.yaml are preserved verbatim.
• format-convert event logged with lossless: true.

References

• @agentic/code/frameworks/training-complete/schemas/example-record.yaml — source-of-truth schema
• ADR-022 D7 — canonical + adapter strategy (this adapter is the canonical serialization)

Delegation

• @agentic/code/addons/semantic-memory/skills/memory-log-append/SKILL.md — logging the format-convert event