Model Source Utilities

Utilities for inspecting and recovering delivered model archives live under ras_commander.sources.federal.

Streaming ZIP Recovery

StreamingZipReader walks ZIP local-file headers in order, so it can survey or extract archives whose central directory is missing. It verifies each extracted member against its recorded CRC and reports truncation and unreadable members instead of silently skipping them.

Python

from pathlib import Path

from ras_commander.sources.federal import StreamingZipReader

archive = StreamingZipReader(Path("delivered-model.zip"))
survey = archive.probe()
print(len(survey.members), survey.truncated)

When an archive has neither a central directory nor sizes in its local headers, probe() must scan compressed member data to locate each trailing descriptor. The per-member scan is bounded by max_probe_member_size (64 GiB by default).

Deflate and stored members use the Python standard library. FEMA deliveries compressed with Deflate64 require the optional dependency:

Bash

pip install "ras-commander[ebfe]"

The ebfe extra installs the decoder on Python 3.10, the newest interpreter for which its publisher provides wheels. It is intentionally excluded from the broad all extra so Python 3.11+ installations do not depend on an unqualified native source build.

Deflate64 members require an authoritative compressed size from a local or central-directory record. Deferred-size Deflate64 members in archives with no central directory are reported as unsupported because the dependency does not expose the exact compressed boundary.

This is a direct recovery API. RasEbfeModels does not automatically invoke it. Archive member names are untrusted input. A sink_factory must normalize each name and verify that its output path remains below the intended destination.

Audit Rendering

ebfe_audit turns an existing _audit.json bundle and its JSONL sidecars into a consistent review document covering the delivery verdict, inventory, supporting data, repair actions, missing inputs, acquisition needs, and provenance. It renders captured audit evidence; it does not generate the audit bundle itself.

Python

from pathlib import Path

from ras_commander.sources.federal.ebfe_audit import (
    load_audit_bundle,
    render_audit_markdown,
)

audit_folder = Path("model-audit")
bundle = load_audit_bundle(audit_folder)
(audit_folder / "AUDIT.md").write_text(
    render_audit_markdown(bundle),
    encoding="utf-8",
)