HKUDS/DeepTutor · error · ValueError

preferences.md is not audited automatically

Error message

preferences.md is not audited automatically

What it means

The audit mode mirrors the consolidator: the preferences.md L3 document is maintained by hand and is never auto-audited. _run_audit_l3 raises immediately when slot == 'preferences', before touching the file.

Source

Thrown at deeptutor/services/memory/consolidator/modes/audit.py:278

        edits_rejected=edits_rejected,
    )


# ── L3 ──────────────────────────────────────────────────────────────────


async def _run_audit_l3(
    slot: L3Slot,
    *,
    language: str,
    user_label: str,
    budget: int,
    llm_selection: dict | None,
    on_event: OnEvent | None,
    settings,
) -> AuditResult:
    if slot == "preferences":
        raise ValueError("preferences.md is not audited automatically")

    l3_path = paths.l3_file(slot)
    if not l3_path.exists():
        await emit(on_event, {"stage": "done", "no_doc": True})
        return AuditResult(layer="L3", key=slot, chunks_processed=0, no_doc=True)

    doc = load_doc(l3_path, default_title=f"{slot} memory")
    l2_lookup = _build_l2_entry_lookup()
    prompt = load_prompt("audit_l3", language)
    focus, _sections = slot_focus(language, slot)

    annotated_text, _ranges = _build_annotated_l3(doc, l2_lookup)
    chunks = chunk_with_boundary(
        annotated_text,
        budget=budget,
        overlap_ratio=settings.chunking.overlap_ratio,
        min_chunk_chars=settings.chunking.min_chunk_chars,
        max_chunk_chars=settings.chunking.max_chunk_chars,

View on GitHub (pinned to 3e82f13042)

Solutions

  1. Exclude 'preferences' from L3 audit sweeps
  2. Audit preferences.md manually (read and edit the file) if review is needed
  3. Centralize the exclusion in a helper like is_auto_managed(slot) so consolidator and audit stay in sync

Example fix

# before
for slot in l3_slots:
    await run_audit("L3", slot, ...)
# after
for slot in l3_slots:
    if slot != "preferences":
        await run_audit("L3", slot, ...)
Defensive patterns

Strategy: validation

Validate before calling

if slot == "preferences":
    skip("preferences.md is maintained manually")
else:
    await run_audit("L3", slot, ...)

Type guard

def is_auditable(slot: str) -> bool:
    return slot != "preferences"

Try / catch

try:
    await run_audit("L3", slot)
except ValueError as e:
    if "not audited automatically" in str(e):
        continue
    raise

Prevention

When it happens

Trigger: Calling run_audit(layer="L3", key="preferences") (directly or through a batch driver like go()) — the guard fires before the l3_file existence check.

Common situations: Running an audit sweep over every L3 slot and forgetting to exclude preferences; porting a consolidation loop to the audit API without carrying over the same exclusion.

Related errors

AI-assisted analysis of HKUDS/DeepTutor@3e82f13042 (2026-08-27). Data as JSON: /api/errors/adeb58e1a45e04e8. Report an issue: GitHub.