MemPalace/mempalace · critical · EmbedderIdentityMismatchError

collection was built with embedder {stored.model_name!r} but

Error message

collection was built with embedder {stored.model_name!r} but the current embedder is {current.model_name!r}. Searching across a model swap silently degrades recall. Re-embed the palace, or run `mempalace palace set-embedder --model <name> --force` to record the new identity if you know the vectors are compatible.

What it means

Raised as TruncationDetected when the on-disk chroma.sqlite3 reports more embedding rows than the chromadb collection layer returned during extraction. This means the segment metadata is stale (classically after a manual HNSW quarantine), and proceeding with a rebuild would silently destroy the difference. It is a load-bearing safety abort that protects against data loss.

Source

Thrown at mempalace/backends/base.py:239

    dim_conflict = bool(stored.dimension and current.dimension) and (
        stored.dimension != current.dimension
    )
    name_conflict = stored.model_name != current.model_name

    if not dim_conflict and not name_conflict:
        return "known_match"

    if force_model_swap:
        return "known_mismatch"

    if dim_conflict:
        raise DimensionMismatchError(
            f"collection was built with a {stored.dimension}-dim embedder "
            f"({stored.model_name!r}) but the current embedder is "
            f"{current.dimension}-dim ({current.model_name!r}); the stored "
            "vectors are incompatible. Re-embed the palace to switch models."
        )
    raise EmbedderIdentityMismatchError(
        f"collection was built with embedder {stored.model_name!r} but the "
        f"current embedder is {current.model_name!r}. Searching across a model "
        "swap silently degrades recall. Re-embed the palace, or run "
        "`mempalace palace set-embedder --model <name> --force` to record the "
        "new identity if you know the vectors are compatible."
    )


@dataclass(frozen=True)
class HealthStatus:
    ok: bool
    detail: str = ""

    @classmethod
    def healthy(cls, detail: str = "") -> "HealthStatus":
        return cls(ok=True, detail=detail)

    @classmethod

View on GitHub (pinned to 06cb6987f0)

Solutions

Restore the palace from your most recent backup, then re-mine — safest path.
Direct-extract rows from chroma.sqlite3 (extract_via_sqlite exists for this) and rebuild the palace from the recovered rows plus source files.
Only if you have independently verified (direct sqlite3 COUNT query) that the palace truly contains only {extracted} drawers, re-run with --confirm-truncation-ok.

Example fix

# before
repair.check_extraction_safety(palace_path, extracted=5, sqlite_count=10_000)
# raises TruncationDetected

# after (only after verifying via sqlite3)
repair.check_extraction_safety(palace_path, extracted=5, sqlite_count=5,
    confirm_truncation_ok=True)

Defensive patterns

Strategy: validation

Validate before calling

from mempalace import repair
count = repair.sqlite_drawer_count(palace_path)
if count is not None:
    extracted = len(list(repair.extract_via_sqlite(palace_path, 'drawers')))
    assert extracted >= count or independently_verified, 'truncation risk'

Try / catch

try:
    repair.check_extraction_safety(palace_path, extracted, confirm_truncation_ok=False)
except repair.TruncationDetected as e:
    # e carries sqlite_count and extracted — decide recovery vs confirm
    if verified_true_count(e.extracted):
        rerun_with_confirm()
    else:
        restore_from_backup_or_sqlite_extract()

Prevention

Never manually quarantine/move HNSW segment files while chromadb is running.
Keep regular palace backups before repairs.
Run repairs with no other process holding the palace open so the SQLite cross-check works.

When it happens

Trigger: check_extraction_safety() in repair.py finds sqlite_count > extracted without confirm_truncation_ok=True. Typical after an HNSW index quarantine or a chromadb version change that leaves segment metadata inconsistent with the embeddings table.

Common situations: Running `mempalace rebuild` / repair after manually moving HNSW segment files, after a chromadb upgrade with segment drift, or against a palace that was copied while chromadb held it open.

Related errors

AI-assisted analysis of MemPalace/mempalace@06cb6987f0 (2026-08-15). Data as JSON: /api/errors/a68daf7858b2c520. Report an issue: GitHub.