Feedback Loop¶

The feedback loop lets your application report whether a cached query was correct or incorrect. Medha accumulates these signals per entry and can automatically invalidate entries that exceed an error threshold.

Recording Feedback¶

Call feedback() with the original question and a boolean indicating correctness:

async with Medha("demo", embedder=embedder, settings=settings) as cache:
    await cache.store(
        "How many active users do we have?",
        "SELECT COUNT(*) FROM users WHERE active = true",
    )

    # User confirmed the query was correct
    await cache.feedback("How many active users do we have?", correct=True)

    # User reported the query was wrong
    await cache.feedback("How many active users do we have?", correct=False)

feedback() returns True if the entry was found and updated, False if no entry matched (expired, invalidated, or never stored).

Feedback is resolved by exact normalised-question lookup — the same mechanism as invalidate(). It does not perform a vector search.

Reading Feedback Counters¶

After recording feedback, the counters are visible on the CacheResult returned by search():

hit = await cache.search("How many active users do we have?")
if hit:
    print(hit.feedback_correct)    # number of correct signals
    print(hit.feedback_incorrect)  # number of incorrect signals

They are also stored on the underlying CacheEntry and persist across restarts for durable backends (Qdrant, pgvector, etc.).

Auto-Invalidation on Error Threshold¶

Set feedback_incorrect_threshold in Settings to automatically remove an entry once its incorrect count reaches the limit:

settings = Settings(
    backend_type="qdrant",
    feedback_incorrect_threshold=3,  # invalidate after 3 incorrect signals
)

async with Medha("demo", embedder=embedder, settings=settings) as cache:
    await cache.store("How many orders exist?", "SELECT COUNT(*) FROM orders")

    await cache.feedback("How many orders exist?", correct=False)
    await cache.feedback("How many orders exist?", correct=False)
    await cache.feedback("How many orders exist?", correct=False)  # triggers invalidation

    # Entry is gone — next search returns NO_MATCH
    hit = await cache.search("How many orders exist?")
    print(hit.strategy)  # SearchStrategy.NO_MATCH

Correct feedback never invalidates

Only incorrect signals count toward the threshold. Any number of correct feedbacks leave the entry untouched.

When auto-invalidation fires, both the vector backend entry and the L1 cache entry are removed atomically. Calling feedback() again after invalidation returns False without raising an exception.

Threshold via Environment Variable¶

export MEDHA_FEEDBACK_INCORRECT_THRESHOLD=5

Set to None (the default) to disable auto-invalidation entirely — counters accumulate but no entry is ever removed automatically.

Behaviour Reference¶

Scenario	Return value	Side effect
Entry found, `correct=True`	`True`	`feedback_correct` incremented by 1
Entry found, `correct=False`, below threshold	`True`	`feedback_incorrect` incremented by 1
Entry found, `correct=False`, threshold reached	`True`	Entry invalidated from backend and L1
Entry not found	`False`	No change
Entry already invalidated, called again	`False`	No change

Typical Integration Pattern¶

async def handle_user_correction(question: str, was_correct: bool, cache: Medha) -> None:
    updated = await cache.feedback(question, correct=was_correct)
    if not updated:
        # Entry expired or was never cached — nothing to update
        return
    if not was_correct:
        # Optionally log for audit
        logger.warning("Incorrect cache hit reported for: %s", question[:80])