RAG & Knowledge-Base Poisoning

Two different bars. To reach the grounding set at all, the passage only has to beat the current last-placed candidate and clear any absolute score floor. To evict the incumbent docs chunk, it has to push that chunk out of the budget entirely — a much harder win.

Effectively none. From outside, a file discarded at ingest and a file waiting for the next build look identical - the share shows it either way - and no surface reports which happened. The activation date is inferred, never observed.

Very little from inside the index. A planted row and a legitimate one carry self-declared values that pass the same predicate identically, so separation depends on records made outside the documents - and those establish submission events, not correctness.

Overlap duplicates text across adjacent units instead of reconciling it. A paragraph near a boundary is emitted both with its framing sentence and without it, and both units are embedded and separately retrievable, so a coarse aim suffices.

Budget it as dated evidence, not capability. The durable product is the finding that a first stage ranks text nobody read; the tuned passage is a perishable demonstration whose expiry is set by someone else's release schedule.

When the next document costs more than the question it would move is worth. Cheap, thinly covered questions go first; each further one is more crowded, more attributable, and needs upkeep as genuine content keeps arriving.

Editing the row closes one instance, not the class. The only defensible claim is that this sentence is gone; anything broader would require re-verifying every passage that can be retrieved, which nobody has funded or scheduled.

Only a bounded upper limit and a window, labelled as such: the period the planted page was reachable and the number of answers that could have carried the chip. Reach, belief and repetition are unmeasured, and unmeasured is not zero.