Measure retrieval success as a function of context position, distractor count, corpus, model, prompt, and evaluation protocol; no universal model-independent floor is asserted.
Specification version 1Published 2026-08-271 evidence records
01
What we know
Current frontiers derived from accepted Claims.
BEST KNOWN LOWER BOUNDPROVEN UPPER BOUND
protocol-specific measured retrieval accuracy
UNKNOWN GAP
?
Best known lower boundGap: UnknownProven upper bound
02
Claims
Assertions tied to evidence, attribution, and review.
2026
CLM-LR-DRAFT-RETRIEVAL-1>= protocol-specific measured retrieval accuracyCOMPUTATIONAL BOUNDEmpirical retrieval frontier under a fixed evaluation protocolEvidence: 1 linked source records
SOURCE CONFIRMED
03
Timeline
The frontier as it changed over time.
How this frontier is derived
Only accepted Claims matching the current specification contribute to the displayed bounds. Strict inequalities remain open; contradictory Claims require editorial review.
—No breakthrough events recorded yetEvents appear here when an accepted Claim tightens a bound or closes the frontier.
Limits RegistryPUBLIC RECORD