{
 "_doc": "The lab's current claim and limits for one result, copied word for word from the line of its current-claims list named here. A value that is replaced by an object with sha256_of_value and bytes is committed by its hash instead of printed, for the reason it states. Nothing here is typed by hand.",
 "portfolio_id": "inference-limits:counting-wall",
 "title": "A proven demand-policy hit ceiling on a real production trace, with simulated LRU far below it",
 "claim": "On the held-out Mooncake test stream (21,069 references, 16,930 distinct) the hit rate of ANY demand policy is bounded above by 19.645%, machine-checked in Lean for all traces and all demand policies. The bound is attained by simulated Belady only at a 1,693-block capacity, where simulated LRU reaches 5.497%; at the smallest swept capacity of 423 blocks simulated Belady reaches 12.687% and simulated LRU 4.059%. Every policy number is a cold-start simulation over the held-out stream, not a deployed cache.",
 "limits": "A ceiling on DEMAND policies only — prefetching escapes it, and the certificate proves that rather than assuming it. The wall is a property OF THE TRACE, not of any implementation: a different workload has a different wall, and the two traces here differ by 10 points. PROMOTED 2026-08-26 from zero prior ranking: this artifact existed at HEAD since ab032b4 and was cited by NO top-40 entry, while entry 20 (`mooncake-real-trace`) cited the weaker computed-optimum Belady gap on the SAME trace. It does not replace entry 20 (`mooncake-real-trace`); it bounds it. Certificate regenerated BYTE-IDENTICALLY at HEAD on 2026-08-26. A3 2026-09-09: the previous headline spliced two capacities. entropy_ceiling_cert.json traces[0].per_capacity carries four rows; Belady equals the 0.19645 wall only from capacity_blocks 1693 upward, and 0.04059 is the LRU figure at capacity_blocks 423, where Belady is 0.12687 (64.58% of wall). The two numbers were never measured at the same capacity. Nothing here is a deployed LRU: all four policy columns are simulations over the trace's held-out test split.",
 "lab_commit": "21c0c4194d8ab615b585e64efa57c86052eacf81",
 "differs_from_sealed": false,
 "source": {
  "list": "the lab's current claims",
  "line": 67,
  "sha256": "7fe8fc3fcd0f88daef78b7f0d64ad414b8780119d854b9e9ead3e42e03bd8507"
 },
 "result_receipt": "/receipts/inference-limits/results/counting-wall.json",
 "vendored_utc": "2026-09-27T09:21:11Z"
}