Phase 22.E pre-registration: testing the Linear A word-final -o gap against the range spanned by previously unmeasured Knossos series.

Pre-registration for Phase 22.E of a pre-registered computational study of Linear A. Deposited before the data were acquired; the OpenTimestamps anchor of this document precedes the anchor of the corpus it analyses, and the interval between the two Bitcoin blocks is reported as evidence of blindness. Question. Linear A words end in -o at 8.7% (tokens, strict layer) against 49.6% for Mycenaean Greek in Linear B. Phase 22.B established that this gap is not an artefact of word length, but its within-Greek yardstick failed: Knossos and Pylos differ from each other by up to 30 points. Phase 22.D attempted a range test on a Thebes/Mycenae/Tiryns hold-out and returned TEST-VOID for insufficient independent units. The gap has been suspended as evidence since 22.B and remains suspended. Design. The hold-out here is Knossos itself. The Knossos corpus used in all previous phases covers only series D, F and E; every other Knossian series has never been measured in this project. The test asks whether Linear A falls outside the prediction interval spanned by those unmeasured series, stratified by word length, with a power gate evaluated before any rate is computed and a validity control that voids the test if any Greek sub-corpus falls below 15%. Pre-declared outcomes include permanent retraction of the gap as a diagnostic. Thresholds are carried over unchanged from Phases 22.C and 22.D. Terminology. What this project has tested and rejected is the Mycenaean Greek hypothesis. Proto-Greek and the pre-Greek substrate were never tested and are untestable for lack of a corpus. The script of Linear A is demonstrably related to Linear B; what is rejected is the language identification. AI use. The pre-registration, the harvester and the runner were drafted with an AI assistant (Claude, Anthropic). Threshold locking, execution and verdict-taking are the author's.