Trial coverage and gaps

Request	Retained result	Limit
Try Flare	Pinned managed native runtime; two 7/7 protocol preflights	Text-only checkpoint; no vision replacement
Compare benchmarks	Four natural cells per arm, 32/32 responses/canaries each	Different checkpoints; variable output; C4 warm; n8/cell
Controlled decoding	Both arms failed exact 128-word C1/C4 gates	No eligible strict performance ranking; Flare 32K strict cells deferred to fit warm restart within budget
Easy future conversion	Recipe, build, checkpoint, package index hash, cache reuse and warm startup	Full weight-blob hashes not independently collected; local image remains unpublished
Restore current service	Exact retained ID/image/recipe, 14 protected hashes, direct/routed7/7, readmission	Scoped restoration, not a new fleet-wide acceptance campaign
Decide on conversion	Keep incumbent; no promotion	Full frozen comparison, cold-C4, MTP0, SWE, independent quality, full-context, cancellation/endurance and clients remain missing

The bounded scout is finished. The full serving comparison is incomplete. No retroactive frozen-plan claim is made for the earlier incumbent strict diagnostic. Runtime allocated KV slots and configured context are not simultaneous full-window capacity.