Recipe feasibility: glm53-cardillo-adaptive-mtp-dual-sm120-wsl2

Required tokens: 258,192

Candidate	Classification	Estimated policy T_max	Policy VRAM margin	Reason
cardillo-exact-vision-adaptive-mtp5-replayssm10-262k	benchmark-survivor	[272,606, 1,090,609]	[0.160, 4.631] GiB	resource bounds pass; behavioral evidence remains incomplete
tpurtell-text-only-adaptive-mtp5-replayssm10-262k	benchmark-survivor	[366,970, 1,468,128]	[1.210, 6.731] GiB	resource bounds pass; behavioral evidence remains incomplete

Missing evidence

• cardillo-exact-vision-adaptive-mtp5-replayssm10-262k: deterministic_pass_rate, measured_max_stable_context_tokens, no_spec_warm_e2e_seconds, quality_score, reference_quality_score, reference_tasks_per_hour, successful_tasks_per_hour, warm_e2e_seconds
• tpurtell-text-only-adaptive-mtp5-replayssm10-262k: deterministic_pass_rate, measured_max_stable_context_tokens, no_spec_warm_e2e_seconds, quality_score, reference_quality_score, reference_tasks_per_hour, successful_tasks_per_hour, warm_e2e_seconds

Unbounded variables

• tracked_variables.wsl2_runtime_compatibility
• tracked_variables.system_ram_peak_bytes
• candidates[0].metrics.deterministic_pass_rate
• candidates[0].metrics.quality_score
• candidates[0].metrics.reference_quality_score
• candidates[0].metrics.warm_e2e_seconds
• candidates[0].metrics.no_spec_warm_e2e_seconds
• candidates[0].metrics.successful_tasks_per_hour
• candidates[0].metrics.reference_tasks_per_hour

benchmark-survivor and math-qualified do not authorize production promotion.