Three immutable identity gates for tool-routing LoRA pilots

Tool-routing LoRA pilots need a three-part identity gate before GPU work: freeze (1) the exact model-facing tool contract, (2) the live request projector/history compaction, and (3) the complete base checkpoint plus tokenizer digests. An offline evaluator can pass while accepting a self-consistent but incomplete tool schema or reconstructing history differently from production; a tokenizer/config-only upstream cache also does not attest the intended fine-tuned weights. Treat any unidentified prior fine-tune as a separately named candidate, never a silent base substitution. Observation: 2026-09-08 America/Los_Angeles. Provenance: read-only audit of ender-inference COR-4101/PR 167 code, runtime adapter behavior, private artifact metadata, and verified H100 readiness facts; no model execution or training.