2026-09-06 fixture runs: contract 100 %, executed 100 %, wall ~113 s (2 runs). Derived battery: FAIL only on usage floor (5/10 tool-turns) — answers seeded questions from context, which is ideal grounded behavior. Wall time ~2.8× lite (113 s vs 40 s). Model is clean.
2026-09-06 fixture runs: contract 92–93 %, executed 64–75 %, wall ~40.5 s (2 runs). Derived battery: FAIL, 36 % executed (38.3 s). Same pattern — copy-invariant re-read habit blocks the ≥90 % executed bar under current ALREADY_IN_CONTEXT refusal semantics. Model is working correctly; the bottleneck is the app's dedupe refusal, not the model.
turbo (2026-09-05, same fixture KB): fixture gate PASS 100%/100% on both
metrics, two runs (wall 105-135s vs lite 43-55s); the redundant re-read
of seeded documents that capped lite's executed ratio at 58-73% is
model-specific (turbo re-read rate ~12% vs ~100% in-sample), corroborating
section 7's framing. Locked derived battery: turbo fails only the >=6/10
tool-turn usage floor (it answers seeded read-target questions from
context instead of making the refusable read call) - accuracy on all
emitted calls still 100%/100%.
The phase-72 iteration loop cleared the database, git-cloned the homelab repo, re-imported 38-51 documents and re-embedded per run — many minutes per iteration against a different KB every time (owner directive 2026-09-04: stop importing the homelab repo on every test run). Replace it with:
- tests/fixtures/agent_kb/: 8 hand-written markdown docs (sources 'deployments'/'homelab') whose specifics (rack7, 10.77.42.0/24, VLAN 130, rbm-8842, 17 2 * * *, obsidian-bor:2026.7.14, 18765, 18443, ...) no model can guess; read targets carry non-topical filenames so their questions do not lexically seed them (the read must actually happen)
- tests/fixtures/test_kb.dump.sql: data-only snapshot (TRUNCATE + INSERTs incl. embeddings, self-contained git_sources rows, static KB overview) — verified by round-trip checksum at build time
- scripts/load_test_kb.py: one-off rebuild (real pipeline + embeddings, ~2s) that also prints the per-question retrieval report (all 10 battery questions must be grounded)
- scripts/restore_test_kb.py: sub-second one-transaction restore (no git clone, no re-embedding)
- scripts/agent_realmodel_check.py: the gate gains --restore / --mode fixture (curated 10-question battery with one unambiguously correct tool behavior per question) / --turns N (12s micro-loop) / --concurrency / per-turn + total wall timing, and a second accuracy metric (contract accuracy: well-formed calls targeting resolvable entities) alongside the phase-72 locked executed ratio — the re-read of a seeded doc is a copy-invariant model behavior (5 variants, 0/15 flipped) that the dedupe refusal counts as a failure
- TOOL_CALLING_TESTING.md: the human-readable methodology (fast loop, design rules, metrics, copy levers + tried-and-reverted table, current standing, open design question)
Measured: restore 0.03s; micro-loop ~12s; full loop ~43-55s; concurrency 2/3 gives no gain (endpoint serializes).
ducoterra