NLU Pattern Matching (First Tier) Not Running — Skipped Entirely in Favor of LLMIssue on docs

[jazzz2000]

Path: /agent-platform/abl/reference/nlu

Platform: Kore.ai Agent Platform — Artemis
Affected tenants:
• 019f8e9d-08fa-76da-8d4c-2f9dfc355962 (personal workspace)
• 019e8442-8d44-7b0c-adcc-4077560b09c5 (company workspace)

Issue:
The NLU documentation states that PATTERNS defined under the NLU: intents: block are used as the "first tier of classification" via fast substring matching, before falling back to model-based classification.

However, in practice, the runtime skips pattern matching entirely. Debug traces show:

• Event: extraction_ml_capability_resolved
• sidecarUsable: false
• disabledReason: tenant_not_enabled
• tenantMlEnabled: false

All intent classification is handled by the LLM via tool selection (deterministic_intent_routing_deferred_to_reasoning). The NLU PATTERNS and EXAMPLES are never evaluated.

Tested across two separate tenants with the same result.

Expected behavior (per docs):
PATTERNS → fast substring match → candidate intents → then LLM if needed.

Actual behavior:
PATTERNS ignored → all classification deferred to LLM coordination call → LLM picks handoff tool → platform does guard check.

Doc reference: https://docs.kore.ai/agent-platform/abl-reference/nlu
Quote: "Pattern matching is the first tier of classification; it is fast but imprecise."

Questions:

1. Is the NLU ML sidecar supposed to be enabled by default for Artemis tenants?
2. If not, what is the process to enable it?
3. If this is expected behavior, should the documentation be updated to reflect that pattern matching is tenant-gated?

I can provide the full debug trace JSON f