fix(orchestrator): KI-012 critical — transition to free-form after fact-find

P002 transcript revealed: after fact-find completed at turn 11, the bot
returned the SAME readback summary ("Got it — here's what I've understood:
…") to all 19 subsequent free-form policy questions. User asked about
claim settlement, room rent caps, comparisons — bot just echoed the
profile readback every time.

Root cause: on fact-find completion, the code only cleared
`session.awaiting_question_id` but did NOT set
`session.free_form_session = True`. Next turn classifier still routed
through the fact-find branch, `next_question()` returned None, code
hit the "complete" path again, emitted readback summary again. Loop.

Fix is one line: set `session.free_form_session = True` + flush. Now
subsequent turns skip the fact-find branch entirely and go through
retrieval + brain as intended.

Effect on the audit data: every persona's post-fact-find ~20 turns
will now go through real RAG + brain instead of silently looping the
readback. The 100-persona audit data will finally surface real
policy-Q&A behaviour at scale.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

backend/orchestrator.py

@@ -242,9 +242,15 @@ async def handle_turn(
             else:
                 brain_tag = "needs_finder::fact_find_continue" if in_fact_find_continuation else "needs_finder::fact_find_start"
         else:
-            # Fact-find complete — produce a profile readback + invite next step
+            # Fact-find complete — produce a profile readback + invite next step.
+            # CRITICAL (KI-012): flip free_form_session=True so subsequent turns
+            # don't re-enter the fact-find branch and repeat the readback. Before
+            # this, every post-fact-find turn returned the same readback summary
+            # because next_question() kept returning None.
             from backend.needs_finder import readback_summary
             session.set_awaiting(None)
+            session.free_form_session = True
+            session._flush()
             summary = readback_summary(session.profile)
             reply = (
                 f"Got it — here's what I've understood: {summary}. Want me to suggest 2-3 policies that fit your profile, "

docs/40-evaluation/known-issues.md

@@ -209,3 +209,48 @@ empirical signal for whether a fix is actually working in production.
 The standing ratio target: **for every 1 user-facing bug a reviewer
 catches, we should close 5 internal issues from this log before the next
 review.**
+
+### KI-011 — Fact-find re-ask infinite loop under load — **FIXED in `171f2a4`**
+
+**Severity:** P0
+**Source:** `backend/orchestrator.py` fact-find branch + `backend/fact_find_normalizer.py` LLM-only path
+**Discovered:** First persona of the 100-persona audit (P002, verbose style)
+
+When NIM rate-limited the Llama-3.3-70B normalizer under audit
+concurrency, the LLM call raised, the orchestrator marked the answer
+ambiguous, kept `awaiting_question_id` set, and re-asked the same
+question on the next turn. User moved on with answers to OTHER
+questions. Normalizer rejected them. Bot re-asked again. Infinite loop.
+
+**Fix:** Keyword fast-path in `fact_find_normalizer.py` (hand-curated
+substring matchers for 9 metros, 15 tier1 cities, dependents
+combinations, income/budget bands, primary goals, common health
+conditions) bypasses the LLM for ~80% of answers. Re-ask cap in
+`orchestrator.py` gives up after 2 failed normalizations on the same
+question and marks it asked. Production audit on persona P002
+post-fix: 30/30 turns completed in 85s with 0 refusals (vs. infinite
+loop before).
+
+### KI-012 — Bot stuck in fact_find_complete readback loop — **FIXED in next commit**
+
+**Severity:** P0
+**Source:** `backend/orchestrator.py` fact-find-complete branch
+**Discovered:** Reviewing audit transcript of P002 post-KI-011-fix
+
+After fact-find completes, the orchestrator only calls
+`session.set_awaiting(None)` — it does NOT flip
+`session.free_form_session = True`. On every subsequent turn, the
+classifier still routes through the fact-find branch (because
+`free_form_session` is false), `next_question()` returns None (all
+fields captured), and the code path emits the readback summary AGAIN
+instead of going to retrieval + brain.
+
+**Effect:** P002 used 19 of 30 turns repeating the same readback
+"Got it — here's what I've understood: …" instead of answering the
+user's real policy questions. Every persona that completes fact-find
+hits this.
+
+**Fix:** When fact-find completes, set
+`session.free_form_session = True` and flush to disk. Subsequent turns
+skip the fact-find branch entirely and go through retrieval + brain
+as intended.