Background: PR #568

PR #568 (feat: third-party eval metrics adapter (DeepEval + Autoevals) with strands-evals mappers) introduced the DeepEvalAdapter — a generic wrapper that takes any DeepEval metric and runs it inside an AgentCore Lambda evaluator. The adapter uses strands-evals mappers to auto-detect span formats, extract fields, and construct the appropriate DeepEval test case (LLMTestCase for single-turn, ConversationalTestCase for multi-turn).

The Bug

1. At SESSION level, the AgentCore eval service collapses all spans sharing a session.id into one span with multiple span_events (one per turn):

Session
 └─ Trace 1 → ONE span
                └─ span_events: [ {body: turn1}, {body: turn2}, {body: turn3} ]

2. strands-evals CloudWatchSessionMapper folds that one span into a single AgentInvocationSpan (first user prompt, last assistant response),
3. so _session_to_span_map_result() yields only 2 turn entries
4. the len(turns) > 2 guard nulls it
5. conversational metrics raise "only a single turn was found."

PR #568's _extract_from_service_format() fallback couldn't help: it only parsed single-turn gen_ai.* events, and map_spans() only invokes it when the mapper yields no usable input/output — which never happens here, so any multi-turn logic added to it was unreachable.

Impact: Every multi-turn metric failed at SESSION level because no turns were extracted at all, forcing users to bypass the adapter and write custom conversational handlers.

The Fix

Two parts, in registry.py:

1. Reachability: map_spans() now detects the collapsed multi-event shape via a new _has_multi_event_span() helper (any span with >1 span_events) and runs the service-format extraction to populate turns even when the mapper already produced a valid single-turn input/actual_output.
2. Correct decoding: _extract_from_service_format() gained a multi-turn branch that iterates span_events[*].body, and its _extract_message_text() helper now reuses CloudWatchSessionMapper's text parsing (_extract_text_from_content). This decodes the real double-encoded content (a JSON string like '[{"text": ...}]') into plain text instead of returning it literally, and selects the latest message per role, consistent with the CloudWatch single-turn path (PR fix: preserve multi-turn history in Strands ConversationTurn #454).

Backward-compatible: the single-turn gen_ai.* events path is unchanged; the multi-turn path activates only when >1 span_event is present.

Scope

This PR restores turn extraction (role + content) from the service-normalized SESSION format. That is sufficient for the multi-turn metrics that only need the conversation turns:

• GoalAccuracy, RoleAdherence, ConversationCompleteness, KnowledgeRetention, TopicAdherence, TurnRelevancy, ConversationalGEval

Out of Scope / Known Limitation (follow-up)

Metrics that need per-turn side data are NOT fixed by this PR, because the service-normalized span_events[*].body format has no slot for that data — only input.messages and output.messages:

• retrieval_context → TurnContextualPrecision, TurnContextualRecall, TurnFaithfulness
• tools_called → ToolUse

_build_conversational_test_case() constructs Turn(role, content) only, so even if the body carried these fields, they would be dropped. Fixing this requires a format convention for where per-turn retrieval_context / tools_called live in span_events[*].body (e.g. body.retrieval_context, body.tools), then threading them through SpanMapResult.turns[i] into the Turn(...) constructor. Tracked as a follow-up.

Issue #, if available:

Description of changes:

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.