Skip to content

Fix presenter narration for outcome reports and restore Buttonify #391

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
witbrock merged 1 commit into from
Aug 18, 2026
Merged

Fix presenter narration for outcome reports and restore Buttonify #391

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
225 changes: 183 additions & 42 deletions src/backend/server/routes/von_routes.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,6 +54,7 @@
from ...services import chat_prompt_queue_service
from ...services import conversation_management_service
from ...services.adaptive_turn_service import (
build_effect_outcome_spoken_fallback,
execute_adaptive_turn,
)
from ...services.conversation_turn_memory_context_service import (
Expand Down Expand Up @@ -7707,6 +7708,51 @@ def _extract_presenter_channels(text: str) -> dict[str, object] | None:
}


_CANONICAL_MODEL_DRAFT_MARKER = "\n\n### Model draft (non-authoritative)\n\n"


def _canonical_report_terminal_status(screen_text: object) -> str | None:
"""Recognise only the controlled lead at the start of a canonical report."""

if not isinstance(screen_text, str):
return None
authoritative_text, _marker, _draft = screen_text.partition(
_CANONICAL_MODEL_DRAFT_MARKER
)
lines = [line.strip() for line in authoritative_text.splitlines()]
nonempty_lines = [line for line in lines if line]
if not nonempty_lines or nonempty_lines[0] != "## Effect outcome report":
return None
lead = nonempty_lines[1] if len(nonempty_lines) > 1 else ""
return {
"This turn completed only partially.": "effect_partially_completed",
"This turn has at least one unresolved effect outcome.": (
"effect_outcome_indeterminate"
),
"This turn did not complete all requested effects.": "effect_failed",
"At least one requested effect was not started.": "effect_not_started",
"The model did not produce a reliable final answer.": "model_error",
"The model did not produce a usable final answer.": "model_non_answer",
}.get(lead)


def _buttonify_response_text(
response_text: Any,
*,
canonical_outcome: bool,
) -> str:
"""Return only authoritative answer text for quick-reply generation."""

if not isinstance(response_text, str):
return ""
if not canonical_outcome:
return response_text
authoritative_text, marker, _draft = response_text.partition(
_CANONICAL_MODEL_DRAFT_MARKER
)
return authoritative_text.strip() if marker else response_text


def _presenter_tag_present(text: str, tag: str) -> bool:
return _extract_tagged_block(text, tag) is not None

Expand Down Expand Up @@ -11753,6 +11799,19 @@ def _progress_update(info: dict[str, Any]) -> None:
getattr(adaptive_turn_result, "response_authority", "model") or "model"
).strip()
canonical_outcome_response = response_authority == "canonical_outcome"
raw_canonical_spoken_text = getattr(
adaptive_turn_result,
"canonical_outcome_spoken_text",
None,
)
canonical_spoken_text = (
raw_canonical_spoken_text.strip()
if isinstance(raw_canonical_spoken_text, str)
and raw_canonical_spoken_text.strip()
else build_effect_outcome_spoken_fallback(
terminal_status=adaptive_terminal_status,
)
)
adaptive_partial_delivery = (
adaptive_terminal_status == "effect_partially_completed"
and isinstance(response_text, str)
Expand Down Expand Up @@ -11823,15 +11882,18 @@ def _progress_update(info: dict[str, Any]) -> None:
)
rag_trace["tool_results_included_in_prompt"] = bool(tool_messages)

presenter_channels = (
{
"screen": response_text,
"spoken": response_text,
"format": "effect_outcome_report_v1",
}
if presenter_mode_requested and canonical_outcome_response
else _extract_presenter_channels(response_text)
)
if canonical_outcome_response:
presenter_channels = (
{
"screen": response_text,
"spoken": canonical_spoken_text,
"format": "effect_outcome_report_v1",
}
if presenter_mode_requested
else None
)
else:
presenter_channels = _extract_presenter_channels(response_text)
current_turn_messages = [{"role": "user", "content": prompt_text}]
if tool_messages:
current_turn_messages.extend(tool_messages)
Expand Down Expand Up @@ -12663,17 +12725,12 @@ def _coerce_spoken_text(text: object) -> str | None:
False if eligibility_denied else None
),
)
presenter_channels = presenter_channels

spoken_backfill_latency_ms = (
time.perf_counter() - spoken_backfill_started_perf
) * 1000.0
if not presenter_mode_requested:
spoken_backfill_status = "skipped"
spoken_backfill_suppression_reason = "presenter_mode_disabled"
elif canonical_outcome_response:
spoken_backfill_status = "skipped"
spoken_backfill_suppression_reason = "canonical_outcome_report"
elif not needs_spoken_backfill:
spoken_backfill_status = "skipped"
spoken_backfill_suppression_reason = "not_required"
Expand Down Expand Up @@ -12980,6 +13037,10 @@ def _coerce_spoken_text(text: object) -> str | None:
buttonify_suppression_reason = None
buttonify_model_attempted = False
buttonify_filtering_boundary: dict[str, Any] | None = None
buttonify_source_text = _buttonify_response_text(
response_text,
canonical_outcome=canonical_outcome_response,
)

if not buttonify_setting_enabled:
buttonify_suppression_reason = "buttonify_disabled"
Expand All @@ -12993,7 +13054,7 @@ def _coerce_spoken_text(text: object) -> str | None:
buttonify_suppression_reason = "buttonify_disabled"
elif not buttonify_allowed:
buttonify_suppression_reason = "buttonify_not_allowed"
elif not isinstance(response_text, str) or not response_text.strip():
elif not buttonify_source_text.strip():
buttonify_suppression_reason = "empty_response"
else:
buttonify_status = "no_op"
Expand All @@ -13014,7 +13075,7 @@ def _coerce_spoken_text(text: object) -> str | None:
BUTTONIFY_PROMPT_IDS,
variables={
"user_message": prompt_text,
"assistant_response": response_text,
"assistant_response": buttonify_source_text,
},
fallback=None,
max_chars=6000,
Expand All @@ -13038,7 +13099,7 @@ def _coerce_spoken_text(text: object) -> str | None:

if buttonify_prompt_available and not buttonify_options:
buttonify_response = None
if _contains_openai_quota_error(response_text):
if _contains_openai_quota_error(buttonify_source_text):
buttonify_suppression_reason = "quota_exhausted"
else:
buttonify_model_attempted = True
Expand Down Expand Up @@ -13148,9 +13209,7 @@ def _coerce_spoken_text(text: object) -> str | None:
input_summary={
"buttonify_enabled": buttonify_enabled,
"buttonify_allowed": buttonify_allowed,
"response_text_chars": (
len(response_text) if isinstance(response_text, str) else 0
),
"response_text_chars": len(buttonify_source_text),
},
output_summary={
"options": list(buttonify_options),
Expand Down Expand Up @@ -14923,6 +14982,8 @@ def history_backfill_spoken():
except Exception:
owner_user_id = user_concept_id

replace_unsafe_canonical_channels = False

# Fetch the stored assistant message and associated previous user prompt.
try:
chat_history_coll = chat_history_service.get_chat_history_collection_service()
Expand Down Expand Up @@ -14951,6 +15012,7 @@ def history_backfill_spoken():
if not isinstance(screen_text, str) or not screen_text.strip():
return jsonify({"error": "Assistant message has no content"}), 400
screen_text = screen_text.strip()
legacy_canonical_status = _canonical_report_terminal_status(screen_text)

existing_debug = entry.get("llm_debug_data")
existing_channels = None
Expand All @@ -14959,35 +15021,47 @@ def history_backfill_spoken():
if not force and isinstance(existing_channels, dict):
spoken_existing = existing_channels.get("spoken")
if isinstance(spoken_existing, str) and spoken_existing.strip():
existing_display_elements = (
existing_debug.get("display_elements")
if isinstance(existing_debug, dict)
else None
existing_format = str(existing_channels.get("format") or "").strip()
unsafe_canonical_channels = bool(
legacy_canonical_status
and (
existing_format != "effect_outcome_report_v1"
or spoken_existing.strip() == screen_text
or "## Effect outcome report" in spoken_existing
)
)
if not isinstance(existing_display_elements, dict):
existing_display_elements = build_turn_display_elements(
response_text=screen_text,
presenter_channels=existing_channels,
if unsafe_canonical_channels:
replace_unsafe_canonical_channels = True
else:
existing_display_elements = (
existing_debug.get("display_elements")
if isinstance(existing_debug, dict)
else None
)
existing_health = None
if isinstance(existing_debug, dict):
existing_health = existing_debug.get("turn_output_health")
if not isinstance(existing_health, dict):
existing_health = build_turn_output_health(
if not isinstance(existing_display_elements, dict):
existing_display_elements = build_turn_display_elements(
response_text=screen_text,
presenter_channels=existing_channels,
)
existing_health = None
if isinstance(existing_debug, dict):
existing_health = existing_debug.get("turn_output_health")
if not isinstance(existing_health, dict):
existing_health = build_turn_output_health(
{
"presenter_channels": existing_channels,
"display_elements": existing_display_elements,
}
)
return jsonify(
{
"status": "already_present",
"presenter_channels": existing_channels,
"display_elements": existing_display_elements,
"turn_output_health": existing_health,
"updated": False,
}
)
return jsonify(
{
"status": "already_present",
"presenter_channels": existing_channels,
"display_elements": existing_display_elements,
"turn_output_health": existing_health,
"updated": False,
}
)

prompt_text = ""
for i in range(history_index - 1, -1, -1):
Expand All @@ -15006,6 +15080,73 @@ def history_backfill_spoken():
except Exception as e:
return jsonify({"error": f"Failed reading history: {e}"}), 500

if legacy_canonical_status:
spoken = build_effect_outcome_spoken_fallback(
terminal_status=legacy_canonical_status,
)
presenter_channels = {
"screen": screen_text,
"spoken": spoken,
"format": "effect_outcome_report_v1",
}
spoken_backfill_reason = "legacy_canonical_outcome_report"
display_elements_contract = build_turn_display_elements(
response_text=screen_text,
presenter_channels=presenter_channels,
spoken_backfill_second_pass_attempted=True,
spoken_backfill_second_pass_reason=spoken_backfill_reason,
)
turn_output_health = build_turn_output_health(
{
"presenter_channels": presenter_channels,
"display_elements": display_elements_contract,
"spoken_backfill_second_pass_attempted": True,
"spoken_backfill_second_pass_reason": spoken_backfill_reason,
}
)
persistence_attempted = owner_user_id == user_concept_id
persistence_suppression_reason = None
if persistence_attempted:
try:
update_result = (
chat_history_service.upsert_presenter_channels_for_history_message(
user_id=owner_user_id,
session_id=target_session_id,
history_index=history_index,
presenter_channels=presenter_channels,
display_elements=display_elements_contract,
turn_output_health=turn_output_health,
generated_at=datetime.now(timezone.utc),
force=force or replace_unsafe_canonical_channels,
)
)
except Exception as exc:
return (
jsonify({"error": f"Failed persisting safe talk track: {exc}"}),
500,
)
else:
# An accepted invite permits the viewer to read and present the owner's
# conversation. It does not grant authority to mutate the owner's
# history carrier merely to cache a derived talk track.
update_result = {"updated": False, "matched": True}
persistence_suppression_reason = (
"shared_viewer_owner_history_write_not_authorised"
)
return jsonify(
{
"status": "ok",
"source": "deterministic_legacy_canonical_fallback",
"presenter_channels": presenter_channels,
"display_elements": display_elements_contract,
"turn_output_health": turn_output_health,
"updated": bool(update_result.get("updated")),
"matched": bool(update_result.get("matched")),
"persistence_attempted": persistence_attempted,
"persistence_suppression_reason": persistence_suppression_reason,
}
)

# Load narration prompt fragments (best-effort).
narration_prompt_text = None
try:
Expand Down
Loading
Loading