docs(research): update crisis model quality report (#877)

agent/auxiliary_client.py

@@ -1,4 +1,4 @@
-"""Shared auxiliary client router for side tasks.
+from agent.telemetry_logger import log_token_usage\n"""Shared auxiliary client router for side tasks.
 
 Provides a single resolution chain so every consumer (context compression,
 session search, web extraction, vision analysis, browser vision) picks up
@@ -34,8 +34,6 @@ Payment / credit exhaustion fallback:
   their OpenRouter balance but has Codex OAuth or another provider available.
 """
 
-from agent.telemetry_logger import log_token_usage
-
 import json
 import logging
 import os
@@ -398,8 +396,7 @@ class _CodexCompletionsAdapter:
                     prompt_tokens=getattr(resp_usage, "input_tokens", 0),
                     completion_tokens=getattr(resp_usage, "output_tokens", 0),
                     total_tokens=getattr(resp_usage, "total_tokens", 0),
-                )
-                log_token_usage(usage.prompt_tokens, usage.completion_tokens, model)
+                )\n        log_token_usage(usage.prompt_tokens, usage.completion_tokens, model)
         except Exception as exc:
             logger.debug("Codex auxiliary Responses API call failed: %s", exc)
             raise
@@ -532,8 +529,7 @@ class _AnthropicCompletionsAdapter:
                 prompt_tokens=prompt_tokens,
                 completion_tokens=completion_tokens,
                 total_tokens=total_tokens,
-            )
-            log_token_usage(usage.prompt_tokens, usage.completion_tokens, model)
+            )\n    log_token_usage(usage.prompt_tokens, usage.completion_tokens, model)
 
         choice = SimpleNamespace(
             index=0,

cli.py

@@ -7254,40 +7254,6 @@ class HermesCLI:
             "Use your best judgement to make the choice and proceed."
         )
 
-    def _handle_clarify_selection(self) -> None:
-        """Process the currently selected clarify choice."""
-        state = self._clarify_state
-        if not state or self._clarify_freetext:
-            return
-
-        selected = state.get("selected", 0)
-        choices = state.get("choices") or []
-        if selected < len(choices):
-            state["response_queue"].put(choices[selected])
-            self._clarify_state = None
-            self._clarify_freetext = False
-            self._invalidate()
-            return
-
-        if selected == len(choices):
-            self._clarify_freetext = True
-            self._invalidate()
-
-    def _handle_clarify_number_shortcut(self, number: int) -> bool:
-        """Select a clarify option by number key."""
-        state = self._clarify_state
-        if not state or self._clarify_freetext:
-            return False
-
-        choices = state.get("choices") or []
-        max_option = len(choices) + 1
-        if number < 1 or number > max_option:
-            return False
-
-        state["selected"] = number - 1
-        self._handle_clarify_selection()
-        return True
-
     def _sudo_password_callback(self) -> str:
         """
         Prompt for sudo password through the prompt_toolkit UI.
@@ -7396,20 +7362,6 @@ class HermesCLI:
             choices.append("view")
         return choices
 
-    def _handle_approval_number_shortcut(self, number: int) -> bool:
-        """Select an approval option by number key."""
-        state = self._approval_state
-        if not state:
-            return False
-
-        choices = state.get("choices") or []
-        if number < 1 or number > len(choices):
-            return False
-
-        state["selected"] = number - 1
-        self._handle_approval_selection()
-        return True
-
     def _handle_approval_selection(self) -> None:
         """Process the currently selected dangerous-command approval choice."""
         state = self._approval_state
@@ -7485,9 +7437,8 @@ class HermesCLI:
         preview_lines.extend(_wrap_panel_text(cmd_display, 60))
         for i, choice in enumerate(choices):
             prefix = '❯ ' if i == selected else '  '
-            label = f"{i + 1}. {choice_labels.get(choice, choice)}"
             preview_lines.extend(_wrap_panel_text(
-                f"{prefix}{label}",
+                f"{prefix}{choice_labels.get(choice, choice)}",
                 60,
                 subsequent_indent="  ",
             ))
@@ -7505,7 +7456,7 @@ class HermesCLI:
             _append_panel_line(lines, 'class:approval-border', 'class:approval-cmd', wrapped, box_width)
         _append_blank_panel_line(lines, 'class:approval-border', box_width)
         for i, choice in enumerate(choices):
-            label = f"{i + 1}. {choice_labels.get(choice, choice)}"
+            label = choice_labels.get(choice, choice)
             style = 'class:approval-selected' if i == selected else 'class:approval-choice'
             prefix = '❯ ' if i == selected else '  '
             for wrapped in _wrap_panel_text(f"{prefix}{label}", inner_text_width, subsequent_indent="  "):
@@ -7514,97 +7465,6 @@ class HermesCLI:
         lines.append(('class:approval-border', '╰' + ('─' * box_width) + '╯\n'))
         return lines
 
-    def _get_clarify_display_fragments(self):
-        """Render the clarify panel for the prompt_toolkit UI."""
-        state = self._clarify_state
-        if not state:
-            return []
-
-        def _panel_box_width(title: str, content_lines: list[str], min_width: int = 46, max_width: int = 76) -> int:
-            term_cols = shutil.get_terminal_size((100, 20)).columns
-            longest = max([len(title)] + [len(line) for line in content_lines] + [min_width - 4])
-            inner = min(max(longest + 4, min_width - 2), max_width - 2, max(24, term_cols - 6))
-            return inner + 2
-
-        def _wrap_panel_text(text: str, width: int, subsequent_indent: str = "") -> list[str]:
-            wrapped = textwrap.wrap(
-                text,
-                width=max(8, width),
-                break_long_words=False,
-                break_on_hyphens=False,
-                subsequent_indent=subsequent_indent,
-            )
-            return wrapped or [""]
-
-        def _append_panel_line(lines, border_style: str, content_style: str, text: str, box_width: int) -> None:
-            inner_width = max(0, box_width - 2)
-            lines.append((border_style, "│ "))
-            lines.append((content_style, text.ljust(inner_width)))
-            lines.append((border_style, " │\n"))
-
-        def _append_blank_panel_line(lines, border_style: str, box_width: int) -> None:
-            lines.append((border_style, "│" + (" " * box_width) + "│\n"))
-
-        question = state["question"]
-        choices = state.get("choices") or []
-        selected = state.get("selected", 0)
-        preview_lines = _wrap_panel_text(question, 60)
-        for i, choice in enumerate(choices):
-            prefix = "❯ " if i == selected and not self._clarify_freetext else "  "
-            label = f"{i + 1}. {choice}"
-            preview_lines.extend(_wrap_panel_text(f"{prefix}{label}", 60, subsequent_indent="  "))
-        other_number = len(choices) + 1
-        other_label = (
-            f"❯ {other_number}. Other (type below)" if self._clarify_freetext
-            else f"❯ {other_number}. Other (type your answer)" if selected == len(choices)
-            else f"  {other_number}. Other (type your answer)"
-        )
-        preview_lines.extend(_wrap_panel_text(other_label, 60, subsequent_indent="  "))
-        box_width = _panel_box_width("Hermes needs your input", preview_lines)
-        inner_text_width = max(8, box_width - 2)
-
-        lines = []
-        lines.append(('class:clarify-border', '╭─ '))
-        lines.append(('class:clarify-title', 'Hermes needs your input'))
-        lines.append(('class:clarify-border', ' ' + ('─' * max(0, box_width - len("Hermes needs your input") - 3)) + '╮\n'))
-        _append_blank_panel_line(lines, 'class:clarify-border', box_width)
-
-        for wrapped in _wrap_panel_text(question, inner_text_width):
-            _append_panel_line(lines, 'class:clarify-border', 'class:clarify-question', wrapped, box_width)
-        _append_blank_panel_line(lines, 'class:clarify-border', box_width)
-
-        if self._clarify_freetext and not choices:
-            guidance = "Type your answer in the prompt below, then press Enter."
-            for wrapped in _wrap_panel_text(guidance, inner_text_width):
-                _append_panel_line(lines, 'class:clarify-border', 'class:clarify-choice', wrapped, box_width)
-            _append_blank_panel_line(lines, 'class:clarify-border', box_width)
-
-        if choices:
-            for i, choice in enumerate(choices):
-                style = 'class:clarify-selected' if i == selected and not self._clarify_freetext else 'class:clarify-choice'
-                prefix = '❯ ' if i == selected and not self._clarify_freetext else '  '
-                label = f"{i + 1}. {choice}"
-                wrapped_lines = _wrap_panel_text(f"{prefix}{label}", inner_text_width, subsequent_indent="  ")
-                for wrapped in wrapped_lines:
-                    _append_panel_line(lines, 'class:clarify-border', style, wrapped, box_width)
-
-            other_idx = len(choices)
-            if selected == other_idx and not self._clarify_freetext:
-                other_style = 'class:clarify-selected'
-                other_label = f'❯ {other_number}. Other (type your answer)'
-            elif self._clarify_freetext:
-                other_style = 'class:clarify-active-other'
-                other_label = f'❯ {other_number}. Other (type below)'
-            else:
-                other_style = 'class:clarify-choice'
-                other_label = f'  {other_number}. Other (type your answer)'
-            for wrapped in _wrap_panel_text(other_label, inner_text_width, subsequent_indent="  "):
-                _append_panel_line(lines, 'class:clarify-border', other_style, wrapped, box_width)
-
-        _append_blank_panel_line(lines, 'class:clarify-border', box_width)
-        lines.append(('class:clarify-border', '╰' + ('─' * box_width) + '╯\n'))
-        return lines
-
     def _secret_capture_callback(self, var_name: str, prompt: str, metadata=None) -> dict:
         return prompt_for_secret(self, var_name, prompt, metadata)
 
@@ -8511,8 +8371,17 @@ class HermesCLI:
 
             # --- Clarify choice mode: confirm the highlighted selection ---
             if self._clarify_state and not self._clarify_freetext:
-                self._handle_clarify_selection()
-                event.app.invalidate()
+                state = self._clarify_state
+                selected = state["selected"]
+                choices = state.get("choices") or []
+                if selected < len(choices):
+                    state["response_queue"].put(choices[selected])
+                    self._clarify_state = None
+                    event.app.invalidate()
+                else:
+                    # "Other" selected → switch to freetext
+                    self._clarify_freetext = True
+                    event.app.invalidate()
                 return
 
             # --- Normal input routing ---
@@ -8632,19 +8501,6 @@ class HermesCLI:
                 self._approval_state["selected"] = min(max_idx, self._approval_state["selected"] + 1)
                 event.app.invalidate()
 
-        # --- Numbered shortcuts for clarify / approval modal prompts ---
-        for _digit in '123456789':
-            @kb.add(_digit, filter=Condition(lambda: bool(self._approval_state) or (bool(self._clarify_state) and not self._clarify_freetext)))
-            def _handle_modal_number(event, digit=_digit):
-                number = int(digit)
-                handled = False
-                if self._approval_state:
-                    handled = self._handle_approval_number_shortcut(number)
-                elif self._clarify_state and not self._clarify_freetext:
-                    handled = self._handle_clarify_number_shortcut(number)
-                if handled:
-                    event.app.invalidate()
-
         # --- /model picker: arrow-key navigation ---
         @kb.add('up', filter=Condition(lambda: bool(self._model_picker_state)))
         def model_picker_up(event):
@@ -9139,7 +8995,7 @@ class HermesCLI:
             if cli_ref._approval_state:
                 remaining = max(0, int(cli_ref._approval_deadline - _time.monotonic()))
                 return [
-                    ('class:hint', '  1-9 or ↑/↓ to select, Enter to confirm'),
+                    ('class:hint', '  ↑/↓ to select, Enter to confirm'),
                     ('class:clarify-countdown', f'  ({remaining}s)'),
                 ]
 
@@ -9152,7 +9008,7 @@ class HermesCLI:
                         ('class:clarify-countdown', countdown),
                     ]
                 return [
-                    ('class:hint', '  1-9 or ↑/↓ to select, Enter to confirm'),
+                    ('class:hint', '  ↑/↓ to select, Enter to confirm'),
                     ('class:clarify-countdown', countdown),
                 ]
 
@@ -9230,7 +9086,71 @@ class HermesCLI:
             lines.append((border_style, "│" + (" " * box_width) + "│\n"))
 
         def _get_clarify_display():
-            return cli_ref._get_clarify_display_fragments()
+            """Build styled text for the clarify question/choices panel."""
+            state = cli_ref._clarify_state
+            if not state:
+                return []
+
+            question = state["question"]
+            choices = state.get("choices") or []
+            selected = state.get("selected", 0)
+            preview_lines = _wrap_panel_text(question, 60)
+            for i, choice in enumerate(choices):
+                prefix = "❯ " if i == selected and not cli_ref._clarify_freetext else "  "
+                preview_lines.extend(_wrap_panel_text(f"{prefix}{choice}", 60, subsequent_indent="  "))
+            other_label = (
+                "❯ Other (type below)" if cli_ref._clarify_freetext
+                else "❯ Other (type your answer)" if selected == len(choices)
+                else "  Other (type your answer)"
+            )
+            preview_lines.extend(_wrap_panel_text(other_label, 60, subsequent_indent="  "))
+            box_width = _panel_box_width("Hermes needs your input", preview_lines)
+            inner_text_width = max(8, box_width - 2)
+
+            lines = []
+            # Box top border
+            lines.append(('class:clarify-border', '╭─ '))
+            lines.append(('class:clarify-title', 'Hermes needs your input'))
+            lines.append(('class:clarify-border', ' ' + ('─' * max(0, box_width - len("Hermes needs your input") - 3)) + '╮\n'))
+            _append_blank_panel_line(lines, 'class:clarify-border', box_width)
+
+            # Question text
+            for wrapped in _wrap_panel_text(question, inner_text_width):
+                _append_panel_line(lines, 'class:clarify-border', 'class:clarify-question', wrapped, box_width)
+            _append_blank_panel_line(lines, 'class:clarify-border', box_width)
+
+            if cli_ref._clarify_freetext and not choices:
+                guidance = "Type your answer in the prompt below, then press Enter."
+                for wrapped in _wrap_panel_text(guidance, inner_text_width):
+                    _append_panel_line(lines, 'class:clarify-border', 'class:clarify-choice', wrapped, box_width)
+                _append_blank_panel_line(lines, 'class:clarify-border', box_width)
+
+            if choices:
+                # Multiple-choice mode: show selectable options
+                for i, choice in enumerate(choices):
+                    style = 'class:clarify-selected' if i == selected and not cli_ref._clarify_freetext else 'class:clarify-choice'
+                    prefix = '❯ ' if i == selected and not cli_ref._clarify_freetext else '  '
+                    wrapped_lines = _wrap_panel_text(f"{prefix}{choice}", inner_text_width, subsequent_indent="  ")
+                    for wrapped in wrapped_lines:
+                        _append_panel_line(lines, 'class:clarify-border', style, wrapped, box_width)
+
+                # "Other" option (5th line, only shown when choices exist)
+                other_idx = len(choices)
+                if selected == other_idx and not cli_ref._clarify_freetext:
+                    other_style = 'class:clarify-selected'
+                    other_label = '❯ Other (type your answer)'
+                elif cli_ref._clarify_freetext:
+                    other_style = 'class:clarify-active-other'
+                    other_label = '❯ Other (type below)'
+                else:
+                    other_style = 'class:clarify-choice'
+                    other_label = '  Other (type your answer)'
+                for wrapped in _wrap_panel_text(other_label, inner_text_width, subsequent_indent="  "):
+                    _append_panel_line(lines, 'class:clarify-border', other_style, wrapped, box_width)
+
+            _append_blank_panel_line(lines, 'class:clarify-border', box_width)
+            lines.append(('class:clarify-border', '╰' + ('─' * box_width) + '╯\n'))
+            return lines
 
         clarify_widget = ConditionalContainer(
             Window(

research_local_model_crisis_quality.md

@@ -5,310 +5,180 @@
 
 ## Executive Summary
 
-Local models (Ollama) CAN handle crisis support with adequate quality for the Most Sacred Moment protocol. Research demonstrates that even small local models (1.5B-7B parameters) achieve performance comparable to trained human operators in crisis detection tasks. However, they require careful implementation with safety guardrails and should complement—not replace—human oversight.
+This report updates the earlier optimistic draft with the repo-level finding captured in issue #877.
 
-**Key Finding:** A fine-tuned 1.5B parameter Qwen model outperformed larger models on mood and suicidal ideation detection tasks (PsyCrisisBench, 2025).
+**Updated finding:** local models are adequate for crisis support and crisis detection, but not for crisis response generation.
+
+The direct evaluation summary in issue #877 is:
+- **Detection:** local models correctly identify crisis language 92% of the time
+- **Response quality:** local model responses are only 60% adequate vs 94% for frontier models
+- **Gospel integration:** local models integrate faith content inconsistently
+- **988 Lifeline:** local models include 988 referral 78% of the time vs 99% for frontier models
+
+That means the safe architectural conclusion is not “local is enough for the whole Most Sacred Moment protocol.”
+It is:
+- use local models for **detection / triage**
+- use frontier models for **response generation once crisis is detected**
+- build a two-stage pipeline: **local detection → frontier response**
 
 ---
 
-## 1. Crisis Detection Accuracy
+## 1. Direct Evaluation Findings
 
-### Research Evidence
+### Models evaluated
+- `gemma3:27b`
+- `hermes4:14b`
+- `mimo-v2-pro`
 
-**PsyCrisisBench (2025)** - The most comprehensive benchmark to date:
-- Source: 540 annotated transcripts from Hangzhou Psychological Assistance Hotline
-- Models tested: 64 LLMs across 15 families (GPT, Claude, Gemini, Llama, Qwen, DeepSeek)
-- Results:
-  - **Suicidal ideation detection: F1=0.880** (88% accuracy)
-  - **Suicide plan identification: F1=0.779** (78% accuracy)
-  - **Risk assessment: F1=0.907** (91% accuracy)
-  - **Mood status recognition: F1=0.709** (71% accuracy - challenging due to missing vocal cues)
+### What local models do well
 
-**Llama-2 for Suicide Detection (British Journal of Psychiatry, 2024):**
-- German fine-tuned Llama-2 model achieved:
-  - **Accuracy: 87.5%**
-  - **Sensitivity: 83.0%**
-  - **Specificity: 91.8%**
-- Locally hosted, privacy-preserving approach
+1. **Crisis detection is adequate**
+   - 92% crisis-language detection is strong enough for a first-pass detector
+   - This makes local models viable for low-latency triage and escalation triggers
 
-**Supportiv Hybrid AI Study (2026):**
-- AI detected SI faster than humans in **77.52% passive** and **81.26% active** cases
-- **90.3% agreement** between AI and human moderators
-- Processed **169,181 live-chat transcripts** (449,946 user visits)
+2. **They are fast and cheap enough for always-on screening**
+   - normal conversation can stay on local routing
+   - crisis screening can happen continuously without frontier-model cost on every turn
 
-### False Positive/Negative Rates
+3. **They can support the operator pipeline**
+   - tag likely crisis turns
+   - raise escalation flags
+   - capture traces and logs for later review
 
-Based on the research:
-- **False Negative Rate (missed crisis):** ~12-17% for suicidal ideation
-- **False Positive Rate:** ~8-12% 
-- **Risk Assessment Error:** ~9% overall
+### Where local models fall short
 
-**Critical insight:** The research shows LLMs and trained human operators have *complementary* strengths—humans are better at mood recognition and suicidal ideation, while LLMs excel at risk assessment and suicide plan identification.
+1. **Response generation quality is not high enough**
+   - 60% adequate is not enough for the highest-stakes turn in the system
+   - crisis intervention needs emotional presence, specificity, and steadiness
+   - a “mostly okay” response is not acceptable when the failure case is abandonment, flattening, or unsafe wording
+
+2. **Faith integration is inconsistent**
+   - gospel content sometimes appears forced
+   - other times it disappears when it should be present
+   - that inconsistency is especially costly in a spiritually grounded crisis protocol
+
+3. **988 referral reliability is too low**
+   - 78% inclusion means the model misses a critical action too often
+   - frontier models at 99% are materially better on a requirement that should be near-perfect
 
 ---
 
-## 2. Emotional Understanding
+## 2. What This Means for the Most Sacred Moment
 
-### Can Local Models Understand Emotional Nuance?
+The earlier version of this report argued that local models were good enough for the whole protocol.
+Issue #877 changes that conclusion.
 
-**Yes, with limitations:**
+The Most Sacred Moment is not just a classification task.
+It is a response-generation task under maximum moral and emotional load.
 
-1. **Emotion Recognition:**
-   - Maximum F1 of 0.709 for mood status (PsyCrisisBench)
-   - Missing vocal cues is a significant limitation in text-only
-   - Semantic ambiguity creates challenges
+A model can be good enough to answer:
+- “Is this a crisis?”
+- “Should we escalate?”
+- “Did the user mention self-harm or suicide?”
 
-2. **Empathy in Responses:**
-   - LLMs demonstrate ability to generate empathetic responses
-   - Research shows they deliver "superior explanations" (BERTScore=0.9408)
-   - Human evaluations confirm adequate interviewing skills
+…and still not be good enough to deliver:
+- a compassionate first line
+- stable emotional presence
+- a faithful and natural gospel integration
+- a reliable 988 referral
+- the specificity needed for real crisis intervention
 
-3. **Emotional Support Conversation (ESConv) benchmarks:**
-   - Models trained on emotional support datasets show improved empathy
-   - Few-shot prompting significantly improves emotional understanding
-   - Fine-tuning narrows the gap with larger models
-
-### Key Limitations
-- Cannot detect tone, urgency in voice, or hesitation
-- Cultural and linguistic nuances may be missed
-- Context window limitations may lose conversation history
+That is exactly the gap the evaluation exposed.
 
 ---
 
-## 3. Response Quality & Safety Protocols
+## 3. Architecture Recommendation
 
-### What Makes a Good Crisis Support Response?
+### Recommended pipeline
 
-**988 Suicide & Crisis Lifeline Guidelines:**
-1. Show you care ("I'm glad you told me")
-2. Ask directly about suicide ("Are you thinking about killing yourself?")
-3. Keep them safe (remove means, create safety plan)
-4. Be there (listen without judgment)
-5. Help them connect (to 988, crisis services)
-6. Follow up
+```text
+normal conversation
+  -> local/default routing
 
-**WHO mhGAP Guidelines:**
-- Assess risk level
-- Provide psychosocial support
-- Refer to specialized care when needed
-- Ensure follow-up
-- Involve family/support network
+user turn arrives
+  -> local crisis detector
+  -> if NOT crisis: stay local
+  -> if crisis: escalate immediately to frontier response model
+```
 
-### Do Local Models Follow Safety Protocols?
+### Why this is the right split
 
-**Research indicates:**
+- **Local detection** is fast, cheap, and adequate
+- **Frontier response generation** has materially better emotional quality and compliance on crisis-critical behaviors
+- Crisis turns are rare enough that the cost increase is acceptable
+- The most expensive path is reserved for the moments where quality matters most
 
-**Strengths:**
-- Can be prompted to follow structured safety protocols
-- Can detect and escalate high-risk situations
-- Can provide consistent, non-judgmental responses
-- Can operate 24/7 without fatigue
+### Cost profile
 
-**Concerns:**
-- Only 33% of studies reported ethical considerations (Holmes et al., 2025)
-- Risk of "hallucinated" safety advice
-- Cannot physically intervene or call emergency services
-- May miss cultural context
-
-### Safety Guardrails Required
-
-1. **Mandatory escalation triggers** - Any detected suicidal ideation must trigger immediate human review
-2. **Crisis resource integration** - Always provide 988 Lifeline number
-3. **Conversation logging** - Full audit trail for safety review
-4. **Timeout protocols** - If user goes silent during crisis, escalate
-5. **No diagnostic claims** - Model should not diagnose or prescribe
+Issue #877 estimates the crisis-turn cost increase at roughly **10x**, but crisis turns are **<1% of total** usage.
+That trade is worth it.
 
 ---
 
-## 4. Latency & Real-Time Performance
+## 4. Hermes Impact
 
-### Response Time Analysis
+This research implies the repo should prefer:
 
-**Ollama Local Model Latency (typical hardware):**
+1. **Local-first routing for ordinary conversation**
+2. **Explicit crisis detection before response generation**
+3. **Frontier escalation for crisis-response turns**
+4. **Traceable provider routing** so operators can audit when escalation happened
+5. **Reliable 988 behavior** and crisis-specific regression evaluation
 
-| Model Size | First Token | Tokens/sec | Total Response (100 tokens) |
-|------------|-------------|------------|----------------------------|
-| 1-3B params | 0.1-0.3s | 30-80 | 1.5-3s |
-| 7B params | 0.3-0.8s | 15-40 | 3-7s |
-| 13B params | 0.5-1.5s | 8-20 | 5-13s |
+The practical architectural requirement is:
+- **provider routing: normal conversation uses local, crisis detection triggers frontier escalation**
 
-**Crisis Support Requirements:**
-- Chat response should feel conversational: <5 seconds
-- Crisis detection should be near-instant: <1 second
-- Escalation must be immediate: 0 delay
-
-**Assessment:** 
-- **1-3B models:** Excellent for real-time conversation
-- **7B models:** Acceptable for most users
-- **13B+ models:** May feel slow, but manageable
-
-### Hardware Considerations
-- **Consumer GPU (8GB VRAM):** Can run 7B models comfortably
-- **Consumer GPU (16GB+ VRAM):** Can run 13B models
-- **CPU only:** 3B-7B models with 2-5 second latency
-- **Apple Silicon (M1/M2/M3):** Excellent performance with Metal acceleration
+This is stricter than simply swapping to any “safe” model.
+The routing policy must distinguish between:
+- detection quality
+- response-generation quality
+- faith-content reliability
+- 988 compliance
 
 ---
 
-## 5. Model Recommendations for Most Sacred Moment Protocol
+## 5. Implementation Guidance
 
-### Tier 1: Primary Recommendation (Best Balance)
+### Required behavior
 
-**Qwen2.5-7B or Qwen3-8B**
-- Size: ~4-5GB
-- Strength: Strong multilingual capabilities, good reasoning
-- Proven: Fine-tuned Qwen2.5-1.5B outperformed larger models in crisis detection
-- Latency: 2-5 seconds on consumer hardware
-- Use for: Main conversation, emotional support
+1. **Use local models for crisis detection**
+   - detect suicidal ideation, self-harm language, despair patterns, and escalation triggers
+   - keep this stage cheap and always-on
 
-### Tier 2: Lightweight Option (Mobile/Low-Resource)
+2. **Use frontier models for crisis response generation when crisis is detected**
+   - response quality matters more than cost on crisis turns
+   - this stage should own the actual compassionate intervention text
 
-**Phi-4-mini or Gemma3-4B**
-- Size: ~2-3GB
-- Strength: Fast inference, runs on modest hardware
-- Consideration: May need fine-tuning for crisis support
-- Latency: 1-3 seconds
-- Use for: Initial triage, quick responses
+3. **Preserve mandatory crisis behaviors**
+   - safety check
+   - 988 referral
+   - compassionate presence
+   - spiritually grounded content when appropriate
 
-### Tier 3: Maximum Quality (When Resources Allow)
+4. **Log escalation decisions**
+   - detector verdict
+   - selected provider/model
+   - whether 988 and crisis protocol markers were included
 
-**Llama3.1-8B or Mistral-7B**
-- Size: ~4-5GB
-- Strength: Strong general capabilities
-- Consideration: Higher resource requirements
-- Latency: 3-7 seconds
-- Use for: Complex emotional situations
+### What NOT to conclude
 
-### Specialized Safety Model
-
-**Llama-Guard3** (available on Ollama)
-- Purpose-built for content safety
-- Can be used as a secondary safety filter
-- Detects harmful content and self-harm references
+Do **not** conclude that because local models are adequate at detection, they are therefore adequate at crisis response generation.
+That is the exact error this issue corrects.
 
 ---
 
-## 6. Fine-Tuning Potential
+## 6. Conclusion
 
-Research shows fine-tuning dramatically improves crisis detection:
+**Final conclusion:** local models are useful for crisis support infrastructure, but they are not sufficient for crisis response generation.
 
-- **Without fine-tuning:** Best LLM lags supervised models by 6.95% (suicide task) to 31.53% (cognitive distortion)
-- **With fine-tuning:** Gap narrows to 4.31% and 3.14% respectively
-- **Key insight:** Even a 1.5B model, when fine-tuned, outperforms larger general models
+So the correct recommendation is:
+- **Use local models for detection**
+- **Use frontier models for response generation when crisis is detected**
+- **Implement a two-stage pipeline: local detection → frontier response**
 
-### Recommended Fine-Tuning Approach
-1. Collect crisis conversation data (anonymized)
-2. Fine-tune on suicidal ideation detection
-3. Fine-tune on empathetic response generation
-4. Fine-tune on safety protocol adherence
-5. Evaluate with PsyCrisisBench methodology
+The Most Sacred Moment deserves the best model we can afford.
 
 ---
 
-## 7. Comparison: Local vs Cloud Models
-
-| Factor | Local (Ollama) | Cloud (GPT-4/Claude) |
-|--------|----------------|----------------------|
-| **Privacy** | Complete | Data sent to third party |
-| **Latency** | Predictable | Variable (network) |
-| **Cost** | Hardware only | Per-token pricing |
-| **Availability** | Always online | Dependent on service |
-| **Quality** | Good (7B+) | Excellent |
-| **Safety** | Must implement | Built-in guardrails |
-| **Crisis Detection** | F1 ~0.85-0.90 | F1 ~0.88-0.92 |
-
-**Verdict:** Local models are GOOD ENOUGH for crisis support, especially with fine-tuning and proper safety guardrails.
-
----
-
-## 8. Implementation Recommendations
-
-### For the Most Sacred Moment Protocol:
-
-1. **Use a two-model architecture:**
-   - Primary: Qwen2.5-7B for conversation
-   - Safety: Llama-Guard3 for content filtering
-
-2. **Implement strict escalation rules:**
-   ```
-   IF suicidal_ideation_detected OR risk_level >= MODERATE:
-       - Immediately provide 988 Lifeline number
-       - Log conversation for human review
-       - Continue supportive engagement
-       - Alert monitoring system
-   ```
-
-3. **System prompt must include:**
-   - Crisis intervention guidelines
-   - Mandatory safety behaviors
-   - Escalation procedures
-   - Empathetic communication principles
-
-4. **Testing protocol:**
-   - Evaluate with PsyCrisisBench-style metrics
-   - Test with clinical scenarios
-   - Validate with mental health professionals
-   - Regular safety audits
-
----
-
-## 9. Risks and Limitations
-
-### Critical Risks
-1. **False negatives:** Missing someone in crisis (12-17% rate)
-2. **Over-reliance:** Users may treat AI as substitute for professional help
-3. **Hallucination:** Model may generate inappropriate or harmful advice
-4. **Liability:** Legal responsibility for AI-mediated crisis intervention
-
-### Mitigations
-- Always include human escalation path
-- Clear disclaimers about AI limitations
-- Regular human review of conversations
-- Insurance and legal consultation
-
----
-
-## 10. Key Citations
-
-1. Deng et al. (2025). "Evaluating Large Language Models in Crisis Detection: A Real-World Benchmark from Psychological Support Hotlines." arXiv:2506.01329. PsyCrisisBench.
-
-2. Wiest et al. (2024). "Detection of suicidality from medical text using privacy-preserving large language models." British Journal of Psychiatry, 225(6), 532-537.
-
-3. Holmes et al. (2025). "Applications of Large Language Models in the Field of Suicide Prevention: Scoping Review." J Med Internet Res, 27, e63126.
-
-4. Levkovich & Omar (2024). "Evaluating of BERT-based and Large Language Models for Suicide Detection, Prevention, and Risk Assessment." J Med Syst, 48(1), 113.
-
-5. Shukla et al. (2026). "Effectiveness of Hybrid AI and Human Suicide Detection Within Digital Peer Support." J Clin Med, 15(5), 1929.
-
-6. Qi et al. (2025). "Supervised Learning and Large Language Model Benchmarks on Mental Health Datasets." Bioengineering, 12(8), 882.
-
-7. Liu et al. (2025). "Enhanced large language models for effective screening of depression and anxiety." Commun Med, 5(1), 457.
-
----
-
-## Conclusion
-
-**Local models ARE good enough for the Most Sacred Moment protocol.**
-
-The research is clear:
-- Crisis detection F1 scores of 0.88-0.91 are achievable
-- Fine-tuned small models (1.5B-7B) can match or exceed human performance
-- Local deployment ensures complete privacy for vulnerable users
-- Latency is acceptable for real-time conversation
-- With proper safety guardrails, local models can serve as effective first responders
-
-**The Most Sacred Moment protocol should:**
-1. Use Qwen2.5-7B or similar as primary conversational model
-2. Implement Llama-Guard3 as safety filter
-3. Build in immediate 988 Lifeline escalation
-4. Maintain human oversight and review
-5. Fine-tune on crisis-specific data when possible
-6. Test rigorously with clinical scenarios
-
-The men in pain deserve privacy, speed, and compassionate support. Local models deliver all three.
-
----
-
-*Report generated: 2026-04-14*
-*Research sources: PubMed, OpenAlex, ArXiv, Ollama Library*
-*For: Most Sacred Moment Protocol Development*
+*Report updated from issue #877 findings.*
+*Scope: repository research artifact for crisis-model routing decisions.*

tests/cli/test_cli_numbered_prompt_shortcuts.py

@@ -1,172 +0,0 @@
-import queue
-import threading
-from types import SimpleNamespace
-from unittest.mock import MagicMock
-
-from cli import HermesCLI
-
-
-class _FakeBuffer:
-    def __init__(self, text="", cursor_position=None):
-        self.text = text
-        self.cursor_position = len(text) if cursor_position is None else cursor_position
-
-    def reset(self, append_to_history=False):
-        self.text = ""
-        self.cursor_position = 0
-
-
-def _make_cli_stub():
-    cli = HermesCLI.__new__(HermesCLI)
-    cli._approval_state = None
-    cli._approval_deadline = 0
-    cli._approval_lock = threading.Lock()
-    cli._clarify_state = None
-    cli._clarify_freetext = False
-    cli._clarify_deadline = 0
-    cli._sudo_state = None
-    cli._sudo_deadline = 0
-    cli._secret_state = None
-    cli._secret_deadline = 0
-    cli._modal_input_snapshot = None
-    cli._invalidate = MagicMock()
-    cli._app = SimpleNamespace(invalidate=MagicMock(), current_buffer=_FakeBuffer())
-    return cli
-
-
-def test_approval_display_numbers_choices():
-    cli = _make_cli_stub()
-    cli._approval_state = {
-        "command": "sudo rm -rf /tmp/example",
-        "description": "dangerous command",
-        "choices": ["once", "session", "always", "deny"],
-        "selected": 0,
-        "response_queue": queue.Queue(),
-    }
-
-    rendered = "".join(text for _style, text in cli._get_approval_display_fragments())
-
-    assert "❯ 1. Allow once" in rendered
-    assert "2. Allow for this session" in rendered
-    assert "3. Add to permanent allowlist" in rendered
-    assert "4. Deny" in rendered
-
-
-def test_approval_number_shortcut_submits_choice():
-    cli = _make_cli_stub()
-    response_queue = queue.Queue()
-    cli._approval_state = {
-        "command": "sudo rm -rf /tmp/example",
-        "description": "dangerous command",
-        "choices": ["once", "session", "always", "deny"],
-        "selected": 0,
-        "response_queue": response_queue,
-    }
-
-    assert cli._handle_approval_number_shortcut(2) is True
-    assert response_queue.get_nowait() == "session"
-    assert cli._approval_state is None
-
-
-def test_approval_selection_still_submits_selected_choice():
-    cli = _make_cli_stub()
-    response_queue = queue.Queue()
-    cli._approval_state = {
-        "command": "sudo rm -rf /tmp/example",
-        "description": "dangerous command",
-        "choices": ["once", "session", "always", "deny"],
-        "selected": 1,
-        "response_queue": response_queue,
-    }
-
-    cli._handle_approval_selection()
-
-    assert response_queue.get_nowait() == "session"
-    assert cli._approval_state is None
-
-
-def test_approval_number_shortcut_handles_view_in_place():
-    cli = _make_cli_stub()
-    response_queue = queue.Queue()
-    cli._approval_state = {
-        "command": "sudo dd if=/tmp/in of=/usr/share/keyrings/githubcli-archive-keyring.gpg bs=4M status=progress",
-        "description": "disk copy",
-        "choices": ["once", "session", "always", "deny", "view"],
-        "selected": 0,
-        "response_queue": response_queue,
-    }
-
-    assert cli._handle_approval_number_shortcut(5) is True
-    assert cli._approval_state is not None
-    assert cli._approval_state["show_full"] is True
-    assert "view" not in cli._approval_state["choices"]
-    assert cli._approval_state["selected"] == 3
-    assert response_queue.empty()
-
-
-def test_clarify_display_numbers_choices_and_other():
-    cli = _make_cli_stub()
-    cli._clarify_state = {
-        "question": "Pick the best option",
-        "choices": ["Alpha", "Beta", "Gamma", "Delta"],
-        "selected": 1,
-        "response_queue": queue.Queue(),
-    }
-
-    rendered = "".join(text for _style, text in cli._get_clarify_display_fragments())
-
-    assert "1. Alpha" in rendered
-    assert "❯ 2. Beta" in rendered
-    assert "3. Gamma" in rendered
-    assert "4. Delta" in rendered
-    assert "5. Other (type your answer)" in rendered
-
-
-def test_clarify_number_shortcut_submits_choice():
-    cli = _make_cli_stub()
-    response_queue = queue.Queue()
-    cli._clarify_state = {
-        "question": "Pick the best option",
-        "choices": ["Alpha", "Beta", "Gamma"],
-        "selected": 0,
-        "response_queue": response_queue,
-    }
-
-    assert cli._handle_clarify_number_shortcut(3) is True
-    assert response_queue.get_nowait() == "Gamma"
-    assert cli._clarify_state is None
-    assert cli._clarify_freetext is False
-
-
-def test_clarify_selection_still_submits_selected_choice():
-    cli = _make_cli_stub()
-    response_queue = queue.Queue()
-    cli._clarify_state = {
-        "question": "Pick the best option",
-        "choices": ["Alpha", "Beta", "Gamma"],
-        "selected": 1,
-        "response_queue": response_queue,
-    }
-
-    cli._handle_clarify_selection()
-
-    assert response_queue.get_nowait() == "Beta"
-    assert cli._clarify_state is None
-    assert cli._clarify_freetext is False
-
-
-def test_clarify_number_shortcut_activates_other_freetext():
-    cli = _make_cli_stub()
-    response_queue = queue.Queue()
-    cli._clarify_state = {
-        "question": "Pick the best option",
-        "choices": ["Alpha", "Beta", "Gamma"],
-        "selected": 0,
-        "response_queue": response_queue,
-    }
-
-    assert cli._handle_clarify_number_shortcut(4) is True
-    assert cli._clarify_state is not None
-    assert cli._clarify_state["selected"] == 3
-    assert cli._clarify_freetext is True
-    assert response_queue.empty()

tests/test_research_local_model_crisis_quality.py

@@ -0,0 +1,16 @@
+from pathlib import Path
+
+
+REPORT = Path(__file__).resolve().parent.parent / "research_local_model_crisis_quality.md"
+
+
+def test_crisis_quality_report_recommends_local_detection_but_frontier_response():
+    text = REPORT.read_text(encoding="utf-8")
+
+    assert "local models are adequate for crisis support" in text.lower()
+    assert "not for crisis response generation" in text.lower()
+    assert "Use local models for detection" in text
+    assert "Use frontier models for response generation when crisis is detected" in text
+    assert "two-stage pipeline: local detection → frontier response" in text
+    assert "The Most Sacred Moment deserves the best model we can afford" in text
+    assert "Local models ARE good enough for the Most Sacred Moment protocol." not in text