Twin Identity setup, Simulation prompt refinement, and twin accuracy roadmap

README.md

@@ -52,6 +52,7 @@ The first usable twin mode is a **native RAG twin**:
 - Full-text search powered by Tantivy.
 - Backlinks, outgoing links, and graph-aware retrieval.
 - Conversation import from ChatGPT, Claude, Grok, Gemini, and Codex-style exports.
+- Structured vault migration with preview, apply, and rollback.
 
 ### Knowledge Graph And Hub Clustering
 
@@ -72,6 +73,12 @@ The first usable twin mode is a **native RAG twin**:
 - Twin context mode using reviewed user records.
 - Smart web search detection for prompts that need current information.
 - Save Canvas sessions as notes.
+- **Local model support** via Ollama — run Canvas and twin answers on local models alongside OpenRouter cloud models.
+
+### Vault Automation
+
+- **Background link discovery** — scans the vault continuously for potential `[[wikilinks]]` using keyword extraction and similarity, surfaces suggestions in the Link Suggestion inbox.
+- **Vault Optimizer** — background service that queues and applies structural improvements to notes (tag cleanup, sidecar metadata, full rewrites). Budget caps and per-change rollback keep it safe to enable.
 
 ### Twin Capture And Review
 
@@ -165,6 +172,27 @@ Grafyn's data can later support stronger personal models, but those are not v1:
 - **Local adapters or fine-tuning** - adjusts a capable base model using reviewed examples.
 - **Scratch-trained personal model** - research path only. Prompts alone are not enough; it would require large volumes of personal writing, decisions, outcomes, corrections, and domain evidence.
 
+## Project Status
+
+| Area | Status |
+|------|--------|
+| Knowledge vault (notes, wikilinks, full-text search) | ✅ Stable |
+| Knowledge graph + topic hub clustering | ✅ Stable |
+| Multi-LLM Canvas with semantic note context | ✅ Stable |
+| Conversation import (ChatGPT, Claude, Grok, Gemini) | ✅ Stable |
+| Native RAG twin (Advisor + Simulation modes) | ✅ Stable |
+| Twin Identity, Constitution, Decision Mirror | ✅ Stable |
+| Twin evidence capture and review dashboard | ✅ Stable |
+| Local model support via Ollama | ✅ Stable |
+| Background link discovery | ✅ Stable |
+| Vault Optimizer (background vault improvements) | ✅ Stable |
+| Structured vault migration (preview/apply/rollback) | ✅ Stable |
+| MCP server (`grafyn-mcp`) for Claude Desktop / Codex Desktop | ✅ Stable |
+| Preference / ranking model from export bundles | 🔲 Not started |
+| Local adapters or fine-tuning from reviewed evidence | 🔲 Not started |
+
+Current version: see [Releases](https://github.com/WKJBryan/Grafyn/releases/latest).
+
 ## Quick Start
 
 ### Download
@@ -260,7 +288,7 @@ Tauri Desktop App
 | Backend | Rust |
 | Search | Tantivy |
 | Graph | petgraph + local graph algorithms |
-| LLM Runtime | OpenRouter via reqwest |
+| LLM Runtime | OpenRouter via reqwest; Ollama for local models |
 | MCP | rmcp over stdio |
 | Storage | Local markdown vault + JSON data files |
 | Updates | Cloudflare R2 + Workers |

frontend/src-tauri/src/commands/canvas.rs

@@ -2930,7 +2930,7 @@ fn build_twin_context_prompt(
         ),
         TwinAnswerMode::Simulation => prompt.push_str(
             "Answer in first person from the configured Twin Identity. Use approved records as stronger style and preference evidence; mention candidate influence as tentative when relevant. \
-             Write as a natural continuation of my documented reasoning pattern, not a report. Lead with my likely reasoning or judgment, show the tradeoff logic, and ask sharp reflective questions when useful. \
+             Write as a natural continuation of my documented reasoning pattern, not a report. Lead with my likely reasoning or judgment, show the tradeoff logic, and do not append questions unless the user's request asks for them. \
              If the evidence packet does not contain enough basis, say so naturally in first person. Use light citations or brief source mentions only where they help; avoid turning the answer into an evidence workflow.\n",
         ),
     }
@@ -2956,14 +2956,7 @@ fn build_twin_context_prompt(
                  In Constitution Check, separate stated values, revealed behavior, taste, somatic signal, and constraints. In Action Gap Risk, state whether past intention-action gaps could change the next step. \
                  Recommendation must be derived after the Constitution Check and Evidence From Grafyn sections, not before them.\n",
             ),
-            TwinAnswerMode::Simulation => prompt.push_str(
-                "\n## Decision Mirror Simulation Style\n\n\
-                 This is a Decision Mirror simulation session. Do not return numbered headings or a card template. \
-                 Return a natural first-person reflection that feels like my reasoning pattern thinking through the decision. \
-                 Prefer concise paragraphs over sections. Start with my likely judgment or hesitation, then explain the filters, tradeoffs, and next question. \
-                 Keep light citations or parenthetical source mentions where helpful, but keep evidence backstage. \
-                 Treat every self-model claim as evidence-constrained. If a useful claim is weakly supported, say so naturally.\n",
-            ),
+            TwinAnswerMode::Simulation => {}
         }
 
         if let Some(metadata) = decision_metadata {
@@ -3555,6 +3548,11 @@ mod tests {
         assert!(prompt.contains("My role/context is founder deciding from product evidence."));
         assert!(prompt.contains("Use reviewed notes and uploaded interviews only."));
         assert!(prompt.contains("Continue my documented reasoning pattern"));
+        assert!(prompt.contains(
+            "do not append questions unless the user's request asks for them"
+        ));
+        assert!(!prompt.contains("reflective questions"));
+        assert!(!prompt.contains("next question"));
         assert!(!prompt.contains("not the user's actual view"));
         assert!(prompt.contains("## Reviewed Constitution"));
     }
@@ -3602,7 +3600,7 @@ mod tests {
     }
 
     #[test]
-    fn decision_simulation_prompt_uses_natural_reflection_instructions() {
+    fn decision_simulation_prompt_uses_base_simulation_without_decision_style_block() {
         let metadata = DecisionPromptMetadata {
             decision: "Should Grafyn build Decision Mirror first?".into(),
             options: vec!["Decision Mirror".into(), "Topology layer".into()],
@@ -3626,8 +3624,13 @@ mod tests {
         assert!(!prompt.contains("Reflection Card"));
         assert!(!prompt.contains("Evidence From Grafyn"));
         assert!(!prompt.contains("Blind Spot Hypothesis"));
-        assert!(prompt.contains("natural first-person reflection"));
-        assert!(prompt.contains("light citations"));
+        assert!(!prompt.contains("Decision Mirror Simulation Style"));
+        assert!(!prompt.contains("Decision Mirror simulation session"));
+        assert!(!prompt.contains("natural first-person reflection"));
+        assert!(!prompt.contains("numbered headings"));
+        assert!(prompt.contains(
+            "do not append questions unless the user's request asks for them"
+        ));
         assert!(!prompt.contains("not the user's actual view"));
         assert!(prompt.contains("Decision: Should Grafyn build Decision Mirror first?"));
     }

frontend/src/__tests__/unit/components/CanvasContainer.spec.js

@@ -2,7 +2,7 @@ import { describe, expect, it, beforeEach, vi } from 'vitest'
 import { mount, flushPromises } from '@vue/test-utils'
 import CanvasContainer from '@/components/canvas/CanvasContainer.vue'
 
-const { store, getOpenRouterStatus, getStatus, getSettings, updateSettings, listOllamaModels, getConstitutionSetup, toastSuccess } = vi.hoisted(() => ({
+const { store, getOpenRouterStatus, getStatus, getSettings, updateSettings, listOllamaModels, getConstitutionSetup, saveConstitutionSetup, toastSuccess } = vi.hoisted(() => ({
   store: {
     currentSession: {
       id: 'session-1',
@@ -58,6 +58,7 @@ const { store, getOpenRouterStatus, getStatus, getSettings, updateSettings, list
   updateSettings: vi.fn(),
   listOllamaModels: vi.fn(),
   getConstitutionSetup: vi.fn(),
+  saveConstitutionSetup: vi.fn(),
   toastSuccess: vi.fn()
 }))
 
@@ -74,7 +75,8 @@ vi.mock('@/api/client', () => ({
     listOllamaModels
   },
   twin: {
-    getConstitutionSetup
+    getConstitutionSetup,
+    saveConstitutionSetup
   },
   isDesktopApp: () => true
 }))
@@ -186,6 +188,21 @@ function mountContainer() {
                 twinLlmProvider: 'ollama',
                 webSearch: false
               })" />
+              <button class="identity-submit-stub" @click="$emit('submit', {
+                prompt: 'Simulate my likely response',
+                promptType: 'standard',
+                models: ['llama3.1:8b'],
+                systemPrompt: null,
+                temperature: 0.4,
+                contextMode: 'twin',
+                twinAnswerMode: 'simulation',
+                twinLlmProvider: 'ollama',
+                webSearch: false,
+                twinIdentitySetup: {
+                  twin_name: 'Alex Chen',
+                  twin_role: 'founder deciding from product evidence'
+                }
+              })" />
             </div>
           `
         }
@@ -280,6 +297,7 @@ describe('CanvasContainer', () => {
       twin_name: 'Alex Chen',
       twin_role: 'founder deciding from product evidence'
     })
+    saveConstitutionSetup.mockResolvedValue({})
     toastSuccess.mockReset()
   })
 
@@ -295,6 +313,55 @@ describe('CanvasContainer', () => {
     expect(wrapper.text()).not.toContain('OpenRouter API Key Required')
   })
 
+  it('saves inline Twin Identity before submitting a simulation prompt', async () => {
+    getOpenRouterStatus.mockResolvedValue({ has_key: true, is_configured: true })
+    getConstitutionSetup.mockResolvedValueOnce({
+      values: ['evidence-backed work'],
+      tastes: [],
+      constraints: [],
+      somatic_cues: [],
+      action_tendencies: []
+    }).mockResolvedValue({
+      twin_name: 'Alex Chen',
+      twin_role: 'founder deciding from product evidence',
+      values: ['evidence-backed work'],
+      tastes: [],
+      constraints: [],
+      somatic_cues: [],
+      action_tendencies: []
+    })
+    const wrapper = mountContainer()
+    await flushPromises()
+
+    await wrapper.find('[data-guide="canvas-prompt-btn"]').trigger('click')
+    await flushPromises()
+    await wrapper.find('.identity-submit-stub').trigger('click')
+    await flushPromises()
+
+    expect(saveConstitutionSetup).toHaveBeenCalledWith(expect.objectContaining({
+      twin_name: 'Alex Chen',
+      twin_role: 'founder deciding from product evidence',
+      values: ['evidence-backed work']
+    }))
+    expect(store.sendPrompt).toHaveBeenCalledWith(
+      'Simulate my likely response',
+      ['llama3.1:8b'],
+      null,
+      0.4,
+      null,
+      null,
+      null,
+      'twin',
+      'simulation',
+      false,
+      undefined,
+      'standard',
+      null,
+      'none',
+      'ollama'
+    )
+  })
+
   it('blocks API submit if OpenRouter becomes unavailable after the dialog opens', async () => {
     getOpenRouterStatus
       .mockResolvedValueOnce({ has_key: true, is_configured: true })

frontend/src/__tests__/unit/components/PromptDialog.spec.js

@@ -117,7 +117,7 @@ describe('PromptDialog', () => {
     expect(wrapper.find('.btn-primary').attributes('disabled')).toBeDefined()
   })
 
-  it('blocks Twin Simulation until identity name and role are configured', async () => {
+  it('asks for Twin Identity inline before running Simulation', async () => {
     const wrapper = mountDialog({
       twinLlmProvider: 'ollama',
       ollamaModel: 'llama3.1:8b',
@@ -127,8 +127,23 @@ describe('PromptDialog', () => {
     await wrapper.find('#prompt').setValue('Simulate my likely response')
     await wrapper.find('#contextMode').setValue('twin')
 
-    expect(wrapper.text()).toContain('Set up Twin Identity with a name and role before running Simulation')
+    expect(wrapper.text()).toContain('Twin Identity')
+    expect(wrapper.text()).toContain('Who should this twin be?')
+    expect(wrapper.text()).toContain('What work or role should this twin reason from?')
     expect(wrapper.find('.btn-primary').attributes('disabled')).toBeDefined()
+
+    await wrapper.find('#inlineTwinName').setValue('Alex Chen')
+    await wrapper.find('#inlineTwinRole').setValue('founder deciding from product evidence')
+    await wrapper.find('.btn-primary').trigger('click')
+
+    expect(wrapper.emitted('submit')[0][0]).toMatchObject({
+      contextMode: 'twin',
+      twinAnswerMode: 'simulation',
+      twinIdentitySetup: {
+        twin_name: 'Alex Chen',
+        twin_role: 'founder deciding from product evidence'
+      }
+    })
   })
 
   it('allows Twin Simulation after identity name and role are configured', async () => {

frontend/src/components/canvas/CanvasContainer.vue

@@ -1221,6 +1221,26 @@ async function refreshTwinIdentity() {
   }
 }
 
+async function saveInlineTwinIdentity(identitySetup) {
+  const existing = await twinApi.getConstitutionSetup()
+  const nextSetup = {
+    ...existing,
+    twin_name: identitySetup.twin_name,
+    twin_role: identitySetup.twin_role,
+    source_boundaries: existing?.source_boundaries || [],
+    values: existing?.values || [],
+    tastes: existing?.tastes || [],
+    constraints: existing?.constraints || [],
+    somatic_cues: existing?.somatic_cues || [],
+    action_tendencies: existing?.action_tendencies || []
+  }
+  await twinApi.saveConstitutionSetup(nextSetup)
+  twinIdentity.value = {
+    name: nextSetup.twin_name || '',
+    role: nextSetup.twin_role || ''
+  }
+}
+
 async function loadCanvasPreferences() {
   try {
     const settingsData = await settingsApi.get()
@@ -1358,7 +1378,8 @@ async function handlePromptSubmit({
   twinAnswerMode = 'simulation',
   webSearch,
   reasoningEffort = 'none',
-  twinLlmProvider: selectedTwinLlmProvider = null
+  twinLlmProvider: selectedTwinLlmProvider = null,
+  twinIdentitySetup = null
 }) {
   const usesLocalTwinRuntime = (promptType === 'decision' || contextMode === 'twin') &&
     selectedTwinLlmProvider === 'ollama'
@@ -1376,6 +1397,20 @@ async function handlePromptSubmit({
     }
   }
 
+  if (twinIdentitySetup) {
+    try {
+      await saveInlineTwinIdentity(twinIdentitySetup)
+    } catch (err) {
+      console.error('Failed to save Twin Identity before submit:', err)
+      saveMessage.value = {
+        type: 'error',
+        text: err.message || 'Failed to save Twin Identity'
+      }
+      setTimeout(() => { saveMessage.value = null }, 5000)
+      return
+    }
+  }
+
   showPromptDialog.value = false
   const activeBranchContext = branchContext.value