feat(analysis): dampen initial/bulk commits in Automation axis

Add weighted averaging for the Automation Heaviness axis to prevent
initial project commits and bulk operations from skewing results:

- First commit: weight = 0.25 (likely scaffolding)
- Bulk commits (>50% of max files, when max > 20): weight = 0.5
- Normal commits: weight = 1.0

This addresses feedback where users with small repos got high Automation
scores due to a single large initial commit.

Documentation updated:
- docs/how-vibed-works.md
- docs/architecture/vibe-metrics-v2.md
- apps/web/src/app/methodology/page.tsx

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: devakone

apps/web/src/app/methodology/page.tsx

@@ -54,7 +54,7 @@ export default async function MethodologyPage() {
                 id: "A",
                 name: "Automation",
                 description:
-                  "Large, wide changes: high files-changed per commit, big commits, big PRs.",
+                  "Large, wide changes: high files-changed per commit, big commits, big PRs. Initial/bulk commits are dampened to avoid skewing small repos.",
               },
               {
                 id: "B",
@@ -186,8 +186,12 @@ export default async function MethodologyPage() {
               copy/paste, refactors between commits), so we infer from what lands in Git.
             </li>
             <li>
-              Different projects can pull you into different modes; aggregation may “average you
-              out”.
+              Different projects can pull you into different modes; aggregation may "average you
+              out".
+            </li>
+            <li>
+              We dampen initial and bulk commits in the Automation axis, but repos with very few
+              commits (&lt;10) may still show unusual scores.
             </li>
           </ul>
         </section>

docs/architecture/vibe-metrics-v2.md

@@ -41,11 +41,26 @@ How "agentic" the workflow looks.
 
 **Score formula:**
 ```
-40% - Commit chunkiness (avg_files_changed, commit_size_p90)
-40% - PR chunkiness (pr_files_changed_p90, commits_per_pr_p90)
-20% - Text templating score
+50% - Commit chunkiness (weighted_avg_files_changed, commit_size_p90)
+30% - p90 commit size
+20% - PR chunkiness (pr_files_changed_p90)
 ```
 
+**Initial/Bulk Commit Dampening:**
+
+To prevent initial project commits and bulk operations from skewing the Automation score, we apply weight dampening when calculating `weighted_avg_files_changed`:
+
+```typescript
+// Weight assignment per commit:
+weight = 1.0;  // default
+if (isFirstCommit) weight = 0.25;  // scaffolding likely
+else if (files_changed > maxFiles * 0.5 && maxFiles > 20) weight = 0.5;  // bulk op
+
+// Then: weighted_avg = sum(files * weight) / sum(weight)
+```
+
+This ensures small repos with large initial commits (e.g., committing an entire create-react-app scaffold) don't get marked as "AI-heavy" based on that single data point.
+
 **Evidence examples:**
 - "Top 10% PRs change 40+ files"
 - "Median commit touches 6 files"

docs/how-vibed-works.md

@@ -76,13 +76,27 @@ Six deterministic axes (0-100 scores) capture your workflow style:
 
 | Axis | What It Measures |
 |------|------------------|
-| **Automation Heaviness** | How much you rely on AI agents and tools for generating code |
+| **Automation Heaviness** | How much you rely on AI agents and tools for generating code (initial/bulk commits dampened) |
 | **Guardrail Strength** | How closely tests, CI, and docs follow your AI-generated changes |
 | **Iteration Loop Intensity** | How often you run prompt-fix-run loops to refine AI output |
 | **Planning Signal** | How much structure you define before prompting AI to generate code |
 | **Surface Area per Change** | How many parts of the codebase your typical prompt or change touches |
 | **Shipping Rhythm** | Your coding session pattern — steady output vs intense vibe sessions |
 
+#### Automation Axis: Initial Commit Dampening
+
+The Automation Heaviness axis measures how "agentic" your workflow looks based on commit size and breadth. However, initial project commits (scaffolding an entire codebase at once) and bulk operations can skew this metric, especially for small repos.
+
+To address this, we apply **commit weight dampening**:
+
+| Commit Type | Weight | Rationale |
+|-------------|--------|-----------|
+| First commit | 0.25 | Often includes project scaffolding, framework setup |
+| Bulk commits (>50% of max files, when max > 20) | 0.5 | Large refactors, dependency updates |
+| Normal commits | 1.0 | Typical development work |
+
+This means a repo with 7 commits where the first commit touched 800 files won't be marked as "AI-heavy" just because of the initial setup. The weighted average better reflects actual day-to-day patterns.
+
 ### Step 5: Detect Persona
 Based on your axes, we match you to one of 7 Vibe Personas:
 

packages/core/src/vibe.ts

@@ -603,9 +603,51 @@ export function computeVibeAxes(input: ComputeVibeAxesInput): ComputeVibeAxesOut
   const episodes = input.episodes ?? [];
 
   // --- Axis A: Automation Heaviness ---
-  const totalCommits = metrics.total_commits ?? input.commits?.length ?? 1;
+  // We dampen the impact of initial/bulk commits to avoid skewing results for small repos.
+  // - First commit: weight = 0.25 (often includes project scaffolding)
+  // - Bulk commits (>50% of max files, when max > 20): weight = 0.5
+  // - Normal commits: weight = 1.0
+  const commits = input.commits ?? [];
+  const totalCommits = metrics.total_commits ?? commits.length ?? 1;
   const totalFilesChanged = metrics.total_files_changed ?? 0;
-  const avgFiles = totalFilesChanged / Math.max(1, totalCommits);
+
+  const avgFiles = (() => {
+    if (commits.length === 0) {
+      return totalFilesChanged / Math.max(1, totalCommits);
+    }
+
+    // Sort by date to identify first commit
+    const sortedCommits = [...commits].sort(
+      (a, b) => new Date(a.committer_date).getTime() - new Date(b.committer_date).getTime()
+    );
+
+    // Find max files changed to detect bulk commits
+    const maxFilesChanged = Math.max(...commits.map((c) => c.files_changed), 1);
+    const bulkThreshold = maxFilesChanged * 0.5;
+
+    let weightedFilesSum = 0;
+    let totalWeight = 0;
+
+    for (let i = 0; i < sortedCommits.length; i++) {
+      const commit = sortedCommits[i];
+      let weight = 1.0;
+
+      // First commit gets dampened weight (likely project scaffolding)
+      if (i === 0) {
+        weight = 0.25;
+      }
+      // Bulk commits (>50% of max files) get reduced weight, but only if max is substantial
+      else if (commit.files_changed > bulkThreshold && maxFilesChanged > 20) {
+        weight = 0.5;
+      }
+
+      weightedFilesSum += commit.files_changed * weight;
+      totalWeight += weight;
+    }
+
+    return totalWeight > 0 ? weightedFilesSum / totalWeight : totalFilesChanged / Math.max(1, totalCommits);
+  })();
+
   const p90Commit = Number(metrics.commit_size_p90 ?? 0);
 
   const prChangedFiles = prs