diff --git a/.beads/beads.jsonl b/.beads/beads.jsonl
index d2bddb7f..7012cf17 100644
--- a/.beads/beads.jsonl
+++ b/.beads/beads.jsonl
@@ -132,7 +132,7 @@
 {"id":"bd-69","content_hash":"a4e81b23d88d41c8fd3fe31fb7ef387f99cb54ea42a6baa210ede436ecce3288","title":"Replace getStorageForRequest with Direct Access","description":"Replace all getStorageForRequest(req) calls with s.storage","acceptance_criteria":"- No references to getStorageForRequest() in codebase (except in deleted file)\n- All handlers use s.storage directly\n- Code compiles without errors\n\nFiles to update:\n- internal/rpc/server_issues_epics.go (~8 calls)\n- internal/rpc/server_labels_deps_comments.go (~4 calls)\n- internal/rpc/server_compact.go (~2 calls)\n- internal/rpc/server_export_import_auto.go (~2 calls)\n- internal/rpc/server_routing_validation_diagnostics.go (~1 call)\n\nPattern: store, err := s.getStorageForRequest(req) → store := s.storage","status":"closed","priority":1,"issue_type":"task","created_at":"2025-10-28T16:20:02.430127-07:00","updated_at":"2025-10-28T19:20:58.312809-07:00","closed_at":"2025-10-28T19:20:58.312809-07:00"}
 {"id":"bd-7","content_hash":"e88e5d98a2a5bebc38b3ac505b00687bfe78bd72654bd0c756bceee4a01e15f5","title":"Enforce daemon singleton per workspace with file locking","description":"Agent in ~/src/wyvern discovered 4 simultaneous daemon processes running, causing infinite directory recursion (.beads/.beads/.beads/...). Each daemon used relative paths and created nested .beads/ directories.\n\nRoot cause: No singleton enforcement. Multiple `bd daemon` processes can start in same workspace.\n\nExpected: One daemon per workspace (each workspace = separate .beads/ dir with bd.sock)\nActual: Multiple daemons can run simultaneously in same workspace\n\nNote: Separate git clones = separate workspaces = separate daemons (correct). Git worktrees share .beads/ and have known limitations (documented, use --no-daemon).","design":"Use flock (file locking) on daemon socket or database file to enforce singleton:\n\n1. On daemon start, attempt exclusive lock on .beads/bd.sock or .beads/daemon.lock\n2. If lock held by another process, refuse to start (exit with clear error)\n3. Hold lock for lifetime of daemon process\n4. Release lock on daemon shutdown\n\nAlternative: Use PID file with stale detection (check if PID is still running)\n\nImplementation location: Daemon startup code in cmd/bd/ or internal/daemon/","acceptance_criteria":"1. Starting second daemon process in same workspace fails with clear error\n2. Test: Start daemon, attempt second start, verify failure\n3. Killing daemon releases lock, allowing new daemon to start\n4. No infinite .beads/ directory recursion possible\n5. Works correctly with auto-start mechanism","status":"in_progress","priority":0,"issue_type":"bug","created_at":"2025-10-25T23:13:12.269549-07:00","updated_at":"2025-10-27T22:22:23.814937-07:00"}
 {"id":"bd-70","content_hash":"c0b1677fe3f4aa3f395ae4d79bff5362632d5db26477bf571c09f9177b8741ef","title":"Event-driven daemon architecture","description":"Replace 5-second polling sync loop with event-driven architecture that reacts instantly to changes. Eliminates stale data issues while reducing CPU ~60%. Key components: FileWatcher (fsnotify), Debouncer (500ms), RPC mutation events, optional git hooks. Target latency: \u003c500ms (vs 5000ms). See event_driven_daemon.md for full design.","status":"closed","priority":1,"issue_type":"epic","created_at":"2025-10-28T16:20:02.430479-07:00","updated_at":"2025-10-28T16:30:26.631191-07:00","closed_at":"2025-10-28T16:30:26.631191-07:00"}
-{"id":"bd-71","content_hash":"6b2a1aedbdbcb30b98d4a8196801953a1eb22204d63e31954ef9ab6020a7a26b","title":"Create cmd/bd/daemon_watcher.go (~150 LOC)","description":"Implement FileWatcher using fsnotify to watch JSONL file and git refs. Handle platform differences (inotify/FSEvents/ReadDirectoryChangesW). Include edge case handling for file rename, event storm, watcher failure.","status":"in_progress","priority":1,"issue_type":"task","created_at":"2025-10-28T16:20:02.430809-07:00","updated_at":"2025-10-28T16:20:02.430809-07:00"}
+{"id":"bd-71","content_hash":"6b2a1aedbdbcb30b98d4a8196801953a1eb22204d63e31954ef9ab6020a7a26b","title":"Create cmd/bd/daemon_watcher.go (~150 LOC)","description":"Implement FileWatcher using fsnotify to watch JSONL file and git refs. Handle platform differences (inotify/FSEvents/ReadDirectoryChangesW). Include edge case handling for file rename, event storm, watcher failure.","status":"closed","priority":1,"issue_type":"task","created_at":"2025-10-28T16:20:02.430809-07:00","updated_at":"2025-10-29T22:49:21.971735-07:00","closed_at":"2025-10-29T22:49:21.971735-07:00"}
 {"id":"bd-72","content_hash":"a596aa8d6114d4938471e181ebc30da5d0315f74fd711a92dbbb83f5d0e7af88","title":"Create cmd/bd/daemon_debouncer.go (~60 LOC)","description":"Implement Debouncer to batch rapid events into single action. Default 500ms, configurable via BEADS_DEBOUNCE_MS. Thread-safe with mutex.","status":"closed","priority":1,"issue_type":"task","created_at":"2025-10-28T16:20:02.431118-07:00","updated_at":"2025-10-28T16:20:02.431118-07:00","closed_at":"2025-10-28T12:03:35.614191-07:00"}
 {"id":"bd-73","content_hash":"27cecaa2dc6cdabb2ae77fd65fbf8dca8f4c536bdf140a13b25cdd16376c9845","title":"Add docs/architecture/event_driven.md","description":"Copy event_driven_daemon.md into docs/ folder. Add to documentation index.","status":"open","priority":2,"issue_type":"task","created_at":"2025-10-28T16:20:02.431399-07:00","updated_at":"2025-10-28T16:20:02.431399-07:00"}
 {"id":"bd-74","content_hash":"8407a18ee38e96f92e7c7afde2f39b3df6fad409ccd5080243925d8a05fc85c1","title":"Run final validation and cleanup checks","description":"Final validation pass to ensure all cleanup objectives met and no regressions introduced.\n\nValidation checklist:\n1. Dead code verification: `go run golang.org/x/tools/cmd/deadcode@latest -test ./...`\n2. Test coverage: `go test -cover ./...`\n3. Build verification: `go build ./cmd/bd/`\n4. Linting: `golangci-lint run`\n5. Integration tests\n6. Metrics verification\n7. Git clean check\n\nFinal metrics to report:\n- LOC removed: ~____\n- Files deleted: ____\n- Files created: ____\n- Test coverage: ____%\n- Build time: ____ (before/after)\n- Test run time: ____ (before/after)\n\nImpact: Confirms all cleanup objectives achieved successfully","acceptance_criteria":"- Zero unreachable functions per deadcode analyzer\n- All tests pass: `go test ./...`\n- Test coverage maintained or improved\n- Builds cleanly: `go build ./...`\n- Linting shows improvements\n- Integration tests all pass\n- LOC reduction target achieved (~2,500 LOC)\n- No unintended behavior changes\n- Git commit messages document all changes","status":"open","priority":1,"issue_type":"task","created_at":"2025-10-29T20:49:49.131575-07:00","updated_at":"2025-10-29T20:49:49.131575-07:00"}
@@ -156,7 +156,7 @@
 {"id":"bd-91","content_hash":"0744c30a5397c6c44b949c038af110eaf6453ec3800bff55cb027eecc47ab5b5","title":"Fix TestTwoCloneCollision to compare content not timestamps","description":"The test at beads_twoclone_test.go:204-207 currently compares full JSON output including timestamps, causing false negative failures.\n\nCurrent behavior:\n- Both clones converge to identical semantic content\n- Clone A: test-2=\"Issue from clone A\", test-1=\"Issue from clone B\"\n- Clone B: test-1=\"Issue from clone B\", test-2=\"Issue from clone A\"\n- Titles match IDs correctly, no data corruption\n- Only timestamps differ (expected and acceptable)\n\nFix needed:\n- Replace exact JSON comparison with content-aware comparison\n- Normalize or ignore timestamp fields when asserting convergence\n- Test should PASS after this fix\n\nThis blocks completion of bd-86.","acceptance_criteria":"- Test compares issue content (title, description, status, priority) not timestamps\n- TestTwoCloneCollision passes\n- Both clones shown to have identical semantic content\n- Timestamps explicitly documented as acceptable difference","status":"closed","priority":1,"issue_type":"task","created_at":"2025-10-28T17:58:52.057194-07:00","updated_at":"2025-10-28T18:01:38.751895-07:00","closed_at":"2025-10-28T18:01:38.751895-07:00","dependencies":[{"issue_id":"bd-91","depends_on_id":"bd-90","type":"parent-child","created_at":"2025-10-28T17:58:52.058202-07:00","created_by":"stevey"},{"issue_id":"bd-91","depends_on_id":"bd-86","type":"blocks","created_at":"2025-10-28T17:58:52.05873-07:00","created_by":"stevey"}]}
 {"id":"bd-92","content_hash":"e006b991353a26f949bc3ae4476849ef785f399f6aca866586eb6fa03d243b35","title":"Add TestThreeCloneCollision for regression protection","description":"Add a 3-clone collision test to document behavior and provide regression protection.\n\nPurpose:\n- Verify content convergence regardless of sync order\n- Document the ID non-determinism behavior (IDs may be assigned differently based on sync order)\n- Provide regression protection for multi-way collisions\n\nTest design:\n- 3 clones create same ID with different content\n- Test two different sync orders (A→B→C vs C→A→B)\n- Assert content sets match (ignore specific ID assignments)\n- Add comment explaining ID non-determinism is expected behavior\n\nKnown limitation:\n- Content always converges correctly (all issues present with correct titles)\n- Numeric ID assignments (test-2 vs test-3) depend on sync order\n- This is acceptable if content convergence is the primary goal","acceptance_criteria":"- TestThreeCloneCollision added to beads_twoclone_test.go (or new file)\n- Tests 3 clones with same ID collision\n- Tests two different sync orders\n- Asserts content convergence (all issues present, correct titles)\n- Documents ID non-determinism in test comments\n- Test passes consistently","status":"closed","priority":2,"issue_type":"task","created_at":"2025-10-28T17:59:05.941735-07:00","updated_at":"2025-10-28T18:09:12.717604-07:00","closed_at":"2025-10-28T18:09:12.717604-07:00","dependencies":[{"issue_id":"bd-92","depends_on_id":"bd-90","type":"parent-child","created_at":"2025-10-28T17:59:05.942783-07:00","created_by":"stevey"}]}
 {"id":"bd-93","content_hash":"b86d4c406dd6783a00683a31c8729ea08e846e0ddbc54211e1e3d6dedb96def4","title":"Document 3-clone ID non-determinism in collision resolution","description":"Document the known behavior of 3+ way collision resolution where ID assignments may vary based on sync order, even though content always converges correctly.\n\nUpdates needed:\n- Update bd-86 notes to mark 2-clone case as solved\n- Document 3-clone ID non-determinism as known limitation\n- Add explanation to ADVANCED.md or collision resolution docs\n- Explain why this happens (pairwise hash comparison is deterministic, but multi-way ID allocation uses sync-order dependent counters)\n- Clarify trade-offs: content convergence ✅ vs ID stability ❌\n\nKey points to document:\n- Hash-based resolution is pairwise deterministic\n- Content always converges correctly (all issues present with correct data)\n- Numeric ID assignments in 3+ way collisions depend on sync order\n- This is acceptable for most use cases (content convergence is primary goal)\n- Full determinism would require complex multi-way comparison","acceptance_criteria":"- bd-86 updated with notes about 2-clone solution being complete\n- 3-clone ID non-determinism documented in ADVANCED.md or similar\n- Explanation includes why it happens and trade-offs\n- Links to TestThreeCloneCollision as demonstration\n- Users understand this is expected behavior, not a bug","status":"open","priority":2,"issue_type":"task","created_at":"2025-10-28T17:59:21.93014-07:00","updated_at":"2025-10-28T17:59:21.93014-07:00","dependencies":[{"issue_id":"bd-93","depends_on_id":"bd-90","type":"parent-child","created_at":"2025-10-28T17:59:21.938709-07:00","created_by":"stevey"}]}
-{"id":"bd-94","content_hash":"d7c5637527778c5c835f5e4b6e15fbd51a3476d6749ab3155b8aeac08a8ef339","title":"Fix N-way collision convergence","description":"Epic to fix the N-way collision convergence problem documented in n-way-collision-convergence.md.\n\n## Problem Summary\nThe current collision resolution implementation works correctly for 2-way collisions but does not converge for 3-way (and by extension N-way) collisions. TestThreeCloneCollision demonstrates this with reproducible failures.\n\n## Root Causes Identified\n1. Pairwise resolution doesn't scale - each clone makes local decisions without global context\n2. DetectCollisions modifies state during detection (line 83-86 in collision.go)\n3. No remapping history - can't track transitive remap chains (test-1 → test-2 → test-3)\n4. Import-time resolution is too late - happens after git merge\n\n## Solution Architecture\nReplace pairwise resolution with deterministic global N-way resolution using:\n- Content-addressable identity (content hashing)\n- Global collision resolution (sort all versions by hash)\n- Read-only detection phase (separate from modification)\n- Idempotent imports (content-first matching)\n\n## Success Criteria\n- TestThreeCloneCollision passes without skipping\n- All clones converge to identical content after final pull\n- No data loss (all issues present in all clones)\n- Works for N workers (test with 5+ clones)\n- Idempotent imports (importing same JSONL multiple times is safe)\n\n## Implementation Phases\nSee child issues for detailed breakdown of each phase.","status":"in_progress","priority":0,"issue_type":"epic","created_at":"2025-10-28T18:36:28.234425-07:00","updated_at":"2025-10-28T19:49:52.776357-07:00"}
+{"id":"bd-94","content_hash":"d7c5637527778c5c835f5e4b6e15fbd51a3476d6749ab3155b8aeac08a8ef339","title":"Fix N-way collision convergence","description":"Epic to fix the N-way collision convergence problem documented in n-way-collision-convergence.md.\n\n## Problem Summary\nThe current collision resolution implementation works correctly for 2-way collisions but does not converge for 3-way (and by extension N-way) collisions. TestThreeCloneCollision demonstrates this with reproducible failures.\n\n## Root Causes Identified\n1. Pairwise resolution doesn't scale - each clone makes local decisions without global context\n2. DetectCollisions modifies state during detection (line 83-86 in collision.go)\n3. No remapping history - can't track transitive remap chains (test-1 → test-2 → test-3)\n4. Import-time resolution is too late - happens after git merge\n\n## Solution Architecture\nReplace pairwise resolution with deterministic global N-way resolution using:\n- Content-addressable identity (content hashing)\n- Global collision resolution (sort all versions by hash)\n- Read-only detection phase (separate from modification)\n- Idempotent imports (content-first matching)\n\n## Success Criteria\n- TestThreeCloneCollision passes without skipping\n- All clones converge to identical content after final pull\n- No data loss (all issues present in all clones)\n- Works for N workers (test with 5+ clones)\n- Idempotent imports (importing same JSONL multiple times is safe)\n\n## Implementation Phases\nSee child issues for detailed breakdown of each phase.","status":"closed","priority":0,"issue_type":"epic","created_at":"2025-10-28T18:36:28.234425-07:00","updated_at":"2025-10-29T22:22:33.818711-07:00","closed_at":"2025-10-29T22:22:33.818711-07:00"}
 {"id":"bd-95","content_hash":"12cd30dee3c08ba58d03e4468e6fe261a47d58c3b75397d9f14f38ee644fab6e","title":"Add content-addressable identity to Issue type","description":"## Overview\nPhase 1: Add content hashing to enable global identification of issues regardless of their assigned IDs.\n\n## Current Problem\nThe system identifies issues only by ID (e.g., test-1, test-2). When multiple clones create the same ID with different content, there's no way to identify that these are semantically different issues without comparing all fields.\n\n## Solution\nAdd a ContentHash field to the Issue type that represents the canonical content fingerprint.\n\n## Implementation Tasks\n\n### 1. Add ContentHash field to Issue type\nFile: internal/types/types.go\n```go\ntype Issue struct {\n    ID          string\n    ContentHash string  // SHA256 of canonical content\n    // ... existing fields\n}\n```\n\n### 2. Add content hash computation method\nUse existing hashIssueContent from collision.go:186 as foundation:\n```go\nfunc (i *Issue) ComputeContentHash() string {\n    return hashIssueContent(i)\n}\n```\n\n### 3. Compute hash at creation time\n- Modify CreateIssue to compute and store ContentHash\n- Modify CreateIssues (batch) to compute hashes\n\n### 4. Compute hash at import time  \n- Modify ImportIssues to compute ContentHash for all incoming issues\n- Store hash in database\n\n### 5. Add database column\n- Add migration to add content_hash column to issues table\n- Update SELECT/INSERT statements to include content_hash\n- Index on content_hash for fast lookups\n\n### 6. Populate existing issues\n- Add migration step to compute ContentHash for all existing issues\n- Use hashIssueContent function\n\n## Acceptance Criteria\n- Issue type has ContentHash field\n- Hash is computed automatically at creation time\n- Hash is computed for imported issues\n- Database stores content_hash column\n- All existing issues have non-empty ContentHash\n- Hash is deterministic (same content → same hash)\n- Hash excludes ID, timestamps (only semantic content)\n\n## Files to Modify\n- internal/types/types.go\n- internal/storage/sqlite/sqlite.go (schema, CreateIssue, CreateIssues)\n- internal/storage/sqlite/migrations.go (new migration)\n- internal/importer/importer.go (compute hash during import)\n- cmd/bd/create.go (compute hash at creation)\n\n## Testing\n- Unit test: same content produces same hash\n- Unit test: different content produces different hash  \n- Unit test: hash excludes ID and timestamps\n- Integration test: hash persists in database\n- Migration test: existing issues get hashes populated","status":"closed","priority":1,"issue_type":"task","created_at":"2025-10-28T18:36:44.914967-07:00","updated_at":"2025-10-28T18:57:10.985198-07:00","closed_at":"2025-10-28T18:57:10.985198-07:00","dependencies":[{"issue_id":"bd-95","depends_on_id":"bd-94","type":"parent-child","created_at":"2025-10-28T18:39:20.547325-07:00","created_by":"daemon"}]}
 {"id":"bd-96","content_hash":"49aad5fa2497f7f88fb74d54553825b93c1021ed7db04cfb2e58682699d8dca9","title":"Make DetectCollisions read-only (separate detection from modification)","description":"## Overview\nPhase 2: Separate collision detection from state modification to enable safe, composable collision resolution.\n\n## Current Problem\nDetectCollisions (collision.go:38-111) modifies database state during detection:\n- Line 83-86: Deletes issues when content matches but ID differs\n- This violates separation of concerns\n- Causes race conditions when processing multiple issues\n- Makes contentToDBIssue map stale after first deletion\n- Partial failures leave DB in inconsistent state\n\n## Solution\nMake DetectCollisions purely read-only. Move all modifications to a separate ApplyCollisionResolution function.\n\n## Implementation Tasks\n\n### 1. Add RenameDetail to CollisionResult\nFile: internal/storage/sqlite/collision.go\n```go\ntype CollisionResult struct {\n    ExactMatches []string\n    Collisions   []*CollisionDetail\n    NewIssues    []string\n    Renames      []*RenameDetail  // NEW\n}\n\ntype RenameDetail struct {\n    OldID string        // ID in database\n    NewID string        // ID in incoming\n    Issue *types.Issue  // The issue with new ID\n}\n```\n\n### 2. Remove deletion from DetectCollisions\nReplace lines 83-86:\n```go\n// OLD (DELETE THIS):\nif err := s.DeleteIssue(ctx, dbMatch.ID); err != nil {\n    return nil, fmt.Errorf(\"failed to delete renamed issue...\")\n}\n\n// NEW (ADD THIS):\nresult.Renames = append(result.Renames, \u0026RenameDetail{\n    OldID: dbMatch.ID,\n    NewID: incoming.ID,\n    Issue: incoming,\n})\ncontinue  // Don't mark as NewIssue yet\n```\n\n### 3. Create ApplyCollisionResolution function\nNew function to apply all modifications atomically:\n```go\nfunc ApplyCollisionResolution(ctx context.Context, s *SQLiteStorage,\n    result *CollisionResult, mapping map[string]string) error {\n    \n    // Phase 1: Handle renames (delete old IDs)\n    for _, rename := range result.Renames {\n        if err := s.DeleteIssue(ctx, rename.OldID); err != nil {\n            return fmt.Errorf(\"failed to delete renamed issue %s: %w\", \n                rename.OldID, err)\n        }\n    }\n    \n    // Phase 2: Create new IDs (from mapping)\n    // Phase 3: Update references\n    return nil\n}\n```\n\n### 4. Update callers to use two-phase approach\nFile: internal/importer/importer.go (handleCollisions)\n```go\n// Phase 1: Detect (read-only)\ncollisionResult, err := sqlite.DetectCollisions(ctx, sqliteStore, issues)\n\n// Phase 2: Resolve (compute mapping)\nmapping, err := sqlite.ResolveNWayCollisions(ctx, sqliteStore, collisionResult)\n\n// Phase 3: Apply (modify DB)\nerr = sqlite.ApplyCollisionResolution(ctx, sqliteStore, collisionResult, mapping)\n```\n\n### 5. Update tests\n- Verify DetectCollisions doesn't modify DB\n- Test ApplyCollisionResolution separately\n- Add test for rename detection without modification\n\n## Acceptance Criteria\n- DetectCollisions performs zero writes to database\n- DetectCollisions returns RenameDetail entries for content matches\n- ApplyCollisionResolution handles all modifications\n- All existing tests still pass\n- New test verifies read-only detection\n- contentToDBIssue map stays consistent throughout detection\n\n## Files to Modify\n- internal/storage/sqlite/collision.go (DetectCollisions, new function)\n- internal/importer/importer.go (handleCollisions caller)\n- internal/storage/sqlite/collision_test.go (add tests)\n\n## Testing\n- Unit test: DetectCollisions with content match doesn't delete DB issue\n- Unit test: RenameDetail correctly populated\n- Unit test: ApplyCollisionResolution applies renames\n- Integration test: Full flow still works end-to-end\n\n## Risk Mitigation\nThis is a significant refactor of core collision logic. Recommend:\n1. Add comprehensive tests before modifying\n2. Use feature flag to enable/disable new behavior\n3. Test thoroughly with TestTwoCloneCollision first","status":"closed","priority":1,"issue_type":"task","created_at":"2025-10-28T18:37:09.652326-07:00","updated_at":"2025-10-28T19:08:17.715416-07:00","closed_at":"2025-10-28T19:08:17.715416-07:00","dependencies":[{"issue_id":"bd-96","depends_on_id":"bd-94","type":"parent-child","created_at":"2025-10-28T18:39:20.570276-07:00","created_by":"daemon"},{"issue_id":"bd-96","depends_on_id":"bd-95","type":"blocks","created_at":"2025-10-28T18:39:28.285653-07:00","created_by":"daemon"}]}
 {"id":"bd-97","content_hash":"ecebc4a18d5355bafc88e778ee87365717f894d3590d325a97ecf8b3f763d54d","title":"Implement global N-way collision resolution algorithm","description":"## Overview\nPhase 3: Replace pairwise collision resolution with global N-way resolution that produces deterministic results regardless of sync order.\n\n## Current Problem\nScoreCollisions (collision.go:228) compares issues pairwise:\n```go\ncollision.RemapIncoming = existingHash \u003c incomingHash\n```\n\nThis works for 2-way but fails for 3+ way because:\n- Each clone makes local decisions without global context\n- No guarantee intermediate states are consistent\n- Remapping decisions depend on sync order\n- Can't detect transitive remap chains (test-1 → test-2 → test-3)\n\n## Solution\nImplement global resolution that:\n1. Collects ALL versions of same logical issue\n2. Sorts by content hash (deterministic)\n3. Assigns sequential IDs based on sorted order\n4. All clones converge to same assignments\n\n## Implementation Tasks\n\n### 1. Create ResolveNWayCollisions function\nFile: internal/storage/sqlite/collision.go\n\nReplace ScoreCollisions with:\n```go\n// ResolveNWayCollisions handles N-way collisions deterministically.\n// Groups all versions with same base ID, sorts by content hash,\n// assigns sequential IDs. Returns mapping of old ID → new ID.\nfunc ResolveNWayCollisions(ctx context.Context, s *SQLiteStorage,\n    collisions []*CollisionDetail, incoming []*types.Issue) (map[string]string, error) {\n    \n    if len(collisions) == 0 {\n        return make(map[string]string), nil\n    }\n    \n    // Group by base ID pattern (e.g., test-1, test-2 → base \"test-1\")\n    groups := groupCollisionsByBaseID(collisions)\n    \n    idMapping := make(map[string]string)\n    \n    for baseID, versions := range groups {\n        // 1. Collect all unique versions by content hash\n        uniqueVersions := deduplicateVersionsByContentHash(versions)\n        \n        // 2. Sort by content hash (deterministic!)\n        sort.Slice(uniqueVersions, func(i, j int) bool {\n            return uniqueVersions[i].ContentHash \u003c uniqueVersions[j].ContentHash\n        })\n        \n        // 3. Assign sequential IDs based on sorted order\n        prefix := extractPrefix(baseID)\n        baseNum := extractNumber(baseID)\n        \n        for i, version := range uniqueVersions {\n            targetID := fmt.Sprintf(\"%s-%d\", prefix, baseNum+i)\n            \n            // Map this version to its deterministic ID\n            if version.ID != targetID {\n                idMapping[version.ID] = targetID\n            }\n        }\n    }\n    \n    return idMapping, nil\n}\n```\n\n### 2. Implement helper functions\n\n```go\n// groupCollisionsByBaseID groups collisions by their logical base ID\nfunc groupCollisionsByBaseID(collisions []*CollisionDetail) map[string][]*types.Issue {\n    groups := make(map[string][]*types.Issue)\n    for _, c := range collisions {\n        baseID := c.ID  // All share same ID (that's why they collide)\n        groups[baseID] = append(groups[baseID], c.ExistingIssue, c.IncomingIssue)\n    }\n    return groups\n}\n\n// deduplicateVersionsByContentHash keeps one issue per unique content hash\nfunc deduplicateVersionsByContentHash(issues []*types.Issue) []*types.Issue {\n    seen := make(map[string]*types.Issue)\n    for _, issue := range issues {\n        if _, found := seen[issue.ContentHash]; !found {\n            seen[issue.ContentHash] = issue\n        }\n    }\n    result := make([]*types.Issue, 0, len(seen))\n    for _, issue := range seen {\n        result = append(result, issue)\n    }\n    return result\n}\n```\n\n### 3. Update handleCollisions in importer\nFile: internal/importer/importer.go\n\nReplace ScoreCollisions call with:\n```go\n// OLD:\nif err := sqlite.ScoreCollisions(ctx, sqliteStore, collisionResult.Collisions, allExistingIssues); err != nil {\n    return nil, fmt.Errorf(\"failed to score collisions: %w\", err)\n}\n\n// NEW:\nidMapping, err := sqlite.ResolveNWayCollisions(ctx, sqliteStore, \n    collisionResult.Collisions, issues)\nif err != nil {\n    return nil, fmt.Errorf(\"failed to resolve collisions: %w\", err)\n}\n```\n\n### 4. Update RemapCollisions\nRemapCollisions currently uses collision.RemapIncoming field. Update to use idMapping directly:\n- Remove RemapIncoming logic\n- Use idMapping to determine what to remap\n- Simplify to just apply the computed mapping\n\n### 5. Add comprehensive tests\n\nTest cases:\n1. 3-way collision with different content → 3 sequential IDs\n2. 3-way collision with 2 identical content → 2 IDs (dedupe works)\n3. Sync order independence (A→B→C vs C→A→B produce same result)\n4. Content hash ordering is respected\n5. Works with 5+ clones\n\n## Acceptance Criteria\n- ResolveNWayCollisions implemented and replaces ScoreCollisions\n- Groups all versions of same ID together\n- Deduplicates by content hash\n- Sorts by content hash deterministically\n- Assigns sequential IDs starting from base ID\n- Returns complete mapping (old ID → new ID)\n- All clones converge to same ID assignments\n- Works for arbitrary N-way collisions\n- TestThreeCloneCollision passes (or gets much closer)\n\n## Files to Modify\n- internal/storage/sqlite/collision.go (new function, helpers)\n- internal/importer/importer.go (call new function)\n- internal/storage/sqlite/collision_test.go (comprehensive tests)\n\n## Testing Strategy\n\n### Unit Tests\n- groupCollisionsByBaseID correctly groups\n- deduplicateVersionsByContentHash removes duplicates\n- Sorting by hash is stable and deterministic\n- Sequential ID assignment is correct\n\n### Integration Tests\n- 3-way collision resolves to 3 issues\n- Sync order doesn't affect final IDs\n- Content hash ordering determines winner\n\n### Property Tests\n- For any N clones with same content, all converge to same IDs\n- Idempotent: running resolution twice produces same result\n\n## Dependencies\n- Requires bd-95 (ContentHash field) to be completed first\n- Requires bd-96 (read-only detection) for clean integration\n\n## Notes\nThis is the core algorithm that enables convergence. The key insight:\n**Sort by content hash globally, not pairwise comparison.**","status":"closed","priority":1,"issue_type":"task","created_at":"2025-10-28T18:37:42.85616-07:00","updated_at":"2025-10-28T20:03:26.675257-07:00","closed_at":"2025-10-28T20:03:26.675257-07:00","dependencies":[{"issue_id":"bd-97","depends_on_id":"bd-94","type":"parent-child","created_at":"2025-10-28T18:39:20.593102-07:00","created_by":"daemon"},{"issue_id":"bd-97","depends_on_id":"bd-95","type":"blocks","created_at":"2025-10-28T18:39:28.30886-07:00","created_by":"daemon"},{"issue_id":"bd-97","depends_on_id":"bd-96","type":"blocks","created_at":"2025-10-28T18:39:28.336312-07:00","created_by":"daemon"}]}