SWAP_EXECUTION

Swap Execution Skill

Production-ready swap execution with comprehensive safety checks, audit trails, and rollback capabilities.

When This Skill Activates

• Processing incoming swap requests
• Validating swap compliance against Tier 1/2/3 constraints
• Executing validated swaps
• Rolling back problematic swaps
• Investigating swap failures or conflicts
• Escalating complex swap scenarios

Overview

The swap execution system transforms resident/faculty schedule swap requests into safe, auditable database transactions. It operates in 5 distinct phases with safety checks at each stage.

The Five Phases

1. Intake - Parse and normalize the request
2. Safety Checks - Validate against constraints and conflicts
3. Execution - Update database with full audit trail
4. Verification - Confirm expected state
5. Monitoring - Track for 24h rollback window

System Architecture

┌──────────────┐
│ Swap Request │ (Resident/Faculty initiates)
└──────┬───────┘
       │
       v
┌──────────────────────────────────────┐
│ PHASE 1: Request Intake              │
│ - Parse request JSON                 │
│ - Normalize dates/UUIDs              │
│ - Record requestor + timestamp       │
└──────┬───────────────────────────────┘
       │
       v
┌──────────────────────────────────────┐
│ PHASE 2: Safety Checks               │
│ - Constraint validation (3 tiers)   │
│ - Conflict detection                 │
│ - Resilience impact assessment       │
└──────┬───────────────────────────────┘
       │
       ├─[REJECT]─> Record reason + notify requestor
       │
       ├─[FLAG]───> Escalate to faculty for approval
       │
       v [PROCEED]
┌──────────────────────────────────────┐
│ PHASE 3: Execution                   │
│ - Create SwapRecord (EXECUTED)       │
│ - Update assignments                 │
│ - Update call cascade (Fri/Sat)      │
│ - Commit transaction                 │
└──────┬───────────────────────────────┘
       │
       v
┌──────────────────────────────────────┐
│ PHASE 4: Verification                │
│ - Confirm assignments updated        │
│ - Re-run constraint checks           │
│ - Notify affected parties            │
└──────┬───────────────────────────────┘
       │
       v
┌──────────────────────────────────────┐
│ PHASE 5: Monitoring                  │
│ - 24h rollback window active         │
│ - Watch for complaints/conflicts     │
│ - Auto-rollback on critical failures │
└──────────────────────────────────────┘

Key Files

Workflows

Each phase has a dedicated workflow document:

1. swap-request-intake.md - Receive and parse swap requests
2. safety-checks.md - Validate swap doesn't break rules
3. audit-trail.md - Log all swap decisions
4. rollback-procedures.md - Undo if something goes wrong

Reference Documentation

• swap-failure-modes.md - Common failure patterns with detection
• escalation-matrix.md - When to involve faculty/coordinator

Output Format

When processing a swap request, always report in this format:

## Swap Execution Report

**Request ID:** [UUID]
**Type:** [ONE_TO_ONE / ABSORB]
**Status:** [PENDING / EXECUTED / REJECTED / ROLLED_BACK]
**Requested By:** [Person Name/ID]
**Requested At:** [ISO timestamp]

### Parties
- **Source Faculty:** [Name] (ID: [UUID])
  - Week: [YYYY-MM-DD]
  - Current Hours: [X/80]
- **Target Faculty:** [Name] (ID: [UUID])
  - Week: [YYYY-MM-DD or null for ABSORB]
  - Current Hours: [Y/80]

### Safety Check Results
- **Tier 1 (Hard Constraints):** [PASS / FAIL]
  - 80-Hour Rule: [status]
  - 1-in-7 Rule: [status]
  - Supervision Ratios: [status]
- **Tier 2 (Soft Constraints):** [PASS / WARN]
  - Back-to-back conflicts: [details]
  - Coverage gaps: [details]
- **Tier 3 (Resilience):** [PASS / WARN]
  - Utilization impact: [before/after]
  - N-1 contingency: [status]

### Decision
- **Outcome:** [EXECUTED / REJECTED / ESCALATED]
- **Rationale:** [Explanation]
- **Executed At:** [timestamp if executed]
- **Swap ID:** [UUID if executed]

### Audit Trail
- Logged to: `swap_records` table
- Transaction ID: [DB transaction ID]
- Rollback eligible until: [timestamp + 24h]

### Next Steps
1. [Action item]
2. [Action item]

Error Handling

Validation Failures

Always provide specific, actionable error messages:

Bad:

"Swap failed validation"

Good:

"Swap rejected: Target faculty (Dr. Smith) has blocking absence
(TDY) from 2025-01-14 to 2025-01-18, overlapping with source
week 2025-01-15. Suggest alternative weeks: 2025-01-22, 2025-01-29."

Execution Failures

On execution failure:

1. Rollback database transaction immediately
2. Record failure reason in audit log
3. Notify requestor with explanation
4. Log for post-mortem (don't leak sensitive data)

Example:

try:
    # Execute swap
    result = executor.execute_swap(...)
except IntegrityError as e:
    logger.error(f"Swap {swap_id} failed: database integrity violation",
                 exc_info=True)
    return {
        "success": False,
        "error_code": "INTEGRITY_VIOLATION",
        "message": "Schedule conflict detected. Please refresh and try again."
    }

Escalation Triggers

Automatically escalate to human when:

See escalation-matrix.md for full decision tree.

Common Scenarios

Scenario 1: Simple One-to-One Swap (AUTO-APPROVED)

Request:

{
  "source_faculty_id": "uuid-a",
  "source_week": "2025-02-03",
  "target_faculty_id": "uuid-b",
  "target_week": "2025-02-10",
  "swap_type": "one_to_one",
  "reason": "Conference attendance"
}

Process:

1. Intake: Normalize dates, validate UUIDs exist
2. Safety: All checks PASS
3. Execution: Update assignments, create SwapRecord
4. Verification: Confirm state
5. Monitoring: 24h rollback window starts

Outcome: AUTO-APPROVED, executed immediately

Scenario 2: Absorb with Back-to-Back Conflict (REJECTED)

Request:

{
  "source_faculty_id": "uuid-a",
  "source_week": "2025-02-03",
  "target_faculty_id": "uuid-b",
  "target_week": null,
  "swap_type": "absorb",
  "reason": "PTO"
}

Safety Check Result:

ERROR: Target faculty already has FMIT weeks 2025-01-27 and 2025-02-10.
Taking 2025-02-03 would create 3 consecutive weeks (burnout risk).

Outcome: REJECTED, suggest alternative faculty or weeks

Scenario 3: Fellowship Interview (ESCALATED)

Request:

{
  "source_faculty_id": "uuid-a",
  "source_week": "2025-03-17",
  "target_faculty_id": "uuid-b",
  "target_week": "2025-03-24",
  "swap_type": "one_to_one",
  "reason": "fellowship_interview"
}

Safety Check Result:

PASS (all constraints satisfied)

Escalation Trigger:

Reason code "fellowship_interview" requires PD approval
for fairness and confidentiality.

Outcome: ESCALATED to Program Director for approval

Testing Checklist

Before deploying swap execution changes:

• Unit tests for SwapExecutor pass
• Unit tests for SwapValidationService pass
• Integration tests with actual DB pass
• Rollback window enforcement tested
• Constraint validation tested for all tiers
• Audit trail creation verified
• Notification delivery tested
• Escalation routing tested

# Run swap-specific tests
cd backend
pytest tests/test_swap_*.py -v

# Run integration tests
pytest tests/integration/test_swap_workflow.py -v

Rollback Safety

Critical: Swaps can be rolled back within 24 hours.

• Purpose: Catch mistakes, resolve disputes, fix data errors
• Window: 24 hours from executed_at timestamp
• Process: See rollback-procedures.md
• Constraints: Cannot rollback if outside window or if status != EXECUTED

# Check if rollback eligible
executor = SwapExecutor(db)
can_rollback = executor.can_rollback(swap_id)

if can_rollback:
    result = executor.rollback_swap(
        swap_id=swap_id,
        reason="Resident reported schedule conflict",
        rolled_back_by_id=current_user_id
    )

Performance Considerations

• N+1 Query Prevention: Use selectinload() for assignments
• Batch Operations: Group related updates in single transaction
• Validation Caching: Cache constraint results for repeated checks
• Async Notifications: Send emails/Slack messages asynchronously (Celery)

See backend/app/services/swap_executor.py for N+1 optimization examples.

Security Considerations

• Authorization: Only requestor or admins can rollback
• Audit Trail: All actions logged with user ID + timestamp
• Data Sanitization: No PII in error messages
• Rate Limiting: Max 10 swap requests per user per day
• Validation: All inputs sanitized via Pydantic schemas

For detailed phase documentation, see the Workflows/ directory. For failure patterns and escalation rules, see the Reference/ directory.