Regulatory Templates Gate2

Regulatory Templates - Gate 2: Technical Validation

Overview

This sub-skill executes Gate 2 of the regulatory template workflow: validating uncertain mappings from Gate 1 and confirming all field specifications through testing.

Parent skill: regulatory-templates

Prerequisites:

• Gate 1 PASSED
• Context object with Gate 1 results

Output: Validated mappings with test results and validation rules

Foundational Principle

Validation is the checkpoint that prevents incorrect mappings from reaching production.

Gate 2 is the quality gate between analysis (Gate 1) and implementation (Gate 3):

• All uncertainties resolved: Gate 1 analysis ≠ Gate 2 validation. MEDIUM/LOW uncertainties often hide critical issues
• 100% mandatory validation: 95% = 5% of mandatory data could be wrong in BACEN submission
• >90% test pass rate: Test data reveals transformation bugs, data type mismatches, edge cases
• Confirmed mappings: Prevents Gate 3 from generating templates based on assumptions
• Validation rules defined: Gate 3 needs explicit validation logic for template generation

Skipping validation in Gate 2 means:

• Gate 1 assumptions become Gate 3 implementation (no verification layer)
• Uncertainties propagate to production (BACEN submission failures)
• Low-confidence mappings generate incorrect templates (compliance violations)
• No test data validation = edge cases break in production

Gate 2 is not redundant - it's the firewall between analysis and implementation.

When to Use

Called by: regulatory-templates skill after Gate 1 passes

Purpose: Resolve uncertainties, validate field mappings, test transformations, define validation rules

NO EXCEPTIONS - Validation Requirements Are Mandatory

Gate 2 validation requirements have ZERO exceptions. This is the quality firewall before template generation.

Common Pressures You Must Resist

Validation Requirements (Non-Negotiable)

All Uncertainties Resolved:

• ✅ REQUIRED: Resolve ALL Gate 1 uncertainties (CRITICAL + MEDIUM + LOW)
• ❌ FORBIDDEN: "Fix critical only", "Skip low-priority items"
• Why: MEDIUM/LOW uncertainties often reveal systemic issues, cascade to mandatory failures

Test Data Validation:

• ✅ REQUIRED: Test pass rate >90%
• ❌ FORBIDDEN: "88% is close enough", "Skip testing, looks correct"
• Why: Test data reveals transformation bugs, data type mismatches, edge cases

Mandatory Field Validation:

• ✅ REQUIRED: 100% mandatory fields validated
• ❌ FORBIDDEN: "95% is outstanding", "Edge cases don't matter"
• Why: Each 1% gap = potential BACEN submission failure on mandatory data

Alternative Mappings:

• ✅ REQUIRED: Find alternatives for ALL rejected fields
• ❌ FORBIDDEN: "Use workarounds", "Keep original with patches"
• Why: Rejected mappings fail validation for a reason - workarounds bypass the firewall

The Bottom Line

Partial validation = no validation.

Gate 2 exists to catch what Gate 1 missed. Lowering thresholds or skipping validation defeats the purpose. Every PASS criterion exists because production incidents occurred without it.

If you're tempted to skip ANY validation, ask yourself: Am I willing to defend this shortcut during a BACEN audit?

Rationalization Table - Know the Excuses

Every rationalization below has been used to justify skipping validation. ALL are invalid.

If You Find Yourself Making These Excuses

STOP. You are rationalizing.

The validation exists to prevent these exact thoughts from allowing errors into production. If validation seems "redundant," that's evidence it's working - catching what analysis missed.

Gate 2 Process

Check for Template-Specific Validation Rules

Check for template-specific sub-skill at skills/regulatory-{template}/SKILL.md containing:

• Validation rules (VR001, VR002...), business rules (BR001, BR002...)
• Format rules, test data with expected outputs

Agent Dispatch with Gate 1 Context

Dispatch: Task(subagent_type: "finops-analyzer", model: "opus")

CRITICAL: ⚠️ DO NOT MAKE MCP API CALLS - use Gate 1 context ONLY

Prompt structure:

Output: Field resolutions + validation rules + cross-field logic + test data

Validation Process

⚠️ All validation uses Gate 1 context ONLY - no MCP API calls.

1. Field Validation

Per uncertain field: field_code, original_doubt, validation_steps (5), resolution (confirmed/rejected), transformation, test_data (input/expected/actual/status)

2. Validation Rules Definition

3. Test Results Documentation

Per test: field, test_name, input, transformation, output, expected, passed (true/false)

Example: Field 001 CNPJ extraction: "12345678000190" → slice:':8' → "12345678" ✓

Capture Gate 2 Response

Merge with Gate 1: validated_mappings[], validation_rules[], all_uncertainties_resolved, test_summary (total/passed/failed/success_rate)

Red Flags - STOP Immediately

If you catch yourself thinking ANY of these, STOP and re-read the NO EXCEPTIONS section:

Partial Resolution

• "Resolve critical only, skip MEDIUM/LOW"
• "Fix most uncertainties, good enough"
• "ALL is unrealistic, most is pragmatic"

Threshold Degradation

• "88% is close to 90%"
• "95% mandatory validation is outstanding"
• "Close enough to pass"
• "The gap isn't material"

Skip Validation Steps

• "Validation dashboard is redundant"
• "Mappings look correct visually"
• "Testing is busywork"
• "We'll catch issues in Gate 3"

Workaround Thinking

• "Use workarounds for rejected fields"
• "Patch it in Gate 3"
• "Fix in next sprint"
• "This is an edge case"

Justification Language

• "Being pragmatic"
• "Following spirit not letter"
• "Outstanding is good enough"
• "Rarely finds issues anyway"
• "Experience says this is fine"

If You See These Red Flags

1. Acknowledge the rationalization ("I'm trying to skip LOW uncertainties")
2. Read the NO EXCEPTIONS section (understand why ALL means ALL)
3. Read the Rationalization Table (see your exact excuse refuted)
4. Meet the threshold completely (100%, >90%, ALL)

Validation thresholds are binary gates, not aspirational goals.

Pass/Fail Criteria

PASS Criteria

• ✅ All Gate 1 uncertainties resolved (confirmed or alternatives found)
• ✅ Test data validates successfully (>90% pass rate)
• ✅ No new Critical/High issues
• ✅ All mandatory fields have confirmed mappings
• ✅ Validation rules defined for all critical fields

FAIL Criteria

• ❌ Uncertainties remain unresolved
• ❌ Test failures on mandatory fields
• ❌ Cannot find alternative mappings for rejected fields
• ❌ Data type mismatches that can't be transformed
• ❌ Mandatory fields validation < 100%

Mandatory Fields Final Validation

CRITICAL: Execute before Gate 2 completion

Per mandatory field, check:

• mapped (in gate1.field_mappings)
• confidence_ok (≥80%)
• validated (in gate2.validated_mappings)
• tested (in gate2.test_results)
• transformation_ok (works correctly)

Status: All checks PASS → field PASS; any FAIL → field FAIL

Gate 2 Pass Condition: all_mandatory_fields_valid == true required. Coverage must be 100%.

State Tracking

PASS: SKILL: regulatory-templates-gate2 | GATE: 2 | STATUS: PASSED | RESOLVED: {n} uncertainties | RULES: {n} defined | TESTS: {passed}/{total} | NEXT: → Gate 3

FAIL: SKILL: regulatory-templates-gate2 | GATE: 2 | STATUS: FAILED | UNRESOLVED: {n} | TEST_FAILURES: {n} | BLOCKERS: {description}

Technical Validation Checklist

Common Validation Patterns

Output to Parent Skill

Return: gate2_passed, gate2_context (merged), all_uncertainties_resolved, validation_rules_count, test_success_rate, next_action (proceed_to_gate3 | fix_validations_and_retry)