Financial Model Validator

Purpose

1. Validates /fm build outputs after every build operation
2. Checks format completeness — expected files exist and are non-empty
3. Runs metric sanity checks — no negative ARR, reasonable LTV:CAC ratios, cash runway logic
4. Verifies internal reconciliation — P&L net income consistent with cash flow changes
5. Flags missing or anomalous months in time series data
6. Reports validation results to the invoking persona agent

Trigger

Event Type: Post /fm build execution

Source: /fm build <scenario> --format <format> command completion

Blocking: Yes (validation must pass before outputs are delivered to user)

Timeout: 15 seconds

Trigger Condition: Any /fm build command that generates output files

Behavior

When Triggered

1. Receives build output context:

   • Scenario ID and name
   • Format(s) built (xlsx, formula, json, csv, all)
   • Output file paths
   • Build summary metrics (breakeven, M60 ARR, M60 cash, peak burn)

2. Format Completeness Check:

   • Verify each expected output file exists
   • Verify file size > 0 bytes
   • For XLSX: verify expected sheet count (10 for static, 10 for formula)
   • For JSON: verify valid JSON with required top-level keys (metadata, summary, monthly_data)
   • For CSV: verify header row and expected column count

3. Metric Sanity Checks:

   Metric	Valid Range	Flag If
   M60 ARR	> $0	Negative or zero
   Breakeven Month	M1-M60	> M60 (never breaks even)
   Peak Monthly Burn	< $0	Positive (never burns cash)
   LTV:CAC (enterprise)	1.0x - 20.0x	< 1.0x or > 20.0x
   LTV:CAC (individual)	0.5x - 10.0x	< 0.5x or > 10.0x
   Customer Count (M60)	> 0	Zero or negative
   Cash Position (M60)	Any	Warn if negative
   Expense Ratios Sum	< 100% at scale	> 100%

4. Reconciliation Checks:

   • Revenue = sum(tier_customers x tier_price x 12) within 1%
   • Cash flow: opening + inflows - outflows = closing within $1
   • Customer math: starting + new - churned = ending (exact)
   • Month-over-month continuity: no gaps in time series

5. Result Classification:

   Result	Action
   PASS	All checks pass — deliver outputs to user
   WARN	Sanity check flags present — deliver with warnings
   FAIL	Reconciliation error or missing output — block delivery, report error

Output Format

VALIDATION: financial-model-validator
Scenario: {scenario_id}
Format: {format}
Status: PASS | WARN | FAIL

Format Check: ✅ 4/4 files present
Sanity Check: ✅ 8/8 metrics in range
Reconciliation: ✅ 4/4 balances match

Warnings: (if any)
  - M60 cash position is negative (-$234,000) — review burn rate

Errors: (if any)
  - Revenue reconciliation gap: 2.3% (threshold: 1%)

Configuration

Parameter	Default	Description
revenue_tolerance	0.01 (1%)	Max revenue reconciliation gap
cash_tolerance	1.0 ($1)	Max cash balance discrepancy
warn_negative_cash	true	Warn on negative M60 cash
block_on_reconciliation	true	FAIL on reconciliation errors

Implementation

def validate_fm_build(scenario_id, format, output_dir):
    """Post-build validation for /fm outputs."""
    results = {"format": [], "sanity": [], "reconciliation": []}

    # 1. Format completeness
    expected_files = get_expected_files(scenario_id, format)
    for f in expected_files:
        path = os.path.join(output_dir, f)
        if not os.path.exists(path):
            results["format"].append(("FAIL", f"Missing: {f}"))
        elif os.path.getsize(path) == 0:
            results["format"].append(("FAIL", f"Empty: {f}"))
        else:
            results["format"].append(("PASS", f"OK: {f}"))

    # 2. Sanity checks (from JSON summary)
    if "json" in format or format == "all":
        summary = load_json_summary(scenario_id, output_dir)
        check_metric_ranges(summary, results)

    # 3. Reconciliation (from CSV monthly data)
    if "csv" in format or format == "all":
        monthly = load_csv_data(scenario_id, output_dir)
        check_reconciliation(monthly, results)

    return classify_results(results)

Related

• Engine: coditect_fm.py (ADR-177)
• Command: /fm (financial-model.md)
• Skill: financial-model-personas/SKILL.md
• Agents: cfo.md, fpa-analyst.md, finance-controller.md, financial-analyst.md
• Workflow: fm-orchestration-workflow.md

---

Track: N.6.13.4.1 ADR: ADR-177 (Database-Driven Financial Model Engine)