SNOWBALLHARE PLAYBOOK
SnowballHare Graham-Inspired Margin of Safety Score
A repeatable framework for deciding whether a cheap-looking stock deserves further research. It combines balance-sheet resilience, normalized earning power, cash conversion, valuation discount, and bear-case protection.
The six scoring inputs
| Input | Weight | Scoring basis | Main evidence |
|---|---|---|---|
| Balance-sheet safety | 25% | Net cash, liquidity, debt maturity, and solvency | Latest 10-K / 10-Q |
| Normalized earnings | 20% | Five-to-ten-year normalized earnings power | Earnings history |
| Cash conversion | 15% | Free cash flow relative to net income | Cash-flow statement |
| Valuation discount | 20% | Price compared with conservative intrinsic value | Valuation model |
| Downside protection | 15% | Bear-case value, assets, and balance-sheet support | Scenario analysis |
| Catalyst patience | 5% | Whether value can compound without a forced catalyst | Business evidence |
Total score = Σ (factor score from 0–100 × factor weight).
Score each factor only after documenting its evidence and bear-case assumptions. A missing datapoint is not a neutral 50; it is an uncertainty that may justify a lower score or no decision.
Score bands and actions
| Score | Band | Action |
|---|---|---|
| 85–100 | Research candidate | Continue diligence; confirm that price still offers a margin of safety. |
| 70–84 | Watchlist | Identify the evidence needed to resolve valuation, leverage, or normalization risk. |
| 50–69 | Value-trap risk | Do not treat statistical cheapness as protection. |
| 0–49 | Reject | Downside protection is insufficient for this framework. |
A high score is not a Buy rating. It indicates that a candidate merits deeper security analysis. Position size, opportunity cost, portfolio fit, and current price remain separate decisions.
Hard invalidation rules
- Debt refinancing becomes unavailable or materially more expensive.
- Free cash flow persistently fails to support reported earnings.
- The bear case implies dilution, covenant pressure, or permanent impairment.
- Receivables or inventory cannot support their recorded value.
- The apparent discount reflects structural decline rather than temporary unpopularity.
- Governance or related-party behavior prevents minority investors from realizing value.
A hard invalidation overrides the weighted score until new evidence resolves it.
Worked score: hypothetical industrial company
| Factor | Score | Weight | Contribution |
|---|---|---|---|
| Balance-sheet safety | 80 | 25% | 20.0 |
| Normalized earnings | 70 | 20% | 14.0 |
| Cash conversion | 85 | 15% | 12.75 |
| Valuation discount | 75 | 20% | 15.0 |
| Downside protection | 65 | 15% | 9.75 |
| Catalyst patience | 60 | 5% | 3.0 |
| Total | 100% | 74.5 |
The result is a watchlist, not a purchase. Cash conversion and liquidity are sound, but the 65 downside score shows insufficient protection if the cycle weakens. The next step is to test lower volumes, margins, asset recoverability, and refinancing costs.
Evidence workflow
- Read the latest 10-K and 10-Q; map cash, debt, leases, covenants, and maturities.
- Build a five-to-ten-year earnings series and replace peak results with normalized assumptions.
- Reconcile net income with operating cash flow and maintenance capital expenditure.
- Estimate a conservative intrinsic-value range and a separate bear-case value.
- Apply haircuts to questionable receivables, inventory, and other assets.
- Record the evidence that would invalidate each factor score and set a review date.
Limitations
The model can under-score asset-light compounders and over-score companies whose accounting assets are economically weak. Scores also create false precision when evidence is sparse. It is less suitable for early-stage technology, biotech, negative-book-value companies, and businesses valued primarily through network effects or unrecorded intellectual property.
It does not replace valuation judgment, accounting analysis, governance review, portfolio construction, or consideration of alternatives. Use it as a structured research filter, never as an automated recommendation.