Source audit report — skill files vs. sources policy

Date: 2026-05-24 Auditor: AlphaSteve audit subagent Policy version: sources-policy v1

Summary

The skill kit is, with very few exceptions, methodologically sound but systematically uncited. Across all 51 skill files audited, I found zero in-text citations in the policy's <span class="tier-t1" title="...">T1</span> / <span class="tier-t2" title="...">T2</span> format. The frameworks themselves are by and large faithful distillations of the canonical literature the policy whitelists (Graham, Klarman, Marks, Greenwald, Helmer, Damodaran, Mauboussin, Chancellor, Porter, Christensen, Sloan, Kelly, Greenblatt, Buffett, Munger). The problem is not that the agent reasons from Wikipedia; it is that the agent reasons from the right books but never says so. Under the new policy this is a violation by omission — claims read as freehand assertion when they are in fact restatements of named authorities.

Estimated revision scope: large but not deep. Roughly 80% of remediation is mechanical: append a ## Sources block to each file naming the canonical works, then insert ~5-15 inline <span class="tier-t2" title="...">T2</span> markers per file at load-bearing assertions. The remaining 20% requires genuine rewriting: (a) quantitative thresholds and percentage bands that were stated as fact without derivation, (b) historical "examples" lists that are cultural-memory facts the agent did not look up, and (c) a handful of broad generalizations that need either a citation or a hedge to "our interpretation."

Top 3 highest-priority rewrites:

1. 02-philosophy-deep-value.md — this is the doctrine. It quotes Klarman's "uncomfortable but right" framing and uses Marks's variant-perception logic and Graham's margin-of-safety language throughout, but never cites any of them. Because it is the file most often loaded into other reasoning, its uncited assertions propagate. Highest leverage to fix.

2. position-sizing-kelly.md — contains a load-bearing quantitative claim (the simplified f* = edge / odds form, the "25-50% of full Kelly" prescription, and several portfolio limit numbers) that need either Kelly (1956), Thorp's adaptations, or Mauboussin / Poundstone's Fortune's Formula as a source. Currently the math reads as authoritative without authority. Also: the simplified continuous-distribution formula f* = (expected return − risk-free rate) / variance is stated incorrectly as equivalent to the discrete Kelly — it is actually the Merton/Markowitz mean-variance optimum, only equal to Kelly under log-utility with normal returns. This needs surgery, not just citation.

3. base-rates.md — every quantitative base rate stated is offered without a study citation: M&A failure rates ("50-70% destroy value"), turnaround success ("20-30%"), IPO underperformance, spin-off outperformance "(Greenblatt)" (named but not cited), the "Buddenbrooks effect" on family successions. These numbers are the heart of the file. They appear directionally correct and consistent with the literature (McKinsey M&A studies, KPMG turnaround studies, Ritter on IPOs, Greenblatt 1999) — but in a deep-value kit whose entire epistemic claim is "we do the work," uncited base rates fail the test the kit applies to others.

Phase 1 — Enumeration

01-Soul/

02-Business-Quality/Fundamentals/

02-Business-Quality/Moats/

02-Business-Quality/Bottlenecks/

02-Business-Quality/Management/

03-Valuation/

04-Market-Read/

05-Environment/

06-Risk/

07-Sectors/

08-Frameworks/

Phase 2 — Deep audit

02-philosophy-deep-value.md

• Supportable, needs citation:
  • "Buy a dollar of value for fifty cents" — Graham, Security Analysis (1934) and The Intelligent Investor ch. 20; Klarman, Margin of Safety (1991), ch. 4-5
  • Margin of safety scaling with uncertainty (25% utility / 50% cyclical / 70% turnaround) — directional logic is from Klarman ch. 4 and Marks The Most Important Thing ch. 5; the specific percentage bands are AlphaSteve interpretation
  • Volatility-is-not-risk distinction — Klarman ch. 7 and Marks ch. 5 verbatim; this needs <span class="tier-t2" title="Marks, Most Important Thing, ch. 5">T2</span>
  • "Most likely cause of loss is not a black swan; it is that you missed something" — Klarman echo
  • "Cigar butts" and "earnings power" framing — Graham (cigar butts) and Greenwald Value Investing (1999/2001) (EPV)
  • "Being early is indistinguishable from being wrong" — Marks (multiple memos); citable to The Most Important Thing
  • The "twenty punches on the punchcard" allusion → Buffett, attributed in multiple Berkshire letters
• Freehand reasoning, needs rewrite:
  • The numbered "deep-value hierarchy of opportunity" (six categories) reads as taxonomic authority but is AlphaSteve synthesis — needs explicit "our framing, derived from..."
  • "The most likely cause of loss is not a black swan; it is that you missed something" is plausible but is presented as an axiom; either cite Klarman or hedge
  • "Most 'value' you will see in the wild is value trap" — accurate folk wisdom in deep value; should cite Damodaran Investment Fables (2004) or Greenwald ch. 6
• Unsupportable, needs removal: none — the file is doctrinally clean. The work is citation, not surgery.
• Quantitative thresholds needing source: the 25% / 50% / 70% MoS ladder. No primary author specifies these exact bands. Either label as "our calibration" or cite directional analogs in Klarman.
• Priority for revision: P1 (highest)
• Recommended canonical sources: Graham Security Analysis, Intelligent Investor; Klarman Margin of Safety; Marks The Most Important Thing (esp. ch. 5, 13); Greenwald Value Investing; Buffett shareholder letters (1977 on margin of safety; 1989 on cigar butts)

earnings-power-value-greenwald.md

• Supportable, needs citation:
  • The EPV formula, normalization steps, and growth-value framing are direct from Greenwald, Kahn, Sonkin, van Biema Value Investing: From Graham to Buffett and Beyond (Wiley, 2001), ch. 5-6 — every section needs <span class="tier-t2" title="Greenwald et al., Value Investing, ch. X">T2</span>
  • WACC build (CAPM with industry beta, ERP 4.5-6%) — Damodaran Investment Valuation (3rd ed., 2012); the 4.5-6% ERP cite goes to Damodaran's annual ERP updates (NYU Stern dataset)
  • "Growth value requires ROIC > WACC" — Greenwald and Damodaran both; also Mauboussin & Rappaport Expectations Investing
  • Buffett's "owner earnings" framework underlies the normalization — Berkshire 1986 letter
• Freehand reasoning, needs rewrite:
  • The MoS thresholds (25-35% stable / 40-50% moderate / 50-70% cyclical) at end of Step 4 are AlphaSteve calibration, not Greenwald's; label as such
  • Greenwald's actual stated discount-rate range ("8-10% for stable businesses") — paraphrase reasonable, but cite the page
  • "Common errors" list is sound but reads as authoritative; can be presented as "frequent errors in EPV practice (see Greenwald ch. 5)"
• Unsupportable, needs removal:
  • The "US 10Y has hovered in a wide band for 2025-2026; use a multi-year average" is fine guidance but the specific window reference is a freehand date stamp; either cite the FRED series or hedge
  • "Size premium for small/mid caps — empirically debated but typically 1-3 percentage points" — this is contested in the literature (Fama-French SMB has shrunk; some studies show it gone). Either cite Damodaran's discussion of size premium decay or qualify
• Quantitative thresholds needing source: the WACC table is more useful with Damodaran ranges than as-is; the 25-35/40-50/50-70 MoS bands need explicit "our calibration."
• Priority for revision: P1
• Recommended canonical sources: Greenwald et al. Value Investing (Wiley, 2001) — esp. ch. 5-6; Damodaran Investment Valuation; Damodaran NYU ERP dataset; Buffett 1986 letter (owner earnings); Mauboussin & Rappaport Expectations Investing

margin-of-safety-pricing.md

• Supportable, needs citation:
  • Margin-of-safety policy: Klarman Margin of Safety (1991) ch. 4; Graham Intelligent Investor ch. 20
  • "Average down only with thesis intact" — Klarman ch. 6; Buffett (multiple letters)
  • Sell discipline tied to central value — Klarman ch. 17; Marks The Most Important Thing ch. 12
  • The patience reservoir / cash-as-optionality — Klarman ch. 11; Marks memo "It's Not Easy" (Sep 2015)
  • "Being early is indistinguishable from being wrong, until it isn't" — Marks (memo, multiple)
• Freehand reasoning, needs rewrite:
  • The MoS table (25-35% / 40-50% / 50-70% / 70%+) duplicates the EPV file; same calibration disclaimer
  • The position-size tiering (Core 5-8%, Mid 2-4%, Probe 0.5-1.5%, single 10-15% cap, 10-30 holdings) is AlphaSteve's portfolio design — needs to be labeled as such; the Kelly file is the closest thing to a derivation but the connection is hand-waved
  • "Trim at 80% of central value / exit at central value / 15-25% above" — the specific bands have no canonical source. Label as our practice
• Unsupportable, needs removal:
  • Nothing structural; this file is internally consistent and aligned with Klarman/Marks doctrine. The issue is citation density and labeling our calibrations as ours.
• Quantitative thresholds needing source: every numeric band in the file (MoS tiers, position sizes, single-name limits, trim/exit points, the 10-30 position count). None are derivable from a single named author. They are reasonable practitioner conventions and should be labeled as AlphaSteve calibration, with directional pointers to Klarman / Marks.
• Priority for revision: P1
• Recommended canonical sources: Klarman Margin of Safety (esp. ch. 4, 11, 17); Marks The Most Important Thing (esp. ch. 5, 12, 13); Graham Intelligent Investor ch. 20; Buffett 1989 and 2008 letters

dcf-and-reverse-dcf.md

• Supportable, needs citation:
  • Standard DCF build, terminal value formulas, sensitivity analysis — Damodaran Investment Valuation (3rd ed., 2012) and Damodaran on Valuation (2006)
  • Reverse-DCF / market-implied expectations — Mauboussin & Rappaport Expectations Investing (2001) is the canonical source and is uncredited despite being the entire reverse-DCF section
  • Terminal-value cap at long-run GDP growth — Damodaran (multiple)
  • "Growth requires ROIC > WACC" — Damodaran, Mauboussin (multiple), Greenwald
  • "Terminal value typically 50-80% of DCF" — Damodaran's published research on TV share of equity value
  • Marks "second-level" framing for reverse DCF — Marks Most Important Thing ch. 1
• Freehand reasoning, needs rewrite:
  • The WACC range table by business type (6-8% utility / 8-10% IG industrial / etc.) is reasonable practitioner convention but not derived from any specific source. Either match Damodaran's industry tables (NYU Stern, updated annually) or label as AlphaSteve heuristic
  • "Real cash flows in 2040 should be discounted by 2040's rates, not today's" — accurate in theory; almost never done in practice. Either cite a discussion (Fama-French on stochastic discount factors? Damodaran on TV sensitivity?) or call this an "ideal" the file does not actually implement
  • "Margins expand for explicit period — often unjustified, base rate is stable or compress" — true and important; cite Mauboussin & Wang The Base Rate Book (Credit Suisse 2016)
• Unsupportable, needs removal: nothing structural; this file is competent.
• Quantitative thresholds needing source: the WACC ranges table. The "70% confidence band of inputs" prescription.
• Priority for revision: P1 (specifically because Mauboussin & Rappaport on reverse DCF is uncredited — the entire most-useful section of the file rests on their idea)
• Recommended canonical sources: Damodaran Investment Valuation; Damodaran NYU industry/WACC datasets; Mauboussin & Rappaport Expectations Investing; Mauboussin & Wang The Base Rate Book (Credit Suisse, 2016)

liquidation-and-asset-value.md

• Supportable, needs citation:
  • Liquidation analysis, NCAV / "net-net" — Graham Security Analysis (1934), Intelligent Investor (1949), Graham & Dodd ch. on NCAV — direct lift from Graham
  • "Buy below 2/3 of NCAV" — Graham's specific rule; cite Intelligent Investor ch. 15 or Security Analysis
  • Reproduction value — Greenwald Value Investing ch. 4 (the Greenwald asset-value chapter is the canonical source for this concept's application to value investing)
  • "Tobin's Q" allusion in the philosophy file — Tobin 1969 Journal of Money, Credit and Banking
  • Hidden assets / real estate at historical cost — Greenwald ch. 4; also Klarman ch. 8 on asset-based investing
  • "Most managers under-estimate reproduction value" — Greenwald's repeated point (ch. 4)
  • Treatment of goodwill at zero in liquidation — Graham; standard accounting forensics (Schilit Financial Shenanigans)
• Freehand reasoning, needs rewrite:
  • The line-by-line haircut table (AR 80-90%, finished goods inventory 60-80%, raw materials 50-70%, WIP 20-40%, etc.) reads as authoritative liquidation banker convention. These ranges are roughly consistent with bank lending haircuts and turnaround-firm practice, but no source is named. Either cite a turnaround-finance text (e.g., Pratt's Valuing a Business covers liquidation discount ranges) or label as practitioner heuristic
  • "Forced liquidation often 30-60% below orderly values" — directionally correct, no source; could cite ABI (American Bankruptcy Institute) studies or Pratt
• Unsupportable, needs removal:
  • Nothing structural. The Graham NCAV claim that "historical record is strong" needs an explicit citation — Oppenheimer (1986) "Ben Graham's Net Current Asset Values: A Performance Update" Financial Analysts Journal is the standard cite; multiple updates since
• Quantitative thresholds needing source: every percentage in the line-by-line table. Sources will be plural (Pratt; ABI; turnaround banker practice) rather than a single author.
• Priority for revision: P2 (high-quality file, but liquidation haircuts are quant claims and need defensible sources before being used in a real thesis)
• Recommended canonical sources: Graham & Dodd Security Analysis (esp. ch. on NCAV); Graham Intelligent Investor ch. 15; Greenwald et al. Value Investing ch. 4; Oppenheimer (1986) FAJ on Graham NCAV performance; Pratt Valuing a Business (for liquidation discounts); Schilit Financial Shenanigans (for goodwill discipline)

position-sizing-kelly.md

• Supportable, needs citation:
  • Kelly criterion f* = (bp − q) / b — Kelly (1956) "A New Interpretation of Information Rate" Bell System Technical Journal
  • Fractional Kelly (25-50% of full Kelly) — Thorp's writings (Thorp 1969, 2006 Handbook of Asset and Liability Management); MacLean, Thorp & Ziemba The Kelly Capital Growth Investment Criterion (World Scientific, 2011); Poundstone Fortune's Formula (2005)
  • Mauboussin on Kelly in equities — Mauboussin "Size Matters" (Credit Suisse 2006) is the canonical practitioner application to investing
  • The 50-70% drawdown observation at full Kelly — Thorp / MacLean-Ziemba
• Freehand reasoning, needs rewrite:
  • Critical math error / oversimplification: the file presents f* = (expected return − risk-free rate) / variance as a "generalized" continuous Kelly formula and then f* = edge / odds as its "simple form." These are not equivalent. The first is the Merton/Markowitz mean-variance optimum (and equals Kelly only under log utility + lognormal returns); the second is the discrete-bet Kelly. The file conflates them. This needs a rewrite, not just a citation.
  • The conviction-tier table (Core 1 8-12% / Core 2 5-8% / Mid 2-4% / Probe 0.5-1.5%) — no canonical source; labeled "practical structure." Make the "AlphaSteve calibration" label explicit
  • The hard limits (single-position 15%, industry 30%, country 25%) — same. No author specifies these. They are reasonable practitioner conventions
  • "Build at 30-50% of target, average down to target" — practitioner pattern; not Klarman or Marks specific, though both write about scaled entry
• Unsupportable, needs removal:
  • The "f* = edge / odds" simple form as written is too loose for a continuous-return context; either replace with the proper Markowitz form (with a citation) or restrict the discussion to discrete bets where Kelly literally applies
• Quantitative thresholds needing source: all of them. Kelly itself, fractional-Kelly ratio, position-size tiers, hard limits.
• Priority for revision: P1 (math error + load-bearing sizing prescriptions without cite)
• Recommended canonical sources: Kelly (1956) Bell System Technical Journal; Thorp (1969, 2006); MacLean, Thorp & Ziemba The Kelly Capital Growth Investment Criterion (2011); Poundstone Fortune's Formula (2005); Mauboussin "Size Matters" (Credit Suisse 2006) and More Than You Know (Columbia, 2006); Klarman Margin of Safety ch. 6 (scaling in)

permanent-capital-loss.md

• Supportable, needs citation:
  • "Volatility is not risk; permanent capital loss is" — Marks The Most Important Thing ch. 5, 6, and 9 (his most-repeated theme); Klarman ch. 7
  • Buffett's "Rule #1: Don't lose money. Rule #2: Don't forget Rule #1." — Berkshire shareholder letter
  • Compounding asymmetry math (−50% requires +100% to recover) — Mauboussin / standard finance; could cite Bernstein Against the Gods
  • Fraud examples (Wirecard, Enron, NMC Health, Luckin, Theranos) — each has a court-documented or SEC-filing trail; cite the SEC litigation releases as <span class="tier-t1" title="">T1</span> rather than naming the company casually
  • "Net debt / EBITDA above 4× → distress risk" / "Interest coverage below 2× → distressed" — standard credit thresholds; Moody's / S&P methodology documents
  • The "asymmetry rule" — Taleb Antifragile / Black Swan on convexity; Mauboussin on asymmetric payoffs
• Freehand reasoning, needs rewrite:
  • The seven sources of permanent loss are AlphaSteve's taxonomy; label as such (consistent with Klarman framings but not a verbatim list from any author)
  • The "10-15% single position / 25-30% industry / 25% country" hard rules duplicate position-sizing-kelly.md and have the same source-attribution issue
  • The "5 years without recovery + 75-80% nominal loss" definition of "permanent" is a working definition, not anyone's published threshold — label as our convention
• Unsupportable, needs removal:
  • The covenants thresholds and "going concern" auditor language warnings are sound but unsourced. Cite AICPA AU-C 570 or PCAOB AS 2415 on going concern language; cite Moody's / S&P methodology for the credit thresholds
• Quantitative thresholds needing source: all leverage / coverage thresholds (need Moody's, S&P, or Fitch methodology docs).
• Priority for revision: P2
• Recommended canonical sources: Marks The Most Important Thing (ch. 5, 6, 9); Klarman Margin of Safety (ch. 6-7); Buffett shareholder letters (recurring); Taleb Antifragile; Moody's and S&P rating-methodology publications; PCAOB / AICPA auditing standards on going concern; SEC litigation releases for the fraud examples

base-rates.md

• Supportable, needs citation:
  • Inside view / outside view — Kahneman Thinking, Fast and Slow (2011), esp. ch. 23; original work with Lovallo, "Delusions of Success" Harvard Business Review (2003)
  • M&A failure rates "50-70% destroy value" — KPMG, McKinsey, Harding & Rovit (Bain), and many academic studies; canonical practitioner cite is McKinsey's M&A Insights publications and Bruner Applied Mergers and Acquisitions (Wiley, 2004)
  • "Synergies achieved typically 40-70% of management's announced targets" — McKinsey M&A research
  • "Goodwill impairments on > 30% of large deals within 5 years" — academic studies (Gu & Lev 2011 Accounting Review); KPMG goodwill impairment studies
  • IPO underperformance — Ritter (1991) "The Long-Run Performance of Initial Public Offerings" Journal of Finance; updated in Loughran & Ritter (1995) and ongoing on Ritter's website
  • Spin-off outperformance — Greenblatt You Can Be a Stock Market Genius (1997); McConnell & Ovtchinnikov (2004) J of Investment Management
  • "Buddenbrooks effect" (third-generation family business failure) — referenced via Mann's novel; the empirical claim links to Family Business Institute and PwC family-business studies
  • Turnaround success rates — Bibeault Corporate Turnaround (1982) and Slatter & Lovett Corporate Turnaround (1999)
  • Persistent margins regressing — Damodaran on mean reversion; Mauboussin & Wang The Base Rate Book (Credit Suisse 2016) — the latter is the single most relevant uncredited source for this entire file
  • Earnings surprise persistence — Bernard & Thomas (1989, 1990) PEAD research
  • Cyclical recovery base rates — Marathon / Chancellor Capital Returns
• Freehand reasoning, needs rewrite:
  • "Concentration of returns in a small number of winners: extreme" (in IPO section) — true; cite Bessembinder (2018) "Do stocks outperform Treasury bills?" Journal of Financial Economics
  • "Founder-to-professional CEO transitions ~50%" — number is in the right range but I cannot pin to a specific study; either find a Harvard Business Review or McKinsey CEO-transition study or hedge
  • Hot themes / Nifty Fifty / Japan 1989 / tech 2000 examples are factual cultural memory but each historical claim should ideally trace to a specific source (Siegel Stocks for the Long Run covers Nifty Fifty; Ahearne et al. on Japan; etc.)
• Unsupportable, needs removal:
  • "Cyclical industries with capacity above 90% utilization within 2 years: > 60% experience trough-to-peak earnings expansion" — this is a plausible but suspiciously specific number with no obvious published source. Either find a Marathon / Chancellor analog or remove the specific %
• Quantitative thresholds needing source: every base rate in this file — the entire point is that they are empirical claims, and right now none are sourced.
• Priority for revision: P1
• Recommended canonical sources: Mauboussin & Wang The Base Rate Book (Credit Suisse 2016) — the master reference; Kahneman Thinking, Fast and Slow; Kahneman & Lovallo "Delusions of Success" HBR (2003); Ritter (1991) and Loughran & Ritter (1995); Greenblatt You Can Be a Stock Market Genius (1997); Bessembinder (2018) JFE; McKinsey M&A research; Bruner Applied M&A; Chancellor Capital Returns; Damodaran on regression to the mean; Bernard & Thomas (1989/90) on PEAD

financial-statement-forensics.md

• Supportable, needs citation:
  • The whole approach derives from Schilit Financial Shenanigans (5th ed. 2018), Mulford & Comiskey The Financial Numbers Game (2002), and Penman Financial Statement Analysis and Security Valuation (5th ed. 2013) — all on the policy whitelist, none currently cited
  • Channel stuffing, bill-and-hold, round-tripping, related-party revenue — Schilit's taxonomy
  • Capitalized vs. expensed (software, R&D) — Penman ch. 9-10
  • "Stock-based comp is real economic expense" — Buffett (multiple letters, esp. 1992 and 1998); Damodaran (multiple)
  • "Recurring restructuring is operating expense" — Schilit
• Freehand reasoning, needs rewrite:
  • The 22-item red-flag checklist is AlphaSteve's compilation. Most items map to Schilit / Mulford-Comiskey directly; label the checklist as derived from those sources
  • "Three or more warrant deeper work; five or more are usually a pass" — practitioner heuristic; cite Beneish M-Score (1999 Financial Analysts Journal) for a defensible quantitative analog or label as our convention
  • "DSO expanding for 4+ quarters" / "Goodwill > 30% of assets, growing" / "CFO/NI below 0.8 for 2+ years" — these specific numeric thresholds are practitioner ranges; cite either Schilit (where present) or Beneish for the quantitative discipline
• Unsupportable, needs removal: nothing structural — the file is high-quality.
• Quantitative thresholds needing source: the red-flag-count rule of thumb; the specific multi-quarter / multi-year tests on DSO/DIO/DPO.
• Priority for revision: P2 (high quality, just needs the sources block — most claims trace cleanly to Schilit/Mulford-Comiskey/Penman)
• Recommended canonical sources: Schilit Financial Shenanigans (5th ed.); Mulford & Comiskey The Financial Numbers Game; Penman Financial Statement Analysis and Security Valuation; Beneish (1999) "The Detection of Earnings Manipulation" FAJ; Buffett 1992 and 1998 shareholder letters (on SBC)

quality-of-earnings.md

• Supportable, needs citation:
  • Sloan accruals ratio — explicitly named in-text ("Sloan, 1996") but not formally cited. Sloan (1996) "Do Stock Prices Fully Reflect Information in Accruals and Cash Flows about Future Earnings?" The Accounting Review. Excellent example of the policy gap: the author is named, the year is given, but no citation block exists — fails the policy format
  • "Owner earnings" — Buffett 1986 shareholder letter is the canonical source, uncited despite explicit name reference
  • Persistence / cash backing / conservatism framework — Penman ch. 12-13; Schilit
  • Sector-specific quality flags table — derived from Schilit by sector; some entries (SaaS capitalized commissions ASC 606) cite a standard but not the underlying discussion
• Freehand reasoning, needs rewrite:
  • The "<5% / 5-10% / >10% / >15%" accruals ratio bands are roughly Sloan's quintile breakpoints but stated as practitioner ranges; cite Sloan (1996) directly for the quintile work
  • "Persistent FCF/NI below 0.7 deserves explanation" — practitioner heuristic; could cite Mauboussin's writings on cash conversion or label as our convention
  • Sector quality flags table is reasonable but each row needs at minimum a sector-text or CFA curriculum cite to harden
• Unsupportable, needs removal: nothing structural; this is one of the cleaner files.
• Quantitative thresholds needing source: Sloan ratio quintiles (Sloan 1996); FCF/NI thresholds.
• Priority for revision: P1 (because the file already names sources and just needs to format them per the policy — quickest wins)
• Recommended canonical sources: Sloan (1996) The Accounting Review; Buffett 1986 letter (owner earnings); Penman Financial Statement Analysis; Schilit Financial Shenanigans; Mauboussin (various) on cash conversion

roic-decomposition.md

• Supportable, needs citation:
  • ROIC = NOPAT / Invested Capital; ROIC vs. WACC as value test — Stewart The Quest for Value (1991) (EVA framework); Koller, Goedhart & Wessels Valuation (McKinsey, multiple editions) — this is the canonical valuation textbook for the ROIC/WACC framing and is the single most uncited authority across the kit
  • DuPont decomposition — DuPont (1920s) attribution; standard since
  • "Growth at returns below cost of capital destroys value" — Stewart; Koller/McKinsey Valuation; Damodaran
  • ROIIC framework — Mauboussin "What Does an EV/EBITDA Multiple Mean?" and "Calculating Return on Invested Capital" (Credit Suisse / Morgan Stanley papers); Koller/McKinsey
  • Goodwill-included vs. excluded ROIC — Koller/McKinsey ch. 7
  • Mid-cycle ROIC for cyclicals — Chancellor Capital Returns; Marathon framework
• Freehand reasoning, needs rewrite:
  • The decomposition into margin × turnover with the three example profiles (20% × 1.0×, 10% × 2.0×, 5% × 4.0×) is illustrative and could cite Koller/McKinsey or any DuPont treatment
  • "A high ROIC has to be explained structurally" — sound; cite Mauboussin "Measuring the Moat" (Credit Suisse 2002, updated several times) — this is the canonical practitioner treatment of moat-vs-ROIC and is uncredited despite being central
• Unsupportable, needs removal: nothing structural; very clean file.
• Quantitative thresholds needing source: none significant — the file is mostly framework, with examples rather than prescriptive thresholds.
• Priority for revision: P2 (high quality; main fix is adding Mauboussin "Measuring the Moat" + Koller/McKinsey citations)
• Recommended canonical sources: Koller, Goedhart & Wessels Valuation: Measuring and Managing the Value of Companies (McKinsey, 7th ed. 2020); Mauboussin "Measuring the Moat" (Credit Suisse 2002 / updates); Mauboussin "Calculating Return on Invested Capital" (Credit Suisse 2014); Stewart The Quest for Value (1991); Damodaran Investment Valuation; Chancellor Capital Returns

moat-taxonomy-and-identification.md

• Supportable, needs citation:
  • The Greenwald taxonomy (customer captivity / production advantage / scale + local market / regulatory) — Greenwald & Kahn Competition Demystified (Penguin, 2005); also in Greenwald et al. Value Investing (2001)
  • The 7 Powers taxonomy — Helmer 7 Powers: The Foundations of Business Strategy (Deep Strategy, 2016); all seven powers named match Helmer's exact terminology
  • The five moat tests are AlphaSteve's, but each test points to a canonical concept (Mauboussin "Measuring the Moat" 2002 is the obvious cite)
  • Porter Five Forces reference — Porter Competitive Strategy (1980) and Competitive Advantage (1985)
  • "Margin stability through cycles" as evidence of moat — Marathon / Chancellor
• Freehand reasoning, needs rewrite:
  • The "moat power vs. moat width" framing — useful and AlphaSteve-original; should be labeled as our framing while pointing to Mauboussin's CAP (Competitive Advantage Period) work as adjacent
  • The "common false-positive moats" table is sound but un-attributed; cite Mauboussin "Measuring the Moat" or Helmer ch. 1 on the difference between moat and head-start
  • "ROIC > WACC for 10+ years" as moat test — practitioner heuristic; matches Morningstar's wide-moat criteria; cite Morningstar's moat methodology or Mauboussin
• Unsupportable, needs removal: none.
• Quantitative thresholds needing source: the 10+ year ROIC test (Morningstar moat methodology is the standard cite).
• Priority for revision: P2
• Recommended canonical sources: Helmer 7 Powers (2016); Greenwald & Kahn Competition Demystified (2005); Greenwald et al. Value Investing (2001); Porter Competitive Strategy (1980), Competitive Advantage (1985); Mauboussin "Measuring the Moat" (Credit Suisse 2002 / Morgan Stanley updates); Morningstar moat methodology documents

moat-durability-and-erosion.md

• Supportable, needs citation:
  • Technological substitution — Christensen The Innovator's Dilemma (1997); the Kodak/Polaroid/Blockbuster examples are Christensen's
  • Counter-positioning erosion — Helmer ch. 3
  • Capital cycle as long-run gravity — Chancellor Capital Returns (2015); Marathon Asset Management writings
  • Management dissipation / brand dilution — Buffett (multiple letters on the franchise concept, esp. 1991 letter on Berkshire's "see's candies" sustainability)
  • Customer behavior shift erosion — Christensen Innovator's Solution (2003); Porter Competitive Advantage ch. 8 on industry decline
  • Regulatory erosion — Porter Competitive Strategy on government as force
• Freehand reasoning, needs rewrite:
  • The fade-horizon table (15-20 years wide network/regulatory, 10-15 standard, 5-10 process/scale, 3-5 "show-me," immediate no moat) — useful but no canonical source; this is AlphaSteve calibration. Mauboussin's CAP work is the closest published analog; label and cite
  • "Streaming bypassed cable's local-monopoly distribution moat" / "LLMs are currently a threat to search and certain SaaS moats" — the LLM claim in particular is freehand reasoning about current events; either qualify or cite specific sell-side / academic analysis
• Unsupportable, needs removal:
  • The current-events claims about LLMs threatening "search and certain SaaS moats" — speculative and uncited. Either remove or hedge to "our reading is..."
• Quantitative thresholds needing source: the fade horizon years.
• Priority for revision: P2
• Recommended canonical sources: Christensen Innovator's Dilemma (1997), Innovator's Solution (2003); Helmer 7 Powers (esp. ch. 3); Chancellor Capital Returns (2015); Porter Competitive Advantage (1985), esp. ch. 8; Buffett 1991 letter (franchise sustainability); Mauboussin on Competitive Advantage Period

bottleneck-mapping-framework.md

• Supportable, needs citation:
  • The general framing (where economic rent accumulates in a value chain) — Porter Competitive Advantage (1985) ch. 2-3 on the value chain; Stuckey & White "When and When Not to Vertically Integrate" Sloan Management Review (1993)
  • The bottleneck-as-distinct-from-moat distinction is AlphaSteve's framing; the underlying logic comes from Porter and from Mauboussin's writings on industry attractiveness
  • Capital cycle operating through bottlenecks — Chancellor Capital Returns
  • The taxonomy of six bottleneck types is AlphaSteve's compilation; each type can be supported individually
• Freehand reasoning, needs rewrite:
  • The AI infrastructure worked example is current-events analysis. Each row (NVDA monopoly on GPU, TSMC on leading-edge logic, ASML on EUV, etc.) is factually correct as of 2025-2026 but contains real-time market claims with no underlying source. Either cite SEC filings for each named firm or label as "our reading at time of writing"
  • "Time horizon matters — years for semis fab nodes, decades for pipelines / real estate locations, effectively permanent for geography" — directionally sound; could cite specific industry texts
• Unsupportable, needs removal:
  • The AI worked example, as written, is the closest the kit comes to a Wikipedia-style summary — it states current industry structure as authoritative fact without citations. Either rebuild with <span class="tier-t1" title="">T1</span> filings for each named firm or move to "illustrative example, current as of [date]" with appropriate hedge
• Quantitative thresholds needing source: none beyond the AI example's specific structure claims.
• Priority for revision: P2
• Recommended canonical sources: Porter Competitive Advantage (1985) ch. 2-3; Stuckey & White (1993) Sloan Management Review; Chancellor Capital Returns; SEC filings for each company named in the AI example (NVDA, TSM, ASML, ADI/MU/etc. for HBM, AVGO for networking); Mauboussin on industry attractiveness

capital-cycle.md

• Supportable, needs citation:
  • Five-phase capital cycle — Chancellor Capital Returns: Investing Through the Capital Cycle (Palgrave Macmillan, 2015), the entire Marathon framework — not cited despite the file title implying it. This is the most egregious case in the kit
  • "Capex / depreciation > 1.2× over years signals over-building" — Chancellor / Marathon; the specific threshold may be AlphaSteve calibration
  • Sector cycle clocks (E&P 7-10 years, mining 10-15, shipping 4-10 by segment, memory 2-3) — directionally correct; cite Chancellor for the general framework, sector-specific work for the specific cycles (e.g., Marathon's commentary, sell-side cycle research)
  • "Buy at maximum capital retreat" — Marathon / Chancellor; also Klarman ch. 10
  • "Cyclical trough buying requires balance sheet survival" — Klarman; Chancellor
• Freehand reasoning, needs rewrite:
  • The recent-cycles narrative (2014-2016 trough, 2020 Covid trough, etc.) is factual but unsourced — these are historical claims that should cite either FRED data, IEA / EIA reports for energy, LME for metals, etc.
  • Sector cycle clock specific year ranges — cite sector source or label as practitioner heuristic
• Unsupportable, needs removal: none structural; the file just needs to acknowledge its primary source.
• Quantitative thresholds needing source: capex/D&A bands, cycle lengths by sector.
• Priority for revision: P1 (because the file is entirely Marathon framework and currently fails to credit it)
• Recommended canonical sources: Chancellor Capital Returns (2015); Chancellor Devil Take the Hindmost (1999, for cycle history); Marathon Asset Management quarterly letters where available; sector-specific cycle data (EIA, IEA, LME, Clarksons, SEMI/TrendForce, etc.)

second-level-thinking.md

• Supportable, needs citation:
  • The entire framework is Howard Marks. The author is named in the file but no work is cited. Primary source: Marks The Most Important Thing (Columbia Business School Press, 2011) ch. 1; also Marks memo "The Most Important Thing" (Jul 2003)
  • Reverse-DCF as second-level applied to valuation — Mauboussin & Rappaport Expectations Investing (2001)
  • Over-extrapolation / recency / narrative momentum / forced selling — Kahneman Thinking, Fast and Slow; Shiller Irrational Exuberance (2000)
  • Coverage gaps / mid-small cap inefficiency — academic literature on neglected stocks (Arbel, Carvell & Strebel 1983 Journal of Portfolio Management)
  • Career risk — Lakonishok, Shleifer & Vishny (1992) "The Structure and Performance of the Money Management Industry" Brookings Papers on Economic Activity
• Freehand reasoning, needs rewrite:
  • The first-level / second-level examples (AMD, "6× earnings," renewables) are AlphaSteve constructions; they are illustrative not authoritative and read fine as written, but should be labeled as illustrative
• Unsupportable, needs removal: none.
• Quantitative thresholds needing source: none.
• Priority for revision: P1 (Marks needs to be cited — this is his framework verbatim and uncredited)
• Recommended canonical sources: Marks The Most Important Thing (2011), esp. ch. 1; Marks memos (esp. "The Most Important Thing" Jul 2003, "Dare to be Great" 2006, "It's Not Easy" Sep 2015); Mauboussin & Rappaport Expectations Investing; Kahneman Thinking, Fast and Slow; Shiller Irrational Exuberance; Arbel et al. (1983) on neglected stocks

variant-perception.md

• Supportable, needs citation:
  • Variant perception as a concept — Steinhardt No Bull: My Life In and Out of Markets (2001) is the canonical origin; Marks The Most Important Thing ch. 1; Mauboussin "Are You an Expert?" (Credit Suisse) and More Than You Know
  • "Without a variant perception, you do not have a thesis" — Steinhardt's discipline; Mauboussin (multiple)
  • The five-type taxonomy is AlphaSteve's (forecast / structural / time-horizon / quality / catalyst); useful framing, label as our taxonomy
  • Falsification discipline — Popper The Logic of Scientific Discovery (1934); operationalized in investing by Marks, Klarman
• Freehand reasoning, needs rewrite:
  • The five-type taxonomy is AlphaSteve's contribution and should be presented as such, with reference to Steinhardt as the conceptual root
  • "Where variant perceptions die" five failure modes — practitioner observation; can stand as "our taxonomy of failure modes" without specific source
• Unsupportable, needs removal: none.
• Quantitative thresholds needing source: none.
• Priority for revision: P2
• Recommended canonical sources: Steinhardt No Bull (2001) — the originator of "variant perception"; Marks The Most Important Thing; Mauboussin More Than You Know (Columbia, 2006), "Are You an Expert?" (Credit Suisse); Popper Logic of Scientific Discovery

Phase 3 — Light pass

01-identity.md — P3

Internal-doctrine file. Doesn't make external knowledge claims; appropriate for it to be a self-spec without citation. Two minor mentions of Buffett's "twenty punches" idea show up downstream (in 04-intellectual-virtues), not here. Clean.

03-mental-models.md — P2

Twenty-nine numbered models, several of which are direct lifts from named authors: Munger latticework attribution is correct; second-level thinking → Marks (named); inversion → Jacobi (named); capital cycle → Chancellor/Marathon (named); 7 Powers → Helmer (named); Porter's Five Forces (named); DuPont (implicit). The file does name authors better than most files in the kit, but it still lacks the formal <span class="tier-t2" title="...">T2</span> cites the policy now requires. Quick mechanical fix.

04-intellectual-virtues.md — P2

The five-tier evidence taxonomy in section 6 is a pre-existing version of the policy's tier system and now duplicates / conflicts with the official policy's four tiers. Reconcile to the policy's T1-T4 explicitly. Other sections are operating-discipline (humility / patience / courage / etc.) without external knowledge claims and don't need citation.

05-decision-framework.md — P3

Internal-process file. The five gates and the thesis structure are AlphaSteve's framework; no external citations required. The reference to "Buffett's twenty punches" should cite the 1991 Berkshire shareholder letter or The Snowball.

balance-sheet-stress-test.md — P2

Strong file. The leverage thresholds (Net debt/EBITDA < 1.0 fortress, > 5.0 distress; interest coverage > 8x fortress, < 2x distress) need credit-rating-agency methodology cites (Moody's/S&P). Hidden-asset and hidden-liability lists derive from Schilit / Mulford-Comiskey / Penman; cite. The fortress vs. death-spiral checklists are practitioner heuristics; label and cite the underlying credit-methodology where possible.

cash-flow-analysis.md — P2

"Earnings are an opinion. Cash is a fact." — attributed quote (often to Alfred Rappaport or to Pam Pohly; needs verification; safest is to attribute to Rappaport Creating Shareholder Value). Owner-earnings concept needs Buffett 1986 letter cite. Sector translations table is sound but each row should point to a sector-specific source (REIT NAREIT for FFO/AFFO; energy SEC PV-10 rules; etc.). Clean otherwise.

working-capital-cash-conversion.md — P3

Almost entirely mechanical formulas (DSO, DIO, DPO, CCC). The industry-norms table is practitioner convention; could cite CFA curriculum or Penman. Negative-CCC-as-moat insight could cite Bezos shareholder letters or Mauboussin on cash conversion. Low priority.

Moats family — cost-advantages.md, counter-positioning.md, efficient-scale-process-power.md, intangible-assets-brand-ip.md, network-effects.md, switching-costs.md — all P2

Each file is a deep-dive on a power from Helmer / Greenwald. Each cleanly maps to: Helmer 7 Powers for the named power; Greenwald & Kahn Competition Demystified for the analogous Greenwald category; Mauboussin "Measuring the Moat" for the practitioner translation; sector-specific examples should cite filings. cost-advantages.md correctly names Greenwald's local-scale insight in-text but doesn't cite it. intangible-assets-brand-ip.md should also cite Aaker Managing Brand Equity (1991) for the brand framework. network-effects.md is one of the better files in the kit but should cite Eisenmann/Parker/Van Alstyne on platforms and Schmalensee on network effects. switching-costs.md should cite Klemperer's academic work on switching costs (multiple papers in 1980s-90s).

Bottlenecks family — distribution-and-channel.md, regulatory-and-licensing.md, supply-chain-and-capacity.md, talent-and-knowhow.md, technology-and-ip.md — all P2

Generally clean derivative work. The major issue is the same as the AI example in bottleneck-mapping-framework.md: each file contains current-industry-structure tables (app store take rates, PBM concentration, lithography monopolies, EV battery players, etc.) that are correct as of 2025-2026 but are stated as authoritative fact without citing the underlying filings or industry reports. Each table row needs a <span class="tier-t1" title="">T1</span> (filing) or <span class="tier-t2" title="">T2</span> (specialist data firm) cite, or a "current as of" hedge. regulatory-and-licensing.md is the cleanest of the five — its examples are mostly historical regulatory events that can be cited to statute or to legal-historical sources.

Management family — capital-allocation-scorecard.md, communication-quality.md, governance-red-flags.md, incentive-alignment-and-comp.md — all P2

capital-allocation-scorecard.md is excellent and tracks Thorndike The Outsiders (2012) closely; that book + Buffett letters + Cunningham The Essays of Warren Buffett should be the source spine. The Berkshire 2008-2009 deployment example needs the Berkshire letter cite. The shareholder-letter list (Mark Leonard, John Malone, etc.) is practitioner cultural knowledge that should be sourced to specific letters or to Thorndike. governance-red-flags.md has the high-fraud-risk pattern at the end that needs Beneish M-Score (1999) or Dechow et al. F-Score (2011) as quantitative analogs. incentive-alignment-and-comp.md should cite Jensen & Murphy's compensation-and-performance work; the insider-ownership ranges have no specific source and should be labeled as our calibration. communication-quality.md is mostly stylistic / illustrative and is low-risk under the policy but the list of named exceptional letter-writers should cite specific letters.

Valuation family — comparables-and-relative.md, private-market-and-replacement-value.md, sum-of-parts.md — all P2

comparables-and-relative.md: the multiples-by-business-type table should cite Damodaran's industry valuation tables. "Economic comp" framing aligns with Damodaran Investment Valuation ch. 17-18. private-market-and-replacement-value.md: explicitly names Gabelli for PMV — must cite Gabelli writings (Gabelli Funds research; his prospectus discussions). Greenwald is referenced for reproduction value but not cited. sum-of-parts.md: the 10-25% conglomerate discount and 15-35% holdco discount need empirical sources — Berger & Ofek (1995) "Diversification's effect on firm value" JFE is the canonical conglomerate-discount study; multiple academic papers since.

04-Market-Read remaining — cognitive-bias-checklist.md, crowding-and-positioning.md, narrative-cycle.md, pre-mortem-and-inversion.md — all P2

cognitive-bias-checklist.md is essentially Kahneman Thinking, Fast and Slow (uncited) plus Tversky-Kahneman classic biases. Each named bias has a canonical citation (loss aversion → Kahneman & Tversky 1979 Econometrica; anchoring → Tversky & Kahneman 1974 Science; confirmation bias → Wason 1960). Add them. pre-mortem-and-inversion.md names Klein and Jacobi in-text but doesn't cite. Klein Sources of Power (1998) and his HBR article "Performing a Project Premortem" (2007) are the canonical sources. crowding-and-positioning.md is mostly mechanical (13F mechanics, short interest math, options skew); needs cites only where it states behavioral claims (squeeze dynamics, crowding-as-leading-indicator). narrative-cycle.md is the most concerning of the four: the eight-phase arc is AlphaSteve's framing with echoes of Shiller's narrative economics and Soros's reflexivity; should cite Shiller Narrative Economics (2019) and Soros The Alchemy of Finance (1987) at minimum.

05-Environment remaining — currency-and-fx-exposure.md, cycle-positioning.md, geopolitics-country-risk.md, rates-and-discount-rates.md, regulatory-regime.md — all P3

Generally well-organized mechanical files (translation vs. transaction FX, regime-question framing for rates, country-risk taxonomy). Each makes few independent knowledge claims — they're frameworks. The macro-data references (FRED, BLS, FOMC) align with the policy whitelist already. Main fixes: rates-and-discount-rates.md "100 bps move in 10Y moves WACC by 30-70 bps" needs either a derivation or a Damodaran cite. geopolitics-country-risk.md should adopt Damodaran's country-risk-premium dataset as the canonical reference (he updates it annually). regulatory-regime.md historical examples (Glass-Steagall, Dodd-Frank, etc.) should cite the underlying statutes.

06-Risk remaining — tail-risk-and-fat-tails.md — P2

Largely Taleb (The Black Swan, Antifragile) and Mandelbrot (The (Mis)Behavior of Markets) territory — both should be cited. Fat-tail empirics for equity returns should cite Cont (2001) "Empirical properties of asset returns: stylized facts and statistical issues" Quantitative Finance. The recent tail events list (1998 LTCM, 2008, 2020, 2023 SVB) needs FRED data or news cites per event.

07-Sectors — 00-sector-template.md through 11-utilities.md — all P2/P3

The sector files are reference material — KPIs by sub-industry, common moat patterns, sector-specific red flags. They are largely mechanical (NIM for banks, AFFO for REITs, AISC for miners, FFO for REITs, ARR/NRR for SaaS). The KPI lists are CFA Institute / Damodaran / sector textbook material; cite the CFA curriculum, Damodaran's industry pages, and sector-specific authorities (NAREIT for REIT metrics, SEC PV-10 rules for E&P, etc.). The sub-industry economics descriptions are sound but uncited. Sector-specific examples (the "US drug distribution flows through three players" / "memory market is an oligopoly of Samsung/SK Hynix/Micron" / "Class I railroads = 7 in the US" type claims) need filings or industry-source citations under the policy. None of the sector files contains made-up thresholds, but each contains many factual claims about industry structure that should be sourced. Low priority because these are reference rather than reasoning, but the policy does apply.

08-Frameworks — deliverable-suite.md, investment-thesis-template.md, porters-five-forces.md, tam-sam-som.md, unit-economics.md, value-chain-mapping.md — P3 mostly

porters-five-forces.md is a Porter restatement; needs the Porter 1980 / 1985 cites and that's effectively the whole fix. tam-sam-som.md correctly identifies the framework's abuse and its skeptical use — should cite Damodaran's writing on narrative-driven TAM and Mauboussin on base rates of TAM realization. unit-economics.md is mechanical (contribution margin, allocation, LTV/CAC); the LTV/CAC > 3 heuristic should cite either David Skok or the early SaaS academic literature. value-chain-mapping.md is a Porter ch. 2 restatement; cite Porter Competitive Advantage. deliverable-suite.md and investment-thesis-template.md are operational documents and don't need citations (they're our process, not external claims).

Recommended revision sequence

1. 02-philosophy-deep-value.md — doctrine file, propagates downstream. Fix first to set the tone.

2. position-sizing-kelly.md — has a substantive math error (conflating discrete Kelly with the Markowitz mean-variance formula) plus uncited Kelly / Thorp / Mauboussin. Both math and citations need real work, not just a sources block.

3. base-rates.md — load-bearing for variant perception and risk assessment. Every quantitative claim needs a source. Mauboussin & Wang The Base Rate Book (Credit Suisse 2016) is the single most-relevant document; bring it in.

4. capital-cycle.md — the entire file is the Marathon / Chancellor framework, and Chancellor is not credited despite being the source of the file's structure. Egregious by the new policy.

5. second-level-thinking.md and variant-perception.md — Marks's and Steinhardt's frameworks respectively, both named in-text but neither cited. Quick mechanical fixes.

6. earnings-power-value-greenwald.md — file title names Greenwald, file content is Greenwald, no cite. Quickest mechanical fix in the kit.

7. quality-of-earnings.md — already names Sloan in-text; just needs the formal cite per policy. Buffett owner-earnings reference is similar.

8. dcf-and-reverse-dcf.md — the entire reverse-DCF section is Mauboussin & Rappaport uncredited.

9. margin-of-safety-pricing.md — Klarman / Marks doctrine, plus AlphaSteve's calibrated thresholds need to be labeled as ours.

10. financial-statement-forensics.md and quality-of-earnings.md (continued) — Schilit / Mulford-Comiskey / Penman are the canonical sources for the entire forensic-accounting layer; mechanical citation pass will cover most of the file.

11. liquidation-and-asset-value.md — Graham + Greenwald primary; the liquidation-haircut percentages need defensible sources (Pratt or ABI).

12. roic-decomposition.md — Koller/McKinsey Valuation and Mauboussin "Measuring the Moat" / "Calculating ROIC" are the canonical practitioner sources, both uncited.

13. moat-taxonomy-and-identification.md and moat-durability-and-erosion.md — Helmer + Greenwald + Christensen + Porter; mechanical citation pass plus removing the speculative LLM-threat freehand in the erosion file.

14. bottleneck-mapping-framework.md — rebuild the AI worked example with <span class="tier-t1" title="">T1</span> filings or hedge appropriately.

15. permanent-capital-loss.md and tail-risk-and-fat-tails.md — credit-methodology and tail-risk empirical literature pass.

16. All other files — Phase 3 light-pass items. Mostly sources-block addition with occasional inline cites.

Notes for the policy itself

A few amendments / clarifications the policy could benefit from based on what I saw across the kit:

1. Practitioner research papers from sell-side / buy-side firms — the policy whitelist allows "sell-side and buy-side research" at T2 when the report and author are on hand. The kit relies heavily on Mauboussin's Credit Suisse / Morgan Stanley papers ("Measuring the Moat," "Calculating ROIC," The Base Rate Book) which are technically sell-side research from a named author. The policy should explicitly call out Mauboussin's body of work (and similar long-form practitioner research from Damodaran, GMO, Marathon, etc.) as a recognized T2 category — currently it falls under "Specialist data firms" loosely or "Sell-side research" with the date/title/author rule, but the importance of these long-form papers as canonical practitioner research deserves its own line.

2. Federal Reserve datasets and similar (FRED) at T1 — already correctly placed at T1. Confirm that NYU Stern (Damodaran's dataset) and Credit Suisse / Morgan Stanley research datasets fall at T2.

3. "Our calibration" / "our framing" as an explicit category — the policy currently distinguishes T1/T2/T3 sources from "rumor" and "opinion explicitly labeled as such." A large amount of the kit's quantitative content (MoS bands, position-size tiers, fade horizons, credit-distress thresholds) is reasonable practitioner calibration that doesn't fit cleanly into any tier because it isn't sourced — it's the agent's synthesis informed by general practice. The policy should provide explicit language for this case, e.g., a tag <span class="tier-cal" title="our calibration; directional inspiration from Klarman ch. 4">AS-cal</span>. Without that, the auditor (me, today) has to repeatedly re-explain the same situation across files.

4. Treatment of named authors in-text without formal cite — many files (especially mental-models.md, quality-of-earnings.md, second-level-thinking.md) name authors in prose but don't put a formal tier-tagged citation. The policy could specify that author mentions in body text still require a footnote-style <span class="tier-t2" title="...">T2</span> cite at first mention, even when the author's name appears in the sentence.

5. Current-events claims in skill files — several files (the AI example in bottleneck mapping; LLM threats in moat erosion; recent capital cycle vintages) contain claims about current industry structure. The policy is silent on how skill files should handle time-sensitive content. Two options worth considering: (a) require an "as of [date]" hedge on any current-state claim in a skill file; (b) require current-state examples to be moved to thesis / research notes and replaced in skill files with historical examples that are durably citable. The latter is cleaner but more work.

6. Wikipedia-derived cultural knowledge — many of the historical examples in the kit (Glass-Steagall years, NHTSA 1970, Bretton Woods, the Nifty Fifty, Japan 1989, Standard Oil 1911 break-up, the Buddenbrooks effect) are cultural-memory facts. They're true and trivially verifiable, but they did not come from a sourced lookup — they came from general knowledge that the agent has absorbed. The policy is silent on whether "ambient historical knowledge that any educated reader would accept as fact" requires a cite. If yes, the kit needs ~50+ encyclopedia-style cites. If no, the policy should say so explicitly to avoid auditor over-flagging.