Capital allocation judge

Capital-Allocation Judge (the finance-desk "Outsiders analyst")

The v2 conviction rubric scores management quality as part of the conviction grade, but gives no structured method for how to read a management team's capital-allocation discipline. This skill encodes the Thorndike Outsiders cross-CEO scorecard as a reusable procedure. Ground every judgment in the canon (below) — never free-hand a management grade.

The binding Munger/Buffett vault remains the FINAL CIO. This skill supplies the scorecard and method; the vault supplies the judgment (moat, owner-alignment, integrity) and the verdict. Both must clear before conviction rises above 3.

STEP 0 — Pull the canon (binding-safe)

The method now lives in two layers: the Thorndike Outsiders canon atomics (tagged role=capital-allocation) AND the 30 external perspective voices that richly cover the India reinvestment-runway question (Raamdeo QGLP, Saurabh Mukherjea capital-discipline, Terry Smith, Manish Gupta, the Tambade asset-light lens). Do a dual-pull and union the slugs:

# (a) BINDING — the CIO ranking and arbiter (run ALWAYS; these atoms are the verdict)
set -a && source /Users/Dhiraj/dev/invest/.env && set +a && /Users/Dhiraj/dev/invest/.venv/bin/python \
  /Users/Dhiraj/dev/invest/data/scripts/32_consult_brain.py \
  --corpus binding --step capital-allocation \
  --company "<company> capital allocation owner earnings reinvest above cost of capital buybacks dividends" \
  --model general --json-out extracted/grilling/<TICKER>_capalloc_binding.json

# (b) PERSPECTIVES — the India runway voices (context/divergence only, NEVER the verdict)
set -a && source /Users/Dhiraj/dev/invest/.env && set +a && /Users/Dhiraj/dev/invest/.venv/bin/python \
  /Users/Dhiraj/dev/invest/data/scripts/32_consult_brain.py \
  --corpus perspectives --step capital-allocation --k 15 \
  --company "<company> reinvestment runway incremental ROCE QGLP asset light capital discipline long runway" \
  --model general --json-out extracted/grilling/<TICKER>_capalloc_persp.json

cites_principles ⊂ the union of returned slugs.

Retrieval wiring (the fix landed — read this). A dedicated --step capital-allocation now exists, and --role is now INCLUSIVE. But there is a lossy trap that runs opposite to the old warning: the Thorndike canon atoms are tagged role=capital-allocation, the 30 external voices are NOT. So --role capital-allocation filters to the academic-canon layer and hides the India runway perspectives. Do NOT lead with --role capital-allocation. Use the dual-pull above: --corpus binding for the CIO ranking, and --corpus perspectives with NO --role to surface the external voices. (--corpus blended also surfaces the desk-synthesis atomics below alongside the binding layer — use it when you want both in one call.)

• Score the perspectives layer evidence-gated — credit a long runway only on actual incremental ROCE history, never on a projected TAM narrative. The perspectives are promotional (managers and newsletter writers talking their book), so they are context and divergence-detection, never the verdict.
• ABSTAIN on any axis whose atom the consult does not return — do not assign a non-zero score on an axis you could not ground. Thin retrieval ⇒ lower the grade, never fabricate the slug.

THE BINDING PRIOR (what the CIO already holds)

Before the scorecard, anchor to the CIO's own allocation ranking — the scorecard measures HOW WELL management executes against this prior, not WHETHER the prior is right. The binding order of preference, from [[owner-earnings-and-capital-allocation-as-intrinsic-value-engine]]:

reinvest above cost of capital > buy back stock when cheap > tax-efficient buyback > dividend

([[prefer-buybacks-over-dividends-for-tax-efficiency]], [[seek-companies-with-strong-capital-return]].) This ranking is the binding prior; the six-axis scorecard below grades execution quality against it. The Munger/Buffett CIO remains the arbiter — a high scorecard cannot override the binding veto on a business that must constantly reinvest just to stand still ([[avoid-businesses-requiring-constant-reinvestment]]).

WHY this scorecard (the core insight)

The Outsiders CEOs outperformed the S&P 500 by ~20× over their tenures — not by being the best operators, but by being the best capital allocators. Thorndike's lens: the CEO is first and foremost a capital allocator; operational excellence matters, but FCF + what management does with it is what compounds per-share value. GAAP earnings are a distraction. The eight Outsider CEOs shared five structural traits (the scorecard below).

THE SCORECARD — six axes, rate each 0 / 1 / 2

(Axes 1–5 are the Thorndike Outsiders style traits; Axis 6 adds the literal compounding driver — incremental ROCE × runway — so the scorecard grades the economics, not just the style.)

Run through each axis for the company under review. Use only evidence from the dossier, annual reports, and canon-returned principles. Never infer a high score without evidence.

Axis 1 — Per-share value as the CEO scoreboard

Canon slug: per-share-value-optimization

Score 2: Management explicitly tracks per-share owner-earnings growth (not total PAT / revenues / market cap) as the primary internal metric; discusses reinvestment decisions in terms of their per-share IV impact; resists accretive-to-EPS dilutive raises. Score 1: Some awareness of per-share metrics but mixes with revenue/headcount growth language. Score 0: Management scorecard is total profit, EBITDA, revenue, or market cap — not per-share value.

Axis 2 — FCF / owner-earnings over GAAP

Canon slugs: cash-flow-over-earnings, cash-flow-over-earnings-metric

Score 2: Management consistently discusses FCF or owner-earnings (not GAAP net income or EBITDA) in communications; capex and reinvestment are laid out explicitly; the income statement is secondary to the cash flow statement in how they talk to investors. Score 1: FCF mentioned alongside GAAP; some explicit capex vs maintenance distinction. Score 0: Communication anchored to PAT/EBITDA; FCF treated as incidental.

Axis 3 — Value-accretive deployment OR disciplined non-deployment

Canon slugs: buybacks-when-cheap, opportunistic-buybacks-vs-programmatic Desk/binding: [[capital-light-vs-capital-heavy-allocation-is-not-a-style-score]], [[holding-cash-is-a-deliberate-allocation-choice]], [[capital-light-moats-perform-best-in-inflation]], [[avoid-businesses-requiring-constant-reinvestment]]

This axis credits value-accretive action OR disciplined inaction — do NOT penalise an asset-light compounder for low capex. Three Score-2 paths:

• (i) Opportunistic deployment: buybacks / acquisitions / aggressive organic reinvestment in large, irregular blocks when price is demonstrably below IV — then nothing for long periods; management can articulate why the price was cheap.
• (ii) Capital-light high-ROC: the business reinvests little yet earns high ROC and returns the rest — low capex is the feature, not a deployment failure ([[capital-light-vs-capital-heavy-allocation-is-not-a-style-score]]; [[capital-light-moats-perform-best-in-inflation]]).
• (iii) Deliberate cash-holding: management holds cash on purpose to await sub-IV prices ([[holding-cash-is-a-deliberate-allocation-choice]]) — credited ONLY when paired with documented sub-IV opportunity-awareness AND an above-cost-of-capital core business.

Score 1: Some evidence of price-sensitivity in buybacks/M&A, OR capital-light economics without a clear return-of-cash record. Score 0: Programmatic steady-state buybacks regardless of price; acquisitions driven by growth narrative / peer pressure; consistent over-payment for M&A; or idle cash hoarded in a mediocre (below-cost-of-capital) business — that is empire-protection, not discipline, and scores 0.

Axis 4 — Centralized capital + decentralized operations

Canon slug: centralized-capital-decentralized-ops

Score 2: Capital allocation authority sits with the CEO / board (not diffused to divisional heads); operating decisions are pushed to the business unit or branch level; HQ is lean. Score 1: Partial centralization — some capital decisions decentralized or locked in budgets. Score 0: Capital decisions fragmented across divisions; headquarters is bureaucratic; operating autonomy low.

Axis 5 — Leverage matched to cash-flow predictability + contrarian discipline

Canon slugs: leverage-matched-to-predictability, leverage-predictability-test, contrarian-analytical-discipline, contrarian-analytical-temperament

Score 2: Debt (if any) is sized to the predictability of recurring cash flows and stress- tested through economic cycles; management can quantify the FCF cushion. Capital decisions are rooted in independent quantitative analysis, not peer benchmarking or analyst consensus. Score 1: Leverage moderate with some stress-test evidence; or contrarian actions documented but not quantitatively grounded. Score 0: Leverage sized to optimism/guidance, not to cycle-adjusted FCF; or management consistently follows the herd on capital deployment.

Axis 6 — Reinvestment runway × incremental ROCE (the compounding driver)

Desk: [[reinvestment-runway-as-the-sixth-allocation-axis]]

The five style axes measure how management allocates; this axis measures the literal compounding economics — incremental ROCE × length of runway — which is what actually drives per-share value. Score it on realised incremental ROCE history, never a projected TAM narrative (the perspectives layer is promotional; this axis is evidence-gated).

Score 2: High incremental ROCE (>~18–20%; Raamdeo's >15% ROE floor as a minimum, Terry Smith's ~30% as the aspiration) AND a demonstrably long runway — large under-penetrated TAM, QGLP "Longevity", reinvestment opportunity not yet exhausted. Score 1: Solid incremental ROCE but a maturing / partly-penetrated runway; OR a long runway at only moderate incremental returns. Score 0: Incremental ROCE near the cost of capital (reinvestment is not compounding value), OR the runway is nearly exhausted (high ROC with nowhere left to deploy → the engine is stalling).

SCORING + CONVICTION FEED

Sum the six axes (max = 12):

Total	Capital-allocation grade	Conviction feed
11–12	Outsider-class (rare)	+1 to conviction (subject to moat gate)
8–10	Strong allocator	Supports conv 4–5 if moat present
6–7	Adequate / mixed	Neutral; does not lift or cap
4–5	Weak allocator	−1 to conviction ceiling
0–3	Capital destroyer	Cap conviction at 2 regardless of moat

Gates still bind:

• No moat → conviction cap 1, regardless of allocation score.
• Chronically weak/near-zero FCF → conviction cap 2, regardless of allocation score.
• ROIC below ~12% cost of capital → conviction cap 2.
• An Outsider-class score in a commoditized or moat-free business is not a pass — great allocation of bad economics still compounds mediocre results.

OUTPUT (feeds Stage 3 / conviction re-grade)

Return a structured block:

capital_allocation_scorecard:
  per_share_value:        [0|1|2] — <evidence sentence>
  fcf_over_gaap:          [0|1|2] — <evidence sentence>
  value_accretive_deploy: [0|1|2] — <evidence sentence (deployment OR disciplined non-deployment)>
  centralized_capital:    [0|1|2] — <evidence sentence>
  leverage_contrarian:    [0|1|2] — <evidence sentence>
  reinvestment_runway:    [0|1|2] — <incremental-ROCE × runway, realised history not TAM>
  total:                  [0–12]
  grade:                  [Outsider-class | Strong | Adequate | Weak | Destroyer]
  conviction_feed:        [+1 | supports 4-5 | neutral | -1 | cap 2]
  cites_principles:       [slugs returned by the consult ONLY]
  cites_moat:             [moat-gate link — seek-enduring-moats / moat-reinvestment-opportunity-valuation-integration]
  cites_valuation_link:   [reinvestment-rate-terminal-value-consistency — the g = reinvest-rate × ROC identity feeds DCF terminal value]
  portfolio_mirror:       [opportunity-cost-and-capital-allocation — this name's allocation vs the portfolio's best alternative use of capital]
  allocation_kill_criteria: <one sentence — what would prove the grade wrong>

Embed this block in the dossier's "Capital Allocation" section and carry conviction_feed forward into the Stage 3 structured output. The three horizontal edges make the dossier explicitly carry allocation into (a) the moat gate ([[seek-enduring-moats]], [[moat-reinvestment-opportunity-valuation-integration]]), (b) the DCF terminal value — runway × incremental ROCE is the g = reinvestment-rate × ROC identity ([[reinvestment-rate-terminal-value-consistency]]), and (c) portfolio sizing ([[opportunity-cost-and-capital-allocation]]).

Hard rules

1. Never assign Outsider-class (11–12) without documentary evidence on all six axes. Narrative framing from management IR is not evidence — look for actions (actual buyback timing, deal multiples, capex decisions vs guidance, realised incremental ROCE) not words. Axis 6 in particular must rest on realised incremental-ROCE history, never a projected TAM.
2. Always cite only canon slugs the consult actually returns. If principles comes back empty / thin on Thorndike atomics, flag it and score conservatively.
3. The allocation grade feeds conviction but does NOT override the moat gate. A conv 1 (no moat) stays at 1 regardless of allocation score.
4. The Munger/Buffett binding CIO remains the arbiter of moat / verdict / margin-of-safety. This skill supplies the capital-allocation number inside conviction; it does not issue verdicts.
5. Indian listed companies rarely have Axis 3 (buyback) evidence — but Axis 3 now also credits capital-light high-ROC economics and deliberate sub-IV cash-holding, so do not auto-score it 0 for a low-capex compounder. Still score 0 for a routine buyback programme (not opportunism) and for idle cash hoarded in a below-cost-of-capital business.
6. The perspectives layer is context, never the verdict. The dual-pull surfaces 30 external voices (Raamdeo, Mukherjea, Terry Smith, Tambade) who talk their book and frequently agree with each other — they can inform Axis 6 and flag divergence, but they may NEVER override the binding Munger/Buffett CIO on temperament, moat, verdict, or MoS. Always run the separate --corpus binding call; its atoms are the arbiter. A glossy "long runway" narrative does not lift conviction without realised incremental-ROCE evidence, and the binding veto ([[avoid-businesses-requiring-constant-reinvestment]]) sits above the whole scorecard.