Fund thesis format

fund-thesis-format — the 7-part structure

This is a single source of truth for HOW to articulate an investment bet. Every new thesis MUST conform to this structure. The structure renders at three depths — a reference file, a proposal memo, a chat/phone render — using the same skeleton, just at different densities.

The idea is progressive disclosure: the same content, three rendering surfaces, no thesis ever lives at only one depth.

Why this exists

Three problems this solves:

1. Comparability. Without a fixed structure, theses can't be ranked, conviction can't be tracked, kill triggers can't be audited. A standardized skeleton makes any thesis comparable to any other in the book.
2. Falsifiability. Section 4 (INVALIDATION) is mandatory. No thesis enters the book without a written kill condition. This is the discipline that prevents narrative drift.
3. AI-agent + human dual readability. The YAML frontmatter is machine-queryable; the prose body is human-readable; the chat render is phone-skim-friendly. One source, three surfaces.

The 7 sections

Every thesis MUST have these 7 sections, in this order, with these exact section names. Skip a section ONLY by writing one line stating WHY it doesn't apply (e.g., "no asymmetry framing for a hedge — see section 4"). A section silently omitted is a bug.

1. CLAIM

A single declarative sentence stating the directional bet. No qualifiers ("potentially," "should," "may"). Just the bet.

Pattern: [Buy / Sell / Hedge with] [INSTRUMENT] because [SPECIFIC OBSERVABLE WILL HAPPEN] over [HORIZON].

Examples (good):

• "Buy XBI 2% NAV because equal-weight biotech captures the AI-drug-discovery premium that cap-weight IBB dilutes, over 6-18 months."
• "Hedge with IBCI 1.4% NAV because Eurozone inflation will surprise upward as ECB lags AI-capex spillover, over 12-24 months."

Examples (bad — rewrite):

• "XBI looks attractive given biotech tailwinds." (No bet, no observable, no horizon.)
• "Add some infrastructure exposure for diversification." (Diversification is not a thesis.)

2. CAUSAL MECHANISM

The chain of cause-and-effect that connects the observable in §1 to the price moving in your favor. Must be explicit, not assumed.

Format: numbered chain, 3-7 links. Each link names an actor or mechanism, not a vibe.

Example (XBI):

1. Big pharma (LLY/PFE/JNJ) facing patent cliffs 2026-2030 → must replenish pipeline
2. AI-drug-discovery shortens preclinical timeline 4-7 years → 2-3 years
3. Small/mid AI-native biotech (Recursion, Schrödinger) becomes M&A target
4. Equal-weight XBI captures takeover premium across 200+ candidates
5. Cap-weight IBB blunts effect (Amgen/Gilead = 25% of fund, no upside from M&A on themselves)
6. → XBI outperforms IBB; you own the right vehicle for the structural pattern.

This forces "I believe X because Y because Z" articulation. No pattern-matching.

3. PORTFOLIO FIT

Where this slots in the existing book. CITE NUMBERS — correlations to current holdings, cluster impact, sleeve consumption.

Required sub-elements:

• Diversifies what? Specific cluster or position the bet's correlation profile reduces overlap with.
• Duplicates what? Specific cluster or position the bet partially overlaps. State the overlap so you don't pretend it's pure-new.
• Cluster impact: Does this add to an existing cluster (concentration risk), open a new cluster, or sit between clusters?
• Sleeve consumption: which sleeve (e.g. ETF backbone / single-stock / crypto / event) and how much of it does this consume post-add?

A thesis without a portfolio-fit section is incomplete because it doesn't answer "why now, why this, why not something else you already own?"

4. INVALIDATION (kill conditions) — MANDATORY, NUMERIC

The most important section. What concrete, observable, numeric event would PROVE this thesis wrong (in part or full)?

Must be:

• Numeric — "rate spikes >5.5%" not "rates rise"
• Time-bound — "5+ consecutive sessions" not "for a while"
• Distinct — full-exit triggers separated from trim triggers separated from "watch closely" triggers
• Asymmetric to entry — exit triggers should be tighter than entry conviction, not looser

Three classes of trigger:

• HARD STOP — full exit, no judgment call. Usually -20% from entry, or a specific structural break.
• THESIS KILL — full exit because the underlying mechanism (§2) is broken. E.g., "AI-capex guidance reduced at META/GOOGL/MSFT" for an infra bet.
• TRIM TRIGGER — partial exit (typically 50%) on heightened risk that doesn't fully break the thesis. E.g., "10Y rates spike >5.5% sustained" for a rate-sensitive bet.

A thesis without numeric invalidation is faith, not investment.

5. SIZING + TIME HORIZON

Why THIS size, not bigger or smaller. Why this horizon to evaluate.

Required sub-elements:

• Size class: STARTER (1-2% NAV, low conviction or new asset class) / CONVICTION (3-5% NAV, well-understood thesis with track record) / CORE (6%+, only after multi-month positive evidence) / HEDGE (sized to the risk it offsets, not the alpha)
• Why this size: Articulate the bound. "Why not 5% NAV? Because correlation to AI-credit cracks scenario is unproven; need 3 months of behavior before upsize."
• Horizon to evaluate: When do you re-grade conviction? "12 weeks for a starter; if tracking diverges by >5ppts that's the upsize/exit decision."
• Marginal-risk-contribution citation (recommended for any addition >2% NAV, required for single-stock if your mandate demands it).

6. ASYMMETRY

Up-case vs down-case rough magnitudes. Without this, sizing is arbitrary.

Format: Up: +X% over <horizon> if <condition>. Down: -Y% over <horizon> if <condition>. Asymmetry ratio ≈ X:Y.

Example (XBI): "Up: +25-40% over 12-18m if AI-drug-discovery thesis manifests as accelerating M&A flow. Down: -15-20% if rates spike + biotech credit dries up. Asymmetry ≈ 1.5:1 to 2:1."

If asymmetry is less than 1:1, the thesis must explicitly frame as a HEDGE (i.e., you accept negative-EV in exchange for protection in a tail scenario). Don't pretend a hedge is a positive-EV bet.

7. CONVICTION LEVERS

The middle ground between "intact" and "killed." What concrete observables would move conviction up or down WITHOUT triggering full exit/upsize?

Format: bulleted list, each line declarative + observable.

Examples:

• Up-conviction (would prompt upsize proposal):
  • A high-trust outlet specifically corroborates the thesis
  • 3-month price action confirms the mechanism
  • Cross-asset confirmation (e.g., bond market signals same regime)
• Down-conviction (would prompt watch elevation, not full exit):
  • A single high-trust outlet refutes a key link in §2
  • 3-month action diverges meaningfully from the comparable (e.g., XBI vs IBB)
  • Cross-asset divergence (e.g., AI capex narrative confirmed in equity but rejected in credit)

Surface 1: REFERENCE FILE — the canonical implementation

A per-position file (e.g. positions/<TICKER>.md) and per-theme file (e.g. themes/<theme>.md).

YAML frontmatter — machine-queryable. Suggested fields:

---
ticker: INFR:xams
weight_pct: 1.50
conviction: 3
view: hold
opened_at: 2026-04-27T14:23:50Z

thesis_slug: ai-data-center-power-broad
claim: "Buy INFR 1.5% NAV because global infrastructure captures AI-data-center power demand more broadly than URA's narrow nuclear, over 12-24 months."
size_class: starter
horizon_weeks: 52
diversifies: [ai_capex_narrow_nuclear, single-country-utility]
duplicates: [partial: ex_us_equity (25% emerging weight)]
hedge_for: null  # null if directional bet; ticker/scenario name if hedge
asymmetry:
  up_pct: 30
  down_pct: -15
  horizon_months: 18
  ratio: "2:1"
invalidation_triggers:
  - type: hard_stop
    rule: "INFR -20% from entry €33.755 = €27.00"
  - type: thesis_kill
    rule: "Hyperscaler 2027 capex guidance reduced (META/GOOGL/MSFT)"
  - type: trim_50
    rule: "10Y rates >5.5% sustained 5+ sessions"
  - type: structural_break
    rule: "INFR breaks 200d MA (~€30 estimated)"
correlations_at_entry:
  SMH: 0.32
  TLT: 0.30
  SCHD: 0.41
  MCHI: 0.30
  SPY: 0.42
upcase_levers: ["high-trust corroboration of AI-power thesis", "3m positive INFR-IGF tracking", "transmission cap-rate compression"]
downcase_levers: ["single high-trust refutation of capex thesis", "3m INFR underperforms IGF >5ppts", "credit cracks not transmitting to equity"]
---

Body sections — prose 1-7 with these exact headers (use ## 1. CLAIM, ## 2. CAUSAL MECHANISM, etc).

Length: 400-800 words. Above 1000, split detail into a theses/<slug>.md reference file and link.

Surface 2: PROPOSAL MEMO + decision-reasoning

A dated proposal memo (e.g. YYYY-MM-DD-<slug>-proposal.md) and the reasoning field of your decision record.

The proposal memo opens with a 6-line summary block (machine-readable, human-skim-friendly):

CLAIM:        Buy INFR 1.5% NAV because global infra captures AI-data-center power demand broadly, over 12-24m.
SIZE:         €776 = $912 = 1.50% NAV (starter); upsize gate at 12 weeks if INFR-IGF tracks within 2ppts.
HORIZON:      12-18 months evaluation; multi-year hold if thesis confirms.
CONVICTION:   3/5; would move to 4 if hyperscaler capex re-accelerates 2027.
ASYMMETRY:    +30% / -15% over 18m (~2:1 up-skew).
KILL:         Hyperscaler capex cuts (thesis-kill) · 10Y >5.5% 5+ sessions (trim-50) · 200d MA break (structural).

Then sections 1-7 in detail. Keep the memo ~600-1200 words; defer correlation tables and detailed math to the reference note.

The decision-reasoning field gets the 6-line summary block + 2-3 sentences per section. Total ~500-800 chars — long enough for audit-trail completeness, short enough for quick review.

Surface 3: CHAT — 5-line compact render

The phone-friendly version. Use for: idea-scout TOP-N ideas, brief mentions of new bets, formal proposal pings, execution confirmations, weekly retrospective per-position summaries.

Exact format (5 lines + 1 header line):

🎯 INFR (1.50% NAV, conv 3/5, starter · 12-18m)
CLAIM: Capture AI-data-center power demand broadly via global infra basket.
WHY: 2030 +220% data-center power → utilities/transmission/grid pricing power. Not redundant w/ URA narrow-nuclear (corr SMH +0.32).
KILL: Hyperscaler capex cuts · 10Y >5.5% 5+ sessions · 200d MA break.
ASYM: +30% / -15% over 18m (~2:1 up-skew).

Rules:

• Line 1 header: 🎯 <TICKER> (<size>% NAV, conv <X>/5, <size_class> · <horizon>)
• Line 2 CLAIM: the single sentence, copied verbatim from §1 of the reference file.
• Line 3 WHY: condensed §2 mechanism, 1-2 sentences. Cite at least one concrete number (correlation, return, threshold).
• Line 4 KILL: condensed §4 — top 3 invalidators separated by ·. Use the type-prefix format ("trim-50", "thesis-kill") only for clarity.
• Line 5 ASYM: §6 magnitudes + ratio. If hedge, write HEDGE FOR: <scenario> instead of asymmetry numbers.

Total: 5 lines, ~250-350 chars per thesis. Three theses = ~1000-1100 chars — fits inside a typical 4096-char chat message cap with room for other sections.

Failure modes to avoid

• Filling sections with vibes. §2 CAUSAL MECHANISM is the test — if you can't write 3-7 specific links, you don't have a thesis, you have a hunch.
• Vague invalidators. "If it doesn't work" is not a kill trigger. Numeric + time-bound or it doesn't count.
• Sizing without horizon. "2% NAV starter" is meaningless without "evaluated at 12 weeks." Say when you'll re-grade.
• Pretending a hedge is a directional bet. If §6 ASYMMETRY ratio is ≤1:1, the thesis is a hedge. Frame it as one explicitly, including a written "what's it hedging" line.
• Chat render skipping the WHY line. The 5-line format is tight; the temptation is to drop the mechanism. Don't. Pure CLAIM + KILL + ASYM is just a trade alert; you're aiming for analysis someone can challenge.