investing-build-thesis

Given a topic or thesis-id, produce a full v2-process thesis doc: source-aggregated structural argument, locked tickers, R-unit sizing, bear case, and the load-bearing cross-domain translation. Output is the artifact founder reads before authorizing paper deploy.

Status: scaffold v0 (2026-05-17). The implementation (source-fetch driver, devil's-advocate subagent, cross-domain mapping prompts) lands in a future build session. This file is the SOP shape.

Why this skill exists (read every time)

Founder's edge in investing is NOT investing expertise. It's pattern recognition trained on data engineering, systems, and supply chains. The thesis-build process must translate every investing pattern into a structural shape founder's existing muscle can ratify.

If a thesis only reads as "AI demand is growing → buy chips," it has not done the work. The thesis must read as "this is structurally like <known pattern from data eng / systems / supply chain>" — and the founder must be able to nod, "yes, I've seen that shape before, and I know what breaks it."

The Latticium method was rejected 2026-05-17 (wrong-method given the asymmetry of a small paper portfolio + structural-anchor approach). This skill replaces it.

Inputs

• topic-or-id — either a known thesis bucket (memory, power, space, longevity, elon-verse) or a new topic slug (e.g. space-cycle, longevity-lly).

Invocation

/investing:build-thesis <topic-or-id>

Examples:

• /investing:build-thesis memory → produces 2026-MM-DD-memory-v1.md for the memory bucket
• /investing:build-thesis space-cycle → produces a new thesis on space-economy structural read
• /investing:build-thesis longevity-lly → produces a name-specific thesis under the longevity bucket

Procedure

Step 1 — Source aggregation (5-10 primary sources per thesis)

Pull from, in priority order:

1. Vault tracked-author CRM (~/rdco-vault/03-contacts/ + 06-reference/) — find existing reference notes from tracked authors covering this bucket. Diamandis, Stratechery, Pabrai, Druckenmiller, Marathon Asset Mgmt, Not Boring (Packy), Seattle Data Guy when relevant to upstream data infra.
2. WebSearch for primary sources — earnings transcripts, capex guidance, supply-side capacity announcements, regulatory filings. NOT investment-bank PT-target writeups (those are noise).
3. Earnings transcripts for the candidate tickers — pull the last 4 quarters via WebSearch or, if the founder authorizes, a transcript API. Capture forward-guidance language verbatim.
4. Adjacent-discipline reads — supply-chain papers, capacity-planning literature, anything that lets you say "the structural constraint here is X" with primary-source backing.

Cap at 10 sources to keep this from sprawling. Each source must be cited in the final thesis doc.

Step 2 — Bull case synthesis (structural argument)

Write the bull case as a structural argument, not a price-target argument. Required components:

• The binding constraint — what's the bottleneck this thesis exploits? (e.g. for memory: HBM packaging capacity is the constraint that gates AI compute scale-out.)
• Demand-side anchors — what specific demand signals would confirm the thesis is playing out? (e.g. hyperscaler capex guidance increases, HBM ASP increases QoQ, packaging-capacity announcements.)
• Disqualifying conditions — what would break the thesis? (e.g. capacity announcements that close the bottleneck in <18 months, demand-side slowdown signal.)
• Locked tickers — explicit list with rationale per ticker. No more than 5-10 names per bucket.

Cite the supporting sources inline. The argument should be readable by someone who isn't deep in the bucket.

Step 3 — Bear case (devil's advocate, 500 words)

Spawn a fresh-eyes subagent (per feedback_fresh_eyes_subagent_for_own_artifacts.md) with:

• The bull-case synthesis as input
• An explicit bear-case mandate: cite contradicting evidence from the vault + external sources, name the specific assumptions that would have to break, and produce the 500-word counter-argument
• Forbidden from hedging — the subagent must commit to a bear position even if the parent thinks the bull case is strong

Embed the bear-case verbatim in the thesis doc under ## Bear case. Do not soften it before filing.

If the v2 quality process upgrades this to a standalone /investing:devils-advocate skill, this step delegates to that skill instead.

Step 4 — Cross-domain translation (load-bearing)

This is the section that decides whether the thesis is deploy-worthy.

Write 100-300 words that map the investing pattern to a pattern the founder already knows from:

• Data engineering (e.g. "this is structurally like a Snowflake credit-consumption bottleneck — usage scales faster than capacity provisioning, ASPs rise, then capacity catches up over 18-24mo")
• Systems / infra (e.g. "this is like the 2018 SSD shortage — supplier consolidation + demand spike + ~2yr fab cycle = predictable price arc")
• Supply chain (e.g. "this is the auto-grade chip shortage in microcosm — single point of failure in the supply chain that downstream actors did not price in")

If the mapping is forced or generic ("this is like data growing"), the thesis is not deploy-ready. Founder cannot ratify a thesis on muscle he doesn't have. Either find a real structural analog or surface to founder that the cross-domain bridge isn't there.

Step 5 — Position sizing + phase markers

Use the locked v1 risk frame from the project README:

• 1R = $5,000 (5% of $100k paper portfolio)
• Max position per name = 2R
• Max allocation per bucket = 4R
• Tranche accumulation per the memory-cycle template — initial position at 0.5R on thesis-confirmed entry, scale to full size on confirming phase-markers

Specify phase markers explicitly: what observable signals tell us we're in phase 1 vs phase 2 vs phase 3 of the thesis. Exit rules tied to phase-marker flips, not calendar.

Step 6 — Write the thesis doc

File at ~/rdco-vault/01-projects/investing/theses/<YYYY-MM-DD>-<thesis-id>-v1.md.

Mirror the structural shape of 2026-05-17-memory-cycle-thesis-v1-framing.md (the v2 template). Required sections:

• Frontmatter: date, type: investing-thesis, thesis_id, status (start as framing-pending-founder-picks, transitions to locked after founder review), tags
• ## The thesis — one-paragraph statement
• ## Bull case (structural argument) — Step 2 output with cited sources
• ## Bear case — Step 3 verbatim subagent output
• ## Cross-domain translation — Step 4
• ## Locked tickers — table with ticker, rationale, max position size
• ## Demand-side anchors — observable signals that confirm thesis is playing out
• ## Disqualifying conditions — what would archive the thesis
• ## Phase markers — phase 1/2/3 definitions
• ## Position sizing — R-unit table per ticker, tranche schedule
• ## Exit rules — phase-marker flips + profit targets + drawdown stops
• ## Sources — numbered list of the 5-10 primary sources from Step 1
• ## Related — wikilinks to vault docs

Step 7 — Spawn paper-trade-go decision page

Generate a companion decision page at ~/rdco-vault/01-projects/investing/decisions/<YYYY-MM-DD>-paper-trade-<thesis-id>-go.html (or the equivalent HQ /decisions/ route).

4-option click-back:

• APPROVE — deploy paper per locked parameters
• REVISE — return to Ray with parameter notes (specifies which parameters to revisit)
• ARCHIVE — thesis does not deploy; file the doc as status: archived-pre-deploy
• DEFER — hold for next quarterly review

Per the iMessage-as-return-channel pattern, the buttons route via sms:ray@raydata.co?body=<click> so founder clicks close the loop without infra.

Step 8 — Return

One-line summary:

<thesis-id> | <N sources> | <bear-case strength: strong/medium/weak> | <cross-domain analog> | decision page: <url>

Dependencies

• Vault read access (qmd MCP)
• WebSearch / WebFetch for primary sources
• Subagent dispatch for the devil's-advocate step (or, future: /investing:devils-advocate standalone skill)
• Companion skill: /investing:backtest-thesis runs walk-forward against the locked tickers + rules from this output

Caveats

• Cross-domain translation is the load-bearing element. If a thesis can't be translated, do not paper into the rationale — surface to founder that the bridge is missing.
• Source aggregation must include primary sources. PT-target writeups and analyst recommendations are noise; they do not count toward the 5-10 source minimum.
• The bear case must be authored by a fresh-eyes subagent, not by the parent that wrote the bull case (confirmation bias).
• All output stays in paper-trade scope until Stage 3 founder gate clears.

Related

• [[01-projects/investing/README.md]] — v2 quality process + dual-identity framing + locked risk frame
• [[01-projects/investing/theses/2026-05-17-memory-cycle-thesis-v1-framing.md]] — template thesis shape
• [[.claude/state/investing-toolchain-buildout.md]] — Stage 2 will drive this skill across the 5 v1 buckets
• [[.claude/skills/investing-backtest-thesis/SKILL.md]] — companion: walk-forward backtest against this skill's output

Changelog

• 2026-05-17 — v0 scaffold authored. SOP shape locked; implementation deferred to future build session.