Investing — RDCO Project
Changelog
- 2026-05-18 v2 toolchain expansion (post-backtest-critique): Skills added:
/investing:label-historical-phases(Ray-autonomous, builds phase timeline from prior cycles)./investing:backtest-thesisupgraded to v2 (multi-symbol survivorship-free universe, multi-cycle test windows aligned to phases not calendar years, leave-one-out reporting honest about outlier cycles). Driver: founder critic correctly flagged v1 was circular + thin + curated. Parallel deployment path:/investing:smart-money-mirror-v1paper portfolio deploys NOW (mirrors top-N 13F positions), backtest-validated strategies deploy only after v2 infra produces honest tests. - v2 (2026-05-17): Dual-identity framing — added as 6th HQ bet for visibility. v1 scope locked: 5 paper-trade buckets (memory, power, space, longevity, elon-verse) + 3 watchlist items. New quality process: source-aggregated synthesis + devil's advocate + cross-domain translation. Latticium method dropped (founder pushback: wrong-method given asymmetry). Plugin scaffold authored at
plugin/. Sequenced buildout driven by/goalagainst~/.claude/state/investing-toolchain-buildout.md. Risk frame locked: 1R = $5k, $100k paper portfolio, max 2R per name, max 4R per bucket. - v1 (2026-05-12): Initial bootstrap. Diamandis Innermost Loop thesis filed. Alpaca surfaced as brokerage candidate. Strategy/execution boundary table established.
v2 scope (current)
Dual-identity framing (v2)
The investing project occupies an unusual position in the RDCO portfolio:
- NOT an external-revenue bet — no customers, no product surface, no go-to-market motion. Generates no external income.
- IS an HQ bet for visibility — appears on hq.raydata.co
/bets/investing/so it gets the same dashboard treatment as Sanity Check, MAC, Squarely, Ops. The reasoning: this is a real capability being built (like the other bets), with locked strategy, audit trail, KPIs, and quarterly review. Visibility-as-a-bet keeps the operating discipline tight.
Internal-revenue rather than external-revenue — capital efficiency on the balance sheet underneath the operating bets, treated as a first-class RDCO capability.
v1 paper-trade buckets (5)
| Bucket | Thesis frame | Status |
|---|---|---|
| memory | DRAM / HBM / NAND supply-side anchor for AI compute | thesis in build |
| power | Power generation + transmission as binding constraint on Innermost Loop | thesis in build |
| space | SpaceX-adjacent + space-economy beneficiaries | thesis in build |
| longevity | Longevity-biotech beneficiaries of compute-enabled R&D | thesis in build |
| elon-verse | Tesla, X, xAI, Boring, Neuralink direct + indirect exposure | thesis in build |
v1 watchlist (3)
| Item | Why watch | File |
|---|---|---|
| SpaceX IPO | Pre-IPO read so we're not chasing post-listing | candidates/spacex-ipo-watch.md |
| Anthropic IPO | Same shape; founder has unique vantage on the AI-lab landscape | candidates/anthropic-ipo-watch.md |
| Upstream-of-NVDA research | Identify the supply-chain layer NVDA doesn't capture | candidates/upstream-of-nvda-research.md |
Risk frame (v1, locked)
| Parameter | Value |
|---|---|
| Paper portfolio | $100,000 |
| 1R unit | $5,000 (5% of portfolio) |
| Max position per name | 2R ($10k) |
| Max allocation per bucket | 4R ($20k) |
| Bucket count | 5 |
| Max total deployed if all buckets max | 20R ($100k = full portfolio) |
| Drawdown kill-switch | -10% portfolio = pause all auto-execution, route to founder |
What this project is
The RDCO investing project is the operating capability for deploying founder capital that is not earmarked for RDCO operating expenses. It is NOT a bet (no external revenue generated); it is an internal capability that grows the balance sheet underneath the bets.
Founder context (2026-05-12 13:14 ET iMessage): "I've got capital sitting in cash that should be growing. I'd like to improve our capabilities to invest and grow." Followed by the operating-model clarification at 13:19 ET: "Financial investing is a skill like any other profession. We are going to build up that skill." Those messages are the founding mandate for this project: build the investing skill, with Ray executing against founder-locked strategies.
What lives here
| Subfolder | Holds | Cadence |
|---|---|---|
theses/ |
One file per investment thesis. Each thesis names the structural argument, the demand-side anchors, the position-sizing logic, and the disqualifying conditions. | Add when a thesis crystallizes; review quarterly. |
positions/ |
One file per held position (or one ledger file with rolling entries — TBD per founder pref). Tracks ticker, entry date, basis, layer-of-thesis, current state. | Update on every entry/exit. |
candidates/ |
Pre-decision research on tickers not yet held. Each candidate cites the thesis it would fit under. | Add when a candidate surfaces; cull quarterly. |
decisions/ |
Per-allocation decision logs (e.g. "deploy $X into Innermost Loop Power basket"). Each decision links to the HQ /decisions/ page that gates the actual capital deployment. |
Generated as needed when a capital-deployment decision is teed up. |
Current theses
| Date filed | Thesis | Layers | Status |
|---|---|---|---|
| 2026-05-12 | [[2026-05-12-innermost-loop-ai-infrastructure]] | Frontier labs · Chips · Data center infra · Power | active-research, no positions yet |
Strategy vs execution boundary (the operating model)
Founder corrected the original draft of this README on 2026-05-12 13:19 ET. The right division of labor is NOT "Ray researches, founder executes." It's:
| Layer | Owner | Notes |
|---|---|---|
| Strategy definition (what to invest in, why, sizing logic, risk parameters) | Founder + Ray, together | Co-defined in theses/ files. Each thesis specifies the locked parameters Ray will execute against. |
| Capital-deployment authorization (the threshold + the frame to deploy) | Founder, via HQ /decisions/ | Any material deployment requires founder click-through on a /decisions/ page. |
| Strategy execution within locked parameters | Ray | Once a strategy is locked, Ray executes trades within the parameters via brokerage API. |
| Portfolio monitoring | Ray | The actual unlock — Ray has infinite attention while founder focuses elsewhere. |
| Strategy re-validation, parameter changes, thesis archival | Founder + Ray, together | Quarterly. |
This is a real capability we're building like MAC for data or Sanity Check for content. The investing skill is the third pillar.
Disclaimers (load-bearing)
- Capital deployment authorization sits with the founder.
- Ray executes against locked strategy parameters. Outside those parameters, Ray pauses and routes the question back via /decisions/.
- Nothing in this folder is financial advice; this is RDCO's internal operating capability, not external advice.
- The founder can pause Ray's execution at any time via channel ("pause investing", "halt trades", etc.) — Ray stops executing and waits for explicit resume.
Operating discipline
- Thesis-first. Every position must tie to a documented thesis in
theses/. No tickers-because-it's-popular. - Strategy parameters explicit. Each thesis specifies the LOCKED parameters Ray executes against: in-thesis ticker list, max position size per ticker, max daily trade volume, max basket allocation, drawdown kill-switch %, time-window restrictions if any. Anything outside those parameters routes back via /decisions/.
- Position-sizing maps to layer leverage. When a thesis names a binding constraint (e.g. the Diamandis power-layer bottleneck), position size in that layer should reflect the asymmetry.
- Decision pages for capital deployment. Initial capital deployment to a new thesis = founder click-through on a /decisions/ page. Once locked, Ray executes within parameters without per-trade approval.
- Order audit trail. Every executed order logs to vault automatically with timestamp, ticker, side, size, price, thesis attribution. The audit trail is non-negotiable.
- Risk gates. Pre-coded ceilings Ray cannot exceed without explicit founder approval via /decisions/: daily trade volume cap, single-position size cap, total portfolio exposure cap, drawdown kill-switch.
- Kill switch. Founder can pause Ray from any surface. Pause = no trades executed, monitoring continues, alerts continue.
- Quarterly thesis review. Each thesis re-validated quarterly against the demand-side anchors. If anchors broke, thesis archives + Ray exits positions per locked exit logic.
Brokerage / execution substrate
Alpaca Markets is the working candidate (founder + Ray exploring 2026-05-11). Alpaca offers:
- Commission-free equity trading with programmatic API
- Paper-trading sandbox (validate strategies + audit-trail mechanics before live capital)
- Algo-trading-native design (this is the right shape for the operating model)
- API key management via 1Password (per
feedback_no_secrets_on_diskmemory)
Open question: account ownership status (asked founder 2026-05-12 — was yesterday's lookup exploration only, or was the account opened?).
v2 build is sequenced via ~/.claude/state/investing-toolchain-buildout.md, driven by /goal. Stage 1 lands the Alpaca wrapper + the two foundational skills; Stage 2 produces the 5 thesis docs + memory backtest; Stage 3 is founder gate + first paper trades.
Quality process (v2)
Founder picked Path A on the quality question 2026-05-17: skills produce the synthesis, founder judges + approves deploy. Process:
- Source-aggregated synthesis —
/investing:build-thesispulls reference docs across the vault for the bucket, aggregates them into a structural argument with demand-side anchors. - Devil's advocate critic —
/investing:devils-advocateruns fresh-eyes against the synthesis with a bear-case mandate. Cites contradicting evidence from the vault and from external sources. - Cross-domain translation — the synthesis must explain why the structural argument holds in language that crosses the home discipline. (If the thesis only makes sense to people deep in the same niche, it's not load-bearing enough to deploy against.)
- Founder review + deploy authorization — founder reads synthesis + bear case + cross-domain explanation, click-through on
/decisions/to authorize, Ray executes within locked parameters.
Not used: Latticium method. Founder pushed back 2026-05-17 23:20 ET — wrong-method given the asymmetry of the bet (small paper portfolio, structural-anchor thesis approach, not a quant-systematic strategy where Latticium would fit).
Skills + plugin
Skills for this project live in the investing plugin authored at ~/rdco-vault/01-projects/investing/plugin/. Active invocation copies at ~/.claude/plugins/investing/ (and individual skills surface at ~/.claude/skills/investing-<name>/).
| Skill | Purpose |
|---|---|
/investing:build-thesis |
Source-aggregated synthesis for a bucket |
/investing:devils-advocate |
Fresh-eyes bear-case critic on a thesis draft |
/investing:backtest-thesis |
Walk-forward backtest against locked ticker list |
/investing:weekly-pulse |
Re-validate active theses against demand-side anchors |
See plugin/README.md for namespace + install pattern detail.
Related
- [[06-reference/2026-05-12-diamandis-innermost-loop-ai-infrastructure-thesis]] - anchor research piece that founded this project
- [[06-reference/2026-04-30-not-boring-scarce-assets-abundance-driven-scarcity]] - Packy framework on scarce-asset migration; relevant for cross-thesis analysis
- [[06-reference/2026-05-12-stratechery-spacex-anthropic-xai-musks-two-companies]] - frontier-lab compute supply landscape (Layer 1 context)
- [[04-finance/]] - finance + capital state context