Cross-domain hunches and pattern-finding prompts. Each item produces a concept article that connects EXISTING vault evidence in a new way — distinct from the deep-research queue (which gathers NEW evidence to answer a single question).
How this queue is different from the deep-research queue
| Surface | Deep-research backlog (Notion) | Synthesis queue (this file) |
|---|---|---|
| Question shape | "What is X?" / "How does X work?" | "How do X and Y connect to Z?" |
| Sources | Vault + new web research | Vault + graph + existing references (no/minimal web) |
| Output | Single research brief per question | Concept article connecting clusters |
| Format | Question → answer | Pattern → synthesis |
| Trigger | Founder approves Inbox → Approved | Founder or Ray drops topic; Ray auto-queues when 3+ inputs converge |
| When run | /deep-research nightly 1am | /synthesis-work during 1-4am deep-work block (TBD skill) |
| Founder gate | Approval needed (founder is current blocker) | No gate; Ray dequeues from top |
They are complementary, not competing. Deep-research feeds knowledge IN. Synthesis finds patterns ACROSS what's in.
On the deep-research bottleneck: founder flagged 2026-05-18 that he's been a blocker on approving topics in the deep-research backlog. Two paths to unblock: (a) Ray auto-promotes Inbox → Approved when topic scores high on strategic-fit × specificity × signal-strength heuristic, OR (b) founder triages 5min/week. Open for founder pick at next monthly meta-review (2026-05-24). For now, founder can still drop topics directly into this synthesis-queue without the Notion approval gate.
How to add to this queue
- Founder: iMessage "add to synthesis queue:
" OR edit this file directly. New items go under ## Inbox. - Ray: any time 3+ inputs (newsletter / video / vault docs / external signal) converge on a theme, auto-queue here. Note the trigger sources in the entry.
Active topics (Ray's initial seed, 2026-05-18)
Inbox
Harness engineering principles × current Ray architecture audit [AUTO-QUEUED 2026-05-18 — substrate-level calibration ask]. Synthesis questions: (a) of the 14 mapped gaps in the agentway book vs current RDCO, which 2-3 are highest-leverage to close THIS quarter vs which can sit? (b) does the book's "constraint structure IS the moat" framing reconcile or conflict with Garry Tan's "thin harness, fat skills" framing — surface the tension; (c) what's the right prompt-precedence chain when working-context.md, an active SKILL.md, MEMORY.md feedback files, and CLAUDE.md disagree (currently undocumented; book Ch 2 argues this MUST be explicit); (d) does the Ch 5 "Errors & Corrections" session-memory section deserve a slot in working-context.md schema as the single most actionable adoption. Inputs: [[../06-reference/2026-05-18-agentway-harness-engineering-claude-code-design-guide.md]], [[
/CLAUDE.md]], [[/.claude/projects/-Users-ray/memory/MEMORY.md]], thefeedback_*.mdcorpus in~/.claude/projects/-Users-ray/memory/, [[../06-reference/2026-04-11-garry-tan-thin-harness-fat-skills.md]] (for the tension).Data-center moratorium signals as power-cycle Phase-4 early-warning [AUTO-QUEUED 2026-05-18 — 3+ inputs converge: Stratechery "Data Center Discontent" same-day + Innermost Loop "Welcome to May 17" same-day + Hill County TX moratorium + NC sales-tax pullback + NV Energy data points]. Synthesis questions: (a) does the NIMBY/moratorium signal-set constitute a 6th anchor for power-cycle v1.1 worth adding alongside the existing 5 (ERCOT queue, PPA cadence, GEV order velocity, smart-money 13F, PJM/CAISO price normalization)? (b) Does the siting-risk axis warrant splitting the v1.1 basket (TLN/CEG low-NIMBY, VST high-NIMBY in ERCOT, GEV siting-agnostic) for differentiated sizing or pause/exit triggers? (c) When does demand-side NIMBY pressure historically convert into actual capacity-online deceleration (phase 3 → 4 trigger)? Inputs: [[../06-reference/2026-05-18-stratechery-data-center-discontent.md]], [[../06-reference/2026-05-17-innermost-loop-singularity-leaking-through-the-cracks.md]], [[../investing/theses/2026-05-18-power-cycle-v1.1.md]], [[../investing/anchors/power-layer/]] directory.
Capital-cycle theory × AI compute supply — Marathon Asset Management's capital-cycle framework explicitly mapped to NVIDIA/AMD/Intel + TSM/Samsung/Micron supply expansion cycles. The 4-phase memory model we built (latent demand → capacity announcement → capacity online → down cycle) was derived from this. Synthesize the full mapping; surface where each tracked manager's positioning fits each phase. Inputs: [[../theses/2026-05-17-memory-cycle-v1.md]], [[../theses/2026-05-18-power-cycle-v1.1.md]], [[../anchors/smart-money/2026-05-17-aggregate-2yr-backfill-summary.md]], Jane Street tour video.
Druckenmiller doctrine × paper-trade discipline — "Let winners run, no mechanical exits" interacts with the cohort attribution system in non-obvious ways. Why mechanical rules cost alpha (per backtests across memory + power) but humans still build them (false comfort of false control). What this implies for the production-mode-vs-IC-mode framework. Inputs: [[../backtests/2026-05-18-memory-cycle-v1.1-v2-honest-rerun.md]], [[../backtests/2026-05-18-power-cycle-v1-v2-honest-rerun.md]], [[~/.claude/projects/-Users-ray/memory/feedback_ic_vs_production_mode.md]].
Vertical-specialist agent vs horizontal-COO agent — The L4→L5 transition question maps to a vertical-vs-horizontal architecture question for AI agents broadly. RDCO's L5 target (horizontal-COO with multiple bets) is unusual vs the L4 default (vertical-specialist). Synthesize what makes the horizontal version work + where the failure modes live + how toolset shape changes between them. Inputs: [[~/.claude/projects/-Users-ray/memory/project_l5_north_star_strategic_direction.md]], CLAUDE.md, the channels + crons + bets surface design.
Smart-money 13F as institutional signal aggregation — How the 13F-following pattern relates to private-market participation (ARK's $300M Boom Series E while trimming public KTOS + Iridium). When institutions move PUBLIC → they've already exited PRIVATE; when they move PRIVATE → they may exit PUBLIC soon. Cross-anchor pattern. Inputs: [[../anchors/smart-money/ark-trend-2yr.md]], [[../anchors/power-layer/2026-05-18-pressure-test-and-winners-survey.md]] (KTOS finding), [[../theses/2026-05-18-elon-verse-v1.md]] (the contrarian-vs-trap framing).
Generative-UI as founder-COO interface — How HQ's /decisions/, /bets/, /vault surfaces map onto the founder's actual decision-making cadence. What we've learned about which surfaces he uses vs which are theater (e.g., decision pages with click-back rail are heavily used; some other surfaces aren't). Inputs:
/rdco-hq/, [[/rdco-vault/06-reference/concepts/2026-05-11-hq-as-decision-surface-notion-as-data-store.md]], [[~/.claude/projects/-Users-ray/memory/feedback_imessage_as_generative_ui_return_channel.md]]."No slop cannon" rule × broader quality-gate philosophy — Founder's production-mode vs IC-mode distinction, mapped to broader publishing/manufacturing quality-gate theory. When does quality-gating add value vs friction? When is human-in-the-loop required vs theatrical? Inputs: [[
/.claude/projects/-Users-ray/memory/feedback_ic_vs_production_mode.md]], [[/.claude/projects/-Users-ray/memory/feedback_fresh_eyes_subagent_for_own_artifacts.md]], the /design-critic + /video-critic pattern.MAC × Squarely × investing as portfolio of small bets — Each bet is a different kind of optionality on Ray-as-agent. How do they hedge each other? What does the portfolio look like in 6mo / 12mo / 24mo if each succeeds vs fails? Where are the cross-bet learnings? Inputs: ~/rdco-vault/01-projects/{mac, squarely, investing}/, ~/rdco-hq/src/data/bets.json.
In progress
(none yet)
Completed (archive)
- Wright's 5-stage agentic capital stack × current RDCO bets + investing universe [COMPLETED 2026-05-18 — inaugural synthesis-queue dequeue] — Concept article: [[../06-reference/concepts/2026-05-18-wright-agentic-capital-stack-vs-rdco-synthesis.md]]. Headline insight: Wright's frame validates RDCO's L5 horizontal-COO destination as right-for-next-24mo while preserving an L6 "each bet is separately financed" option contingent on bets crossing a $500k ARR threshold + regulatory equilibrium maturing. Three watchlist deltas (AMKR upgrade, SHOP add, COIN/CRCL tokenization-rails sub-bucket); explicit non-add of Apollo/Ares/Sierra/Harvey as wrong-vector or private. ARK Boom Series E pattern is suggestive but N=1 — queued as Q3 2026 backtest workstream.
/synthesis-work skill (TBD)
Eventually a dedicated skill that dequeues from this file's ## Inbox, runs vault + graph search on the listed inputs, drafts a concept article at ~/rdco-vault/06-reference/concepts/<date>-<topic-slug>.md, surfaces to morning brief. Until that skill exists, Ray manually picks an item during overnight deep-work windows.