Skip to main content

Three Lenses — Belief, Upside, Sequencing

The frame. Three questions for evaluating any product, epic, or flow. Together they form a decision system: the first asks what's the bet, the second asks how big is the prize, the third asks what to do first.

Origin. Erik's 2026-05-18 strategy session. The Investor lens is a Vinod-Khosla / classic-VC frame; the Pincus lens is Mark Pincus (Zynga); the Job-Shop lens is this house. The novelty is using all three together to evaluate the catalog at once.

What this page does. It (1) names the three lenses, (2) applies them to every ranked Forge and Lab bet, and (3) synthesizes a recommendation that re-orders the work by the real bottleneck, which turns out to be distribution.

The lenses​

1. The Investor lens (Vinod / classic VC)​

What are you asking me to believe? What experiment verifies the belief? How much does the experiment cost?

The three sub-questions are the whole job:

  • The belief is the load-bearing claim. Strip away everything else and you have one or two sentences that, if true, justify the work. If you can't write it, the bet isn't a bet — it's a sketch.
  • The experiment is the cheapest test that materially updates your confidence in the belief. Not a perfect test — a cheap one that moves a probability needle.
  • The cost is the actual cost in time, money, attention. Not the hand-wave cost — the real one.

A good Investor-lens result fits on three lines.

2. The Pincus lens (Mark Pincus, Zynga)​

What if everything went right?

The Pincus lens forces a concrete maximalist scenario. Not a fantasy ceiling — the actual operational shape of “we executed, the market responded, the math worked.”

Most strategy work goes wrong by imagining the floor (downside) and ignoring the ceiling. Pincus's question forces you to articulate the ceiling in operational terms. The result tells you which bets have asymmetric upside (low downside / huge upside) versus which bets have a real but bounded prize.

3. The Job-Shop lens (this house, 2026-05-18)​

Given finite hands and a shared substrate, how should we sequence?

Job-shop scheduling is the operations-research framing of we have multiple jobs, each consumes shared resources, each has dependencies, what's the optimal sequence?

Applied to product strategy: which bet should we work on now vs. in parallel vs. later, given that everything pulls on the same engineering hands, the same design hours, the same Erik-attention, the same V3 substrate maturity, the same distribution channels?

Spoiler that the rest of this page unpacks: the bottleneck isn't ingredients. It's distribution.

Per-product application​

Engine​

ProductBeliefExperimentIf everything went right
🐝 Bike4Mind platformA composable AI workshop with all six Deep Agents ingredients beats single-modal chatbots for non-trivial cognitive work, and customers pay for the composition.Already running & passing. Polaris, IonQ B1, TwinSpires, Exacta are the live test.B4M is the AI workshop for any organization that does cognitive work. Verticals spin out under their own brands (Polaris-class) and feed back into the platform.
Polaris (TFG)The Briefcase pattern sold as a vertical works.Already passing.Polaris becomes the analyst-intelligence canon, not just inside TFG but as a platform other analyst orgs license.
IonQ Bucket 1Quantum-domain Briefcase = same recipe, different vertical. If it works for analysts, it works for quantum sales engineers.10 IonQ staff are the live test. This is the customer acid test for V3.Briefcase becomes the V3 reference deployment. Every future enterprise vertical (oncology, robotics, finance) inherits the shape.
TwinSpires / Exacta (zero-line)Services revenue from these accounts funds the substrate. The fact that they don't compose the six pillars is the point — they are cash, not recipe.Already operational.Continuing cash that funds Lane C (long arcs) without burning equity.

Forge — ranked​

#1 — BedrockNews​

  • Belief. News audiences will choose “reliable, authentic, readable, clean” over the noisy default if the curation and writing land.
  • Experiment. Launch under the new name; measure return-visit rate and time-on-site over 30 / 60 / 90 days. Threshold: > 20% returning weekly by day 60.
  • Cost. Days of positioning work plus a content cadence we already have.
  • If everything went right. Durable owned audience. And here's the strategic re-frame: BedrockNews isn't just #1 by readiness — it's the distribution flywheel for everything downstream. The audience that reads news from B4M is the same audience that signs up for StocksAndVibes, NovelClaims, K2Kanji. News is a distribution asset, not a content business.

#2 — K2Kanji​

  • Belief. Thermodynamic-inspired spaced repetition produces meaningfully higher learning rates than Anki / Duolingo-class apps, and the algorithm is patent-worthy IP.
  • Experiment. A/B test K2Kanji vs. Anki on a single domain (kanji acquisition) with N=100 users over 30 days. Measure retention curves.
  • Cost. Three weeks of focused work plus an acquisition budget for the test pool.
  • If everything went right. The most underrated upside on this list. Erik framed it as “a small piece of language learning” — but Duolingo trades around $15B and Anki has a six-figure paid-app revenue line. 1% of language learning is a real outcome. And the algorithm becomes licensable IP into music education, medical certification prep, finance certs. The customer-zero story: Erik teaches himself Chinese in a year at a flow rate that surprises him.

#3 — StocksAndVibes​

  • Belief. Retail traders will pay for an LLM-augmented learning + paper-trading experience that demystifies options and equities without bleeding real money.
  • Experiment. Port the BedrockNews login system; funnel BedrockNews readers into StocksAndVibes paper-trading; measure activation and 30-day retention.
  • Cost. Login port (~2 weeks) plus content cadence. Cheap because we're routing existing audience.
  • If everything went right. Retail traders moving from income class to asset class using StocksAndVibes as their education-and-execution layer — without paying it forward to wealth managers. Subscription LTV + affiliate flow into brokerages. Democratization of asset-class transition is a real product category, and the LLM augmentation is the wedge.

#4 — Starlight​

  • Belief. AI-native DPC EMR with parent-app + voice-cloned AI scribe + synthetic-persona dev beats Atlas.MD because Atlas can't ship those features without rebuilding from scratch.
  • Experiment. Get Dr. Yogini using it daily; expand to 3–5 pediatric DPC practices through her network.
  • Cost. The wedge features in active build + the DPC sales motion.
  • If everything went right. Starlight owns DPC pediatrics; expands to other DPC specialties; eventually a real Atlas replacement; eventually a real EMR competitor for the broader concierge-medicine market. The doctor's own cloned voice writing visit summaries is what no incumbent can replicate.

#5 — LibreOncology​

  • Belief. Master the data-lake + GenAI primitive well enough to put publishing tools back in the hands of the people writing the content. Bespoke professional publishing today is captured by aggregators; LibreOncology is the proof that the captured can be uncaptured.
  • Experiment. Build LibreOncology on V3 packages once V3 lands. Time-to-prototype is the test — if V3 collapses 6 months of custom work into 6 weeks of composition, the bet is proved.
  • Cost. Sunk in V3 work; LibreOncology's marginal cost is small once V3 lands.
  • If everything went right. LibreOncology becomes the reference for a B4M publishing primitive sold into bespoke-publishing markets (medical, legal, scientific, professional certs). Massive upside if the primitive generalizes.

#6 — NovelClaims​

  • Belief. A weekly cadence of agentic paper reproductions creates a scientific clearinghouse of record that researchers return to and cite.
  • Experiment. Ship 4 reproductions in 4 weeks; measure inbound (links, citations, repeat readers, time-on-page).
  • Cost. Editorial cadence + Agentic Research compute (funded anyway).
  • If everything went right. NovelClaims and Agentic Research are one bet, not two. Just like MarsWiki + MoM V2. NovelClaims is the legible face; Agentic Research is the engine. If NovelClaims becomes the scientific clearinghouse, Agentic Research becomes the way scientific knowledge propagates in the agent era. That's a generational outcome. Sequencing note: worth catalog-merging these the way Mars was merged.

#7 — Horde​

  • Belief. One pane of glass for invoices + ledger + cap table + pro-forma beats the QuickBooks + Carta + Excel default, especially for SMB / startup operators who do all four jobs themselves.
  • Experiment. Harden for a second B4M-adjacent user beyond Erik / Kyle (Michael? Nimai?); measure time saved vs. the previous stack.
  • Cost. MFA + multi-tenancy + design polish. Meaningful but bounded.
  • If everything went right. Real B2B SaaS replacing QuickBooks + Carta for the SMB / startup segment. Tessera pro-forma engine spins out as its own product. Horde becomes the operations OS for any B4M-class company.

Agentic Research (unranked)​

  • Belief. Agentic paper reproduction at scale becomes the new substrate for how scientific knowledge gets verified.
  • Experiment. Ship N reproductions per week via NovelClaims; measure researcher inbound.
  • Cost. Compute + Isao's UI maturity.
  • If everything went right. Everyone's chasing this — Anthropic, OpenAI, the open-source crowd. Our wedge is the consumer / editorial integration via NovelClaims. If we own the public-facing reproduction layer, we own the legible face of agentic science.

Lab — ranked​

#1 — PotionQuest​

  • Belief. An RPG runtime where AI plays AI generates a big tools + world data lake that becomes valuable IP — both as a game and as a generic agentic-game runtime others license.
  • Experiment. Ship a playable vertical slice; threshold question: would a human RPG GM find the agent-generated content interesting?
  • Cost. Game design + content polish.
  • If everything went right. PotionQuest is the canonical “agentic game runtime” reference. The world data lake spins out as a licensable RPG primitive.

#2 — Tavern  ⚠️ reframe​

  • Belief (the gift in Erik's verbal prompt). “Manage your agents and employees in a cozy RPG.” This isn't a Lab toy. This is the daily B4M client for power users.
  • Experiment. Integrate the PotionQuest API; let the B4M team manage their actual quests as Tavern characters; measure daily-active use among internal users for 30 days.
  • Cost. API integration + UX layer. Bounded.
  • If everything went right. Tavern leapfrogs from Lab #2 to a Forge platform surface — the cozy-RPG daily-driver UI for Bike4Mind. Every B4M Teams user has a Tavern view of their agents, quests, and team. Potentially the most strategically valuable insight of this whole session — and it came out of an offhand verbal sentence.
  • Sequencing note. Re-evaluate Tavern's tier in 30 days. Lab #2 today; if the dogfood proves out, Tavern becomes a Forge bet inside the Bike4Mind platform-product within a quarter.

#3 — OptiHashi​

  • Belief. Natural language → QUBO → solver-race conversational optimization unlocks business optimization for non-PhDs.
  • Experiment. File the patent; ship a joint demo with IonQ; measure enterprise inbound.
  • Cost. Patent filing + demo polish.
  • If everything went right. OptiHashi becomes the optimization layer B4M sells to enterprise. “ChatGPT for business optimization” lands.

#4 — GoPetto​

  • Belief. Modern LLM + agentic memory + on-device AR makes the GoPets concept genuinely alive — orders of magnitude more present than the 2004 original.
  • Experiment. Prototype one pet with strong personality + persistent memory + AR placement. Test with 10 users for 30-day retention. Threshold: do users name and miss the pet?
  • Cost. Mobile + AR + agent-runtime composition.
  • If everything went right. Cozy-RPG mobile game with retention curves that match cohort-leading mobile games. An owned consumer audience that doesn't depend on news or trading flywheels — a third B4M audience pillar.

#5 — MarsWiki / MoM V2​

  • Belief. Civ-class strategy gameplay powered by agentic AI on a curated MoM canon creates a generation-defining game.
  • Experiment. Ship a vertical slice of Mars colonization with agentic NPCs reading from MarsWiki. Threshold: can you tell the NPCs are agents, and does that matter to the experience?
  • Cost. Long. Years.
  • If everything went right. MoM IP becomes a cultural artifact at the Civilization / Strange Loop scale. Slow burn, biggest upside on this list.

#6 — D&D Room​

  • Belief. The all-max demo (every B4M pillar at heavy intensity) creates an artifact unique enough to stand as a recruiting and partnership magnet — not a product.
  • Experiment. Build the physical room demo for an evening; measure response from the people who experience it.
  • Cost. Custom dev + physical setup.
  • If everything went right. The D&D Room becomes the visit that makes new hires, strategic partners, and investors say “I have to be part of this.” Indirect upside via talent and partnerships, not product revenue.

#7 — DFAL​

  • Belief. Discovery-first quantum research surfaces primitives that pain-first methodology would miss.
  • Experiment. Run a single discovery sprint; see what falls out that we wouldn't have looked for from a customer-pain start.
  • Cost. Erik + a researcher's time.
  • If everything went right. Novel quantum primitives that drive OptiHashi and Agentic Research forward. Long-arc IP.

The synthesis​

The bottleneck is distribution, not ingredients​

The columns of the connections map are the ingredients — Data Lake, GenAI Window, ReAct, Tools, Memory, UX. We have all six at heavy intensity. We have demonstrated composition across 25 products. The substrate is abundant.

The rows are the products. Every product is a recipe. And every product needs to reach humans for the cash loop to close.

Cash loop = product + recipe + audience. Two of the three are paid for. The third — audience — is the bottleneck for everything on the Forge ranking that isn't already Engine.

Erik named the symptom in the same session: “distribution marketing channels is definitely our bottleneck going forward to success.” This page makes the consequence explicit.

Distribution-establishing vs. distribution-consuming bets​

Once you see the bottleneck, the Forge bets reorganize themselves into two classes:

ClassWhat they doWhich bets
Distribution-establishingCreate an audience B4M can reuse for everything downstream. The audience is the asset.BedrockNews · K2Kanji · NovelClaims · Tavern (reframed)
Distribution-consumingLand via channels that other bets establish (or via partnerships / direct sales).StocksAndVibes · Starlight · Horde · LibreOncology · OptiHashi · Agentic Research

This is the strategic re-frame the three-lens analysis unlocks. Readiness-to-market told us which bets are shippable soonest. The distribution view tells us which bets we should ship soonest to unblock the rest. Those are different questions, and the answers are different.

The job-shop sequencing​

Three concurrent lanes, each running at a different cadence:

Lane A — Establish owned distribution (highest velocity priority)​

Until at least one B4M-owned channel is real, every downstream bet has to find its own audience the hard way. This lane is the choke point.

  • BedrockNews launch + audience build — news as flywheel
  • K2Kanji customer-zero identification — a school, an org, a music program
  • Tavern as the B4M-team daily driver — dogfood + internal proof + the seed of a Teams surface

Lane B — Land via distribution (kicks off as Lane A produces channels)​

  • StocksAndVibes activates the moment BedrockNews can funnel
  • Starlight presses on wedge features (AI scribe, parent app, auto-summary) + the DPC industry channel via Yogini
  • Horde finds beachhead user #2
  • NovelClaims + Agentic Research as a paired weekly cadence (catalog-merge candidate)
  • LibreOncology activates the moment V3 lands

Lane C — Build the deep capability (background, long-arc)​

These are about what we are, not what we sell next quarter.

  • Mars project — MarsWiki + MoM V2 / Civ4x.com
  • OptiHashi + DFAL — quantum IP
  • GoPetto — third audience pillar (cozy-RPG mobile)
  • D&D Room — recruiting + partnership demo

Two specific moves this analysis recommends​

1. Reframe Tavern from Lab toy to Forge platform surface​

Erik's verbal sentence — “manage your agents and employees in a cozy RPG” — is potentially the most strategically valuable insight of this entire session. Tavern isn't a Lab game. It's a candidate to become the daily-driver UI for the Bike4Mind platform.

Treat it accordingly:

  1. Integrate the PotionQuest API
  2. Dogfood Tavern with the B4M internal team for 30 days
  3. If daily-active use among internal users is real, Tavern leapfrogs into Forge as a Bike4Mind platform surface within a quarter

Cost is bounded; upside is qualitatively different from the rest of the Lab cards. Run the experiment.

2. Merge NovelClaims and Agentic Research in the catalog​

They are one project, the same way MarsWiki + MoM V2 are one project. NovelClaims is the legible consumer face; Agentic Research is the substrate engine. The current split makes them look like two competing bets when they are a single paired bet.

Render them as one row in the next SVG update. Replace the two cards with a single “NovelClaims / Agentic Research” card in the catalog. Treat the weekly arXiv reproduction as the bet, and the audience + clearinghouse position as the upside.

Distribution is the bottleneck — what's the actual move?​

Erik named the symptom: “distribution marketing channels is definitely our bottleneck.”

Four remedies. Updated 2026-05-18 to reflect Riley Greer's hire and the BedrockNews Reddit win.

RemedyWhat it looks likeStatus
BuildBedrockNews + K2Kanji + Tavern audience surfaces. This is what Lane A is for.In flight. BedrockNews became overall B4M DAU leader after a Reddit ad campaign — 201 DAU record on Friday 2026-05-15. First proof that Lane A is generating real audience.
HireA marketing / distribution leader who treats distribution as a product.Done. Riley Greer is in the seat as of 2026-05 and already shipping (the Reddit campaign above is his work). The forward question isn't whether to hire — it's whether the marketing mechanics are excellent. See Operating Principle #10: “Show up and do the work — 85% is mechanics.”
PartnerIonQ for OptiHashi; DPC industry orgs for Starlight; bespoke publishers for LibreOncology; brokerages for StocksAndVibes.Per-product, opportunistic. Erik handles these naturally as relationships mature.
BuyAcquire a small media property if BedrockNews's organic growth proves slower than the bottleneck demands.Reserve option. The 201 DAU result suggests organic + paid is working; revisit only if measurement at day 60 / 90 stalls.

The forward question, restated. With Riley in the seat and a first audience-establishing win on the board, the strategic question is no longer “who runs distribution” — it's “are the marketing mechanicals excellent?” Google Analytics tagging that's actually wired up. A/B tests that actually run. Daily posts that actually post. The unsexy mechanical layer that makes the difference between “we ran a campaign” and “we ran a campaign and the numbers moved and we know why.” See Operating Principle #10 for the belief that drives this.

The framework doesn't choose for you. It makes the choice legible.

How to use this page​

The three lenses scale down to any decision:

  • Applied to a product, you get the analysis above.
  • Applied to an epic inside a product, you get a faster version.
  • Applied to a flow inside an epic, you get a faster-yet version.

The point isn't the format — it's the discipline.

When you're stuck on a roadmap question, run the three lenses on the candidates:

  1. What am I asking the team to believe?
  2. What does it look like if this goes right?
  3. Where does this fit in the sequence?

Most strategy mistakes are first-question mistakes — the belief is fuzzy. The other two lenses can't save a fuzzy belief. Sharpen the belief first, every time.

Framework added 2026-05-18 by Erik + Claude, working at high temperature. The doc itself is the experiment — does framing the catalog through three lenses generate decisions we wouldn't have generated otherwise? Re-evaluate after 30 days. The two specific moves above (reframe Tavern, merge NovelClaims/Agentic Research) are the first instances of decisions that fall out of this lens system — if those land cleanly, the lens has earned its keep.