# Fellow Path Router > Routes a fellow through the Icarus module by situation type, not by tearing down one idea. Fires on "where do I start", "I already have a product / traction — which stages apply to me", "onboard me to Icarus", "which stages should I skip", "I have a mature product, what's my path". Classifies the fellow as Type A (blank page), B (traction, no moat), or C (mature product) on the evidence ladder, then returns a keep / trim / subtract / leap stage ledger and a think:build:test ratio tied to that type's dominant risk. Do NOT fire to tear down or grade one idea (use concept-council or so-what-stress-test), to assess the fellow's own skill level (use fellow-level-ladder), or to draw the platform build-vs-buy line (use platform-vs-venture). - Skill: `the-utopia-studio/fellow-path-router` (Agent Skill, multi-file: 13 files) - Install (CLI): `npx skillmds@latest add the-utopia-studio/fellow-path-router` - Raw SKILL.md: https://api.skillmd.com/api/skills/the-utopia-studio/fellow-path-router/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: The-Utopia-Studio (https://skillmd.com/u/the-utopia-studio) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/the-utopia-studio/fellow-path-router --- # Fellow Path Router ## What it does Takes a fellow's whole situation — not an idea — and returns their route through Icarus. It classifies them as Type A (blank page), Type B (traction, no moat), or Type C (mature product), grades that type on the evidence ladder, and hands back two things: a stage ledger that marks each of the nine Icarus sections KEEP / TRIM / SUBTRACT / LEAP for this fellow, and a think:build:test ratio derived from the one risk most likely to kill this type. It is the first thing a fellow runs at onboarding, before any single idea is torn down. It answers "which stages are mine and in what order", not "is my idea good". ## The Icarus reframe Generic onboarding pushes everyone down the same funnel. Icarus has an opinion: the biggest waste is not skipping a stage, it is re-running a stage whose answer you have already earned with behaviour or money. So the router SUBTRACTS stages you have evidenced past and REFUSES to let you LEAP a stage you only feel you have done. The type is not self-declared — it is graded to the highest rung of evidence the fellow can actually show, and it rounds down, never up. A founder who says "mature product" but cannot show flattening retention is Type B, not C, and the router says so. The ratio is not a vibe: each type has one dominant failure mode, and the split falls straight out of it — a blank page dies from building the wrong thing, so build is capped near zero; a traction- without-moat product dies from being copied, so build is capped and the moat is tested. A router that gives two types the same ledger has done no work. ## When to use / When NOT Use it at first contact with a fellow, when they ask where to begin, which stages apply, what to skip, or how to onboard. Use it whenever a fellow's stage is unclear and effort could go to the wrong section. | Not this skill | Use instead | Why | |---|---|---| | "Tear my idea apart / what am I missing / what should I build first" | `concept-council` | That is a five-lens teardown of one raw idea ending in a probe. This routes the fellow's path across sections; it does not judge the idea. | | "Is this idea worth doing / pressure-test it / so what" | `so-what-stress-test` | That gates one idea across five business questions. This sequences stages, it does not grade a concept. | | "How is this fellow doing / assess them / level them up" | `fellow-level-ladder` | That grades the person's mastery (Literate→Author). This routes the product's path. A Frontier-level fellow can still be on a Type A product. | | "Should we build or use the platform / what does Daedalus give me" | `platform-vs-venture` | That draws the build-vs-buy line for platform primitives. This sequences the fellow's own stages. | If the fellow gives no product signal and no problem signal, do not fabricate a type. Ask the one classifying question (below) and stop. ## Method Copy this checklist and work top to bottom. Fill [template.md](template.md) as you go. ``` Fellow Path Router progress: - [ ] Step 1: Restate the situation in one line + name the highest evidence rung shown - [ ] Step 2: Classify by the evidence gate (round down over-claims; round up under-claims) - [ ] Step 3: Fill the stage ledger — keep/trim/subtract/leap per section, with why + tripwire - [ ] Step 4: Name the dominant risk for this type - [ ] Step 5: Set the think:build:test ratio from the dominant risk, with rationale - [ ] Step 6: Name the entry point + the first artefact to produce - [ ] Step 7: Run the kill-line self-check before returning ``` ### Step 1 — Situation and evidence rung Write the fellow's situation back in one line. Then name the single highest rung of the evidence ladder they can show: money moved (1.0), behaviour observed (0.7), artefact shown (0.5), verbal commitment (0.3), opinion (0.1). If they name no product and no problem, ask **the one classifying question** — "Do real users already touch something you shipped, or is this still a domain and a hunch?" — and stop. Invent no product, no usage, no number. ### Step 2 — Classify by the evidence gate The type is the highest rung the fellow can **evidence**, not the one they claim. | Highest rung the fellow can show | Type | One-line meaning | |---|---|---| | No shipped product; best is a problem opinion or artefact (≤0.5) | **A — blank page** | Still deciding what to build | | A product real users touch at behaviour or money (≥0.7), but the moat is still a plan (≤0.3) | **B — traction, no moat** | Has demand, at risk of being copied | | Product plus money AND a compounding asset with evidence it is accruing — flattening retention or a growing proprietary corpus at ≥0.7 | **C — mature product** | Has a moat forming, must keep it compounding | Round the claim down, never up: - Claims C, cannot show flattening retention or a compounding asset → regrade **B**. - Claims B, product usage is free or opinion only (≤0.5) → regrade **A**. - Claims A but already has paying users → up-grade **B**; do not waste them on a front funnel they have earned past. State the down-grade or up-grade out loud, with the rung that forced it. The most common correction is C→B. ### Step 3 — Fill the stage ledger Mark each Icarus section for THIS fellow's type. Verbs: **KEEP** run in full; **TRIM** run a light version; **SUBTRACT** skip, with the tripwire that would re-add it; **LEAP** jump ahead and start here; **GATED** run only once a probe clears 0.7; **not yet** blocked until an earlier stage lands. | Section | A — blank page | B — traction, no moat | C — mature product | |---|---|---|---| | 00 concept-council | KEEP — entry point | SUBTRACT · tripwire: pivoting to a new problem → re-enter here | SUBTRACT · tripwire: launching a new line → that line re-enters as its own Type A | | 01 first-principles-problem | KEEP | TRIM — traction answers "is it real"; re-run only `product-as-decision` to sharpen | SUBTRACT | | 02 customer-discovery | KEEP | TRIM — mine your own users via `trace-to-interview`, don't cold-start | TRIM — continuous / trace-driven, not fresh interviews | | 03 yoda-your-own-data | KEEP — start the corpus | **KEEP** — this is your missing moat piece | KEEP — the corpus is the compounding asset | | 04 probes | KEEP — earn the right to build | KEEP — probe the moat wedge, not the next feature | SUBTRACT — you are past demand proof | | 05 frame | TRIM — light frame + first mocks | SUBTRACT `first-mocks` (live product); KEEP `usability-test-protocol` | TRIM — frame the next bet only | | 06 invent-wedge-moat | TRIM — sketch the moat hypothesis, don't over-build it yet | **KEEP** — the whole point for this type | KEEP — defend and deepen the moat | | 07 commit-build-v1 | GATED — not until a probe clears 0.7 | KEEP — `eval-first-spec` + `guardrail-design` + `value-based-pricing` | KEEP — tighten guardrails and pricing | | 08 refine-flywheel | not yet | TRIM — start the retention-flattening check now | **KEEP** — the core; the loop is where value compounds | ### Step 4 — Name the dominant risk Each type dies one way. Name it in one sentence; the ratio in Step 5 must answer it. | Type | Dominant risk (the thing most likely to kill it) | |---|---| | A | Building something nobody needs. A killed concept costs a week; a built wrong product costs a quarter. | | B | Commoditisation. Every week shipping features is a week the moat does not compound, and a copyable product loses. | | C | The flywheel stalls, or autonomy gets promoted by feel instead of by eval result. | ### Step 5 — Set the think:build:test ratio Split the fellow's next 4–6 weeks of effort across three modes, summing to 10. **Think** = deciding what to build and why it is defensible (00, 01, 02, 03, 05-frame, 06-invent). **Build** = shipping the v1 / agent / workflow (07, 09). **Test** = getting evidence back — probes, usability, evals, flywheel measurement (04, 05-usability, 07-eval-first, 08). The split is not free; it must be justified by the dominant risk. | Type | think : build : test | Why this split (must tie to the risk) | |---|---|---| | A | **6 : 1 : 3** | Build is capped near zero — the only expensive mistake is building. Test (probes) is how you earn each unit of build. Build stays low until a probe clears behaviour 0.7. | | B | **4 : 2 : 4** | Think is re-pointed at the moat, not the problem (traction already answered that). Build is capped at 2 — feature-shipping is the trap that starves the moat. Test proves the moat hypothesis. | | C | **2 : 4 : 4** | Think is cheap now (discovery is trace-driven). Build is shipping-behind-signoff. Test is re-running the whole eval set on every model release — the discipline most pilots skip. | The **build** number is a cap, not a target. Effort above it requires a probe result at 0.7. ### Step 6 — Entry point and first artefact | Type | Open this stage first | First artefact to produce | |---|---|---| | A | 00 `concept-council` | A Gap Map + one probe under a week | | B | 03 `yoda-your-own-data`, then 06 `invent-wedge-moat` | An explicit/tacit ledger + a moat-design canvas | | C | 08 `refine-flywheel` | A flywheel cadence doc + the re-run-evals-on-model-release schedule | ### Step 7 — Kill-line self-check Run the checks at the foot of [template.md](template.md). If any fails, the route is wrong, not the template. ## Evidence standard The type itself is scored on the ladder — money 1.0 → behaviour 0.7 → artefact 0.5 → commitment 0.3 → opinion 0.1. A fellow cannot self-declare a type; each carries an evidence bar. B requires behaviour or money (≥0.7) on the product. C requires a compounding asset with evidence it is accruing (≥0.7). When the evidence is missing, the default is the lower type, never the flattering one. Tag every situation claim `[Fact]`, `[Assumption]`, or `[Hypothesis]`; an untagged "we have traction" is an opinion until behaviour or money backs it. Done means: type graded with the rung that set it, a ledger that differs by type, and a ratio whose build cap is justified by the named risk. Never let an opinion promote a type or lift a build cap. ## Gotchas - **Type is claimed, not shown.** Fellows round up. "Mature product" often means a free beta with churn; "traction" often means five friends on a free tier. Grade to the highest rung they can evidence and name the down-grade with the rung that forced it. - **Subtracting a stage is not skipping its evidence.** A Type B subtracts concept-council because the concept survived contact, not because teardown is beneath them. Every SUBTRACT carries a tripwire; a pivot to a new problem fires it and sends them back to 00. - **The ratio is a cap, not a suggestion.** Type B's build:2 is the whole point. A router that lets a Type B spend 6 on build has failed the type — it has funded the commoditisation it was meant to prevent. - **Same flow for everyone is the failure mode.** If two types come out with the same ledger or the same ratio, the classification did no work. Differentiation is the deliverable; a generic funnel is the auto-fail. - **A working tool is not always a venture.** The mundane case — a live internal tool with real users — can still be a feature, not a business. When the moat section returns "nothing compounds here", route to `so-what-stress-test` before funding more build. ## Examples See [examples/sample.md](examples/sample.md) for a full run on Mentix: a factory-copilot pilot the fellow calls "traction-ready" is graded to Type B (behaviour 0.7 on usage, moat still a plan at 0.3, so not C), routed to LEAP into YODA and the moat sections with 00/01 subtracted, and handed a 4:2:4 ratio whose build cap is justified by the commoditisation risk. ## Related skills - Sits in section 00 beside `concept-council` and `so-what-stress-test`. Those tear down or gate ONE idea; this routes the fellow's whole path across sections 00–08. Run this first at onboarding; run those when a specific idea needs attacking. - Feeds every downstream section by naming the entry point and first artefact. - Not `fellow-level-ladder` (section 10): that grades the person's mastery; this routes the product's path. The two are orthogonal — a fellow of any level can be on any product type. - Not `platform-vs-venture`: that draws the build-vs-buy line for platform primitives; this sequences the fellow's own stages. - Supersedes nothing. It is the module's front door.