agentleFS
pando-bpando-b/summer-solstice/skills/demand-confidence-gate/SKILL.md

demand-confidence-gate

Skill0 starsChanged 46 days agoMITView on GitHub

Decide fast whether an idea earns a build week — by killing most of them. Use this skill after an opportunity has been scored and you're deciding whether to build, whenever the user asks "should I build this", "is this worth building", "validate this idea", "demand test", "will anyone pay", "kill or build", or needs a pre-build go/no-go. This is Recipe #5 of the Summer Solstice recipe book — the gate between scoring and building. Its default verb is KILL: it earns confidence from data you already have (revealed preference, existing spend, engagement intensity) in hours, not weeks, and only lets insanely-high-confidence ideas through. Use it instead of slow "validation" theater (smoke pages, week-long waitlists, cold-community posting), which is too slow for a one-week build cycle.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 7 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    After a candidate clears the Scorecard (≥75) and before it enters a build week.

    line 26
  2. Distinctive:

    Run them in order. Most ideas die in Tier 1.

    line 30
  3. Very specific:

    Look for revealed preference, not opinions.

    line 34
  4. Distinctive:

    Use access you already have — never gate on building new community standing (earning Reddit karma is slow and unreliable.

    line 57
  5. Distinctive:

    Pre-register the kill bar: before looking (e.g., "needs ≥2 independent proofs of existing spend + emphatic intensity").

    line 100
  6. Distinctive:

    No cold-community standing-building: If a probe needs karma/clout you don't have, it's not fast — skip it.

    line 132
  7. Distinctive:

    Cap the whole gate at ~1–2 days: If you can't get to High or a clean warm-probe result quickly, the answer is kill.

    line 133
  8. Distinctive:

    Don't rationalize a weak signal into a build: When unsure, kill — there's another candidate behind it.

    line 134

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Demand Confidence Gate (Recipe #5)

The gate between scoring (Recipe #2) and building (Recipe #6). A high Scorecard score says a problem is real and painful; it does not say anyone will pay. This skill's job is to earn — fast and cheaply — a real signal of demand, and to kill most ideas so only the highest-confidence ones consume a precious build week.

Philosophy (why this works)

Mark Pincus's bar: a click is not validation — retention, usage, and an emphatic "10 out of 10 would use this" are. And "your instincts are right ~95% of the time, but your ideas are wrong ~75% of the time" — so we don't trust the idea, we look for evidence the demand already exists. In the AI era the build is nearly free and fast, which flips the old logic: faking demand with a smoke page often costs more time than just building the thin real thing and watching whether people use it.

So this gate is falsification-first (try hard to kill the idea) and signal-first (use data you already have before running any new test). Confidence must be earned; the default outcome is kill.

When to use

After a candidate clears the Scorecard (≥75) and before it enters a build week. Input: the scored candidate from Recipe #1 — problem, persona, venue, and the evidence already gathered.

The three tiers

Run them in order. Most ideas die in Tier 1. Escalate only when needed.

Tier 1 — Signal proof (default; hours; ~$0)

Look for revealed preference, not opinions. Reuse last30days and targeted pulls to check, with dated, linkable evidence:

  • Existing spend: are people already paying for inferior tools or hacking workarounds? (Competitor pricing pages, revenue/Latka, install counts, app-store review volume, "I pay for X and it still sucks.") Money already moving is the strongest signal.
  • Demand intensity & recency: high-engagement complaints in the last ~30 days (upvotes, comments, views) — and is it emphatic, the Pincus "10/10 fuck-yes," or merely lukewarm?
  • Search/market pull: keyword/"alternatives to X" demand, rising interest.
  • Spend-by-proxy: people hiring or doing the task manually = budget exists.

Convergent, strong, recent signals → High confidence. Anything thin or lukewarm → kill. This tier alone should kill most candidates.

Hard cap (External Review #01): category evidence can never clear the gate alone in a crowded market. Competitor spend proves the category wins — A2X's 13k customers validate A2X, not a new entrant. In any market with entrenched incumbents (Competitive Wedge ≤ 3 on the Scorecard), Tier 1 caps at Medium regardless of how strong the category signals are, and the switch test (Tier 2) is mandatory. In a crowded space, heavy existing spend is evidence against an easy entry, not for one.

Tier 2 — One warm probe (if Tier 1 is ambiguous, or mandatory as the switch test; 1–2 days max)

The switch test (mandatory when Wedge ≤ 3): find ≥3 target-persona humans who currently pay a competitor and get a real commitment from each — a pre-order, a paid pilot, or an explicit, specific "I'd switch because ___" in a Mom-Test-style conversation. Politeness doesn't count; a named reason to switch or money does. Fewer than 3 commitments within 2 days → kill. This is the only cheap signal that de-risks your build rather than the incumbent's category.

Use access you already have — never gate on building new community standing (earning Reddit karma is slow and unreliable; don't depend on it).

  • Pre-sell / paid pilot to known prospects — take a deposit or a card. Money committed is the cleanest signal.
  • Targeted warm outreach to the exact people who voiced the pain. Recipe #1 already surfaced specific complainers; a contact-lookup CLI (e.g., Printing Press's contact-goat: LinkedIn → warm-intro graph → verified email) can reach them. Send a short, specific, personal offer of a fix — not a blast.
    • Compliance guardrail (do not skip): cold email to found contacts is regulated. Keep it permission-aware and personal, honor CAN-SPAM (identify yourself, easy opt-out) and GDPR/region rules, and respect each platform's ToS on scraping/automation. Before doing this at any scale, run the legal compliance-check. A handful of hand-written, relevant messages is fine; an automated blast is both legally risky and a weaker signal.
  • Offer to your own audience (LinkedIn/email/network) where you already have reach.

A real commitment (a pre-order, a paid pilot, a "yes, take my money / when can I have it") → pass. Silence or politeness → kill.

Tier 3 — Build-as-the-test (the AI-era move)

For ideas that are high-confidence on pain but unproven on willingness-to-pay, the cheapest real test is to build the thin real thing (one week, agent-native) and launch it where the audience already lives — an existing app store/ecosystem, the Printing Press Library, or your own audience — then measure the only signals that count: usage, retention, and first dollars. Building beats faking when building is this cheap. (This hands off directly into Recipe #6 with a pre-set "alive" metric.)

Tier-3 eligibility rule (added 2026-07-07, owner directive — the no-outreach path). Tier 3 REPLACES outreach as the demand test when ALL of these hold:

  1. The pre-chosen channel has organic discovery (directory/store search, SEO surface) — buyers find the product without us messaging anyone. Build-as-the-test measures our demand, not the category's, which is the very failure the switch test protects against — but only if discovery is organic.
  2. An instant payment rail exists — charge from day one; first dollars are the zero-interpretation signal.
  3. The pre-registered alive metric + kill window ship in v1 (instrumented, read in Week B). The falsification discipline moves post-launch; it does not disappear. No bar → no Tier-3 pass.
  4. The build is cheap and inference-light/within the build week (true by construction after S2's gates).

Boundary (External Review #01 protection stands): at Wedge ≤ 3 the switch test remains mandatory and Tier 3 cannot substitute for it. A Wedge ≤ 3 candidate whose channel is outreach-dependent (forum-DM, FB-group, 1:1 motions — no organic discovery) is BANKED for a cell where outreach is accepted — never rationalized through the no-outreach path.

No-outreach demand tactics (use in place of, or alongside, warm probes):

  • Upgrade-intent doors: free tier ships with Pro visible; a click on "Upgrade" before Pro is finished is purchase-intent telemetry from a real user inside a real workflow — stronger than any stated "I'd pay."
  • Real checkout from day one: a hosted checkout (Stripe, Paddle, or whatever billing rail your channel supports) costs about an hour to add.
  • Self-qualifying lead magnet: a free diagnostic whose output quantifies the user's own pain ("this is costing you $X a month") — discovery converts itself into demand measurement.
  • Public revealed-preference data: competitor install/download curves (most app stores and package registries publish counts), competitor waitlists (a waitlist IS a demand queue), search volume. Zero contact, already public.
  • Async broadcast ≠ outreach: one launch post or demo video in the venue is publishing (S11's job), not 1:1 solicitation — the channel pulls or it doesn't.

Gate logic

  1. Pre-register the kill bar before looking (e.g., "needs ≥2 independent proofs of existing spend + emphatic intensity"). This stops you rationalizing later.
  2. Score confidence High / Medium / Low from Tier 1. If Wedge ≤ 3, Tier 1 caps at Medium — no exceptions, no "signal proof may clear it without a test."
    • High → advance to build (Tier 3), with a pre-set alive-metric and kill window.
    • Medium → one Tier 2 warm probe (the switch test, if Wedge ≤ 3); pass → build, fail → kill.
    • Low → kill, log why.
  3. Bias to kill. The sprint's scarce resource is build-weeks; protect them. Killing fast is winning.

Output format

DEMAND CONFIDENCE GATE — <candidate / persona>
Wedge score (from Scorecard): <n>  → switch test mandatory? <Y/N>
Pre-registered kill bar: <what would have to be true to pass>
Tier 1 signals:
  • Existing spend: <evidence + link + date>
  • Intensity/recency: <evidence + engagement metric + link>
  • Search/market pull: <evidence>
  • Spend-by-proxy: <evidence>
Economics check (S5 rung): <does what these people ACTUALLY pay today support the S4 pricing corridor?
  If the revealed spend sits below the corridor floor, the margin_floor PASS is stale — flag before BUILD>
Confidence: High | Medium | Low
Decision: BUILD | ONE WARM PROBE | KILL
If BUILD → pre-set "alive" metric + kill window: <e.g., ≥1 paying customer or ≥X activations within 14 days>
If WARM PROBE → exact probe + pass/fail threshold + deadline (≤2 days)
If SWITCH TEST → the 3+ named prospects, what each currently pays for, and each one's commitment/reason
Rationale: <2–3 lines>

Speed rules / anti-patterns

  • No smoke pages as proof. A click measures curiosity, not demand.
  • No week-long waitlists. They burn the build week they're meant to protect.
  • No cold-community standing-building. If a probe needs karma/clout you don't have, it's not fast — skip it.
  • Cap the whole gate at ~1–2 days. If you can't get to High or a clean warm-probe result quickly, the answer is kill.
  • Don't rationalize a weak signal into a build. When unsure, kill — there's another candidate behind it.

Compounding (do this every run)

Log the predicted confidence vs. the actual outcome (did built ideas get usage/retention/revenue? did killed ideas turn out to be missed winners elsewhere?) in sourcing-log.md. Over time this calibrates the gate — Pincus's whole point is to sharpen instinct with feedback, not to trust it blindly.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.