The Window Problem

Fifty files of proven demand

In the first week of July 2026 we built what is probably an above-average specimen of a common artifact: a research corpus on startup opportunities. Fifty files covering YC and accelerator cohorts, seed and Series A signals, indie-hacker revenue stories, overpriced incumbents, Reddit complaint mining, and app-store monetization mechanics. From it, a ranked list of 25 plays scored on three axes: proven demand, ease of copying for a small team, speed to first revenue.

The list looked authoritative. Every play carried named companies with revenue figures, documented pain points, and a specific differentiation wedge — flat pricing against per-seat incumbents, one-time payment against subscription traps, a vertical niche against generalists.

Then we asked the two questions the ranking never asks. Is the evidence true? And is the window still open? The rest of this report is what happened when we checked.

The numbers are softer than they read

We traced the 31 revenue and traction claims doing the most work in the ranking — the “$10M ARR bootstrapped” and “$0 to $5M in 30 days” figures that make a play look inevitable — back to their sources, and graded each one.

Read strictly, about a quarter of the evidence base can support a build decision on its own. The failures follow one narrow pattern: a single founder tweet or podcast figure, laundered through five to ten SEO “case study” blogs that cite each other and check nothing. The strong claims share the opposite shape — they are anchored to a hard corporate event confirmed by a party with no incentive to inflate: an acquisition confirmed by the acquirer, a public company’s SEC filing, a named-investor round, a company’s own shutdown notice.

The D grades are instructive:

• Icon, the AI ad-creative startup, claimed $0→$5M ARR in 30 days in a founder tweet. The company shut down in February 2026, eleven months after its Founders-Fund-backed launch, and pivoted to reselling human-shot video. The viral number still circulates as proof the category prints money.
• n8n “$40M ARR at a $5.2B valuation” pairs two moments ten months apart. By the time SAP set the valuation, n8n’s own disclosed ARR had passed $100M. The stale pairing overstates the revenue multiple by roughly 2.5×.
• The claim that Zapier→n8n migration runs “12× the reverse” appears in no source we could find. The one independent dataset on record shows roughly 2:1.
• Candy.ai “$25M+ ARR” coexists with a $50–120M estimate built from the same traffic data, and the company has never published a figure. A 2–5× spread between guesses is not a data point.

One correction ran the other way: a widely-cited $15.4M portfolio figure for one consumer-app founder turned out to be a significant undercount — a single app in that portfolio was later independently reported at $30–40M. Self-reported numbers are not systematically inflated. They are systematically unverified, which is worse, because you cannot even sign the error.

Every window was already shut

Evidence quality is the smaller problem. For each of the fifteen Tier-1 and Tier-2 plays we measured the market as it stood in the second week of July 2026: named competitors, current price floors, and whether the corpus’s proposed wedge was still unoccupied.

“Flat pricing” against Canny is UserJot’s homepage headline. “No subscription” QR codes is GetQRCodePro’s. “Your team size shouldn’t determine signing costs” is Zignt’s. “Free scheduling app for one-person service businesses” is FieldVibe’s. The wedges weren’t hypotheses to test. They were other people’s SEO pages.

Three kill mechanisms recur, and only one of them is ordinary competition:

• Platform absorption. MyFitnessPal shipped GLP-1 dose tracking free to its entire base in April 2026, erasing a thirty-app tracker category’s reason to exist. Shopify shipped native B2B wholesale free on every paid plan the same month. A platform giving the feature away is faster than any competitor.
• Open-source pre-emption. The “SOC 2 at $199/mo” play is undercut by two open-source products, one of them free to self-host — and an identical free-for-startups play already ran and failed to dislodge the incumbents in 2022. Flat-rate e-signature faces three open-source signing stacks.
• Demand decay under the story. Two plays sell to seed-stage founders, a population whose deal count fell 15–27% year over year. The demand the stories “proved” is shrinking under them.

The uncomfortable part: demand in most of these categories is genuinely still growing. GLP-1 prescriptions hit 8% of all US scripts in March 2026. AI-search visibility is growing at 40–50% a year. Sub-100-employee field-service shops sit at 29% software adoption — real whitespace. Supply simply grows faster, because the same published evidence that proves the demand also recruits the competition.

Why idea lists self-destruct

Put the two audits together and a mechanism falls out: the supply chain that produces idea lists consumes the opportunity as a side effect of documenting it.

“Published proven demand” is therefore a lagging indicator by construction. A list compiled from success stories describes the windows of 6–18 months ago. We measured the lag empirically, and it is long enough that fifteen out of fifteen wedges were occupied before the list existed.

The decay is accelerating. When a Shopify tool vendor announced its wind-down this June, three purpose-built replacements were live within three weeks. When we sent full diligence at eight “residual openings” our own saturation pass had rated conditionally open days earlier, three of the first four had already been closed by events our keyword searches had missed: an acquisition that absorbed the leading church-transcription product into a 17,000-church platform in late 2024, and vertical incumbents that never market themselves under the category term — the pest-control operations leader does not call itself “field service management,” so category-term searches count it as absence. Thin competition, on inspection, is often a measurement artifact.

Diligence on the residue

The saturation pass left eight conditional openings. We ran each through full diligence: demand structure, competitor map, unit economics, and the distribution asset an entrant would actually need. Six died. One narrowed to a licensing play. One survived with a deadline attached.

Three shelves of live opportunity

After five passes of destruction, what remains sorts into three shelves by the clock that governs each.

The urgency shelf — live signals under 90 days old, shelf life measured in weeks. A one-pass sweep of shutdown orphans, price-hike outrage, and fresh complaint clusters surfaced five actionable-now entries: customer-feedback tooling native to Salesforce (two incumbent products sunset this year, orphaning their base), an AI voice and scribe layer for veterinary practice software, document AI for property-management lease abstraction, single-vertical EU AI Act compliance, and answer-engine-optimization tooling for the llms.txt era. Each needs a narrow wedge; each decays if unclaimed.

The calendar shelf — windows that open on a schedule, where being early is structural rather than lucky. These are the strongest finding of the whole exercise: dated, forced migrations with thin dedicated supply.

The barrier shelf — plays protected by regulation or proprietary data rather than by being unknown. Mapping the space our own corpus had excluded as “too hard for a small team” found it roughly 85% claimed by companies holding $50M to $1B+ in funding — but the openings that do remain are real precisely because a clone flood cannot follow: AI-native professional-services firms in marketing and PR (the be-the-firm model is proven in law and accounting, unclaimed there), full-workflow AI operations for med spas, large-animal veterinary work, and church administration, and credit infrastructure for agent-initiated payments. One caution generalizes: barriers hold when they are regulatory or data-based, and dissolve when they are merely technical.

What compounds: gates, calendar, cadence

The list above is already decaying. The durable output is the instrument that produced it, and anyone can run it against any idea from any source:

• Gate 1 — evidence. Trace every load-bearing claim to a hard event: an acquirer’s confirmation, a filing, a named round, an official dataset. A founder tweet repeated by forty blogs is one source.
• Gate 2 — saturation. Search your intended differentiator as a phrase. If it is someone’s tagline, it is gone. Count open-source occupants. Check whether the platform beneath the play has announced the feature. Check demand direction, not just existence. Remember the two blind spots that fooled us: acquisitions that absorb a category leader, and incumbents that do not use the category’s name.
• Gate 3 — distribution. Name the channel asset you already hold that incumbents lack. “SEO plus communities” does not count in a category with two-year-old comparison pages.
• Gate 4 — primary demand. A landing test, a paid-traffic probe, ten buyer conversations with a price attached. Published demand is other people’s funnel.
• Gate 5 — a pre-registered kill. Decide the week-N threshold and the shutdown rule before writing code. The base rates are unforgiving: half of indie products earn nothing, and the median new app-store entrant earns exactly $0.

Behind the gates, two standing processes: a weekly sweep of the six live-signal classes (about three hours with the search operators in hand), and the dated-event calendar, extended as new deadlines are announced. The calendar clusters are worth planning around now — six events land in the third quarter of 2026, and six more converge in the fourth quarter of 2027.

Opportunity itself is fine — demand kept growing in almost every category we audited. What has changed is that the free information environment now converts publicly legible opportunity into crowded opportunity within months. What is left rewards verification, calendars, and speed — the unglamorous instruments — over lists. Including this one.