MarkosWebWeb intelligence for the agentic economy

New startup ideas · AI for people who run the AI themselves · Analysts, researchers and creators running their own stack

startup concept

Citegrid

Every number in your published research links to a source snapshot you verified

A verification layer for independent analysts who sell their own research.

12

similar startups, last 2 years (39 all-time)

yes

8 matching federal grants and programs

Direction supported by government programs and grants

Test it before you build it

$300 · 4 weeks · 25 prospects

$300 and four weeks to prove analysts who sell their own research will pay $79 to have every number in an issue verified before publish, and come back for a second issue.

Riskiest assumption · An analyst selling paid research trusts an outside claim check enough to pay $79 and hold an issue for 48 hours before publishing, and to come back for the next issue, in a relationship where one missed error ends it.

1Focus group: who and where

An independent analyst publishing paid financial or policy research under their own byline, typically a Substack with 500-5,000 subscribers at $10-30 a month, who drafts with Claude or ChatGPT and knows one traceably wrong number in a paid issue costs subscribers and reputation.

where to find 25 · Substack's public Finance, Business and U.S. Politics leaderboards, which list paid publications by size (directory); Seeking Alpha's contributor pages (list); FinTwit on X, where these analysts promote every issue (channel); FinCon, the annual conference for money media creators, for its published speaker and creator lists (event).

2Sell first, build later

Verification of one research issue before it publishes: every factual claim flagged, matched to a source in the writer's own library or the public record, with timestamped evidence snapshots behind a link they can share with readers. 48-hour turnaround from receiving the draft, performed by hand.

the ask · $79 per issue for the first, then a $59 per issue founding rate locked for six months

a real yes · A real yes is $79 moved before delivery, and the decisive yes is a second purchase after the first report. A thank-you reply to the free audit, a request for another free one, or praise on FinTwit is not a yes.

3Small experiments

The first one attacks the riskiest assumption; each ends with a number that says whether to run the next.

  1. 1. Free audits of published issues

    $200 · 10 days

    Subscribe to 10 paid newsletters picked from the Substack finance and politics leaderboards. Hand-audit one recent issue each: flag every factual claim, trace it to a source, snapshot the evidence, and send the author the claim-by-claim report, including the 2-4 claims that could not be traced. This tests whether their own unsupported numbers move them at all.

    keep going if · 6 of 10 authors reply substantively; 4 of 10 book a call

  2. 2. Paid pre-orders on calls

    $0 · 10 days

    Run 15 20-minute calls with audit repliers plus analysts reached through FinTwit and r/Substack. Offer verification of their next issue before it publishes at $79, 48-hour turnaround, 10 slots, done by hand. Doing it as a founder-run manual service is the right instrument here because the deliverable is a report that can be produced without software, and paying for it is the exact behavior the product needs.

    keep going if · 5 of 15 pay $79 before or within 24 hours of the call

  3. 3. Repeat-purchase trust test

    $100 · 14 days

    Deliver every paid verification by hand within 48 hours, with timestamped source snapshots behind a shareable link, then offer a 3-issue pack at $199. Repeat purchase is the only honest measure of the trust premise, since one missed error should end the relationship.

    keep going if · 3 of the first 5 buyers purchase a second verification or the 3-pack

4Collect a deposit up front

Tesla took $1,000 refundable reservations for the Model 3 and $100 for the Cybertruck before building either: the deposit is the measurement, not the revenue.

$79

per prospect, refundable

how · $79 prepaid at booking through a card checkout on the slot-reservation page, the same way this buyer already pays for their newsletter tooling. Payment reserves a dated 48-hour verification window ahead of their publish date; the founder confirms scope in one email. set up: Stripe Checkout ↗

what it reserves · A 48-hour verification slot tied to their next publish date and the $59 founding per-issue rate locked for six months.

refund · Full refund if the report arrives late, or if the author finds a numeric claim marked verified that its cited source does not support.

target · 5 payments of $79 from 15 calls within 30 days, with 3 of the first 5 buying again

Go: build it if

5 prepaid verifications within 30 days and at least 3 of the first 5 buyers paying for a second issue: the trust premise holds, build the parser and snapshot store.

Kill: stop if

Fewer than 2 payments from 15 calls and 10 free audits, or fewer than half of first buyers returning: a verification product nobody re-buys has already failed on trust, stop.

5 Scripts to run itoutreach message, landing copy, deposit terms · click to open

outreach message

I read your latest issue and hand-checked every number in it against a source. Three claims I couldn't trace; the report is attached, with snapshots of the evidence behind everything else. I'm building a pre-publish verification service for analysts who sell their own research, and I'm doing the first ones by hand. Want me to verify your next issue before it goes out, every claim matched, 48-hour turnaround? 20 minutes this week and I'll walk you through the report.

landing page

Every number in your next issue, traced to a source $79 per issue: a claim-by-claim report with evidence snapshots, 48-hour turnaround Book a verification slot ahead of your next publish date

deposit terms

$79 paid now reserves a 48-hour verification slot ahead of your next publish date and locks a $59 founding rate per issue for six months. You get every factual claim matched to a source, with timestamped snapshots behind a link you can share with readers. Full refund if the report is late or if you find a claim we marked verified that its source does not support.

Would you run this test?

One tap. The yes-share feeds the Demand pillar of this idea's score; nobody sees who answered.

Budgets are out-of-pocket estimates for a team of one to three, US market. Size the deposit to the deal, and check the terms before taking money in a regulated line.

Scorecard

Ranked against every idea in the catalog: trend, demand and 100x potential from the corpus, competition relative to the other ideas. A generated concept has no judges or swipes yet, so its pillars use the data signals only.

63

Idea Score, 0-100 · raw 39.2 x 1.61

Crowded

competition: more crowded than 83% of ideas · headwind x0.59

+4.8

government priorities, secondary (152 matching grants)

Trend

40

Is the wave forming now? 2025-26 entrants vs 2023-24, rounds since 2025, the sector's live-batch direction, the 2026 trend analyst.

  • Entrants 2025-26 vs 2023-24 (similar companies)70
  • Rounds announced 2025+ in the sector0
  • Sector direction (live batch)50

Demand

65

Does anyone want it? YC's current RFS, companies already paid for something similar, the operator judge, founders' yes-rate in decks, readers who would run the test.

  • YC asks for it (current RFS: idea / sector)30
  • Someone already pays (similar companies, recent / all-time)100

100x potential

77

Can it return a fund? The venture judge (double weight), market-size and moat axes, neighbours still alive, the technologist judge.

  • Neighbours still alive77

Score = 100 x cbrt(Trend x Demand x 100x) x (1 - 0.5 x crowding) + government bonus (max 5), calibrated so the 95th-percentile idea scores 90 (order never changes). A geometric mean: a weak pillar cannot be papered over. Percentiles are among the 382 ideas in the catalog; the terms matched were number, published, research, links, source, snapshot, verified, verification.

The concept in full

What
A verification layer for independent analysts who sell their own research. It parses a draft, flags every factual claim, matches each to a source in the writer's library, and stores a timestamped snapshot of the evidence behind a public link. In the first hour a solo analyst connects their document folder, runs one draft through it, and gets a claim-by-claim report they fix before hitting publish.
Grounded in (2025-2026 signals)
Anthropic Economic Index (November 2025 data, reported January 2026): 52% augmentation vs 45% automation on Claude.ai, with the February 2026 report showing augmentation rising again. Emerging term 'knowledge work': 4 companies in 2025-2026 vs 0 before. 2025 cohort companies Percival (yc X25, 'Accelerating data analysis for research') and Hanji (yc X25, 'Turn your hardest documents into reliable data') prove the extraction side is being built while the verification-before-publishing side is not.
What it rides
Rides 'Augmentation edges ahead of automation on Claude.ai': the Anthropic Economic Index shows the iterate-with-the-model population growing, and those people publish under their own name, so trust in their output is the asset they sell. Verification the user controls is exactly what the model alone does not give them.
Why now
The January 2026 Anthropic Economic Index puts augmentation at 52% on Claude.ai, and MIT NANDA (published August 2025) counts more than 40% of knowledge workers running personal AI tools; people who publish AI-assisted analysis under their own byline now need proof they checked it, because their readers know the drafting was done with a model.
Wedge: first customer and entry point
Independent financial and policy analysts who already pay for Claude Max or ChatGPT Pro and sell paid research; entry point is a single 'verify this draft' command that works on one Markdown or Docs file, no integration.
Closest real companies, as the generator saw them
Percival accelerates the analysis itself and Hanji turns documents into data; neither checks a finished draft against sources. Confident AI does LLM evals for engineering teams building products, not for a writer verifying their own claims. Citegrid sits at the publish step, after all of them.
Main risk
Claim-to-source matching that misses even 10% of errors destroys the product's entire premise, which is trust.

Similar startups in the directory

Companies whose pitch matches most of the concept's terms (number, published, research, links, source, snapshot, verified, verification).

  • Ritivelyc W26 · 2026 · Vertical AI agentsalive

    AI-native platform for Life-Sciences Documentation

  • Cognitoyc S14 · 2014 · Security and complianceacquired

    Frictionless, modern identity verification that starts with just a…

  • Epistra Inc.plugandplay · Healthcare and bioalive

    Epistra provides software to accelerate the R&D process using AI.

  • Boatyc F26 · 2026 · Agent infrastructurealive

    The best cloud VMs for billions of agents

  • Datoricyc S26 · 2026 · Data for AIalive

    The secure training data R&D engine for physical AI

  • Perceptron MLyc S26 · 2026 · Vertical AI agentsalive

    We make law firms AI-native.

  • Cautionyc S26 · 2026 · Security and compliancealive

    Hosting platform for software you can't afford to have hacked

  • hiloopyc S26 · 2026 · Agent infrastructurealive

    Autoresearch as a service: we solve your hardest measurable problems

  • Harboryc X26 · 2026 · Healthcare and bioalive

    AI-native CRO for clinical trial operations

  • Kitayc W26 · 2026 · Fintechalive

    Underwrite borrowers around the world in minutes

  • Logos Researchsosv SOSV IBSF 25H2 2025 · 2025 · Developer toolsalive

    AI-native verified algorithmic synthesis

  • Firayc W25 · 2025 · B2B SaaSalive

    Financial research platform for investment firms

Public money in this direction

US federal grants and open opportunities matched to the concept's terms.

Other concepts in this collection

Fictional concept generated 2026-08-26 by claude-fable-5 from the collection's brief and MarkosWeb data. Treat it as a research prompt, not a plan.