Scored ideas.Evidence attached.
650 curated software ideas, each carrying a ProvenScore built from real search, SERP and App Store evidence — and a 13-section build prompt you paste straight into Claude Code, Cursor or Codex.
ProvenScore rates the evidence we could verify, not the outcome of a business. Validate before you build.
A preview of the ProvenTools dashboard: a sidebar with Ideas, Assistant, Smart Search, This Month's Drop, My Projects, Validate My Idea and Guides, beside a list of curated ideas sorted by ProvenScore, each row showing its score and evidence tier.
── The ProvenScore ──
A list of ideas is cheap.Knowing which one to trust is not.
Hype lists rank ideas by how exciting they read: no demand data, no look at who already owns the search results, no note on whether one person could ship it, and no date telling you when any of it was last true.
So every curated idea here carries one 0–100 score instead — rolled up from three weighted subscores, each built from source-appropriate evidence, and stamped with a confidence level and the date it was last checked.
Example — how a scored idea reads
Current evidence is consistent across demand, gap and build fit.
- Demonstrated demand45% of the score
- Evidence that people already look for this: buyer search behaviour, marketplace search, comparable products that exist and charge for it.
- Opportunity gap30% of the score
- How much room is left. Strong demand served well by entrenched incumbents scores lower than the same demand served badly, or not at all.
- Builder feasibility25% of the score
- Whether a solo builder or small team can realistically ship and run it. A hard builder-fit failure caps the whole score below 40.
Where the evidence comes from
Different kinds of software leave different traces, so each opportunity type is scored against sources that actually apply to it — plus a manual review.
- Web SaaS
- Google keyword and SERP dataBuyer-intent search evidence, plus what the current first page already answers.
- iOS app
- Apple App Store searchWhat the store actually surfaces for the query, and how well the incumbents are rated.
- Developer tool
- Workflow and ecosystem evidenceWhether the workflow the tool serves is one developers demonstrably already have.
- Platform app
- Marketplace search and API feasibilityWhat the marketplace ranks, and whether the platform API can support the build at all.
What a score does not claim
ProvenScore rates the evidence we could verify, not the outcome of a business. Validate before you build.
Current source-appropriate evidence passed its contract. That describes the evidence, not the outcome of a business.
Weak evidence stays visible
Little usable source evidence. The score is deliberately conservative and is not a measurement of demand.Thin evidence is pulled toward the midpoint instead of inflated, an input with no usable source reads as “not measured” rather than zero, and older records scored under a previous method stay labelled as such.
Read the full methodologyHere's one.Fully unlocked.
This is a real build prompt from the library, not an excerpt written for a landing page: tech stack with reasoning, database schema, file structure and the first 10 tasks. Copy it, paste it into Claude Code, and your agent starts with the decisions already made.
A B2B SaaS fraud detection layer that sits between affiliate program activity and payout execution. It ingests raw click/conversion event streams via webhooks, runs heuristic risk scoring across multiple fraud signals (time-to-convert, IP clustering, conversion rate anomalies), and gates payouts behind a hold-and-review queue. The defensible moat is the configurable rule engine—every program has different fraud patterns, so partner-level scoring tuned to your program beats generic solutions.
Partnership managers at bootstrapped-to-Series-B SaaS companies running affiliate programs through platforms like PartnerStack, Impact, or FirstPromoter—typically a single person or a two-person team managing 50–500 affiliates. Their current workflow: export a CSV from the affiliate platform every two weeks before payout, eyeball conversion counts in a spreadsheet, Google suspicious partner names, and approve payouts on gut instinct. They've already paid out fake leads at least once and have no audit trail when disputing with a bad partner.
1 of 650. Want the other 649, each with its own score and evidence? See pricing →
Sort 650 scored ideas by ProvenScore, then narrow by category, difficulty or a plain search.
One click copies that idea's full 13-section build prompt to your clipboard.
Paste it into your AI coding agent. It starts building with the decisions already made.
All three work in the browser with a Library you own outright. Add ProvenTools Live and the assistant will also search on evidence confidence, validation state and budget — or run the first two steps for your agent.
── What you own ──
One payment.The whole library, permanently.
$99 once buys every scored idea, every build prompt and every drop that ships after it. There is no tier above this one.
Curated ideas, with the score attached
Full breakdowns — problem, audience, revenue model, build scope — and on every one a ProvenScore with its three subscores, an evidence confidence level, a validation state and the date it was last checked. Sort the library by score, then narrow it by category and difficulty.
A build prompt per idea
Written to be pasted whole into Claude Code, Cursor or Codex, so the agent starts with decisions already made instead of asking you for them.
A drop every month
50 newly scored ideas each month with a written release report. Every drop that ships after you buy is already paid for.
Read the drop historySmart Search
Describe the thing you want to build in your own words and get matching ideas from the curated library, the idea libraries and the platform databases — not keyword matching, actual meaning.
And the boring parts
Execution guides and launch tooling, already written.
AI search optimization · MVP building · Growth tactics · Marketing · Business models
Launch Checklist · Get Cited by AI · Progress Graph
Underneath the curated library sit the original volume databases, kept because they are useful for niche hunting — not because the total is the point.
116k+ records in total
- Tool ideas
- 99k+
- Chrome extensions
- 9,289
- Shopify apps
- 3,008
- Discord bots
- 1,996
- Build ideas
- 188
- Idea libraries
- 74
── Optional layer ──
Let your coding agentquery the library itself.
ProvenTools Live is a $9/month add-on for Library owners. It opens the library to your agent over MCP, CLI or REST, adds the assistant, and lets you re-check evidence or score an idea of your own.
Optional add-on
$9/month
100 credits each paid month. Cancel any time — everything in your Library stays yours.
The ProvenTools assistant
0 creditsAsk for a shortlist in plain language, compare two ideas side by side, or ask why an idea scored what it scored. It searches on the fields the browser cannot: evidence confidence, validation state and budget.
MCP, CLI and REST access
0 credits to readConnect Codex, Claude Code, Cursor or your terminal with official local clients. Search ideas, inspect evidence and pull build prompts at 0 credits.
Evidence refresh
3 creditsRe-run the sources behind any idea and get a fresh score, a new checked date and an append-only history entry showing what moved and why.
Validate My Idea
10 creditsSubmit your own idea and put it through the same ProvenScore pipeline as the library: subscores, confidence, validation state and the evidence behind it.
codex mcp add proventools -- npx --yes @proventools/mcp@0.1.0-beta.2
The official local MCP and CLI betas are live. Both are read-only, pin exact package versions and reuse a key saved through the secure CLI login.
What actually costs credits
Browsing, searching and every stored Library read over the web, MCP, CLI or REST costs 0 credits. Only evidence refresh and Validate My Idea spend credits.
| Action | Available in | Credits |
|---|---|---|
| Browsing, Smart Search and stored Library reads | Web, MCP, CLI, REST | 0 |
| Assistant recommendations, shortlists, comparisons, explanations | Web, MCP, REST | 0 |
| Evidence refresh | Web, REST | 3 |
| Validate My Idea | Web, REST | 10 |
Every credit-bearing action shows its exact cost and your projected balance before it runs, and refunds itself if the job produces no usable result. Included credits reset each Live month; purchased credits last 12 months.
Live is genuinely optional. Nothing in the $99 Library is held back for it: every idea, score, subscore, evidence snapshot and build prompt is readable without a subscription, forever.
Compare Library and Live── Check it yourself ──
No testimonials.Just things you can open.
Idea lists are easy to fake and hard to verify, so the useful proof is not a wall of quotes. It is whether the work is visible before you pay for it.
The drop history is public
A new batch of scored ideas has shipped every month since February 2026, each with its own release report. Read the changelog before you buy anything.
See every monthly dropThe scoring method is public
The three weighted subscores, the tier bands, the confidence levels and what "validated" is allowed to mean are all written down and unchanged behind the paywall.
Read the methodologyThe ideas themselves are public
Curated ideas have an open page covering the problem, the market and the shape of the opportunity. The 13-section build prompt and the deep sections are what you pay for.
Browse the idea pagesSample it before you decide
A preview set of curated ideas is free — the details unlock with your email. And the build prompt further down this page is the real thing, not an excerpt written for a landing page.
Get the free ideasPricing
Buy the library once
650 curated ideas, a 13-section build prompt for each one, and every future monthly drop — for a single payment. The live layer is a separate $9/month add-on you can switch on and off.
ProvenTools Library
Own the library. No subscription.
$99one-time
- 650 curated ideas with a 13-section build prompt each
- Every future monthly idea drop, permanently
- The stored ProvenScore, subscores, evidence confidence and checked date for every idea
- 99,781+ tool ideas and 4 more idea databases
- 74 idea libraries (1,480+ seeds)
- Smart Search, saved projects, guides and launch tools
One payment. Yours permanently, including every future monthly drop.
ProvenTools + Backlink Rocket
Library forever, plus Backlink Rocket fulfilment.
$299one-time
- Everything in ProvenTools Library, permanently
- Backlink Rocket fulfilment — 12 dofollow backlinks for one project (DR 20–39)
- 3 months of ProvenTools Live included (100 credits each)
Backlink Rocket is fulfilled separately from ProvenTools. The three included Live months are a grant, not permanent Live access.
Optional add-on
ProvenTools Live$9/month
Requires ProvenTools Library. Every paid month includes 100 credits, which reset at the end of that month and do not roll over. Cancel any time — your Library stays.
- 100 credits every paid month (they reset monthly and do not roll over)
- The ProvenTools assistant: recommendations, shortlists, comparisons and score explanations
- Official MCP, CLI and REST access for Codex, Claude Code, Cursor and other agents
- Refreshed evidence and append-only ProvenScore history
- Evidence refresh (3 credits) and Validate My Idea (10 credits)
Available once you own the Library.
Browsing, searching and every stored Library read over the web, MCP, CLI or REST costs 0 credits. Only evidence refresh and Validate My Idea spend credits. Extra credits are 100 for $9 or 500 for $29, and purchased credits last 12 months.
Backlink Rocket, and what it has actually moved
12 dofollow backlinks for one project across DR 20–39 sites, plus 3 months of ProvenTools Live. Domain Rating movement after joining the network, verified in Ahrefs:
Backlink Rocket is fulfilled separately from ProvenTools. The three included Live months are a grant, not permanent Live access.DR metrics verified via Ahrefs. ScreenEdit was added at no charge and did not purchase the bundle. Network results are from Elysium-owned sites. Results vary depending on your site's existing profile, content and other factors.
Unlock all 10 ideas
Enter your email for instant access. We will also send the ideas to your inbox.
ProvenTools Library is $99, once. It gives you the curated idea library, every build prompt, every future monthly drop and all 116,000+ ideas, permanently. ProvenTools Live is an optional $9/month add-on for the assistant, agent access and credit-bearing actions. ProvenTools + Backlink Rocket is $299 once and adds Backlink Rocket fulfilment plus 3 months of Live.
No. The Library is a single payment and access does not expire. Only ProvenTools Live is recurring, and cancelling Live never touches your Library.
Browsing, searching and every stored Library read over the web, MCP, CLI or REST costs 0 credits. Only evidence refresh and Validate My Idea spend credits. An evidence refresh costs 3 credits and Validate My Idea costs 10. Every Live month includes 100 credits, and those expire at the end of that month rather than rolling over. Extra credits are 100 for $9 or 500 for $29; they last 12 months from purchase, and if Live lapses they stay recorded but cannot be spent until Live is active again.
Yes, with ProvenTools Live. Official read-only MCP and CLI betas are available for Codex, Claude Code, Cursor and terminal workflows, alongside the REST API. Save your key once through the secure CLI login; the local MCP server reuses it without putting the secret in client configuration. Every exposed MCP and CLI action currently costs 0 credits.
ProvenTools Live can be cancelled at any time from your billing page and stays active until the end of the paid period. One-time Library and bundle purchases are final because of the digital nature of the product. If something is broken, contact support and we'll make it right.
Yes. A new idea drop ships every month with a release report — 50 newly scored ideas — plus ongoing database and feature work. Every future drop is included in the one-time Library price.
Stop researching.Start shipping.
Pick an idea by what the evidence says, then hand your agent a build prompt that already answers the stack, the schema and the first ten tasks.650 scored ideas. 50 more every month.