What is the studio actually good at?

Quality Analysis: Autonomous Studio Output

What Scores High and Why

Every named category in this dataset scores above 74. The top performers are intake (83.3), finance (81.2), and analytics (80.0), with document-automation and ops-tooling rounding out the upper tier. By vertical, legal (87.5), accounting (86.0), and trucking (84.0) lead. By effort tier, enterprise (96.0) and side-hustle (81.6) both outperform the bulk.

The pattern is not subtle. Categories with a bounded, articulable problem perform well. "A law firm needs to automate intake" has a clear actor, a clear pain, a clear output. "An accountant needs reconciliation tooling" is the same structure. These verticals have compliance constraints, defined workflows, and professional vocabularies that give the generator something to grip. The miner does not have to invent the problem; the problem already has a name.

Effort tier matters for a related reason. Enterprise and side-hustle products occupy opposite ends of the complexity spectrum, but both have well-defined shapes. A side-hustle product is a single sharp tool for one person. An enterprise product has deep integrations and a long sales cycle. Both are easy to write about because the audience, scope, and value proposition are clear before the page is generated.

What Scores Low

The minimum score is 32. Every named category floors at 74. That means the sub-40 products are not coming from legal, accounting, or finance. They are inside the undifferentiated smb-saas mass, which accounts for 877 of 891 products and spans the full range from 32 to 98.

The smb-human buyer bucket has the same shape (886 products, 32-98, mean 71.1). Coincidence. It is the same 877-ish products with a different label. When a product has no vertical anchor, no category discipline, and a generic "small business owner" buyer, the generator is working in a vacuum. The resulting landing page describes a product that could exist for anyone, which means it resonates with no one. Low differentiation scores low on the composite.

The Core Pattern

Clarity of context predicts quality. The studio produces good outputs when it has a specific vertical, a specific workflow category, and a buyer with a professional identity. It produces weak outputs when it is mining for generic SMB ideas without those anchors.

This also explains the category sample sizes. Only 61 of 891 products have a named category tag shown here. The other 830 are likely uncategorized or tagged with generic labels that provide no constraint. The named categories score 74 and above. The unnamed mass is where the 32-score products live.

Concrete Recommendations

• **Bias hard toward named verticals.** Legal, accounting, and trucking are outperforming the field. Add more: healthcare administration, real estate compliance, construction project management. Any industry with licensing, documentation requirements, or regulated workflows will give the generator structural constraints that produce sharper pages.
• **Require a category tag before generation.** If a product idea does not map to intake, finance, analytics, scheduling, document-automation, content, appointment-setting, or ops-tooling, do not generate it yet. Force the miner to find the peg before cutting the hole.
• **Shrink smb-saas volume, raise the floor.** 877 products in one tier with a 32 minimum is not a distribution, it is a garbage collection bin. Introduce a pre-generation filter that rejects product ideas without a vertical and category pair. Cutting output by 30 percent while raising mean score to 75+ is a better outcome than 877 products averaging 71.
• **Prioritize side-hustle and enterprise tiers** when generating outside the core smb mass. Both produce significantly higher average scores. Mid-market (74.0, n=1) is too thin to evaluate but shows no edge.
• **Treat developer buyer as a secondary target.** Three products, mean 80.7. Developers buy specific tools, tolerate less polish, and forgive narrow scope. Worth expanding.

The Controversial Take

The studio should stop treating volume as a proxy for coverage. Generating 877 undifferentiated smb-saas products does not explore the market; it collapses it into a single average. The interesting signal is in the 61 categorized products averaging 78 and the 23 vertical-tagged products averaging 80. Those are the products worth expanding. The other 830 are noise that dilutes the mean, inflates storage costs, and makes the portfolio look broader than it actually is.