Erik Marutian :Flashcrafter.ai

Solving Service Business Growth and Operations needs at 1/10x cost

[Erik Marutian :Flashcrafter.ai @aitechguy1]

Check to try: compare review counts and visibility separately. Record a baseline, choose one change and measure again. Treat any improvement as a result to investigate.

Measured 2026-08-10. https://t.co/I7xvIb8Mxy

[Erik Marutian :Flashcrafter.ai @aitechguy1]

Use the fillable worksheet on the research page to choose one action, record a baseline and set a review date. Want to understand your own business? Explore its free Growth Brief there.

[Erik Marutian :Flashcrafter.ai @aitechguy1]

This is a pattern observed in one sample, not proof that review count causes ranking; other unmeasured factors could also be at play.

[Erik Marutian :Flashcrafter.ai @aitechguy1]

Sacramento, CA — appliance repair service.

Higher-ranked listings tend to have more reviews

[Erik Marutian :Flashcrafter.ai @aitechguy1]

Among the 25 listed businesses, the median review count for the first three ranked was 40, versus 15 for the rest.

[Erik Marutian :Flashcrafter.ai @aitechguy1]

Check to try: write down your actual diagnostic fee, warranty and service hours, including conditions. Can a customer find and understand each before calling?

Measured 2026-08-10. https://t.co/TW6mMaE5pu

[Erik Marutian :Flashcrafter.ai @aitechguy1]

Franklin, TN — Glass replacement.

Quote options differ, so compare what each offer includes

[Erik Marutian :Flashcrafter.ai @aitechguy1]

Use the fillable worksheet on the research page to choose one action, record a baseline and set a review date. Want to understand your own business? Explore its free Growth Brief there.

[Erik Marutian :Flashcrafter.ai @aitechguy1]

@trq212 amazing!

[Erik Marutian :Flashcrafter.ai @aitechguy1]

@claudeai @_Borriss_ Thank you. Long due. TBH. Never understood why those were different

[Erik Marutian :Flashcrafter.ai @aitechguy1]

@adriablancafort @livekit In our tests, it’s the first model that actually exceeds human operator experience.

It’s truly is amazing

[Erik Marutian :Flashcrafter.ai @aitechguy1]

@petergyang yes, similar experience - especially on saying it would do it, instead of just doing it.

[Erik Marutian :Flashcrafter.ai @aitechguy1]

@bindureddy Same with 3.7 in our tests. Didn’t hold well at all

[Erik Marutian :Flashcrafter.ai @aitechguy1]

@MatthewBerman @bot I’ll test it. Pls!

[Erik Marutian :Flashcrafter.ai @aitechguy1]

For @AnthropicAI to really fumble this bad with all the negative news before their upcoming IPO means either they are incredibly mismanaging everything... or they don't care because they have something up their sleeves anyways... so I really hope it's the latter

[Erik Marutian :Flashcrafter.ai @aitechguy1]

This may reveal a gap between benchmark performance and production reliability.

A model can be excellent at synthesizing data while still being too confident about what the evidence proves.

Has anyone compared Gemini 3.7 Flash and GPT-5.6 Luna on similarly evidence-heavy work?

[Erik Marutian :Flashcrafter.ai @aitechguy1]

Gemini 3.7 Flash didn’t outperform GPT-5.6 Luna in our evidence-heavy data analysis tests—even though public benchmarks suggest it should.

Has anyone else observed this?

We sent both models the exact same prompt and data. Here’s what happened 🧵

[Erik Marutian :Flashcrafter.ai @aitechguy1]

So my conclusion isn’t that Luna is universally better at data analysis.

It’s that Luna was more reliable for our specific task: analyzing large dossiers where provenance, uncertainty, and defensible recommendations matter more than confident presentation.