Can AI build guitar resources players actually use?

Counted from the git history and content on disk at generation time — including the corrections.

1. 2026-03-16

   An empty app and a question

   The first commit is a bare create-next-app. The question behind it: can AI build guitar-tone resources good enough that real players actually use them — exact per-rig settings, not another “set everything to 5 and trust your ears” content farm?

2. 2026-03-22

   It becomes a product

   Ten recipes, user accounts, favorites, album art, and a signal chain you can read like a pedalboard. One tone, translated across platforms.

3. 2026-03-25

   First downloadable presets

   The library hits 30 recipes and ships its first .hlx files. A recipe you can read is nice; a file you can load is the point.

4. 2026-03-26

   First public corrections

   A tone audit catches 12 recipes with flat-out wrong settings. They get fixed in one commit, in the open. The generate → check → correct loop starts here and never stops.

5. 2026-03-28

   Fader & Knob is born — and so are the writers

   ToneRecipes becomes Fader & Knob. Same day, the strangest branch of the experiment: instead of one anonymous blog voice, a roster of AI authors — each with a name, a rig, formative bands, a personality profile — so the writing sounds like the actual range of players in the community. Every post gets rewritten in their voices.

6. 2026-03-29

   The engine goes autonomous

   The content pipeline starts running itself: five posts, a SERP analysis, and five new target topics per day, committed automatically. Dozens of those runs now sit in the history.

7. 2026-04-05

   Making the files actually load

   The least glamorous, most important stretch: reverse-engineering the Helix .hlx format from real HX Edit exports. One day’s haul: 38 wrong amp/effect model IDs found and fixed, verified against 250+ real presets. A patch that lists the right settings is useless if it loads the wrong amp.

8. 2026-04-07

   Fourteen homepages in two days

   The hero section gets rebuilt fourteen times in a row — v4 r1 through r14 — chasing a look that could stand out on an AI-saturated web. Most of the fourteen were bad. All of them are in the log.

9. 2026-04-08

   The transparency page ships

   Pricing restructures, and the site gets its first AI-transparency page — what the AI does, in plain language. The disclosure stops being a footer whisper.

10. 2026-04-26

    Ground truth for the generator

    The Helix factory-preset corpus gets harvested into a structured dataset, and the preset generator is rebuilt against it — no more speculating model IDs. Every artist preset is regenerated from verified templates.

11. 2026-05-01

    Every recipe gets its homework

    All 50 flagship tones receive deep-dive rig research — 50 documents on what the artists actually played through — and the corrections flow back into the recipes: Hetfield, Iommi, Van Halen, Gilmour, and the rest.

12. 2026-05-03

    The redesign becomes the site

    After eleven public feedback rounds, the editorial v3 design cuts over to become faderandknob.com. One commit from that week says it all: “drop count-flex eyebrows” — the day bragging about volume stopped being the strategy.

13. 2026-06-26

    Axl picks up the phone

    The tone chatbot launches with a rockstar persona, built specifically to reason about gain staging and block order instead of bluffing like a general model. The Sunday Setlist newsletter starts sending itself.

14. 2026-07-06

    The experiment starts reporting on itself

    A content-authority audit forces a hard look in the mirror: seeded ratings and comments — fake social proof from the early days — get deleted from production. Real ratings become server-rendered. And this page ships, with every number generated from the repository instead of typed by anyone. What you’re reading is the correction.

• 01The dashboard is generated from the repository, not written by hand. If a number is wrong, the code that counts it is wrong — and that gets fixed in public too.
• 02The writers are AI. They're named, they're distinct, and they're labeled. The method is documented, not implied.
• 03Corrections are part of the record. Fixed settings, deleted fake data, dead-end redesigns — they stay in the log, because the log is the experiment.
• 04When something isn't known, it says so. No invented verification steps, no borrowed credentials.

The full method — what the AI does, where it fails, how corrections happen — lives at how we work. Spot something wrong? That's the experiment working — tell us and the fix ships in public.