I nerfed my own product before anyone was playing

Games usually nerf things after players complain. I nerfed mine before launch, when the only players were my wife and me, and nobody had complained about anything. If anything, the numbers felt great. That was the problem.

Context in one line: Hotari turns your daily life into an RPG. You log what you did, an AI Scribe scores it into five life stats (Wisdom, Body, Heart, Career, Bonds) and writes a short narrative of your day. The launch post has the rest.

The inflation

In the early builds, an ordinary day earned around +60 stat points. Every task came back from the AI with +2s and +3s sprayed across multiple stats. Settling your day felt like hitting a piñata.

Then I saw two lines in one settlement that broke the spell for me:

Fed the baby lunch — Bonds +3, Heart +3 Solo-parented the entire Saturday — Bonds +3, Heart +3

Anyone who has done both knows these are not the same day. One is fifteen minutes. The other is a marathon that starts before you’re awake, and you finish it a different person. My app looked at both and shrugged: +3.

Why an AI scores everything the same

The root cause was embarrassingly simple. My prompt told the AI to be “generous but honest” and gave it no scale. An LLM with no rubric does what a nervous first-time manager does in performance reviews: rates everyone a 4 out of 5. All scores cluster in the safe middle, and the safe middle plus generosity means everything inflates.

And in a stats system, inflation isn’t cosmetic. Hotari’s whole premise is that the numbers are a mirror of your real effort. The moment “fed the baby” equals “parented all Saturday,” the mirror is lying, and a mirror that lies is worthless no matter how pretty the frame is. Worse, attribute points accumulate permanently, so every inflated day compounds into a permanently inflated character. Confetti today, meaningless character sheet forever.

The nerf

The fix had two layers. First, the rubric: instead of “generous but honest,” the AI now scores against four anchored tiers. Routine and light effort is a 1, and the prompt says explicitly that most tasks belong there. Normal deliberate effort is a 2. Significant investment is a 3. And 4 is reserved for rare, heavy days, with the prompt naming the criteria: duration, scale, difficulty, and whether it’s routine. The feeding-versus-Saturday pair is literally in the prompt as a worked example: one scores 1, the other scores 3 to 4.

Second, I took the numbers away from the AI entirely. It no longer returns scores. It returns a tier (1 to 4), which stats the task exercised (at most two), and which skills it fed. The server derives every number from the tier: the primary stat gains the tier value, the secondary gains one less. There is nothing to clamp, because nothing above 4 can be produced.

The result: a good day now lands around +12 instead of +60. XP follows the same tiers, so leveling slowed down along with it. I shipped that without any compensation buff, and existing test users kept their inflated historical points; the new rules only govern the future, which felt more honest than retroactively rewriting anyone’s history.

Nerfing yourself is a strange discipline

Every instinct says bigger numbers are better onboarding. Number goes up, dopamine flows, user returns. The entire mobile gaming industry runs on that loop, and it works.

But Hotari isn’t a game you play to escape your life. It’s an instrument you check to see your life. Instruments have one non-negotiable property: they have to be calibrated. A bathroom scale that reads five kilos light is more pleasant than an accurate one, exactly once.

The same principle drove other choices in the app. The skill-fusion AI refuses combinations that don’t correspond to a real ability instead of inventing a cool name for nonsense. New skills always start at the lowest rarity, no matter how impressive the AI finds your day. Watching your day get scored a modest, defensible +12 and reading a narrative that saw the real texture of it turns out to be more satisfying than any piñata, because it’s the only version you can believe.

Big honest numbers don’t exist. Small honest numbers compound.

Hotari is at hotari.app, seven days free, no card. If you settle a heavy day and the score still feels wrong, tell me. Calibration never really ends.