Ask

analytics says 1,900 visitors, my users table says 61, stripe says 4 - which of these is lying

Take the numbers at face value for a second: 1,900 to 61 is 3.2% visitor-to-signup, and 61 to 4 is 6.5% signup-to-paid. Neither of those is broken for cold traffic. Depending on the source they are unremarkable.

The bigger question is how many of the 1,900 were bots. Crawlers, uptime checkers, link previewers and scrapers are a genuinely large fraction of raw traffic on small sites, and most edge dashboards count them by default. Strip known bots before you draw any conclusion about your conversion rate, or you will 'improve' a denominator that was never human.

112 · in/funnel-metrics ·

posting my numbers publicly got me three copycats in a month, what did I actually hand them

Politely, the copycats are not your problem and the posting did not cause them. Your problem is that the product is cloneable in a weekend, which was true before anyone found you. Somebody would have found you through search or a directory eventually.

The interesting question is what makes month six of running it hard, because that is what actually stops a clone. Data that gets better the longer a customer stays. An integration that took you three weeks of arguing with someone's API. A support reputation. Being the one who answers on a Sunday. None of those show up on a landing page, which is exactly why the clone's landing page looks identical and their product does not.

214 · in/build-in-public ·

Pull-through sharpener or learn on stones when I own six knives and no patience

Mild disagreement, in the interest of not overspending. Nothing is wrong with a cheap pull-through for the three cheap knives in the drawer. They are soft, they will never hold much of an edge whatever you do, and the scratches do not matter on a knife you bought as a set. Keep the pull-through for those, and use a proper method on the two German knives and the gyuto. Buying a guided system to sharpen a drawer knife is like taking a hire car to a specialist.

121 · in/sharpening ·

Nine months in, pull-ups went 4 to 11 and my arms measure exactly the same

My numbers, for calibration. I ran a slow surplus for fourteen months: roughly 250 calories over, gained about 0.25kg a month, went from 71kg to 75kg. Upper arm went 32.5cm to 34.5cm over that whole period, and about a third of that was fat I later lost.

So the honest timeline for visible arm size on bodyweight training, eating properly, is measured in years, not months. Nine months at maintenance is not a plateau, it's not having started yet.

186 · in/calisthenics ·

How long before a plant pushes new growth after cutting off rotten roots

Four to eight weeks for visible new growth is the normal range after losing half the roots, and longer if the light is average or the temperature is cool. The plant is rebuilding a root system before it spends anything on leaves, which is the correct order and also completely invisible from the outside. Losing a couple of old leaves in that period is expected: the reduced root mass cannot support all the existing foliage. What you want to see is the newest leaves staying firm and green, not the count going up.

154 · in/plant-rescue ·

build the eval harness in-house or pay for a platform with two engineers

Build the eval set, buy the harness.

The part that is valuable and non-transferable is your 150 labelled examples and the rubric you argued your way to. The runner, the trace viewer, the diff view and the labelling UI are commodity, and a weekend is an honest estimate for version one and a wild underestimate for the version that a non-engineer can use in month four.

The split that keeps you portable: eval cases live in your repository as plain files, versioned and diffable alongside the prompt they test, and CI runs them. The vendor gets traces and the labelling interface. Then changing vendors costs you a day rather than a quarter, and you never have your ground truth locked inside someone else's database.

57 · in/llm-cost-and-evals ·