About Kaizen64

Why another tactics trainer?

Because while other websites like Lichess or ChessTempo are great, they have limited functionality when it comes to puzzles. I wanted to build a training tool that has everything an ambitious chess player could possibly ever want or need. Making my own means I can implement cool ideas that others just don't have, e.g. Routines, self-specified daily goals, Teams and Homework, a powerful and customizable open-ended mode, Blindfold mode, and more.

The Kaizen approach

Kaizen (改善) is a Japanese concept meaning "continuous improvement through small, consistent steps." The idea here is simple: set a realistic daily goal, follow through with it, and let the rating system give you honest feedback over time. And of course I take the same approach w.r.t. building the tool in the first place.

Who is it for?

Players of any strength. Beginners can start with Mate in 1 or 2 and a modest daily goal; experienced players can enable adaptive difficulty and use the consistency score to maintain discipline in the long term.

Trainer

Puzzle selection

Puzzles come from a large open-source database of positions derived from real games (the mighty lichess puzzle database!). Each position has a rating so the trainer can serve you puzzles at the right difficulty, regardless of whether you choose a fixed range or let the system adapt.

Puzzles you have already attempted are excluded from future draws, so you will encounter fresh positions every session. Once the pool of unseen puzzles at your target range thins out, the trainer widens the search window gradually before eventually allowing repeats.

Rating system (Glicko-2)

Your rating is maintained using the Glicko-2 algorithm — the same system used by Lichess and FIDE for online ratings. It tracks not just your current rating but also a rating deviation (RD) that reflects how confident the system is in the estimate.

A high RD means the system is uncertain about your true strength and will update your rating more aggressively after each attempt. As you solve more puzzles, RD decreases and your rating stabilises. Periods of inactivity cause RD to rise again — the system acknowledges that it has less information about where you stand.

Difficulty modes

Two modes are available from the settings panel:

• Adaptive — puzzles are drawn within ±150 rating points of your current puzzle rating and the window follows your rating as it changes.
• Custom — you pick a fixed rating range from a slider. Useful for targeted theme practice or deliberate over/under-rating training.

You can also filter by theme (pin, fork, endgame, etc.) independently of the difficulty mode.

Routines

What is a routine?

A routine is a named sequence of puzzle sections that resets every day at midnight. Each section has its own difficulty range, theme filter, and puzzle count. Sections are meant to be completed in order, and the puzzle set for a given day is deterministic — two players running the same routine on the same date face identical positions.

This makes routines useful for group training, for example. A team or study group can all attempt the same positions and compare their times and mistake counts on the routine's leaderboard.

Creating your own

Pro users can create custom routines and share them with the community. A routine can consist of up to ten sections with any combination of difficulty ranges and themes. Once published, other logged-in users can run it and appear on its daily, weekly and all-time leaderboards.

Consistency

What it measures

The consistency score (0–100) captures how reliably you hit your chosen daily puzzle goal. It is intentionally independent of accuracy — a day on which you attempt your target number of puzzles counts fully, regardless of how many you solved. The idea is to separate two distinct habits: showing up (consistency) and performing well (rating).

Daily goals

You set your own daily goal between 3 and 100 puzzles. Once set, it can be adjusted every five days, and changes only affect future days — past history is frozen at the goal that was active on each day. This prevents retroactive gaming of the score.

Until you set a goal, your consistency appears as "?" on your profile and on the leaderboard. Goal-setting and tracking are available to all logged-in users.

How the score is calculated

Each calendar day, you earn a credit between 0 and 1: the number of attempts divided by your goal for that day, capped at 1. Missing a day earns 0.

The score is the product of two terms:

• Base — a recency-weighted average of your daily credits over the last 28 days. Recent days count more than older ones, so a week of inactivity noticeably reduces the score, but a single missed day has only a modest effect.
• Maturity — a factor that starts slightly below 1 when you first set a goal and slowly approaches 1 as your lifetime accumulated credit grows. It ensures very new users start with a slightly lower ceiling and cannot immediately claim a perfect score.

About the Dev

I am IM Jakob Pfreundt. In 2025 I started building my first chess-related tool, ChessMint, which facilitates creating puzzle sheets and is more geared toward coaches as well as players who prefer to train offline and/or off-screen. You could say Kaizen64 is the natural counterpart, trying to make online training as fun and effective as possible.

Training

Trainer

The open-ended solving page. Puzzles are served one at a time with no fixed end — you stop whenever you like. Positions you have already attempted are not served again until the pool at your difficulty runs thin.

Training mode

Which rating an attempt counts toward. Each mode keeps its own independent Glicko-2 rating starting at 1500, so strength in one does not carry over to another.

Standard

Untimed. An attempt simply counts as solved or not.

Blitz

Speed counts. Every puzzle has a target time: hitting it scores a full point, a slower solve scores a fraction of one, and a fast solve can score above it, up to 1.25. The score is what the rating system sees, so a Blitz rating answers "how hard a puzzle can you solve quickly" rather than simply "can you solve it".

Blitz target time

The target is (5.3 + 2.9 x (moves - 1)) seconds, doubled for every 400 rating points the puzzle is above yours and halved for every 400 below. "moves" counts only the moves you play, not the replies. So a 2-move puzzle rated the same as you allows 8.2 seconds; a 4-move one 14.0; a 4-move puzzle 400 points above you, 28.0.

Why longer solutions get more time

Puzzle ratings come from how often people solve a position, not from how long they take. A long solution that is easy to find is therefore not rated any harder, even though it takes longer to calculate and to play out. Before this allowance existed, a typical four-move solve scored 0.47 where a typical one-mover scored 1.03 — a gap of more than half a point that had nothing to do with the solver. The two numbers were fitted to several thousand recorded solves and scaled so that the average score is unchanged: the correction moves points between short- and long-solution solvers rather than handing them out.

Blindfold

The pieces are hidden. Only the squares immediately around the cursor are revealed, and only once the cursor has rested on a square briefly — so the position has to be held in your head rather than scanned.

Difficulty mode

How the trainer chooses the rating range it draws puzzles from.

Adaptive

Puzzles are drawn within ±150 points of your current rating in that mode, and the window follows your rating as it moves.

Custom

You fix the range yourself with the difficulty slider — useful for deliberately training above or below your level.

Theme

A tactical-motif filter (pin, fork, endgame, mate in two, …) applied on top of the difficulty range.

Rating and figures

RatingR

A Glicko-2 estimate of your puzzle strength, kept separately for each training mode and starting at 1500. It rises and falls with each attempt, weighted by the difficulty of the puzzle.

Rating deviationRD

How uncertain the system is about your rating. A high RD means bigger swings after each attempt; it falls as you solve and rises again during inactivity.

Personal bestPB

The highest rating you have reached in that mode.

Average puzzle ratingAPR

The mean difficulty of the puzzles you have been served, every attempt counted once however long it took.

Time-weighted puzzle ratingTWPR

The same mean, but weighted by the time each puzzle took, so a five-minute puzzle counts five times a one-minute one. Thirty quick warm-ups outnumber three hard puzzles on APR yet take far less of the clock, so TWPR is the better measure of where your effort actually went. The unit of time does not matter: it is a weighted average, so it cancels out.

Solve rateSR

The share of your attempts solved without a wrong move, as a figure between 0 and 1.

Time on solvedTOS

The share of your playing time that went into puzzles you got right — the time-weighted counterpart to SR. A low figure means most of your clock is going into puzzles you miss.

Performance ratingPerf

The rating a set of results implies, calculated FIDE-style from the average difficulty faced and the share solved. Unlike your rating it has no history: it describes one batch of puzzles rather than tracking you over time.

Hover countHC

Blindfold only: the average number of squares you reveal per puzzle. A low count means you are visualising rather than sweeping the board.

Consistency

Consistency

A score from 0 to 100 for how reliably you meet your daily goal. Deliberately independent of accuracy — a day on which you hit your target counts fully regardless of how many puzzles you actually solved. It measures showing up; rating measures playing well.

Daily goal

The number of puzzles you commit to per day, from 3 to 100. It can be changed every five days and only affects future days: past days stay frozen at the goal that was active then, so the score cannot be gamed retroactively.

Base

A recency-weighted average of your daily credits over the last 28 days. A missed day costs little; a quiet week is clearly visible.

Maturity

A factor that starts just below 1 and approaches it as your lifetime credit grows, so a brand-new account cannot immediately claim a perfect score.

Routines

Routine

A named, ordered sequence of puzzle sections that resets every day at midnight. Routines are meant to be worked through in order, and each has its own daily, weekly and all-time leaderboard.

Section

One block within a routine, with its own rating range, theme filter and puzzle count. A routine may have up to ten.

Fixed

A routine whose sections use absolute rating ranges, so everyone running it on a given day faces exactly the same positions. This is what makes routines comparable across a group.

Adaptive

A routine whose sections are defined as offsets from the solver's own Standard rating, so the same routine lands at the right difficulty for each person. The rating is fixed when you first open the routine each day, so solving it cannot move your own target mid-run.

Team routine

A routine owned by a team rather than published to the community; only its members can see and run it.

Teams

Team

A group of players with a shared leaderboard, routines and homework. A team can be open, password-protected, or joinable by request.

Founder

Whoever created the team. Has full control over it and handles join requests.

Coach

A member promoted by the founder. Can create team routines and homework, and grade submitted homework.

Homework

A set of positions a coach assigns to the team, with a due date. Members submit their answers, the coach awards points, and the totals form the team homework leaderboard. Homework exists only inside a team.

Account and community

Leaderboard

Ranks players by rating, one board per training mode. To appear you need at least 10 attempts in that mode and a solved puzzle within the last 14 days; a single puzzle is enough to return.

Collection

Your saved puzzles. Any position you have attempted can be bookmarked, optionally with a private note only you can read.

Title

A FIDE title (CM through GM, and the women's titles) shown next to your name once set in your profile.

Pro

An account tier intended to gate creating routines and teams. Every account currently has it.

Moderator

A user who can edit, delete and set the status of feedback posts.

Feedback

The page for bug reports and feature requests. Anyone signed in can post and vote with a like; each post carries a status of Considering, Accepted, Shipped or Rejected.