Changes and corrections

This page lists the corrections made to the English tools, games and labs, with links to the pages, and how many pages were added each month.

Fixes and corrections

Most of these came out of a review of the English pages in October 2026. How the pages are made and tested is described on How these pages are made and tested.

4 November 2026: Gravity Simulator and page footer

Gravity Simulator: no sideways scrolling at window widths from 1000 to 1140 px, and on phones and narrow windows the Pause, Step, Reset and Clear buttons sit directly above the simulation instead of below it.
The page footer now has a “Report a problem” link that opens an e-mail already filled in with the address of the page.

3 November 2026: Brain & Memory Tests

Stroop Test, Go/No-Go Test, Change Blindness Test, Mental Rotation Test, Visual Search Test, Task Switching Test, Verbal Memory Test, Memory Test, Attention Span Test, Attentional Blink Test, Multiple Object Tracking and Time Perception Test: a double click on Start, Stop, Pause or Resume now does only one thing; after Stop the buttons wait 0.7 s, so the second click no longer starts a new run.
Stroop Test, Memory Test, Task Switching Test, Visual Memory Test, Go/No-Go Test, Verbal Memory Test, Number Sense Test and Multiple Object Tracking: switching the language now changes every message, button and chart label.
Task Switching Test, Visual Search Test, Verbal Memory Test and Time Perception Test: on a touch screen the hints name the buttons (left and right, Yes and No, Old and New, Stop) instead of the keys F, J and Esc.
Task Switching Test, Visual Search Test, Verbal Memory Test, Multiple Object Tracking and Number Sense Test: a run marked less precise says whether the screen ran below 50 Hz or its frames were uneven.
Brain Profile: every number and 95% interval is written as on its test page (for example 14% to 48% on the Go/No-Go Test and 0.19 ms/° without a plus sign on the Mental Rotation Test), and a run is marked less precise exactly when its test page warned about its timing, also for runs saved earlier. The tip to compare the keyboard with the keypad of a test now appears only on a computer.
Change Blindness Test, Time Perception Test and Go/No-Go Test: This run, History and the Brain Profile now show the same rounded number; before, a value saved with too few decimals could read −1.2 s in This run and −1.1 s in History.
All Brain & Memory Tests and the Brain Profile write a 95% interval the same way, as “95%: 0.15 to 0.28”, never with a dash or dots between the values. Chart X axes label steps such as 2.5 and 0.25 exactly and number runs with whole numbers; chart legends draw a dot for points and a dashed line for a dashed series; the points of the Visual Memory Test and Number Sense Test have their own legend colour and are no longer cut off at 100%.
Attention Span Test: the countdown shows its first number, scrolling over the test box no longer starts a run, and Start fits the first screen of a 1280 × 800 window. Attentional Blink Test: the letter pad fits small phones, the less precise note gives the measured time, and the Sperling chart puts the first arrow at the run’s real time. The results of the Verbal Memory Test and Mental Rotation Test fit narrow phones in Slovak.
Task Switching Test: Trails show the screen warning, the break shows the right trial number, and a late tap no longer hides the block result. Number Sense Test: the result line shows the planned flash, as in the settings, with the measured mean beside it, and the note about the repeat limit counts practice and test apart. Dual N-Back notes a slow or uneven screen under its result.
Visual Memory Test, Attentional Blink Test and Dual N-Back: Enter or Space on the test field starts the test. Attentional Blink Test, Visual Memory Test, Number Sense Test and Multiple Object Tracking: at the end of a run a phone scrolls to This run, where the status line points. Mental Rotation Test writes long times as 1,801 ms, and the Slovak frame counts of the Change Blindness Test use the right plural.

3 November 2026: Reaction Time Test

Reaction Time Test: new Choice test with 1, 2, 4 and 8 boxes that fits your own Hick's law slope in milliseconds per bit; every result and the history now show a device mark (the input you answered with and the measured screen refresh rate), and the page gives how simple and choice reaction time change from age 20 to 65 (Woods and colleagues, 2015).

3 November 2026: Time Perception Test

Time Perception Test: new Precise mode for the tapping task, which times your taps and the clicks inside one microphone recording, so sound delays cancel.
Time Perception Test: Precise mode no longer stops now and then with “did not hear the clicks” or “not evenly timed” when the browser drops a few milliseconds of the recording. Failure messages show in full on a phone, the display no longer jumps when the practice ends, a setting tapped during a run says right under it that settings are locked, and counts such as “1 of 1 tap” use the right plural.

2 November 2026: Brain & Memory Tests

Number Sense Test: repeats of mistimed trials now stop at 12 per task, so a run on an uneven screen always ends, and a run that reaches the limit is marked less precise.
Brain & Memory Tests: chart axis labels now show exact values at steps such as 2.5 and 0.25 instead of rounding them.
Brain Profile: Stroop Test and Memory Test runs on a screen with uneven frames are now marked less precise, including runs saved earlier.

2 November 2026: Brain Profile

Brain Profile: Number Sense task B now shows the same range and interval as its test page, cut to 0 to 9 dots, in the view, the chart, the table and the CSV. Task A is now marked unreliable fit when its page warns that the fit did not settle or the interval is very wide. Attention Span runs on a screen below 50 Hz or with uneven frames, and Dual N-Back runs with uneven frames, are now marked less precise.
Brain Profile: corrected sentences. Dual N-Back, the Mental Rotation Test and the False memory task of the Verbal Memory Test keep part of a stopped run, and the profile counts it; a run marked less precise can also come from uneven frames; 0.60 and 0.66 come from two studies, not one. The idea to compare a phone with a computer could not work, because each browser keeps its own results, so it now compares keyboard and keypad. Chart axis labels no longer round 2.5 to 3.

31 October 2026: history chart

Dual N-Back: the date axis of the history chart had labels on days without a session and past the last session, even a date in the future. Now only days with a session have a label; when labels would overlap, older ones are left out and the newest stays.

29 October 2026: memory and attention tests

Memory Test: in the digit and block tasks the top row of a Slovak keyboard, which types letters such as “ľ” and “š” without Shift, and the number pad now enter digits; before, those answers were not taken.
Visual Memory Test: after a click on the language link during a run, the focus goes back to the test field, so Space no longer scrolls the page and Enter no longer switches the language back.
Memory Test, Visual Memory Test and Dual N-Back: the history filter no longer says “the same refresh rate”; runs on the same input count as matching when their refresh rates are within 10% of each other or both below 50 Hz, as the filter on Attention Span Test already said, where the label now writes 10% without a space.

28 October 2026: new features

SSTV Decoder: pictures now also start by themselves when the VIS code was missed, from the timing of the lines; the page cuts out the short silences that web receivers drop into their sound and follows their changing speed, so pictures from a web receiver come out straight.
Signal Identifier and RTTY Decoder: they now link the new NAVTEX Decoder and APRS Decoder; the Identifier’s table lists where to listen for both, and the RTTY answer about NAVTEX no longer says that it cannot be read on this site.

28 October 2026: game menus

Bot Arena: Protocol Survive: the start menu was headed “Neural combat simulation // Training network 07”, but the robots are not a neural network, so the line is gone. The menu labels are now plain words in English and in Slovak (Mode, Map, Weapon, Difficulty, Start, Paused, Results) instead of spaced capitals, and the text under the game names the button Start.

27 October 2026: new features

DCF77 Decoder: now also decodes MSF, the British time signal on 60 kHz, and recognises the station by itself; the page explains the MSF bits and how to receive 60 kHz.
DCF77 Decoder: the page now also reads the phase code of DCF77; a clear phase bit replaces an uncertain amplitude bit, a minute that needed the phase code is accepted only when the next minute confirms it, and a new panel shows the code, its agreement with the amplitude bits and the scatter of the second start.
Signal Identifier: the table now lists MSF on 60 kHz, and the advice for a time signal no longer says that the site has no decoder for time codes; it links the DCF77 Decoder, which reads DCF77 and MSF.
RTTY Decoder: Baud rate now starts on Auto, which measures the rate of the signal, switches to it together with Reverse and decodes the last 20 seconds again; “Unshift on space” is set from what the station sends after a space in figures. A new answer explains why NAVTEX and SITOR cannot be read here.
Weather Fax Decoder: every chart now gets its own black and white points, worked out from the chart itself, so noise on the white paper is drawn white and a centre set a few tens of hertz off is evened out; moving either slider switches this off.

27 October 2026: sources

Eighteen pages that named a person, a year, a standard or a broadcast schedule now link their sources: Auditory Illusions, Chladni Patterns, Doppler Effect, Galton Board, Gravity Simulator, Game of Life, Langton’s Ant, PSK31 Decoder, SSTV Decoder, Weather Fax Decoder, QR Code Generator, Regex Tester, Unit Converter, Instrument Tuner, Hourly to Salary Calculator, 15 Puzzle, 3D Cube Puzzle and Periodic Table Quiz.
Game of Life: the invention is no longer dated to 1970; the page now says the game became known through Martin Gardner’s Scientific American column of October 1970.
Langton’s Ant: two details without a source were removed, the names attached to ants with more colours and the exact step count before the highway, which now reads about 10,000.
PSK31 Decoder: the 40 m note now gives the US band plan’s data segment, 7.080 to 7.125 MHz, instead of 7.070 MHz.
Weather Fax Decoder: the tones are now the standard 1500 Hz for black and 2300 Hz for white, with 1475 and 2325 Hz only at the German weather service; the Northwood alternative frequencies have their own times; and the second example of a German surface analysis is at 10:50 UTC, not 12:56.
Instrument Tuner: the answer on cents no longer quotes an unsourced 5 to 6 cent limit of hearing.
3D Cube Puzzle: the classic cube is now dated by its Hungarian patent application of 1975 instead of an unsourced 1974.
15 Puzzle: the craze is now dated around 1880, because a journal note of December 1879 already describes it.
Code Breaker: Donald Knuth’s proof that the opening 1 1 2 2 wins in at most five guesses is now dated 1976 and cited; the average of about 4.5 guesses, which his paper does not give, was removed.

25 October 2026: lab layout on wide screens

Lab pages: on a wide landscape screen the picture or graph will stay in view on the left while the controls sit in one or two columns on the right. Phones, upright tablets and printing keep the layout they have. The shared layout is ready; the lab pages move over one by one and are listed here as they do.
Double Pendulum: on a wide screen the pendulum stays in view on the left while the readouts, presets and sliders sit beside it, so you can see what each change does; phone layout unchanged.
Caesar Cipher: on a wide screen the cipher wheel with the result preview and the status line stays in view on the left, while the text, the alphabet strip and the crack panel (verdict, charts, all 26 shifts) sit in two columns on the right.
Synthesizer: on a wide screen the instrument (oscilloscope, spectrum and keyboard) stays in view on the left while you change the settings on the right.
Neural Network Playground: on a wide screen the decision plane stays in view on the left while the training bar, loss chart, data, network diagram with its weight editor and settings sit in columns beside it; phone layout unchanged.
Light Sources & Spectra, Physics Sandbox, Polarization of Light, Projectile Motion and Thin-Film Interference: on a wide screen the display stays on the left while you use the controls beside it (Polarization: the Polarizers mode). Light Sources & Spectra, Physics Sandbox, Polarization of Light, Projectile Motion, Thin-Film Interference.
Cryptography labs (Playfair Cipher, Rotor Cipher Machine, Vernam Cipher, Vigenère Cipher): on wide screens the key square, the signal path, the randomness check and the letter-frequency chart stay in view beside the controls; the phone layout is unchanged.
Chladni Patterns, Doppler Effect, Lissajous Figures, Music Sequencer and Sound Picture: on a wide screen the display and its controls stay on screen together; phone layout unchanged.
Sound Visualizer, Theremin, Tone Generator, Tuning Systems and Hearing Loss Simulator: on a wide screen the display stays in view next to its controls; the phone layout is unchanged.
Cloth Simulator, Color Mixing, Double-Slit Experiment, Electric Field and Ideal Gas: on a computer screen the simulation stays in view next to its controls; the phone layout is unchanged.
Fourier Drawing, Galton Board, Matrix Transformations, Spiral Gears and Strange Attractors: on a wide screen the drawing, board, plane or 3D view stays in view beside the controls, so you can see what each setting does while you change it.
Digit Recognizer, Genetic Algorithm and Pathfinding Visualizer: on a wide screen the drawing area with its answer, the rocket world and the search grid stay in view beside their controls, so a change and what it does are on screen together; the phone layout is unchanged.
Life & Emergence labs (Epidemic Simulator, Flocking Simulator, Game of Life, Langton’s Ant, Particle Life): on a wide screen the simulation stays in view beside its controls.
DCF77 Decoder, Morse Decoder, PSK31 Decoder and RTTY Decoder: on a wide screen the waterfall or spectrum and the decoded text (on DCF77 the decoded clock, the bits of the minute and the amplitude chart) stay in view on the left while the audio source, readouts and settings sit beside them; the phone layout is unchanged.
Double Pendulum: at page zoom 110 % the pendulum shows again; the canvas no longer resizes in a loop and the view no longer shakes.
Neural Network Playground: on a wide screen the Settings button stays in place when you hide or show the settings.
Caesar Cipher: on a wide screen the cipher wheel keeps its size while you type.

25 October 2026: new features

Percentage Calculator: a percentage change from a negative starting value now divides by the size of that value, so going from −10 to −5 is +50%, not −50%, and a change from 0 shows the change in absolute terms. Two new calculators work out a chain of changes (net change and the change back to the start, e.g. +20%, −20%, +5%) and percentage points vs. relative change (5% to 8% is +3 percentage points and +60%). The page now links its sources.
Final Grade Calculator: added a points mode, a chart of your overall grade for every final score, and an option for a final that replaces your lowest test.
Memory Game: rebuilt with 4×4, 4×5 and 6×6 boards, four symbol sets (animals, shapes, numbers, letters) that screen readers announce by name, a two-player mode with a score per player, best moves and time per board size, and arrow-key play; the text now explains the rules, what a perfect game means (with averages from a simulation of perfect recall) and a systematic turning order.
CSV to JSON Converter, CSV to SQL Converter and CSV Viewer now read CSV with one shared reader. A file gets the same delimiter and the same column names on all three pages: column_N for an empty or missing name and for an extra field, and _2, _3 for repeated names. CSV to JSON Converter, CSV to SQL Converter, CSV Viewer
Sudoku: Hint now names the technique that finds the digit (naked or hidden single, pointing pair, box/line reduction, naked or hidden pair, naked triple), shades the row, column or box it used, and tries the cell you selected first; when no technique applies it fills a cell as before.
Klondike Solitaire: every deal now has a number from 1 to 1,000,000. You can type a number to replay a deal or copy a link that opens it in the same Draw mode. An optional Winnable deals only switch, off by default, hands each shuffle to a solver first; on 25 October 2026 it proved 669 of game numbers 1 to 1,000 winnable in Draw 1 and 620 in Draw 3.
Minesweeper: a new option, No-guess boards, places the mines after your first click as before but keeps a board only if the game’s own solver can open every safe square from the numbers and the mine total, so you never have to guess; if no such board turns up within 3 seconds, you get an ordinary random board and a message saying so. Classic random boards stay the default, and best times on no-guess boards are kept separately. A script tests the solver against a brute-force check on 600 small boards and on 500 boards per level; run on 25 October 2026, it accepted about 9 in 10 random Beginner boards, 7 in 10 Intermediate boards and 1 in 6 Expert boards, which the page now quotes.
Weighted Grade Calculator: the letter grade now comes from the unrounded grade, the same rule as the Grade Curve Calculator, so 89.996% (shown as 90.00%) is a B with an A cut-off of 90; both pages explain how to mimic a school that rounds first.
Unit Converter: new quantity Volume with millilitres, litres, the metric cup and the US and imperial fluid ounce, cup, pint, quart and gallon, each from its exact definition (the US gallon is 231 cubic inches, the imperial gallon 4.54609 litres), with sources.
Word Counter: buttons for common limits fill in the limit (text message 160 or 70 characters, search description 155 characters, short post 280 characters, abstract 250 words), and the default reading speed is now 238 words per minute from Brysbaert’s 2019 review.
Online Notepad: back up all notes to one file and restore them on another device, either added to the notes there or replacing them, with Undo; reading time now uses the same 238 words per minute as Word Counter.
Random Group Generator: optional Keep apart and Keep together rules. The draw repeats until the rules hold, and rules that cannot be met get a message that says why.
Typing Practice and Typing Speed Test counted net speed in two different ways: Practice took one word off for every wrong key press, the test only for errors left in the text. Both now use the test's rule. In Practice a wrong key is never inserted, so no error is left and the net speed equals the gross speed: the page now shows one speed, and wrong key presses lower the accuracy. Lesson bests saved earlier stay comparable, because the best is now the gross speed, which the change does not affect.
Game of Life: now plots the population and says when a pattern has settled: a still life, an oscillator with its period, or still changing at generation N.
JSON Formatter: an optional Repair button fixes trailing commas, single quotes, unquoted keys, comments and NaN/Infinity/undefined, lists each change with its line number and can be undone; the Comments example in the table now uses a block comment.
Standing Waves & Harmonics: a Guitar A string button sets the length, tension and mass per metre of a guitar’s fifth string, so the readout shows 110.1 Hz (A2) and the string formula can be checked in one click.
Paintball Bot Arena 3D: replacement bots now enter around one spot picked at random, 60 to 85 m from you, at the start of each match, so they no longer appear behind you; the first eight still spread out around you.
Resistor Color Code Calculator: orange, yellow and grey tolerance bands now follow IEC 60062:2016 (±0.05 %, ±0.02 %, ±0.01 %); a grey band notes the older ±0.05 % value.
Correlation & Covariance Calculator: an optional 95% confidence interval for r (Fisher z) and the p-value of the t-test for zero correlation, shown from 4 pairs on; the page now lists its sources.
Tic Tac Toe: an optional Show outcomes button labels every free square Win, Draw or Loss under perfect play, and in Impossible mode a line under the board says why the computer played each move.
Calcudoku: a new optional Cage combinations panel under the buttons (key K) lists every set of numbers that makes the target of the selected cage, including a repeated number when the cage bends so that its copies sit in different rows and columns, and strikes out the sets that no longer fit the numbers already in the cage, its rows and its columns. It starts closed; the grid and the number pad are unchanged.
Paddle Duel: after a click on Sound, the focus stayed on that button, so the next Space switched the sound again instead of serving. A mouse or touch click on Sound, Pause or Menu now gives the keys back to the game.
Neon Trails: when full screen was switched on during a match, the arena kept the shape of the window and left the sides of the screen empty. Each round now takes the shape of the screen. During the countdown the arena is redrawn at once; in the middle of a round the page says that the next round fills the screen.
Air Combat: when the pointer came back into the view away from the centre, the aircraft rolled into a full turn at once. The turn now builds up over about a second.
3D Maze: scrolling the page while half of the start screen was out of view removed the focus from the selected button, even while the button was still on screen. The focus is now released only when that button leaves the window.
Bug Blaster: the probe follows the mouse as soon as you move it over the board (no Controls switch any more), and Space starts a game.
Missile Defense: Space on the start screen continues a saved game like Enter instead of starting over.
Snake: on phones the Full screen button sits beside Walls, so the Speed buttons stay on screen.

25 October 2026: small corrections

Five lab pages, Signal Identifier, Polarization, Michelson Interferometer, Electric Field and Flocking Simulator: the first sentence was reworded so that each now states a concrete detail of what the app does.
Contact: the dashes in the list of what you can write about were replaced with colons.
Reaction time test: the result note no longer quotes a typical range, so it agrees with the page text that there is no pass mark, and like the page text it asks you to compare the medians of your own earlier runs.
List Compare: the unmeasured claim about how fast 50,000-line lists compare was removed.
Instrument Tuner: an unmeasured accuracy figure was removed from the FAQ.
Typing Speed Test and Typing Practice: an unsourced speed tip and an unsourced practice-time claim were removed; the study figures now appear only on the test page.
Sources: the MySQL manual, MathWorld and IANA entries on the CSV to SQL, Extended Euclidean, GCF/LCM and Time Zone pages now carry an access date.
Rotor cipher machine: the FAQ said the rotor wirings were “documented by historians” without a source for them; it now says that the listed sources cover the codebreaking history, not the wiring tables.
Weather fax decoder: the intro now says that the page cuts short gaps out of the stream.
Pitch test: the intro says that the first tone shifts from trial to trial.
Particle life: the year 2022 is now attributed to the source code, not to the idea.
Genetic algorithm: the FAQ called the fitness function a fixed formula, but its exponent, the selection pressure, is a slider you set; the answer now says so.
Final Grade Calculator: with the lowest test at 100% and the replacement option ticked, the “not reachable” message quoted 106.67% instead of 110% for Example 1’s numbers. A final of at most 100% can’t replace a 100% test, so the score needed now comes without the replacement, and the message says why.
Sources on four labs (rotor cipher machine, sound picture, spiral gears, hearing test): Wikipedia and MacTutor entries now carry an access date, and the hearing test names the years of its method.
Sudoku: a naked pair now reads as two cells whose notes are only the same two digits, and the Expert generator is described as trying every cell.
Hangman: a word comes back only after every word in its category has come up (lead, rules and FAQ now say the same).
Precision Range: the FAQ names the DOPE columns Drop (cm) and Scale (DIV) as on screen.
Snake: Controls now say an arrow key, WASD or Enter resumes a paused game.
Paintball Bot Arena: the 15-point rule is stated for Arena mode only.
Game opening lines rewritten on 9 game pages so each states a concrete rule of the game: block-puzzle, crate-pusher, missile-defense, star-fighter, tunnel-flight, maze-chase, tic-tac-toe, pyramid-solitaire, sea-battle.
Wikipedia and Puzzles.wiki sources on maze-chase, drone-delivery, nonogram, nurikabe, slitherlink, light-up, bridges now carry an access date.

22 October 2026: text and sources review

Flocking Simulator: the first sentence promised “thousands of tiny rules”; the model has three. The page said Craig Reynolds introduced it in the 1980s; he published it in 1987. It now links his paper and his page about boids.
Particle Life: the page now says where its force rule comes from, the open-source code of Tom Mohr’s Particle Life, which grew out of Jeffrey Ventrella’s Clusters, and links both.
Epidemic Simulator: the page now says that its SIR equations are the simplest case, with constant rates, of the model that Kermack and McKendrick published in 1927, and links the paper.
Nine physics and optics labs now name and link the sources they rely on: Double-Slit Experiment (Young’s Bakerian Lecture), Light Sources & Spectra (Planck, the NIST Atomic Spectra Database), Michelson Interferometer (Michelson, and Michelson and Morley), Lenses, Eye & Optical Instruments, Polarization of Light (Brewster), Thin-Film Interference, Electric Field (Feynman’s lectures), Ideal Gas (Maxwell, Brown, Einstein) and Physics Sandbox (Erin Catto’s papers on the solver).
The first sentences of six labs were generic. They now say what each page does: Michelson Interferometer (count the fringes to find a hidden wavelength), Polarization of Light (the electric field in 3D), Ray Optics (the lens equation checked against where the traced rays cross), Thin-Film Interference (every reflection inside the film added up), Electric Field (a test charge that leaves a curved field line) and Physics Sandbox (stepping through a collision 1/60 s at a time).
Four more labs now name and link the sources they rely on, and say in the text who published each method and when: Genetic Algorithm (Holland’s book, and the “weasel” program of Dawkins that its Word evolution follows), Digit Recognizer (the MNIST paper by LeCun and colleagues, and Rumelhart, Hinton and Williams on backpropagation), Pathfinding Visualizer (Dijkstra, and Hart, Nilsson and Raphael on A*) and Sorting Visualizer (Shell, Hoare, and Sedgewick on Knuth’s gaps for Shell sort).
The first sentences of four more labs were generic. They now say what each page does: Logic Gate Simulator (a half adder is already wired up, and you can turn a Boolean formula into gates), Neural Network Playground (set the weights by hand in Challenges, then let gradient descent train the same network), Pathfinding Visualizer (six searches on one grid, with g, h and f for A*) and Sorting Visualizer (exact counters and live pseudocode).
Hearing Test: the page now names the method behind its steps, the modified Hughson–Westlake method, and says that the “down 10, up 5” procedure is described in the 2005 guidelines of the American Speech-Language-Hearing Association. Its sources now say that Amendment 1 (2024) of ISO 7029 replaces only Tables 2 to 5, the spread around the median, so the median curves of the age comparison are not affected, and they give the year of the advice on sudden deafness.
Equal-Loudness Contours: the source line for the 2024 paper by Suzuki and colleagues now says what it backs, the history of the contours from the 1933 headphone measurements to the 1956 loudspeaker measurements, and the 2023 edition of ISO 226.
Tinnitus Matcher: the NHS and NICE entries in the sources now say which advice they back: urgent advice for tinnitus that beats in time with your pulse, and referral for tinnitus in one ear only that persists or that comes with sudden hearing loss.
Pitch Discrimination Test: the first sentence now says what makes the test hard to game, that the first tone is shifted at random from trial to trial so that memory cannot help.
Seven maths labs now name and link the sources they rely on: Factor Tree (Stein’s free textbook Elementary Number Theory, for the fundamental theorem of arithmetic and the sieve that stops at √n), Fourier Drawing (Cooley and Tukey on the fast Fourier transform, and MIT OpenCourseWare on Gibbs’ phenomenon and its overshoot of about 9%), Fractal Explorer (Mandelbrot’s 1980 pictures of the set, Douady and Hubbard’s Orsay notes, and Shishikura on the dimension of its boundary), Matrix Transformations (Strang’s MIT OpenCourseWare lecture summaries on the 2×2 inverse, the determinant as an area and the eigenvalue equation), Pseudorandom Generators (Park and Miller for MINSTD, Matsumoto and Nishimura for the Mersenne Twister, O’Neill for PCG), Spiral Gears (the MacTutor pages on the hypotrochoid and the epitrochoid, whose equations the page uses) and Strange Attractors (Lorenz, Rössler, and Viswanath for the Lyapunov exponent and the dimension of the Lorenz attractor).
JSON Formatter: an answer said that a trailing comma is flagged “right after the comma, not at the comma itself”. In Chrome the position depends on the input: after a comma at the end of an object the arrow points at the closing brace, which may be on the next line, and after a comma at the end of an array the page marks the comma. The answer now says both. Two claims about multi-megabyte files were removed and replaced by what the page does: input of up to 50,000 characters is checked 0.35 seconds after you stop typing, longer input only when you click a button. The page also links the specification it follows, RFC 8259, which defines an object as an unordered collection of members.
TDEE Calculator: the 1,200 and 1,500 kcal limits now cite Harvard Health Publishing (2024), which gives them as the intake not to go below without a health professional’s supervision, and the Frankenfield review now says what it backs, that the Mifflin–St Jeor equation is the most reliable of the common ones and falls within 10% of measured resting metabolic rate for most people.
GPA Calculator: the page said that most colleges recalculate applicants’ GPAs onto an unweighted scale; it now says some do, or use their own scale. The general advice on raising a GPA was replaced by a worked example (3.20 over 45 credits plus a perfect term of 15 credits gives 3.40) and the formula for the term GPA you need to reach a target.
Final Grade Calculator: the page called its formula the standard one used by most US and UK gradebooks and named three commercial learning platforms; neither claim could be backed, so the page now says only that it is a weighted average.
Percentage Calculator: the page said that results are formatted so that you never see floating-point noise. It now states the real precision, at most four decimal places (2 as a percent of 3 shows as 66.6667%).
Percentage Calculator: an answer said that a decrease cannot go below −100%. The calculator accepts negative values and shows −150% for a change from 10 to −5, so the answer now says that −100% is the limit only for a value that cannot go below zero, such as a price or a count.
RTTY Decoder: the table of presets said that the German weather service programme on 11.039 and 14.4673 MHz (call signs DDH9 and DDH8) has the same content as the first one and is transmitted only from 05:00 to 22:00 UTC. According to the service’s schedules of September 2023 it is a second programme, mostly in German, and it is transmitted 24 hours a day; the first programme (DDK2, DDH7 and DDK9) is mostly in English. The page now also names and links the sources it relies on: ITU-T S.1 for the five-unit code, the two programmes of the Deutscher Wetterdienst and the IARU Region 1 band plan.
Three more radio labs now name and link the sources they rely on: Morse Decoder (ITU-R M.1677-1 for the lengths of dots, dashes and gaps, the NCDXF and IARU beacon schedule, and the IARU Region 1 band plan), DCF77 Decoder (the PTB brochure About Time and the paper by Piester and colleagues on the DCF77 time code) and Signal Identifier (ITU-R M.1677-1 for the 1:3 ratio of dot to dash, and M.476-5 for the synchronous 7-unit code).
The first sentences of five radio labs were generic. They now say what each page does: SSTV Decoder (pictures that start by themselves on the mode header and are timed by their own sync pulses), RTTY Decoder (amateur contacts and German weather service bulletins, with both tones and the polarity found by the page), PSK31 Decoder (click a station’s trace in the waterfall and read its text), Morse Decoder (each dot and dash against the noise threshold) and Weather Fax Decoder (the speed changes and short gaps of web receivers).
Noise Generator: the answer to “Does noise help you sleep?” said only that some studies find small benefits and others none. It now cites the 2021 systematic review of 38 studies by Riedy and colleagues, which rated the evidence that continuous noise improves sleep as very low quality, with some studies finding better sleep and others worse. The page also names and links the method it uses to build the noise, that of Timmer and König (1995).
Tuning Systems: the page now names and links the two sources it relies on, Ellis (1885), who introduced the cent, and Barbour (1951) for the Pythagorean and syntonic commas and the wolf fifth.
Lissajous Figures: the history in the text (Lissajous’s 1857 memoir with mirrors on two tuning forks, and Bowditch’s curves of 1815) now links its source, the MacTutor biography of Lissajous.
Molar Mass Calculator: the note under the table called its values “conventional standard atomic weights”. It now says what the table holds, the abridged values of the IUPAC 2021 table with hydrogen kept at 1.008, and the text explains that a 2024 revision changed zirconium from 91.224 to 91.222, which the page has not taken over. The page also names and links its sources: the 2021 IUPAC technical report by Prohaska and colleagues, the CIAAW table of abridged atomic weights and the NIST value of the Avogadro constant.
Resistor Color Code Calculator: the page now names and links the sources it relies on, IEC 60062:2016 for the colour code and its tolerance values, IEC 60063:2015 for the E series of standard values, and the Wikipedia article that lists both the older and the newer tolerance for grey.
Heart Rate Zone Calculator: the sources now include Karvonen, Kentala and Mustala (1957), the study behind the Karvonen method, which sets training intensity as a share of the heart rate reserve.
Pregnancy Due Date Calculator: the two ACOG committee opinions in the sources now give their journal, volume and pages, and the entry for Committee Opinion 700 says what it backs, the 263 and 261 days counted from day 3 and day 5 embryo transfers and the limits for changing a due date after an ultrasound (its Table 1).
Grade Curve Calculator: the page said that many teachers prefer shifting the grade boundaries to the other curves; the claim could not be backed, so it was removed; the reason given before it, that the points stay as earned, remains. The page now also says how it differs from the Weighted Grade Calculator: this one curves the scores of a whole class, that one combines one student’s scores into a class grade.
Weighted Grade Calculator: the first paragraph now says what the page does that a plain average does not, that categories without scores are left out rather than counted as zero, and that it shows the average you need on the rest for your target and for each letter grade. It names its source for the weighted mean (Wikipedia) and says how it differs from the Grade Curve Calculator.
RSA Encryption: the page said that the largest challenge number factored in public, in 2020, had 829 bits. The public record is now RSA-896, an 896-bit challenge number factored in September 2026, and the page says so and links the announcement. It also links the RSA paper of 1978 and the paper by Sorenson and Webster, the source of the bound below which the page’s Miller–Rabin test is a proof of primality.
Theremin: the history said that the instrument was invented in the early 1920s, called it one of the earliest electronic instruments and the best known one played without contact, and said that it has been used in film music, without a source. It now gives what its sources support: Lev Termen first showed it to a group of physicists in 1920 and received US patent 1,661,058 in 1928, Clara Rockmore became a celebrated virtuoso on it, and its gliding tone became familiar from the soundtracks of early science-fiction films. The page links the patent and the Lemelson-MIT profile.
Six more labs now name and link the sources they rely on: Caesar Cipher (Suetonius on the letters of Caesar and Augustus, and Lewand’s table of English letter frequencies), Vigenère Cipher (Friedman on the index of coincidence, and Lewand), Vernam Cipher (Vernam’s paper of 1926, Shannon’s paper of 1949 on perfect secrecy, and the Venona monograph on the duplicated one-time pad pages), Playfair Cipher (Lyon Playfair’s own account of showing Wheatstone’s cipher to cabinet ministers in January 1854), Rotor Cipher Machine (the MacTutor biography and the Wikipedia article on Marian Rejewski, and the museum page on the bombe), and Metronome (Wilson on scheduling sounds on the audio clock). In the Rotor Cipher Machine answer on how the machine was broken, Welchman’s diagonal board no longer makes the bombe “much faster”; it now says what the museum page says, that it sharply reduced the false stops.
Three labs had a generic sentence at the top or in the FAQ. Drum Machine now says that each drum is built from oscillators and noise as it plays, not from recordings, and answers how to make a beat longer than one bar (chain the patterns A to D); Music Sequencer now answers whether you can export MIDI (you cannot: Export WAV saves the sound and Copy share link packs the notes into a web address); Rotor Cipher Machine now says that you can follow the current through every rotor, stage by stage.
Reaction Time Test: the text gave laboratory values of 180 to 250 ms for a light and 140 to 200 ms for a sound without a source, and said that online tests read higher “by tens of milliseconds” and that a colour test often lands between 200 and 300 ms. It now gives about 180 to 200 ms for a light and about 140 to 160 ms for a sound (Jain and colleagues, 2015), and the delay of a browser measured with a robot that pressed keys at known moments, about 80 ms on average and up to about 100 ms on some setups (Anwyl-Irvine and colleagues, 2021), so the text says that the page has no pass mark. The first sentences now say what the page shows, the median next to the average, and that a reaction under 100 ms counts as a guess and is repeated. Three overlapping sections became one, two of four worked examples and two of five FAQ answers were removed because the rest of the page already says the same, and both papers are linked.
Typing Practice and Typing Speed Test: the pages said that the average typing speed is around 40 WPM, and Typing Practice gave bands of 20 to 35, 50 to 70 and 80 or more WPM that could not be backed. Both now quote one measured study: 168,960 volunteers typed 15 English sentences in an online test and averaged about 52 WPM, with the slowest tenth under about 26 WPM and the fastest tenth over about 78 WPM (Dhakal and colleagues, 2018). The pages say that the volunteers were mostly teenagers and young adults, not a cross-section of all adults, and both link the paper. On the Typing Speed Test the ranges in the table are now described as the page’s rough guide, not measured values. Each page now says in one sentence how it differs from the other: in Typing Practice a wrong key is never inserted, so every wrong key press lowers the net speed, while in the Typing Speed Test you can fix mistakes with Backspace and only the errors left in the text count.
BPM Counter: the note on accuracy said that on synthetic test tracks with a steady tempo between 60 and 200 BPM the file analysis came within 0.1 BPM of the true value. The claim could not be backed, so it was removed; the note now says only that the analysis is most exact on music with a steady tempo. The page now names and links the source of its method for analysing a file, Ellis (2007), whose approach it follows: an onset curve, its autocorrelation, and a preference for tempos near 120 BPM.
CSV to JSON Converter, CSV to SQL Converter and CSV Viewer: each page now says in one sentence how it differs from the other two. They detect the delimiter by their own rules, so on an unusual file they may choose different ones, and they name an empty or extra column differently (fieldN and _extraN, column_N, Column N); CSV to SQL now also describes its own rule, which reads the first 20 records. The CSV to JSON page said that in a test a 5 MB file with 50,000 rows converted in about one second, and the CSV Viewer said that it was tested with 100,000 rows; no test was recorded for either, so both statements were removed. All three name and link RFC 4180, and the converters name their other sources: RFC 8259 for the number syntax of JSON, the MySQL manual for backslash escapes, and Microsoft Learn for the limit of 1,000 rows in one VALUES list of SQL Server.
Password Generator: an answer called twelve characters from all four sets, about 77 bits, “a reasonable minimum” without a source. It now gives the 77 bits and says that the NIST guidelines (SP 800-63B-4, 2025) ask services to require at least 15 characters when a password is the only login factor. The strength labels are now described as the page’s own scale, not a standard, and the page links NIST and the Diceware page.
Project to AI Text: the page said that every pattern in its .gitignore table was checked against version control’s own ignore check. It now says that each row gives the same result as git check-ignore on the same files, which was run again on all six rows, and it links the gitignore documentation. The answer on why a file is missing from the list moved into step 3 of the instructions.
Four more tools now name and link the sources they rely on: Hash Generator (FIPS 180-4 for the algorithms and block sizes, RFC 2104 for HMAC, and the paper on the first SHA-1 collision), Subnet Calculator (RFC 1918 for the private blocks, RFC 3021 for the /31 prefix, RFC 6890 for the special-purpose blocks), UUID Generator (RFC 9562) and YAML to JSON Converter (the YAML 1.2.2 specification and RFC 8259).
The first sentences of four tools were generic. They now say what each page does: Hash Generator (paste a checksum in hex or Base64 and the row it matches is marked, whichever algorithm it uses), Password Generator (strength in bits, counted exactly even when every character set must appear), UUID Generator (v7 values that stay in strict order even within one millisecond) and CSV Viewer (prices written as 3,50 sort as numbers, Central European code pages are in the encoding list, and Object lines mode turns toString() output into columns).
Matrix Calculator: the first sentence said that the page finds the eigenvalues of matrices up to 6 × 6 and that every answer is exact. Eigenvalues are offered only for 2 × 2 and 3 × 3 matrices, and for a 3 × 3 matrix whose characteristic polynomial has no rational root they are rounded numbers, labelled as such. The sentence now says which sizes are covered and that the arithmetic uses exact fractions. The page also names and links its source, Strang’s Introduction to Linear Algebra (6th edition, 2023).
Cups to Grams Converter: the page described its table of weights as typical figures of the kind found in food-composition tables and baking charts. It now says plainly that the values do not come from one named source: 125 g for a cup of all-purpose flour, 200 g for granulated sugar and 244 g for whole milk match USDA FoodData Central, and for other ingredients published tables differ by a few percent. The page now also links the source of the cup, spoon and fluid-ounce sizes (NIST Special Publication 811) and the USDA entry for the weight of flour.
Four more tools now name and link the sources they rely on: Prime Factorization Calculator (Sorenson and Webster for the bound below which the Miller–Rabin test is a proof, Baillie and Wagstaff and Baillie, Fiori and Wagstaff for the Baillie–PSW test), Time Zone Converter (the IANA Time Zone Database, EU Directive 2000/84/EC and 15 U.S.C. § 260a for the dates of daylight saving time), GCF & LCM Calculator and Extended Euclidean Algorithm Calculator (chapter 2 of the Handbook of Applied Cryptography, and Lamé’s theorem on MathWorld). The last two now also say in one sentence how they differ from each other.
Color Contrast Checker: the page described 0.03928 as the threshold in an older printing of WCAG 2.0. It now says that the WCAG definition used 0.03928 until May 2021 and 0.04045 since, and that no 8-bit colour value lies between the two, so no hex colour changes its result. A sentence said that several well-known style guides specify #767676; it now says what was computed, that #767676 (4.54:1) is the lightest neutral grey that passes AA for normal text on white and #777777 (4.48:1) fails. A sentence about designers softening body text, with no source, was replaced by this site’s own figure: its body text is 18.08:1 on white. An answer on colour blindness said that a red and a green of matching luminance can pass; two colours of the same luminance have a ratio of 1:1, so it now says that colours of similar luminance can both pass against the same background and still look alike. The page also names and links its sources: WCAG 2.2, the WCAG 3.0 working draft, which has not settled on a contrast algorithm, and the U.S. Access Board’s Section 508 standards.
Quadratic Equation Solver: the first sentence was a list of features and a promise of a reliable answer. It now says that decimals and fractions stay exact (0.1 is 1/10) and that decimal roots use a form of the formula that keeps its digits when b² is much larger than 4ac. The page links its source for the cancellation in the textbook formula, Goldberg’s paper on floating-point arithmetic (ACM Computing Surveys, 1991).
Word Counter: the paragraph on reading and speaking time gave a range of “commonly put at” speeds without a source. It now gives the figures of Marc Brysbaert’s 2019 review of reading rate (about 238 words per minute for non-fiction, 260 for fiction, 183 aloud), says that 130 to 160 words per minute for a talk is a rule of thumb, and places the page’s defaults, 225 and 150, against them. The first sentence now says how characters are counted (an emoji with a skin tone counts once, which the page does), and the page links Unicode Standard Annex #29 for it.
Unit Converter: the first sentence promised that you never see rounding noise such as 0.30000000000000004. It now says what the page does: each unit uses its exact definition and the arithmetic is exact, so 1 lb gives exactly 0.45359237 kg and 0.3 m gives exactly 30 cm.
Voice Dictation: the first paragraph said that you can say punctuation aloud, but spoken punctuation works in five of the 12 languages, and it now says so. The note under the tool and the answer on privacy began with “This page does not upload your audio or your text” and only then named the browser’s speech service, which in most browsers receives the audio; both now begin with that fact. The heading “Voice commands for every language” became “Voice commands by language”, because the table covers five.
List Randomizer: the page now names and links the sources for its shuffle, Durstenfeld’s 1964 algorithm, the paper by Bayer and Diaconis on how many riffle shuffles it takes to mix a deck, and the W3C Web Cryptography API. It is now the one page that explains why a shuffle is fair. Random Group Generator, Random Name Picker and Random Number Generator no longer repeat that explanation in a FAQ; each links to it.
Maze Generator: the page now names and links the sources for Wilson’s algorithm (Wilson 1996) and for the six algorithms and circular grids (Buck 2015). The table of difficulty figures was computed again with the page’s own code on 22 October 2026; the numbers did not change, and the page now shows the date.
Online Notepad: the page now names and links the sources for its 5 MB storage limit and for private windows forgetting what you wrote (MDN Web Docs), and for the Markdown marks it understands (Gruber 2004).
The first sentences of two more tools were generic. They now say what each page does: Random Number Generator (every number in the range, every die face and both coin sides equally likely, also with values left out, with exact decimals and in a draw without repeats between 1 and 10¹⁸) and QR Code Generator (the code holds the content itself, so it never expires, and Encoding details shows the segments and the penalty score of all eight masks).
Sudoku: the page described its four levels in a few generic lines and gave general tips. It now says how it builds a puzzle (a complete grid filled at random, clues removed in pairs placed symmetrically around the centre for as long as the built-in grader can still solve it with the techniques of the chosen level, and no symmetry on Expert) and shows a worked example for Easy, Medium and Hard: a naked single, a hidden single and a pointing pair. The tips now explain the number pad, the shading of the row, column and box, the highlighting of copies of a digit and the Hint button. The first paragraph says how Sudoku differs from Killer Sudoku, Calcudoku, Futoshiki and Kakuro: each game is a new grid built on your device, while the others draw on collections made in advance. Three answers that repeated the text above them were removed.
2048: the page now says where the game comes from, Gabriele Cirulli’s open-source game of 2014, and links its code. The text gives the numbers from the code: a new tile is a 2 nine times out of ten and otherwise a 4, and Undo steps back up to 200 moves. The list of seven tips was replaced by a short paragraph on the corner strategy.
Block Puzzle: the first sentence was “Place the blocks, clear rows and columns”. It now says what the page does: an 8 × 8 board, a tray of three pieces, and every new set of three contains at least one piece that fits.
The first sentences of two more games were generic. They now say what each page does: Snake (switch the walls off and the snake wraps around the edges) and Missile Defense (place blasts in the warheads’ path, because one blast can set off the next).
Bubble Pop: the first paragraph now begins with the goal, clearing the coloured bubbles at the top by popping them in groups, and says that every colour also carries its own symbol, so you can match bubbles by shape when colours are hard to tell apart.
Kanji Trainer: the page said that its set follows the lists that most N5 textbooks share, a comparison that was never made. It now says that the exam has published no official kanji list since its 2010 revision, that unofficial N5 lists range from about 80 to 103 kanji, and that all 100 kanji of the trainer are on an unofficial list of 103; only 午, 気 and 立 are left out. The page now links the JLPT FAQ and the two lists.
The first sentences of three more games were generic. They now say what each page does: Crate Pusher (40 levels, or random puzzles you can share by seed, each with a par found by a complete search), Hangman (no word comes back until you have played its whole category) and Kana Trainer (characters you miss come back a few questions later; the sentence used to promise “smart repetition”).
Mahjong Solitaire: three of the five answers in the FAQ were removed because the rules already say the same. The one fact that only the FAQ held, that every deal is built by playing a game backwards, placing tiles in pairs that would be free at that moment, moved into the rules.
Six puzzle games now say where the puzzle comes from and link the sources they rely on: Light Up (invented in 2001, and McPhail’s proof that deciding whether a grid has a solution is NP-complete), Nurikabe (first published in March 1991, and the NP-completeness proof by Holzer, Klein, Kutrib and Ruepp), Slitherlink (first published in June 1989, and Yato’s NP-completeness proof), Tents (invented around 1995 by Leon Balmaekers, and Hall’s 1935 paper behind the marriage theorem the page uses), Nonogram (Non Ishida’s idea of 1987 and her first puzzles of 1988, and the independent invention by Tetsuya Nishio) and Bridges (first published in 1990, after an earlier form in 1989). On Light Up, Nurikabe, Slitherlink and Tents, the two answers that every one of these pages repeated, on guessing and on what Check and Hint do, now sit in the text of the page itself.
Light Up: the description of Hint said only that it makes the easiest step it can find. It now also says what the code does when no simple step is left, that Hint adds one bulb of the solution.
Nonogram: the page said that before a puzzle is shown the game solves it with the same line-by-line reasoning. The text now says that a random grid is shown only after the game’s own line solver has finished it from the clues, and that the 31 pictures were run through the same solver on 22 October 2026; it finished all of them from the clues alone. The six bold labels in front of the tips were removed.
Bridges: the first sentence only restated the rules; it now also says that each level is graded by the reasoning it needs, from counting to what-if steps. The answer on whether a puzzle has only one solution was removed because “How the puzzles are made” says the same.
TriPeaks Solitaire: the page said that with Winnable deals only switched on, the game gives you only a deal that has at least one winning line, and tries another deal if checking takes too long. In the code, each shuffle gets up to 0.3 seconds; when the solver finds no winning line in that time, the game shuffles again, and after five tries you get the shuffle in which it cleared the most cards, without proof that it can be won. The text now says exactly that.
Dots and Boxes: the page said that against Medium, Hard won 1,188 of 1,200 test games, drew 11 and lost one. No script was behind the figures and a new run did not reproduce them. A script now plays 240 games on each board size from 3×3 to 7×7; run on 22 October 2026, it gave 1,194 wins for Hard, 2 draws and 4 losses, which the page now quotes. The page also lists two sources, Berlekamp’s book on the game and Peterson’s article in Science News on the long chain rule, and its first sentence now says what the page teaches, that giving away two boxes on purpose usually wins.
Sea Battle: the averages of the simulated games (Easy 86.8 shots, Normal 50.0, Hard 38.5, and 95.5, 60.1 and 44.4 with touching allowed) had no script behind them. A script now plays 2,000 games per level with a fixed seed; run on 22 October 2026, it gave 86.4, 50.2 and 38.4 shots, and 95.3, 59.7 and 44.9 with touching allowed, which the page now quotes. The page also cites the source of the formula for the mean of the largest of n numbers drawn without replacement, Siegrist’s chapter on order statistics, and its first sentence says that the heat map uses the same probability count as Hard.
The first sentences of six more games were generic. They now say what each page does: Klondike Solitaire (21 hidden cards, Draw 1 or the harder Draw 3), FreeCell (all 52 cards face up, four free cells, any deal replayed by its number), Spider Solitaire (two decks, 1, 2 or 4 suits), Pyramid Solitaire (pairs that add up to 13, with a solver that checks each deal by default), TriPeaks Solitaire (each card of an unbroken streak scores more than the last) and Tic-Tac-Toe (against the Impossible computer, a draw is the best you can do). Each of the five solitaire pages also says in one sentence how it differs from the others.
The five solitaire games repeated two FAQ answers, on whether every deal can be won and on how moves and time are counted. They now answer questions about the game alone: what the Out of moves panel means (Klondike), which deals cannot be won (FreeCell, whose first million deals include a handful of unwinnable ones, for example #11982), whether a supermove counts as one move (FreeCell), whether some deals are impossible (Spider), whether turning the waste over counts as a move (Klondike) and whether drawing from the stock counts as a move (TriPeaks). On Pyramid and TriPeaks the question about winnable deals was removed because the section on winnable deals already answers it.
Air Combat: the page now names and links the sources for its maths, the NASA Glenn Research Center page on the lift equation and Palumbo, Blauwkamp and Lloyd (2010) on proportional navigation, the guidance law the missiles use. Its first sentence now says what sets the game apart, the yellow lead circle that shows where to aim and the speed that hard turns cost. The four bold labels in front of the items in How to play were removed, and the page links the Flight Simulator page instead of calling it “our” game.
Flight Simulator: the page now names and links the sources for its figures, the NASA Glenn page on the lift equation, chapter 5 of the FAA Pilot’s Handbook of Aeronautical Knowledge (a 60° bank means a load factor of 2 g, and the stall speed grows with the square root of the load factor) and the FAA Aeronautical Information Manual (the approach path lights, two white and two red on a 3° glide path). The first sentence now says that the landing modes score sink rate, centre line and touchdown point. The four bold mode names in the list were turned into sentences.
The first sentences of three more games were generic. They now say what each page does: Star Fighter (a diamond shows where to aim, and a warning comes before any shot from behind), Tank Arena (each robot’s barrel glows before it fires, and its shells are slow enough to dodge) and Tunnel Flight (a dashed circle on the next obstacle shows whether you will fit, and every gap is within reach). The Tank Arena section on drawing in 3D shared its heading with Star Fighter; it is now called Projection and hidden lines.
Moon Lander: the rule for hovering said that the engine must run at about g divided by the engine power. It now says g divided by the full-thrust acceleration, 4.2 m/s² in lunar gravity, which is the 39% of full thrust the text already gave. The page also says that the lander does not get lighter as it burns fuel, so full thrust always gives the same acceleration, which a real lander does not do. It now names and links the sources for its gravity values, the NASA Moon Fact Sheet (1.62 m/s²) and the NIST value of standard gravity (9.806 65 m/s²).
Maze Chase: the page now says where the genre comes from, the arcade game that Toru Iwatani designed and that was released in Japan in 1980, and links two Wikipedia articles about it and about maze games.
Drone Delivery: the page now names and links the sources for the physics behind its numbers, the OpenStax chapter on the period of a pendulum (the swing of the parcel on its rope) and the Wikipedia article on momentum theory (the power a hovering rotor needs).
The first sentences of four more games were generic. They now say what each page does: Maze Chase (four drones that hunt you in four different ways), Moon Lander (one tank of fuel for the whole run, and readouts that warn you when your speed or tilt is unsafe), Neon Trails (on Hard the rivals count the space each move wins and try to cut you off; the section on how they think now follows the rules) and Cave Copter (a press stops a fall in about 0.04 seconds, and the flight is the same on a slow or a fast screen).
Road Hopper: the shared answer on sound was removed from the FAQ. The text under Controls now says that sound effects are off at first, that the Sound button turns them on and that the game remembers the setting.
Target Hunter 3D and Paintball Bot Arena 3D: the hint shown in the game said that the mouse wheel zooms the scope between 2.5× and 6.0×. The code allows 2.0× to 6.0×. The hint is corrected in English and in Slovak.
Paintball Bot Arena 3D: the menu promised “smart AI opponents”. It now says what the game has, eight bots that the computer controls. The first sentence of the page says that the bots always know where you are and that only trees, rocks and hills stop their shots.
Bot Arena: Protocol Survive: the menu said that the robots “learn to read your movement”. They do not learn anything; they hunt you by sight and by sound. The sentence is corrected in English and in Slovak. The page now also says that turning the view and firing work only while the browser has captured the mouse, so in some remote-desktop sessions you can walk but not aim, and a new answer explains how a robot finds you after you hide.
Precision Range: a new answer explains why the bullet drops more than a real rifle’s. The game multiplies gravity by 1.25 for the pistol, 2.35 for the DMR and 2.75 for the .308, and the two drop columns of the DOPE table show the real drop and the hold the game needs.
3D Maze: the page now names and links the sources for its two algorithms, Amanatides and Woo (1987) on the grid walk behind every ray and Jamis Buck (2010) on the recursive backtracker that carves the mazes.
Maze Blaster: the page repeated the explanation of raycasting from 3D Maze, with its formula. It now links 3D Maze for it. The FAQ now answers two questions about the robots, why one finds you after you hid and whether one can see you from behind.
Target Range: the first sentence was generic. It now says that you turn on the spot and hit targets on every side in precision, speed and tracking rounds, with the mouse, the keyboard alone or touch.
Target Hunter 3D, Paintball Bot Arena 3D, Bot Arena: Protocol Survive and Precision Range now say that Simpli, who runs this site, made the game, and that the Slovak texts are the original ones.
Seven shooter and maze games repeated answers about the mouse pointer, the screen edge or a slow game that are already under Controls or How it works. Their FAQs now answer questions about the game alone, for example what TARGET LOCKED means in Target Hunter 3D.
Two statistics tools now name and link the sources they rely on: Confidence Interval Calculator (Wilson’s 1927 paper, which introduced the score interval the page uses for a proportion, and Brown, Cai and DasGupta (2001) on why the Wald interval is unreliable in small samples) and Histogram Maker (Sturges (1926), Scott (1979) and Freedman and Diaconis (1981) for its three rules for the bins).
The first sentences of three logic tools were lists of features. They now say what each page does and which neighbour to use instead: Boolean Algebra Simplifier (an expression or minterms with up to 8 variables, the result in both notations, checked row by row in a truth table; Karnaugh Map Solver is for clicking cells on a map of up to 5 variables), Number Base Converter (four bases at 8, 16, 32 or 64 bits, a bit grid and named bit flags) and Two’s Complement Calculator (any width from 2 to 64 bits, adding and subtracting with an overflow flag, and binary fractions).
Twelve labs had five answers in their FAQ. Each now has at most four: the answer that only repeated the text above it was removed from Double Pendulum, Double-Slit Experiment, Epidemic Simulator, Fourier Drawing, Game of Life, Langton’s Ant, Projectile Motion, Strange Attractors and Tinnitus Matcher. On Flocking Simulator, Fractal Explorer and Noise Generator the one fact that only the FAQ held moved into the text first: how the Order figure is calculated, that the colour palette repeats in cycles, and that a very attentive listener may notice the 11.9-second loop. Number Base Converter and Two’s Complement Calculator also had five answers; the one on fractions and other widths, which the first sentence of the converter now covers, and the one on why 128 does not fit in 8 bits, which the section on range covers, were removed.
Pyramid Solitaire: the page said that in “our tests” with 600 random deals about two in three could be won with 2 or unlimited redeals, but only about one in twenty with none. No script was behind the figures. A script now runs the game’s own solver on 600 random deals; run on 22 October 2026 with 2 seconds per deal, it proved 390 deals winnable with 2 redeals, 398 with unlimited redeals and 27 with none, which the page now quotes.
Hash Generator: when its first sentence was rewritten, the sentence that nothing is uploaded went with it. The note under the text box, which already says that files and the HMAC key are never stored, now says it again.

22 October 2026: controls

Prime Factorization Calculator: the English / Slovenčina switch at the top of the tool became a small Slovenčina link under the input box, and the line under the heading is now translated too.
Factor Tree: the English / Slovenčina switch in the header of the lab became a small Slovenčina link under the lab, and the line under the heading is now translated too.
Rotor Cipher Machine: letters typed on the keyboard reached the machine only while the focus was on the keys or on the signal path. After you pressed a sample button, a setting or an experiment tab they were ignored. They now work wherever the focus is in the app, except in a text field.

21 October 2026: facts and sentences

Hangman: the description promised 20 word categories; the game has 18 themes plus a random mix.
Retirement Calculator: a tip said that the return in Example 3 is below inflation. It equals inflation before retirement and falls below it afterwards.
Grade Curve Calculator: the example about rounding at a grade boundary was wrong. On a 100-point test a curved score of 89.996, shown as 90.00, still gets a B.
Pregnancy Due Date Calculator: the examples now write every date the same way, a percentage without a source was removed, and the rule for moving the due date after an early scan follows ACOG Committee Opinion 700.
Pathfinding Visualizer: the description named five search methods; the page has six, including bidirectional BFS.
Cups to Grams Converter: the description said 60 ingredients and the first sentence more than 60; the table has 59.
Caesar Cipher: success rates for short texts from a test that was never published were removed, and the sentence about al-Kindi now says what he described: frequency analysis.
Links on three pages said that the Unit Converter converts volume, and links on two pages said that the Percentage Calculator converts fractions and probabilities. Neither page does that, and the link texts were corrected.
Resistor Color Code Calculator: a note now explains that IEC 60062:2016 gives the orange, yellow and grey tolerance bands values that this calculator does not use.
Color Contrast Checker: an undated sentence about WCAG 3 was replaced with what the current draft (10 September 2026) says about its contrast method.
QR Code Generator: a broken sentence in the link to the Color Contrast Checker was rewritten.
Word Counter: it now says why its reading time differs from the one in the Online Notepad (225 against 200 words per minute).
Small corrections on Typing Game, Kanji Trainer, Missile Defense, 15 Puzzle, Killer Sudoku and Periodic Table Quiz: grammar in a description, a button label that did not match its table, mixed spelling, a page title that did not say what the page is, and list headings.

21 October 2026: sources and limits

The Loan, Compound Interest, Retirement, Heart Rate Zone, TDEE and Pregnancy Due Date calculators now name and link the sources of their formulas.
Seven hearing labs now list the standards and studies they rely on, and the Hearing Loss Simulator says how its age profiles compare with ISO 7029.

21 October 2026: what the pages say about your data

On more than 100 pages, sentences such as “nothing is sent anywhere” or “never leaves your device” were replaced with “nothing you type or record is uploaded”. The site counts visits with Cloudflare Web Analytics and may show ads, so the old sentences were not exact. See Your data.
The Voice Dictation tool and the Privacy Policy now say that Read aloud may use an online voice of your browser, which then receives the text it reads.

21 October 2026: controls and full screen

On 36 pages with keyboard controls, the keyboard instructions were hidden on computers with a touch screen. The pages now show the controls for the mouse, keyboard or touch you actually use, for example Snake, Synthesizer and Typing Speed Test.
Twelve arcade games got a play area that fills most of the window, a Full screen button and game keys that no longer scroll the page, and most of them now explain the controls on the start screen: Alien Wave, Brick Breaker, Bubble Pop, Bug Blaster, Maze Chase, Missile Defense, Moon Lander, Neon Trails, Paddle Duel, Road Hopper, Snake and Space Rocks.
Space Rocks can now be played with the mouse.
Alien Wave, Neon Trails, Road Hopper and Space Rocks: the title screens stand still when your device asks for reduced motion.
Air Combat: steering comes back as soon as the pointer returns to the game. Before, the pointer had to reach the centre circle first.
Tank Arena: a first game starts on Easy.
3D Maze: Space starts the game from the start screen, and the arrow keys no longer scroll the page there.
Bot Arena: Protocol Survive: the menu background stands still when your device asks for reduced motion.

Added pages

Counted from the day each page first appeared on the site.

September 2026: 34 pages, 13 tools and 21 games.
October 2026, 1 to 29 October: 179 pages, 53 tools, 46 games and 80 labs, the last five being the Brain & Memory Tests on 29 October: the Memory Test, Visual Memory Test, Dual N-Back, Stroop Test and Attention Span Test.
November 2026, 1 and 2 November: 11 labs, which brings the labs to 91, all of them more Brain & Memory Tests: the Go/No-Go Test, Task Switching Test, Visual Search Test, Multiple Object Tracking, Number Sense Test, Time Perception Test, Attentional Blink Test, Change Blindness Test, Mental Rotation Test and Verbal Memory Test, and the Brain Profile, which shows the saved results of all of them in one place.