Releases 1.139.0 – 1.170.2

Newest first. Three numbers, and which of them moves says what happened to you rather than how much work it was.

The first moves when something that worked stops working: a query that used to answer and now does not, a saved list that will not load, a link you shared that no longer opens what it opened. It has never moved, and that is the whole of what it is for.

The second moves when the page answers something it could not answer before, or answers it better than it did. New grammar is the usual reason. A correction is the other, and a correction is not a break even though the answer changes underneath somebody: "days be christmas" used to say 154 days and now says nothing, because it was reading Belgium out of the word "be", and taking that away left the page more right rather than less useful. An answer that gains a drawing of what it just said counts here too.

The third moves when nothing you type comes back any different.

Numbers get skipped when it takes more than one commit to introduce a release, so only the release from the last commit goes out.

1.170.21 August 2026

Hex digits say what they are

ff in decimalnothing255 · ff base 16 in base 10
beef in decimalnothing48879
A prefix says what base a number is written in, and nobody types one: "ff in decimal" says it by the letters it is made of, since base ten has no f in it.
The echo says the reading out loud rather than assuming it silently, and a target of hex is left alone, because "ff in hex" would be answering the question with the question.

1.170.11 August 2026

Two to the tenth

2 to the 10thnothing1,024
e to the 2nothing7.38905609893
"to the power of" has been read since the spoken signs went in, and nobody says the word "power" out loud. The ordinal is what marks a number rather than a noun, so "to the moon" is still nobody's sum.

1.170.01 August 2026

The fractions people say instead of writing

half of 60nothing30
a third of 90nothing30
3/4 of 200nothing150
double 42nothing84
"half of 60" is the commonest sum there is out loud, and the one shape of it this file could not read: it has read "60 / 2" since the first week and "half" is the word for the same thing.
The periods keep their words. A quarter of a year is still a quarter of a year, because these are read only where a number follows.

1.169.01 August 2026

The whole, asked for instead of the part

20% of what is 30nothing150 · 20% of 150 is 30
20% of what is 30 cmnothing150 cm
5 is 10% of whatnothing50
Five orders of the percentage question were readable and this was the sixth: every other rule takes the whole and hands back a part, and this is the one where the part is the number you have.
A unit rides along on the part, because the answer is in whatever the part was in. Dropping the "cm" would hand back a number that has lost what it counts.

1.168.01 August 2026

Told what to do rather than what happened

increase 200 by 15%nothing230
take 20% off 200nothing160
add 20% to 200nothing240
"200 increased by 15%" has answered since the spoken signs went in, and that is the sentence somebody writes about a number that has already moved. The imperative is what they type when they are telling the calculator what to do.
Both orders of the two halves are one sum, so "add 20% to 200" and "take 20% off 200" arrive at the rules that were already there.

1.167.01 August 2026

Which way it is

bearing from tokyo to londonnothing336° NNW · arriving 212° SSW
which way is london from tokyonothing336° NNW
It has been printed on the detail line of every distance answer since the coordinates went in, and it is the whole question for anybody pointing an aerial, a telescope or themselves.
Both ends are named, because a great circle does not hold one course: Tokyo to London leaves north-north-west and arrives south-west, which is why the route looks bent on a flat map.
Asked with the verb in it, the far place is said first and answered second: a course is measured from where you are standing.

1.166.11 August 2026

Two more ways of asking how far

how many miles is tokyo from londonnothing5,943 miles
midpoint tokyo londonnothing69.02, 89.75
The verb between the unit and the first place, and the joining word inside the pair rather than in front of it: two ways of writing a question that was already answered in one way.
The second word of a unit is a second word of the unit -- nautical miles -- and not the joining word behind it, which the greedy reading took and then failed to look up.

1.166.01 August 2026

The moon, asked as a question

how many days until the full moonnothingin 5 days · Wed, 29 Jul 2026 · 10:37 am
is it a full moon tonightnothingno · waxing gibbous, 78% lit · the full moon is 29 Jul 2026
is it a full moon on 29 jul 2026nothingyes · full at 10:37 am
These are the two ways the moon actually gets asked about, and both were silent. Handing back the date and the instant answers a question nobody quite asked.
Same arithmetic, different headline: a countdown where a countdown was asked for, and a yes or a no where the question was a yes-or-no one.

1.165.01 August 2026

The moon on a night that has not happened yet

moon phase tomorrownothingwaxing gibbous · 86% lit
moon phase on 25 dec 2026nothingwaning gibbous · 97% lit
full moon in decembernothingWed, 23 Dec 2026 · 8:29 pm
The moon was only ever asked about tonight. The commonest night to ask about is not tonight: a walk, a photograph or a sky worth going outside for is planned for one that has not happened yet.
Whatever the date engine reads is the day, the way the sun already takes one, so every date this calculator knows can be the night.

1.164.01 August 2026

The solstices and the equinoxes

june solstice 2026nothingSun, 21 Jun 2026 · 4:24 am
when is the equinoxnothingTue, 22 Sep 2026 · 8:16 pm
winter solsticenothingMon, 21 Dec 2026 · 3:54 pm
The longest day has been answerable since the sun went in, and it is not the same fact: the solstice is an instant, the day is the day it lands in, and a calendar prints the first.
The season names are answered as the reader's own hemisphere means them. "summer solstice" asked from Sydney is December, because telling somebody the season they are not in is not an answer.
Good to about a quarter of an hour, which the answer says out loud. The perihelion the sunrise equation counts from is held at its value for the year 2000, and half a degree of drift is half a day of sun, so the same equation read the usual way came out eleven hours late.

1.163.01 August 2026

The golden hour, and the blue one

golden hour in tokyonothing6:17 pm → 7:11 pm (54m) · morning 4:26 am → 5:20 am
blue hour in londonnothing9:28 pm → 9:45 pm (17m)
Both are elevations rather than events, which is why they were the two sun questions this file could not answer while answering the rest. The sunrise equation takes whatever angle it is asked for and has since it went in.
Golden light is the sun between four degrees below the horizon and six above it; the blue hour is the two degrees below that. Neither is an hour, and both say how long they actually last.
The evening one leads, because that is the one people plan around: the morning one asks you to be standing there before dawn. Both are printed, since the question does not say which.

1.162.01 August 2026

How long until sunset

how long until sunsethow long until isn't a place I knowin 3h 41m · 8:21 pm in New York
how long until sunset in tokyonothingthe same, for the clock there
how long since sunrisenothing6h 14m ago
What somebody standing outside wants to know is not when the sun sets but how long they have got. The instant has been computable since the sun went in and the clock has counted to a time of day since the spans did, and nobody had put the two together.
A sun time already gone is tomorrow's, the way "hours until 9am" asked at ten in the morning means tomorrow morning. Asked backwards, one still to come is yesterday's.

1.161.01 August 2026

First light, nightfall, and the other names for the same instant

first light in londonnothing4:30 am in London
nightfall in parisnothing10:21 pm in Paris
twilight in tokyonothing7:22 pm in Tokyo
hours of daylight in londonnothing15h 50m
The civil twilights have been computed here since the sun went in, under the names "dawn" and "dusk". These are the same two instants said the way people say them out loud, and they cost no arithmetic at all.

1.160.11 August 2026

A span of clock time with a place on the end

3pm to 4pm in tokyoto 4pm isn't a place I know1 hour
9am in new york to 5pm in tokyonew york to 5pm isn't a place I know19 hours
Two clocks with a range word between them is a span of time whatever places are hung on either end, and the clock grammar was reaching the second half of the line and blaming it for not being a city.
A place said once at the end is meant for both ends: an afternoon written that way is one afternoon in one city, not an afternoon spread across two.

1.160.01 August 2026

A shift, written the way a rota writes it

9am to 5pmnothing8 hours
15:00 to 17:30nothing2 hours 30 minutes
10pm - 6amnothing8 hours
"hours between 9am and 5pm" has answered since the clock spans went in, and nobody writes the word "hours" on a rota.
Both ends have to be a clock that says it is one, an am or a pm or a colon, which is what keeps this off "23 jul to aug" and every other range the date grammar has always read.

1.159.01 August 2026

The name somebody writes in front of a line

Price: 6 times 5nothing30
Deposit: 20% of 4500nothing900
Fee: 20 km in milesnothing12.4274 miles
That is how a calculation gets written down when there is more than one of them, and the words in front of the colon are for the person reading it back tomorrow rather than for the calculator.
A clock keeps its colon. Letters and spaces only in front of it, and not too many of them, so "3:30 in tokyo" is a time and a colon halfway down a long sentence is punctuation.

1.158.01 August 2026

Names for things

rent = 1200nothingrent = 1200 · kept in this browser
rent * 12nothing14,400
speed = 20 km, then speed in milesnothing12.4274 miles
varsnothingthe names you have saved
clear rentnothingrent forgotten
A calculator is used twice in a row more often than once, and the second question is nearly always about the first answer. Everything else here is a rule; this is the one thing that has to be remembered.
What is stored is the text on the right of the equals rather than the answer it came to, so "budget = 1200 usd" uses the rate at the moment you ask rather than the rate when you typed it.
A name is written out before any reader sees the line, which is what makes one substitution serve the whole grammar: a saved name works in a sum, a conversion, a date and a clock alike.
A name this calculator already uses is not free. "km", "may", "max", "pi", "usd" and "tokyo" all go on meaning what they meant, because a variable that took one of those would hand back a wrong answer to a sentence with nothing wrong in it.

1.157.01 August 2026

Two facts about your screen, typed rather than assumed

ppi = 320nothing320 px per inch · 1 cm is 125.984 px
1 cm in px37.7953 px125.984 px, once ppi is 320
em = 14pxnothing14 px per em · 16 px is 1.14286 em
The screen in front of you is probably not 96 pixels to the inch and its text is probably not 16, so the two numbers the table cannot know are typed rather than assumed.
Nothing is applied while the line is being typed: halfway through "ppi = 320" the box says 3, and a setting that took that would have to be argued out of it. Enter applies it, and this browser keeps it.
Every row already on the page is re-read underneath it, so "1 cm in px" written before the setting answers for the new screen.

1.156.01 August 2026

The screen is a length once you say how dense it is

1 cm in pxpx isn't a unit I know37.7953 px · 96 px per inch
16 px in emnothing1 em
2 em in pxnothing32 px
1 pica in mmnothing4.23333 mm
A CSS pixel is 1/96 of an inch by the specification, which is a definition exactly like the inch, and a browser's default text is 16 of them. The arithmetic was already here; the two units were not.
Both numbers are assumptions about somebody else's screen rather than measurements, so every answer that leans on one says which it leaned on.
"pt" stays a pint. The CSS point is spelled csspt here, because a unit that is sometimes a drink is a unit that is sometimes wrong.

1.155.11 August 2026

The question asked in the middle of the sentence

9am in tokyo is what time in londontokyo is what time isn't a place I know1:00 am in London
what time is 3pm est in pstis 3pm est isn't a place I know12:00 pm in Los Angeles
convert 9am tokyo to londonnothing1:00 am in London
Spoken aloud, the question sits in the middle of the sentence rather than at either end, between the clock and the place it is wanted in. Read as part of a place name it produced a refusal built out of the question itself.

1.155.01 August 2026

Here, said in words

3pm tokyo in my timemy time isn't a place I know2:00 am in New York
how far is tokyo from meme isn't a place I know10,851.1 km
what time zone am i innothingEastern Daylight Time
Somebody asking this does not think of where they are sitting as a place with a name. "me", "here", "local time", "my timezone": every one of them was handed to the place resolver and blamed for not being a city, which is a true sentence and a useless one.
They resolve rather than being added to the zone table, so none of them turn up in the suggestions: they are a way of saying a place, not a place.

1.154.01 August 2026

The two-letter zones a meeting is written in

3pm et to ptpt isn't a place I know12:00 pm in Los Angeles
3pm ct to mtnothing2:00 pm in Denver
ET, CT, MT and PT are how a meeting is written down more often than EST and PST are, because they do not commit to a season: an invitation says 3pm ET precisely so it need not say EST in January and EDT in June.
PT is a pint to the converter, and a name the units own is kept out of the zone table on purpose. It is read as a zone only where a clock time is standing in front of it, so "3 pt in litres" is still a conversion.

1.153.11 August 2026

The word now, said before the place instead of after it

time now in tokyonow isn't a place I know9:00 am in Tokyo
time right now in tokyoright now isn't a place I know9:00 am in Tokyo
"time in tokyo now" answered and "time now in tokyo" did not. The word was dropped off the end of a line and not off the front, so it ended up inside the place name and was blamed for not being a city.

1.153.01 August 2026

Is tokyo ahead of london

is tokyo ahead of londonnothingyes · Tokyo is 8 hours ahead of London
is los angeles ahead of new yorknothingno · Los Angeles is 3 hours behind New York
is tokyo aheadnothingyes · Tokyo is 13 hours ahead of New York
The number has been answerable since the zones went in, and this is the question underneath it: somebody deciding whether to send a message now wants a yes, not thirteen hours.
A no is never only a refusal. The line underneath says the true direction whichever way the answer falls, because being told "no" and not being told "behind by three" is half an answer.

1.152.01 August 2026

The other end of an ahead-or-behind question

how many hours ahead is tokyo of londonnothingTokyo is 8 hours ahead of London
how far behind is los angeles from new yorknothingLos Angeles is 3 hours behind New York
The place compared against was only read in front of the subject. Said out loud the sentence puts it behind: "ahead is tokyo of london", which read as one long place name and answered nothing.

1.151.01 August 2026

A time difference with no word between the places

time difference tokyo londonnothingLondon is 8 hours behind Tokyo
time difference new york tokyonothingTokyo is 13 hours ahead of New York
time difference tokyo narnianothingnarnia isn't a place I know
The joining word was required and is not always typed. Two place names with nothing between them is how this gets asked once somebody has asked it twice, and it read as one long place name that does not exist.
It splits the way the distance reader splits: the joining word first, then every plain gap, and the first split that names two places wins.

1.150.01 August 2026

When is christmas

when is christmasnothingin 154 days · Fri, 25 Dec 2026
when is thanksgiving 2027nothingThu, 25 Nov 2027
when was easternothing110 days ago · Sun, 5 Apr 2026
"christmas" has answered since the holidays went in and "when is christmas" has not, which is the same question with the word everybody actually asks it with on the front.
The past tense is read as one: "when was easter" is not a question about the Easter that has not happened yet, so it answers with the last one rather than the next.

1.149.031 July 2026

The word people ask a flight in

flight tokyo to londonnothing11h 8m
flight from vladivostok to dubai timenothing8h 21m
tokyo to london flight timenothing11h 8m
how many hours is the flight from tokyo to londonnothing11h 8m
flight to tokyonothing12h 33m · New York → Tokyo
The sentence this was written for was "how long is the flight from tokyo to london", which is the longest way anybody says it. The noun on its own is the question, wherever in the line it lands and whether or not "time" is said after it.
The trailing keyword is the weak one, the way the trailing "distance" is: it answers only when both halves turn out to be places, and blames nobody when they do not.
how far apart are tokyo and londonapart are tokyo isn't a place I know9,564.34 km
when does the sun rise in tokyonothing4:43 am in Tokyo
sunrise tomorrow in tokyotomorrow in tokyo isn't a place I know4:44 am in Tokyo
1000 days agonothingSat, 28 Oct 2023
what is the date in 90 daysnothingThu, 22 Oct 2026
how long ago was 1 jan 2020nothing2,396 days ago
Four more silences in the same family, found by typing the questions rather than the grammar: the word "apart", the word "when", a day said before the place it is about, and a count of four digits, which was being read as a year with the unit left over behind it.

1.148.231 July 2026

The source is twenty-seven modules; the page still gets one file

Nothing you can type changes, and not one byte of what the page serves moved: public/calc.js is now built by concatenating src/, and the build is a concatenation and nothing else.
One file is still what gets delivered on purpose. Serving the modules separately would cost about a tenth of the compressed size again, and would stop the page opening from disk, since browsers refuse to load a module over file://.
A name is now something a module owns: the build refuses to produce a file where two modules declare the same thing at the top level. That had happened twice, silently, and the suite passed both times.
Fixed one of those two: a bill formatter and the currency formatter were both called fmtMoney, so bills were being written by the currency one with no currency to write them in.

1.148.131 July 2026

The README catches up with twenty-five releases

Nothing you can type changes. The page had shipped twenty-five additions and four fixes without the README naming one of them, which makes it a description of a calculator that no longer exists.
The table gains the moon, the day counts, the constants, the bills, the bits, the words, the formats and the rest; four new sections carry the arguments that needed more than a row.
The test count on that page said 4,108 and the suite runs 13,754, a number that had been quietly wrong for long enough that nobody was going to notice it.

1.148.031 July 2026

The moon

moon phasenothingwaxing gibbous · 73% lit
next full moonnothingWed, 29 Jul 2026 · 10:37 am
next new moonnothingWed, 12 Aug 2026 · 1:38 pm
The sun has been answerable here since the coordinates went in, and the moon is the other light people plan around: a walk without a torch, a photograph, a tide, a night worth going outside for.
Lunar New Year and Diwali are refused in this file because the arithmetic everybody ships disagrees with the published dates. That refusal is about a calendar built on a moon, not about the moon: the instant of a full moon computes to the minute, and these agree with the published times to within one.
The full moon is computed rather than assumed to sit halfway between two new moons, since the moon runs fast at perigee and can fall half a day off the middle of its cycle.

1.147.031 July 2026

The other two means, and two names that were missing

geometric mean of 2 8nothing4
harmonic mean of 30 60nothing40
truncate 4.9nothing4
A geometric mean averages a growth rate or a ratio and a harmonic mean averages a speed over a fixed distance. In both of those the arithmetic mean is not a different number, it is the wrong answer.
"trunc" answered and "truncate" did not, which is four letters of a programming language standing in for a word.
"3.14159 to 3 sig figs" needs no verb in front of it: the phrase already names what it wants done.

1.146.031 July 2026

A date written back out in the shape you need

today as dd/mm/yyyynothing24/07/2026
christmas as mm/dd/yyyynothing12/25/2026
today as iso weeknothing2026-W30-5
This file reads a date in every shape anybody writes one and could write it in exactly one, which is a translator that only speaks its own language.
Half the times a date gets looked up it is being copied into something: a form that wants the American order, a filename that wants the sortable one, a field that will not take a spelled-out month.
The ISO week date carries its own surprise, its year not always being the year its days are in, so the plain date goes underneath it.

1.145.031 July 2026

The Julian day, and the spreadsheet serial

julian daynothing2,461,246
excel serial 45000nothingWed, 15 Mar 2023
excel serial 60nothing29 February 1900, a day that never happened
Ten calendars answer here and neither of these is one. They are a running count of days from a fixed morning, which is what you reach for when you want to subtract two dates without months getting in the way.
The Julian day begins at noon, which is its whole point: an astronomer works through a midnight and does not want the date changing halfway.
The spreadsheet count is one out on purpose. Lotus treated 1900 as a leap year, Excel copied the bug for compatibility, and everything since has copied Excel, so day 60 is a 29th of February that never happened.

1.144.031 July 2026

Money over time

1000 at 5% for 10 yearsnothing1,628.89
mortgage on 300000 at 6% for 30 yearsnothing1,798.65 a month
doubling time at 7%nothing10 years 3 months
All three are one line of arithmetic and all three are the arithmetic nobody does. A percentage is easy; a percentage applied ten times over is not, which is why compound interest is the thing everybody is told to understand and few ever work out.
Compounded once a year unless the line says otherwise, and the answer says which, since the same rate compounded monthly is a different number.
The rule of 72 goes under the exact doubling time, because the rule is what most people are carrying and the use of an exact figure is knowing where the shortcut stops being close enough.

1.143.031 July 2026

The constants, which are rules rather than facts

speed of lightnothing299,792,458 metres per second
speed of light in mphnothing670,617,000 mph
avogadro numbernothing6.02214076 × 10²³ per mole
Since 2019 the SI is defined the other way up: the second and the metre are fixed by giving these numbers exact values. The speed of light is not measured at 299,792,458 metres per second, it is that.
Which makes them the same kind of thing as a unit factor or a leap-year rule, and they belong here for the same reason: nothing about one of them can go stale.
The two that are measured say so, since an answer that will not admit which of its digits are settled is claiming more than it has.

1.142.031 July 2026

A number written out in words, both ways round

1234 in wordsnothingone thousand two hundred and thirty-four
1004 in wordsnothingone thousand and four
two million in digitsnothing2,000,000
Roman numerals have read both ways round since they went in, which is the rarer notation by a long way. The one everybody has to write is the one on a cheque, where a figure is repeated in a form that cannot be altered with a pen.
Said the way it is said, with the "and" a hundred takes in front of what follows it, and a decimal read digit by digit, since "point fourteen" says nothing about where the point is.
Read before the rewrites, since the first thing those do to "two million in digits" is turn its words into the digits it was asking for.

1.141.031 July 2026

A tip, and a bill split between the people who ate it

20% tip on 60nothing72.00 · 12.00 tip
split 120 between 4 with 20% tipnothing36.00 each
split 91 between 3nothing30.33 each · 1 penny short
Both of these are percentages and division, which this file has done since the first week, and neither is the sum anybody wants: "20% of 60" is twelve and the number wanted is seventy-two.
The arithmetic was never the difficulty. The answer is two numbers and a calculator gives one, so the second gets done in the head at the table, which is where it goes wrong.
Two decimal places, because a bill has to be paid in pennies, and the pennies left over by a split that does not divide are counted rather than left to be discovered.

1.140.031 July 2026

Fixed: a conversion that handed the question back

100 mbps in mb/s100 Mbps100 Mbps · the same unit twice
100 km in kilometres100 kilometres100 kilometres · the same unit twice
The number came back unchanged with nothing said about why, which reads as a conversion that ran rather than as two spellings of one unit.
For a data rate it says the reason as well: a capital B is the byte and nothing else is, so "mb/s" is megabits and "MB/s" is eight times fewer. That eight is the whole reason anybody is confused about how long a download takes.

1.139.031 July 2026

Fixed: the age reader answering questions it had not read

retirement date if born 196066 years 6 months 23 daysnothing
pension age if born 196066 years 6 months 23 daysnothing
Anything at all could stand in front of "born", so every line with a birth date somewhere in it was answered as an age, whatever else it asked for.
A question asking for a date, answered with a number of years, by a reader that never looked at the first two words and gave no sign of it. A question this file cannot answer has to look unanswered.
The words that do belong in front of it are listed, and every way of asking an age keeps its answer.