Ruler Calibration, Error Bands and Limits

What is measured, saved and checked

What is measured. Calibration measures one ratio: how many of your display's pixels cover a millimeter. No browser is told the physical size of the screen it draws on, so you supply that ratio by matching an outline to a clean, smooth-edged reference held just above the display without touching it. Every millimeter, inch and fraction the calibrated ruler draws comes from it. Marks sit at their exact fractional pixel positions. Snapping them to whole pixels would move a mark by up to half a pixel, which is 0.13 mm at the uncalibrated density.

How it is kept. The tool saves the match as device pixels per millimeter, together with the display readings the browser reported at the time, and compares those readings on later visits. A pixel-ratio change is corrected by arithmetic, and the readout names the correction. Any other change it detects takes the readout out of green and asks for a re-check. The browser-specific cases are below. No measurement is sent anywhere: the calibration stays in your browser, and the site's content security policy blocks connections to outside origins.

What it cannot tell you. The tool cannot see which object you are holding, and it cannot tell two screens apart when both report the same resolution and settings. Casting, mirroring and magnifiers change the size of the final image without changing any reading the page takes. All of these fall outside the calibration's error band.

What this site claims

The calibrated ruler makes four claims.

A green reading applies your match. It means the saved record passed every state check, and a centimeter drawn on screen is a centimeter to within the band printed beside it, as long as you matched the right reference on this same display.
The band is worked out for the reference and edge you used. It adds the object's permitted dimensional spread to a fixed alignment allowance and covers those two sources of error only.
Closely spaced marks are omitted. A graduation level whose pitch falls below three device pixels is dropped. The readout names the omitted level and the density needed to show it.
Your measurements stay on your device. Your calibration is held in one record in your own browser's storage, including the density, reference and display readings. There are no accounts or uploads, and the site's cookieless page counter never reads the record.

How a calibration is taken

Use a bank card, ID card or driving license in ISO/IEC 7810 format ID-1, with nominal dimensions of 85.60 by 53.98 mm. Use one that is clean, flat, undamaged and smooth-edged; hold it just above the display without touching it, do not press or slide it on the glass, and look square-on while you drag the outline until the edges agree. Any extra gap or tilt is outside the printed band. For the long edge, the tool divides the matched width in CSS pixels by 85.60 mm to get pixels per millimeter. The outline travels from 129 to 1369 CSS pixels across the long edge, which is 1.5 to 16 pixels per millimeter: enough to cover a living-room television used as a browser at one end and a 13-inch 4K laptop at 100% scaling at the other. A match left at either end of that travel is refused, because the control has run out of room and the real edge could lie farther out.

The control chooses a step small enough to stay within 0.35% of the outline being matched. For a card near the uncalibrated desktop scale, a single-pixel step is 0.31%. A shorter reference needs a finer step to provide the same relative adjustment.

The edge and orientation are separate choices. The edge determines the reference length and error band; the orientation determines whether the outline runs across or down the screen. Changing direction does not change the calculation. On a phone, the 85.60 mm long edge may fit down the portrait screen even when it is too wide to fit across it.

The band, in full

The calibration band uses this formula:

band% = object tolerance% + (0.7 mm / reference length in mm) × 100

The first term is the dimensional tolerance recorded from the object's standard. The second is an assumed alignment allowance of 0.7 mm across the pair of edges you line up, the same for every reference, so two references differ only by their lengths and their published tolerances. The terms are added, not combined in quadrature, so the band allows for both errors pushing in the same direction. The result is rounded up for display. The standard's tolerance describes a conforming object under the standard's conditions, and a worn card or sheet can fall outside it.

Reference and edge	Nominal	Object tolerance	Alignment slop	Printed band
Bank card, long edge	85.60 mm	0.351%	0.818%	±1.2%
Bank card, short edge	53.98 mm	0.371%	1.297%	±1.7%
A4 sheet, long edge	297.00 mm	0.673%	0.236%	±1.0%
A4 sheet, short edge	210.00 mm	0.952%	0.333%	±1.3%

The same 0.7 mm alignment allowance is 1.59 times larger as a share of the card's 53.98 mm short edge than of its 85.60 mm long edge. That makes the printed short-edge band half a percentage point wider. Use the long edge when it fits; use the short edge with the wider band when it does not.

What an uncalibrated screen shows

CSS defines an inch as exactly 96 pixels, and a millimeter as 96/25.4 = 3.7795 pixels. Browsers use those figures as fixed layout conversions and do not check them against the screen. Before calibration, this ruler draws at that 96 pixels per inch, so its real size depends on the panel's density and on the scaling between CSS pixels and the display. The table below works that size out for seven displays from the panel density and pixel ratio in each row.

Display	Panel density	Pixel ratio	CSS px per real inch	A CSS inch renders as
24-inch 1920×1080	91.8 ppi	1	91.8	1.046 in (+4.6%)
27-inch 2560×1440	108.8 ppi	1	108.8	0.88 in (−12%)
MacBook Pro 14-inch, default scaling	254 ppi	2	127	0.76 in (−24%)
13.3-inch 1920×1080 at 100%	166 ppi	1	166	0.58 in (−42%)
iPhone 15	460 ppi	3	153	0.63 in (−37%)
Pixel 7	416 ppi	2.625	158.5	0.61 in (−39%)
15.6-inch 4K at 100%	282 ppi	1	282	0.34 in (−66%)

The examples run both larger and smaller than physical size, so no single correction factor fixes them. A ruler drawn at 96 pixels per inch looks just like a calibrated one, which is why the ruler prints its calibration state beside every reading. The uncalibrated state is neutral: it means no match has been made yet. Keep hard or abrasive objects away from the display; use the printable ruler or another physical ruler for those jobs.

Measuring a distance within the page itself is a different job. The CSS pixel ruler measures layout distances, such as the span between its handles, in CSS pixels and uses no physical calibration.

How a calibration is kept

The ruler draws in CSS pixels, but it stores the matched density in device pixels per millimeter and converts with the browser's reported pixel ratio:

D_phys = ppmm_css × devicePixelRatio, and on any later visit ppmm_css = D_phys / devicePixelRatio.

Page zoom in Chrome, Edge and Firefox changes that ratio, and so does an operating-system display-scaling change on Windows and GNOME. Dividing the stored density by the current ratio gives the right CSS scale, so the calibration survives both. That was measured: one calibration, four operating-system display scales, and a 100 mm span drawn as 378.5 device pixels at every one of them. Safari page zoom and macOS scaled resolutions work differently and get separate checks, described below.

The saved display snapshot contains the pixel ratio, reported native pixel count, browser chrome width and height, viewport width in device pixels, layout scale, pinch scale, the measured width of a CSS inch, and, when available, the ratio of pointer travel on the screen to travel within the page. The privacy policy names all ten of those fields individually and lists the rest of the calibration record. The page compares these readings on each visit, whenever the browser fires a relevant event, and on a timer. The timer is there because a pixel-ratio change can happen without any event: in testing, the ratio went from 1 to 3 and back with the viewport fixed, and the browser fired nothing.

The stored record has checks of its own:

Matching and storage use different ranges. The matching control covers 1.5 to 16 CSS pixels per millimeter. The stored density is checked against 1.2 to 28 device pixels per millimeter, roughly 30 to 711 pixels per inch. Because the stored range is the wider one, a good calibration read back at a different pixel ratio is kept even when its CSS-pixel value falls outside the matching range.
An invalid record is discarded, and the page says so. A record that fails its schema version or its value bounds is thrown away and not repaired. The ruler goes back to the uncalibrated 96-pixel assumption and explains why the saved match was dropped.
A reading the page could not take is stored as missing. An unavailable CSS-inch or pointer measurement is compared as absent. If it were stored as zero, it would look like a large disagreement with the display and drop a valid calibration out of green for no reason. A missing reading also cannot confirm that the display stayed the same.

Once a calibration is more than about six months old, the readout leaves green and asks for another twenty seconds with a card. The match may still be right; after that long, the page has no way to know what has changed.

Where those claims stop

These are the limits of the ruler, its saved ratio and its printed band.

Moving to another monitor

This page cannot detect a move to a different monitor of the same resolution. Take a 1080p calibration from a 24-inch panel to a 32-inch one: the reported resolution, pixel ratio, window, browser chrome and CSS-inch width all stay the same, while the glass is a third larger. The ruler is then 33% wrong, in green, with a ±1.2% band printed beside it. After any move to another screen, calibrate again, even though nothing on the page has asked you to. This case is also why a calibration is re-offered after about six months.

Casting, mirroring and magnifiers

A tab cast or mirrored to a television or projector is rescaled after the browser has drawn it, and an operating-system magnifier, such as the accessibility zoom on macOS and Windows, enlarges the final image the same way. None of the page's own readings change, so the readout stays as it was. Use the ruler on the display where you matched the reference, without casting or magnification. If the saved record is copied to another machine, match again there: the page notices that move only when the new display reports different readings.

Marks this ruler will not draw

Each graduation level needs a pitch of at least three device pixels, center to center: one pixel of line and two pixels of clear gap. Closer than that, the marks blur together or alias into a flat gray band with nothing left to count, so the ruler leaves that level out. The threshold depends only on the calibrated density.

Graduation	Step	Device px per mm needed	Density needed	24-inch 1080p (3.61)	27-inch 1440p (4.28)
Millimeters	1 mm	3.0	76.2 ppi	yes	yes
Sixteenths of an inch	1.5875 mm	1.9	48.0 ppi	yes	yes
Thirty-seconds of an inch	0.79375 mm	3.8	96.0 ppi	no	yes
Half-millimeters	0.5 mm	6.0	152.4 ppi	no	no

At the densities shown, neither display meets this ruler's threshold for half-millimeter marks. The lower-density example also omits thirty-second-inch marks. When a level is dropped, the page names it and prints the density that would have kept it, rounded up to the next whole pixel per inch so the printed figure is never below the real threshold.

Which reference objects are offered

The site's data holds sixteen reference objects. One is offered as a primary reference: the ID-1 bank card. A sheet of A4 is offered as a cross-check. The other fourteen are refused by name, each with its reason.

An object becomes selectable only after its dimensional tolerance has been read from a published standard and recorded with the clause, the derivation, at least two independent sources and a verification date. When the engine loads the data, it checks that those fields are present and withholds any row that lacks one. The fields are required because a tolerance guessed too small would understate the band while every other step of the arithmetic still came out right. Banknotes, US Letter sheets and coins are in that position today: their nominal sizes are listed below, but no tolerance has been recorded for them, so they cannot be selected.

Two withheld objects do have recorded tolerances. The older-style ID-2 card would give a narrower modeled band, about ±1.0% against the ID-1 card's ±1.2%, but the two formats are easy to confuse. The ID-2 long edge is 22.7% longer than the ID-1 edge, so choosing the wrong format would put the ruler about 23% out before alignment even enters. A mini-SIM would have a ±3.6% band and can be confused with a nano-SIM, a different form factor with a 103% long-edge size difference relative to the smaller card. The tolerance band does not cover choosing the wrong object. The outline match cannot tell the page which card you are holding.

Paper is accepted only as a cross-check: ask the tool to calibrate from paper and it refuses and asks for a card match first. A sheet's long edge keeps the relative alignment allowance small, but a sheet can bow away from a vertical screen, and that bias is systematic and falls outside the band arithmetic.

What the band does not model

A bowed or damaged card, any gap or tilt while the reference is held above the display, parallax between the cover glass and the pixels, and screen curvature all fall outside the printed band, and any of them can be larger than it. The page cannot measure any of them. Use a flat, undamaged reference, keep it off the glass and look at it square-on. If you cannot hold it steady enough to compare the edges, use a physical ruler.

Browser zoom and altered display readings

The stored-density formula corrects a scale change when the browser reports that change through the pixel ratio. Safari (WebKit, on macOS and iOS) leaves the pixel ratio unchanged on page zoom, so there the tool watches the page layout and, when available, the pointer. When those readings show a zoom, the readout leaves green, says the calibration may be stale and asks for a re-match; it does not guess a correction factor. On macOS scaled resolutions the scaling factor lands in the reported pixel count while the pixel ratio stays put, so the saved density describes the framebuffer, and a change in that count is treated as drift. Fingerprinting-resistance modes, such as those in Firefox, Tor and Brave, can alter or restrict the values these checks read. The tool does not identify those modes. It discards and reports a record that fails validation, but altered values that still pass can put the ruler at the wrong scale.

This site keeps no database of device screen sizes. Resolution and pixel ratio do not identify a panel: Android devices can share those values, and iOS Display Zoom, macOS scaled resolutions and privacy settings change what is reported. The screen diagonal you can type in only starts the outline near the expected size and gives a sanity check. It does not count as a calibration or earn a green badge; the reference still has to be matched.

The cross-check threshold is tighter than the bands

When a second, independent reference agrees with the first to within 1%, the state becomes cross-checked, the only readout state on this site that earns the word verified. Above 1%, the tool reports the disagreement and leaves the original calibration unchanged. The threshold is tighter than the two bands: a card at ±1.2% and an A4 long edge at ±1.0% can differ by 2.1%, using the unrounded model terms, with each reading inside its own band. So a disagreement above 1% means the second object did not confirm the first; it does not show that the first was wrong. Every cross-check uses the same fixed 1% threshold.

How the reference figures are kept

Each verified reference row records its date, its sources, the standard clause and the arithmetic that turns the dimensional tolerance into a percentage. Writing that arithmetic out makes a derivation checkable without fetching the standard again, and it caught a unit error in an earlier draft of the card's short-edge figure: a difference in millimeters had been entered as a percentage, which understated that tolerance by a factor of 1.85.

Each row carries its own verification date, and re-checking a row moves only that row's date. The bank card, the A4 sheet and the withheld ID-2 and mini-SIM rows carry dates and sources; the other rows do not. The sections below print every row, including the refusals and their reasons; the calibration page carries the procedure and the offered table.

The reference objects that are offered

Three things must be on record before you can select an object: its dimensional tolerance, read out of a published standard; at least two independent sources; and a verification date. The ruler checks this itself when it loads: a row switched on without that evidence moves to the withheld list with the missing item named, so editing a flag cannot add a row to this table. The source notes below, and the table on the calibration page, are generated from that data file.

For a bank card, the tolerance comes from the ID-1 row of ISO/IEC 7810:2019, Figure 1, under the heading for personalized and returned cards: 85.37 to 85.90 mm across and 53.82 to 54.18 mm tall. That row governs a card in a wallet: the standard's definition of an unused card excludes anything that has been through a personalization operation, and every bank card, ID card and license has been printed, embossed or encoded. The limits sit unevenly around the nominal size, so the figure quoted is the larger of the two excursions, because a ± sign reads as a bound.

For a sheet of A4, the tolerance comes from clause 7.1 of ISO 216:2007. The ±1.5 mm figure often quoted for A4 applies only to dimensions of 150 mm and under; both A4 edges are larger than that, so both fall in the ±2 mm band. That 2 mm is 0.673% of the 297 mm edge and 0.952% of the 210 mm edge, so each edge has its own recorded figure. Use the 297 mm edge: at ±1.0% it has the tightest band on this page.

The sheet is only ever a second opinion. Ask the tool to calibrate from paper and it refuses by name and tells you to match a card first. Paper bows away from a vertical screen, and that bias is systematic, unlike tolerance and hand slop, so it falls outside the ±1.0%. What the sheet adds is length, and the calibration page shows how the cross-check uses it.

A4 sheet, 210 × 297 mm (ISO 216 trimmed size)

Offered · verified 2026-08-09 · 3 sources · tier crosscheck

Where the tolerance comes from. ISO 216:2007(E), clause 7 "Tolerances", 7.1 "Permissible tolerances", quoted: "Unless closer tolerances are specified at the time of ordering, the permissible tolerances on the dimensions are as follows: a) for dimensions <= 150 mm: upper limit +1,5 mm / lower limit -1,5 mm; b) for dimensions > 150 mm but <= 600 mm: upper limit +2 mm / lower limit -2 mm; c) for dimensions > 600 mm: upper limit +3 mm / lower limit -3 mm." A4's nominal 210 × 297 mm is Table 1 of the same standard. Both A4 edges fall in band (b), because 210 mm and 297 mm are each greater than 150 mm and not greater than 600 mm; band (a) governs A5's 148 mm edge and smaller. The envelope is symmetric, so the bound is the excursion: long edge 2/297 = 0.67340% -> 0.673%; short edge 2/210 = 0.95238% -> 0.952%. Clause 7.1 calls this the range outside which a sheet cannot be regarded as being of a given size, and says a process tolerance "is likely to be more stringent", so the figure bounds the object from above, which is the only safe direction. Clause 7.2 conditions the measurement on the ISO 187 standard atmosphere.

Limits of the band. The ±2 mm is not a manufacturing tolerance, as clause 7.1 itself says: a sheet outside it stops counting as A4. A real ream is almost certainly tighter, by an amount that cannot be known, because mill tolerances are agreed between buyer and supplier and are not published. So the band bounds what ISO 216 permits and says nothing about the particular sheet in your hand. Two further effects fall outside it. Clause 7.2 measures under a controlled atmosphere, and paper is hygroscopic: in an ordinary room a sheet changes size with the humidity, more across the grain than along it, by a further tenth to half a percent, and the figure leaves that out. And a sheet held against a vertical screen bows away from the glass, which is why paper is offered here as a second opinion and never as the calibration itself.

Bank/payment card, ID card, driving license (ISO/IEC 7810 ID-1)

Offered · verified 2026-08-09 · 5 sources · tier primary

Where the tolerance comes from. ISO/IEC 7810:2019(E), Figure 1 "Card size dimensions", block headed "Personalized card and returned card requirements", row ID-1: dimension a (width) max 85,90 / min 85,37; dimension b (height) max 54,18 / min 53,82 mm. Clause 5.2 requires the whole outline to fall between two concentric rectangles at those limits; clause 5.1 conditions them at 23 C ± 3 C and 40-60 % RH. The envelope is asymmetric about the nominal, so the bound is the larger excursion. Long edge: 85,90 - 85,60 = +0.30 mm against 85,60 - 85,37 = -0.23 mm, so 0.30/85.60 = 0.3505% -> 0.351%. Short edge: 54,18 - 53,98 = +0.20 mm against 53,98 - 53,82 = -0.16 mm, so 0.20/53.98 = 0.3705% -> 0.371%.

Limits of the band. The band is a conformance requirement: the size a card must still measure if it were returned for testing. It does not come from measuring worn cards. A card bowed past 85.90 mm can exist, and it is non-conforming, so the band describes conforming cards and makes no claim about every card in circulation. The limits are also defined only at 23 C ± 3 C and 40-60 % RH, so a card measured in a hot car or in dry winter air is outside the conditions the envelope itself assumes. Clause 5.2 sets a zone between two concentric rectangles that the whole outline must sit inside, and it gives no per-edge ± for a single measurement. That is why the band spans the full range between the endpoints, and why uneven limits are bounded by their larger excursion.

Verified, and still withheld

These two were measured against the same standard, and their numbers are sound. They are withheld because a reference can fail in a way no error band covers: it can be confused with a similar object. A tolerance error can be no larger than the figure printed beside it. A mix-up has no such limit, and it shows the same tight band as a correct pick.

ISO/IEC 7810 ID-000 (mini-SIM / 2FF plug-in card)

Verified 2026-08-09 · withheld · would print ±3.6% on the long edge, ±5.8% on the short edge

Why it is not offered. Its numbers are verified, but the band this object earns is ±3.6% on the long edge and ±5.8% on the short, too wide to count as a calibration. The SIM a user is holding is almost certainly a nano-SIM (12.3 × 8.8 mm, an ETSI form factor with no published ISO tolerance), so picking this row for it is 103% wrong. The bevel the standard mandates on one corner also leaves one edge without a clean straight line to match.

Where its tolerance comes from. ISO/IEC 7810:2019(E) Figure 1, "Personalized card and returned card requirements", row ID-000: a max 25,20 / min 24,80; b max 15,15 / min 14,84 mm. The bound is the larger excursion from nominal: long ±0.20 -> 0.20/25 = 0.800%; short +0.15 against -0.16 -> 0.16/15 = 1.0667% -> 1.067%. Clause 5.2 also mandates a 3 ± 0,1 mm 45 degree ± 2 degree bevel on one corner.

What is still missing. ETSI TS 102 221 would be the proper second numeric source for SIM geometry. It was not read, because etsi.org returned HTTP 403, so nothing here rests on it.

ISO/IEC 7810 ID-2 card (older-style national ID card, visa sticker)

Verified 2026-08-09 · withheld · would print ±1.0% on the long edge, ±1.4% on the short edge

Why it is not offered. Its tolerance is verified, but ID-1 and ID-2 are both "a card from my wallet", so picking the wrong one calibrates 22.7% wrong while every on-screen signal looks like a correct match. That mix-up has no bound, and it would still be shown under a ±1.0% band. ID-2 is also a legacy-only format, and the commonest ID-2 object is a visa sticker in a passport, which cannot be laid flat against a screen.

Where its tolerance comes from. ISO/IEC 7810:2019(E) Figure 1, "Personalized card and returned card requirements", row ID-2: a max 105,30 / min 104,80; b max 74,30 / min 73,70 mm. The clause 5.2 concentric-rectangles rule and the clause 5.1 environment are the same as for ID-1. The bound is the larger excursion from nominal: long +0.30 against -0.20 -> 0.30/105 = 0.2857% -> 0.286%; short ±0.30 -> 0.30/74 = 0.4054% -> 0.406%.

What is still missing. The availability claims (Germany 2010, Romania 2021, Romanian ID-2 valid for travel to 2031) rest on a single encyclopaedia source. They would need a government registry behind them before the site relies on that reasoning.

Carried, with sizes but no verified tolerance

The nominal sizes of these objects are not in dispute, and they are printed below. The tolerance column is empty because the data holds no tolerance for these rows. Several of them once had a working figure, and it was dropped, because a small assumed tolerance understates the band while looking exactly like a measured one. This site treats an understated band as a false statement.

Object carried	Nominal	Verified tolerance	Why it is not offered
1 euro coin	23.25 mm	none recorded	Tolerance unverified. Verifying it needs an EU coin specification plus a second source.
2 euro coin	25.75 mm	none recorded	Tolerance unverified. Verifying it needs an EU coin specification plus a second source.
10 euro note	127.00 × 67.00 mm	none recorded	Tolerance unverified. Verifying it needs an ECB technical specification plus a second source.
20 euro note	133.00 × 72.00 mm	none recorded	Tolerance unverified. Verifying it needs an ECB technical specification plus a second source.
5 euro note	120.00 × 62.00 mm	none recorded	Tolerance unverified. Verifying it needs an ECB technical specification plus a second source. Euro notes differ in size by denomination, so each denomination has its own row, and no row covers them all.
50 euro note	140.00 × 77.00 mm	none recorded	Tolerance unverified. Verifying it needs an ECB technical specification plus a second source.
US Letter sheet	279.40 × 215.90 mm	none recorded	Tolerance unverified, because no size-defining standard states one. ASME Y14.1 defines the 8.5 × 11 in sheet and is silent on tolerance; ASTM D3460, the cut-size office paper specification, was withdrawn in 2014 with no replacement; and the only live clause found is a government procurement trim tolerance, which does not govern retail paper and has no independent corroboration. A4 is offered instead, and at 297 mm it is the better cross-check.
US dime	17.91 mm	none recorded	Tolerance unverified. Verifying it needs a US Mint specification plus a second source.
US nickel	21.21 mm	none recorded	Tolerance unverified. Verifying it needs a US Mint specification plus a second source.
US penny	19.05 mm	none recorded	Tolerance unverified. Verifying it needs a US Mint specification plus a second source.
US quarter	24.26 mm	none recorded	Tolerance unverified. Verifying it needs a US Mint specification plus a second source. The quarter is also a last-resort reference by design: at 24.26 mm the alignment slop alone is 2.9%.
US banknote (any denomination)	155.96 × 66.29 mm	none recorded	Tolerance unverified. An early prototype used 0.10% as a placeholder, because §6.4 of the spec says "tight; worn notes bias" and gives no number, and the addendum blocked the row over that placeholder. A small assumed figure understates the band, which §4.1 forbids, so the site's data does not carry the number at all. The row needs a BEP or Federal Reserve dimensional specification, and a second source, before it can be offered.

The screen-size shortcut, and why it never earns a badge

If you know your screen's diagonal, the ruler can use it to move the outline to roughly where it belongs, so there is less dragging. The arithmetic multiplies the reported resolution by the pixel ratio, takes the hypotenuse and divides by the inches to get a pixel density. Marketed diagonals are usually within about half a percent of the truth, so the answer often lands within a percent or two of a card match.

It is not a calibration and the tool will not let it become one. The arithmetic assumes the pixel count your browser reports is the panel's native one. That is false under macOS scaled resolutions, where the framebuffer is not the panel; in Safari under page zoom, which changes the size of a CSS pixel without reporting it; and wherever a fingerprinting-resistance mode rewrites those values on purpose. Once a card has been matched, the page prints the two figures side by side as a sanity check that changes nothing, in this form: your card says 4.02 px/mm; the screen size you gave implies 3.96, which is 1.5% apart. Try it on a 13.3-inch screen, a 15.6-inch one or a 27-inch monitor.

Questions

Why does an on-screen ruler need calibrating at all?

Because CSS fixes an inch at 96 CSS pixels and the browser cannot measure the screen it draws on, only a match against a real object can set the ruler's physical scale. In the seven calculated examples on this page, a CSS inch ranges from 66 percent short to 4.6 percent long, and the six that come out short are typically 25 to 40 percent short: the median and the mean of those six both fall in that range. Each example is worked out from the density and pixel ratio in its row of the table.

Where does the plus or minus 1.2 percent come from?

From two terms added together. The card's permitted spread is 0.351 percent of its 85.60 mm long edge, taken from the personalized and returned card row of ISO/IEC 7810:2019. Lining up two edges by eye gets an assumed allowance of 0.7 mm across the pair, which is 0.818 percent of the same edge. Adding them allows for both errors pushing in the same direction, and the sum is rounded up to ±1.2 percent. A damaged card, a gap or a tilt adds error outside that figure.

What happens to my calibration when I change my screen?

Zoom and display-scaling changes are corrected or caught; a move to another monitor of the same resolution is not. In Chrome, Edge and Firefox, page zoom moves the browser's pixel ratio, as does a display-scaling change on Windows or GNOME, and the saved density divides that back out, so the calibration survives and the readout names the correction. Safari does not move the pixel ratio for page zoom, and a macOS scaled resolution changes the reported pixel count instead, so in those cases the readout leaves green and asks you to match the card again, which takes about twenty seconds. On iOS Safari this was measured on 2026-08-09: a page zoom took the readout out of green, and a pinch showed that nothing on screen is to scale while it is held. A monitor swap at the same resolution changes none of the readings, so a 1080p calibration carried from a 24-inch panel to a 32-inch one draws 33 percent too large while the readout stays green. Whenever you switch screens, match the card again.

Why will this ruler not draw half-millimeter marks on my monitor?

Because the monitor has too few pixels per millimeter to draw them apart. The ruler draws a graduation level only when its marks sit at least three device pixels apart, center to center, so half-millimeter marks need a calibrated density of 6.0 device pixels per millimeter, or 152.4 pixels per inch. The table's 24-inch 1080p and 27-inch 1440p examples have 3.61 and 4.28 device pixels per millimeter. Closer marks would blur together, so the ruler leaves the level out and names it in the readout.

Why is only one primary reference offered?

Because the bank card is the only primary reference that has a verified tolerance and is not withheld. An offered reference needs a recorded dimensional tolerance, its derivation, at least two independent sources and a verification date. A4 paper meets that too, and the tool accepts it only to check a card match. Of the other fourteen objects, the ID-2 and mini-SIM cards also meet it but stay withheld, because each is easy to confuse with another card format. The rest have no verified tolerance, and the tool refuses each unavailable object by name.

Ruler calibration, error bands and limits