The reference
Everything the app assumes you know: how to photograph a page so the analysis works, what each score actually measures, how to sit and hold the pen, and what to do when a number refuses to move.
1 · Photographing a page
The analyzer finds lines of ink, fits a baseline to each, and measures how those lines vary. It is looking at geometry, not meaning — so the photo has to preserve geometry.
- Straight on, not at an angle. A tilted photo shears every letter, which the analyzer reads as slant. Hold the camera parallel to the page, centred.
- Flat paper. A curved page bends the baselines. Flatten it, or use the Drive scanner app, which de-warps for you.
- Even light, no hard shadow. Your own shadow across the page turns into fake ink.
- Dark ink on light paper. Pencil works if it is dark enough; very light pencil does not.
- Crop to the writing. Desk, hands and margins are noise. If the page has printed guide lines and they photograph dark, the analyzer may score them instead of you.
- Two lines minimum. Slant steadiness and size consistency compare lines to each other, so a single line cannot produce them.
The single biggest mistake
Photographing at an angle and then wondering why the slant score dropped. If a score moves sharply for no reason you can feel in your hand, re-shoot the page before believing it.
1b · Scan batches and duplicates
The Google Drive scanner puts a whole stack of pages into one PDF, so a scan is routinely forty pages, not one. Open a PDF — from Drive, or straight off your phone with the file picker; no Google account needed either way — and you get a gallery of every page in it.
- Every page has a thumbnail, so you can find the three you have not logged yet without stepping through a dropdown one render at a time.
- Pages you have already saved are dimmed and marked. This is tracked per page, not per file, so a forty-page batch with one page logged does not read as finished.
- Duplicates are flagged in amber — either a page that repeats an earlier page of the same batch (scanners double-feed) or one matching a page already in your history.
- Import all new analyzes and files every page that is neither saved nor a duplicate, using your current hand, style and alphabet settings. You can stop it part-way.
- It opens instantly. Previews are rendered only as you scroll to them, and kept afterwards, so a forty-page batch opens in a few milliseconds instead of grinding through every page first — and opening it again later costs nothing at all. Importing is the one action that does have to look at every page, and it says which one it is on.
How a duplicate is recognised
Each page gets a 256-bit fingerprint of its layout — the page is shrunk to a 16×16 grid and what is recorded is which cells are darker than their neighbour. Lighting and small shifts barely change it, so the same sheet scanned twice still matches, while two different pages are far apart. The thresholds were measured on real page shapes rather than guessed: a re-scan of one sheet scores around 13 bits different, two different pages around 70, and anything at or under 32 is called a duplicate.
It is a warning, never a block. If you really did write the same drill twice and want both logged, save it anyway.
The fingerprint is taken once, at a fixed size, and kept with the page's preview. That matters more than it sounds: the same page fingerprinted from a small preview and from a full-size render came out 26 bits apart, most of a 32-bit budget, which would have hidden genuine re-scans. Measuring once and reusing the answer removes the problem rather than papering over it.
2 · The anatomy of a written line
Every term the app uses comes from this diagram. The four horizontal lines are what the printable guide sheets draw.
- Baseline — the line letters sit on.
h,a,ntouch it;g,y,phang below it. - x-height — the body height. Adult handwriting is usually 4–8 mm; children's and new off-hand writing starts far bigger, and should.
- Ascender — the tall part of
b d f h k l t. Traditionally about 1.5× the x-height. - Descender — the tail of
g j p q y. - Slant — the angle of the vertical strokes. Anywhere from upright to about 20° forward is normal and none of them is more correct than another.
3 · What each score measures
Every score is 0–100 and every one is about consistency, not beauty. The app cannot tell you your handwriting is attractive; it can tell you it is the same on line nine as on line one, which is most of what people mean when they call handwriting neat.
| Score | What it actually measures | Raises it |
|---|---|---|
| Slant steadiness | The spread of stroke angles across the page. It does not care what your angle is, only that you keep picking the same one. | Slant rails, push-pull strokes |
| Baseline | How far the bottoms of letters scatter around the straight line fitted through them.
Descenders are trimmed out first so g and y don't count against you. |
Ruled-line discipline, long-word landings |
| Size consistency | Variation in the height of each detected line of writing. | x-height rows, boxed letters |
| Spacing | Variation in the gaps between marks. In printing it reads letter gaps; in cursive, where letters are joined, it reads the word gaps instead. Gaps are measured along your own slant, so writing at an angle doesn't cost you. | The 'o' gap, the minimum drill |
| Pen control | Variation in stroke width. Wildly varying width means pressure or speed is changing mid-stroke — usually a grip that's too tight. | Even-pressure lines, loose-grip reset |
| Legibility | Optional. Runs OCR over the page and reports its confidence. A machine reading your handwriting is a harsh, honest proxy for a stranger reading it. | One form per letter, cold read-back |
| Mirror match | Only in mirror-prime mode. How closely the copy line matches the line above it in slant, size and steadiness. | Prime and copy |
Why Overall is only four of them
Overall averages slant steadiness, baseline, size and spacing — and deliberately nothing else. Legibility needs OCR and is often absent; pen control was added later. Folding either into the average would silently re-scale every score already in your history and make the trend line a lie. So Overall means the same thing on your first page as on your hundredth.
3b · Other alphabets
Everything above is written in Latin terms, and most of it does not survive a change of script. Pick your alphabet in the Analyze tab and the app changes what it measures, what it prints, and which OCR language it reads with.
| Alphabet | Organised around | Slant scored? | Guide sheet |
|---|---|---|---|
| Latin, Cyrillic, Greek | a baseline the letters sit on | yes | ruled |
| Hebrew | a baseline; final forms hang below it | yes | ruled, right-to-left |
| Arabic | a baseline; letters join | no — no canonical angle | ruled, right-to-left |
| Devanagari | the headline the letters hang from | yes | headline |
| Thai | a baseline; marks sit above and below | yes | ruled |
| Chinese / Japanese, Korean | staying level in a cell | no | square cells with a centre cross |
What changes, concretely
- The line being measured. Latin letters sit on a baseline, so the app fits a line through the bottom of the writing. Devanagari hangs from the शिरोरेखा, so it fits the top instead — fitting the bottom would measure the ragged underside of the letters and report a page as far worse than it is. Chinese, Japanese and Korean are written in cells with no baseline at all, so the app checks that characters march level through their centres.
- Whether slant is scored. A consistent stroke angle is a real skill in Latin, Cyrillic, Greek, Hebrew, Devanagari and Thai. It is not a thing in Arabic, Chinese, Japanese or Korean, so no slant score is shown rather than a meaningless one.
- Which gaps count. In separated scripts the gaps between letters carry the information. In joined ones — Arabic, Devanagari, and Latin cursive — the pen rarely lifts, so the gaps that matter are between words. In cell-set scripts the app measures the pitch from one character's centre to the next, because the white space inside a single character says nothing about your spacing.
- The drills and their sheets. Alphabet, round-letter, chain and sentence drills print in your script, on the right guide sheet, running the right way across the page.
- OCR. The legibility check loads the language model for your script.
Scores are comparable within a script, not across them
Each alphabet is measured consistently, but the absolute numbers sit at slightly different levels — some scripts simply carry more below-the-line material than others. A Hebrew 80 and a Latin 80 are not the same achievement. Compare yourself against yourself in one script, which is what the trend chart does anyway.
3c · What the app can work out on its own
After analyzing a page the app compares what it sees against your current settings, and says so if they disagree. It never changes a setting for you — a wrong guess would put a mislabelled score in your history, which is worse than not guessing. You get a one-tap switch and the reason.
| It can tell | How | How clearly |
|---|---|---|
| Joined vs separated (cursive vs printing) |
Counts the separate blobs of ink per line. Printed letters stand alone; a cursive word is one continuous blob. | Separated scores about 1.5, joined about 0.16 — nine times apart, and steady across typefaces and sizes. |
| Mirror-prime pages | Checks whether the odd and even lines differ from each other by more than the lines within each group differ. An off-hand copy line is visibly less steady than the line it copies. | Alternating pages score ~37, a page that is simply messy throughout scores under 1. |
| A headline script (Devanagari) |
Looks for a long horizontal stroke running along the top of each line. | Devanagari's headline is unmistakable. A cursive join is also a long stroke, so the app checks it is at the top of the band and not on the baseline. |
What it deliberately does not guess
Which alphabet, within the baseline family. Latin, Cyrillic, Greek and Hebrew are the same shape of writing — letters of mixed widths sitting on a line — and nothing in the geometry separates them. Only reading the text could, and that is what the alphabet selector is for.
Chinese, Japanese and Korean. The obvious signal, how even the character widths are, measures 0.31–0.48 for those scripts and 0.48 for Latin: no separation at all. Rather than ship a coin-flip, the app stays quiet. Set those from the Alphabet menu once and it is remembered.
4 · Posture, grip and paper
Sit before you write
- Both feet flat. Back against the chair. Table at roughly elbow height.
- Both forearms on the table — the non-writing hand anchors and moves the paper. If it is in your lap, the page drifts and your baselines drift with it.
- Eyes about 30 cm from the page. Hunching closer is a sign the writing is too small.
Hold the pen loosely
The standard dynamic tripod: thumb and index finger pinch the barrel, it rests on the side of the middle finger, ring and little fingers curl under and glide on the page. Grip it just firmly enough that it doesn't fall out.
A tight grip is the single most common cause of both uneven strokes and a hand that aches after half a page. The test: if someone could pull the pen out of your hand without a struggle, you're holding it about right.
Move the right joints
Letters are made by the fingers; words are made by the wrist; lines are made by the forearm and shoulder. People who write only from their fingers get tired fast and produce cramped, shrinking lines. The warm-up drills exist to remind the big joints they are involved.
5 · Writing with the left hand
A left hand writing left-to-right is doing a genuinely harder job than a right hand: it moves into the text it just wrote, covering it and dragging through wet ink, and it pushes strokes that a right hand pulls. Almost every left-handed handwriting problem traces back to one of those two facts, and the first is fixable in about ten seconds.
Tilt the paper clockwise. That's the fix.
Rotate the page 30–45° clockwise, so the bottom-right corner points toward your belly — the mirror image of what a right-hander does. Now your hand sits below the writing line instead of on top of it. You can see what you are writing, you stop smudging, and the hook posture — wrist curled over the top of the line — becomes unnecessary.
Also: shift the paper to the left of your body's centre line, and hold the pen slightly further up the barrel than a right-hander would, so your fingers don't block your view of the nib.
The hook
Curling the wrist over the top of the line is a workaround for not being able to see past your own hand. It works, but it is slow, tiring, and hard on the wrist over years. If you already hook, re-tilting the paper removes the reason for it — but the habit will take a few weeks to fade, and it will feel wrong before it feels better.
Push versus pull
Cross-strokes and many joins are pushes for a left hand, which is why the line skips or digs in. Two mitigations: a pen that writes without pressure (rollerball, gel, fibre-tip — not a cheap ballpoint), and letting the letters be slightly more upright or even backward-leaning, which shortens the push.
Expect it to be slow
If you are training a genuinely new hand, the first weeks should look like a child's writing, because motorically that is exactly what it is. Go big — letters two to three lines tall — for the first couple of weeks. Size comes down on its own once control arrives. Trying to write small immediately is the most reliable way to build in a permanent tremor.
6 · How to practise so it sticks
- Short and daily beats long and weekly. Handwriting is motor learning. Fifteen minutes six days a week will beat two hours on a Sunday, and it isn't close. The streak counter is not a gimmick — it is the mechanism.
- Slow before fast. Speed comes back on its own once the movement is right. Practising fast and sloppy rehearses fast and sloppy.
- One target per session. Pick the weakest score and work only on that. Trying to fix slant, spacing and size in the same page usually fixes none of them.
- Warm up. Two minutes of loops and ovals before real writing measurably steadies the first lines of the page.
- Stop before the cramp. Writing through hand fatigue rehearses a tense grip.
- Photograph the same kind of page. Comparing a careful drill sheet against a rushed shopping list tells you about the situation, not about you. Keep one steady benchmark.
Expect a plateau, and a dip
Scores rise fast for two or three weeks, then flatten. That is normal — the easy structural wins are used up and what remains is slower motor consolidation. A dip often follows a change you deliberately made (a new slant, a smaller size), and it usually recovers past the old level within a fortnight. Judge by the three-session average line, never by one page.
7 · Troubleshooting: symptom → fix
| What you see | Usual cause | Do this |
|---|---|---|
| Lines sag toward the right margin | Not moving the paper up as you go; writing from the fingers only | Ruled-line discipline; move the page with your other hand |
| Letters shrink across the page | Grip tightening as you tire | Loose-grip reset; shorter sessions |
| Slant score low but writing looks fine | You have two slants — usually one for slow careful words and one for fast ones | Slant rails; deliberate slant switching |
| Spacing score low in cursive | Word gaps too small — the words run together | The 'o' gap |
| Spacing score low in print | Uneven letter gaps, often after tall letters | The minimum drill |
| Pen control low | Death grip, or a ballpoint needing pressure | Loose-grip reset; switch to a gel or rollerball |
| Legibility low, everything else high | Consistent but ambiguous letterforms — a/o, n/u, r/v collapsing | One form per letter; cold read-back |
| Hand aches after half a page | Grip and finger-only movement | Air writing; loose-grip reset; warm up first |
| Left hand smudges everything | Paper not tilted; overwriter posture | Paper tilt setup; quick-drying pen |
| Scores jump around with no pattern | Inconsistent photos, not inconsistent writing | Re-read section 1 |
8 · Mirror priming
Mirror priming is the one technique here that is specific to training a second hand. You write a line with your dominant hand, then copy it immediately with the other hand on the line directly below.
The point is that your dominant hand already knows what the movement should feel like, at what size and rhythm. Writing it first loads that target into working memory, and the copy has something concrete to imitate rather than a vague intention. The Mirror match score is how close the copy came.
To make it work: alternate lines strictly — dominant, copy, dominant, copy — because the analyzer assumes that pattern. Write the dominant line slowly enough that the other hand can plausibly match its pace. And photograph the whole page in Mirror prime mode so both hands get scored from the same image.
9 · Questions
Does the app upload my pages anywhere?
No. All analysis runs in your browser. Nothing about a photo leaves the device — the image is never stored or transmitted. If you connect Google Drive, only your scores sync, into a private folder that only this app can see.
What are projects for?
A page of practice and a page of a book you are copying out are not the same thing, and averaging them together hides both. Filing pages under a project — a book, a journal, a course workbook — lets you see how a particular body of work is going, track pages against a goal, and filter the trend chart to just that material. Pages you do not file stay under Unfiled, which is a perfectly good place for warm-ups and one-offs. Deleting a project never deletes its pages; they simply become unfiled.
Where do the "Then and now" pictures come from?
When you save a page, a small thumbnail of it is kept so you can look back at your actual handwriting rather than only a number. Those images stay on the device that made them — they are never included in the Drive sync, which carries scores only. The newest 40 are kept and older ones drop off, so this can't quietly fill up your browser storage.
Why did OCR need the internet?
The OCR engine is a few megabytes and downloads on first use, then caches. Everything else works offline, including the drills and the printable sheets.
Can I use lined paper?
Yes, and you should while training baselines. Very dark printed lines can occasionally be detected as writing, so if the scores look strange on a heavily-ruled pad, try the app's own guide sheets, which are printed pale on purpose.
What pen should I use?
Anything that writes without pressure. A gel pen, rollerball or fibre-tip forces a lighter grip than a cheap ballpoint, which needs force to lay ink down and encourages exactly the tension you are trying to remove. This matters more for left hands, which push strokes.
How long until this works?
Legible improvement in an existing hand: two to four weeks of daily practice. A genuinely new off-hand: expect two months before it is comfortable, and to keep improving for a year. The eight-week left-hand program is built around that, not around a promise of a fortnight.