Anchoring

A zone you draw on a sample page is a fixed rectangle. Real scans rarely line up that precisely — the paper feeds in a few millimetres high, the layout shifts between document variants, or the page comes through skewed. Anchoring solves this by tying a zone to a visible reference on the page, so it follows the content instead of staying put.

Why a fixed box drifts

When you draw a zone in the Designer, Scanix remembers where it sits on the sample page. On every later document it reads that same rectangle. That works perfectly when each page is an exact copy of your sample — and almost never is.

A scanner that grabs the next page a little lower, a supplier whose invoice number sits two lines further down, a sheet that went through the feeder at a slight angle: any of these moves the value out of the box. The zone still reads its fixed rectangle faithfully, but now the rectangle covers the wrong text — a neighbouring label, a blank margin, or nothing at all. The fix isn't a bigger box (that just captures extra noise); it's giving the zone something on the page to hold onto.

That something is an anchor: a landmark Scanix can recognise on each page, near the value you actually want. Instead of "read the rectangle at this position," the instruction becomes "find this landmark, then read the value in its usual place relative to it." When the page shifts, the landmark shifts with it, and the zone goes along for the ride.

What a zone can anchor to

You set anchoring per zone, on its Anchoring tab, by ticking Enable. Each anchor has an Anchor target — the kind of landmark Scanix looks for:

For a text anchor you also give an Anchor value — the exact label text to find, such as "Invoice No." — and a Search in region: a box on the sample where Scanix should look first. Keeping that region tight (the header strip, say, rather than the whole page) makes matching faster and less ambiguous.

Matching the way OCR really reads

Scanned text is never perfectly clean, so anchoring is built to tolerate the kinds of mistakes OCR makes. Two options, both on by default, do most of the work:

• Fuzzy match lets a label still resolve when OCR garbles a character or two. It's deliberately OCR-aware: commonly confused characters are treated as free swaps, so a label like "INVOICE NO" still matches when the scan reads "lNVOlCE NO."
• Case-insensitive ignores letter case, so "Invoice No." and "INVOICE NO." are the same landmark.

Multi-word labels are handled for you. Even though OCR splits a heading like "Invoice No." into separate words, Scanix re-assembles adjacent words on the same line into phrases before matching — so a label that reads as several tokens still resolves as one.

When the landmark isn't where you drew it

The Search in region is where Scanix looks first. If the label isn't there — because the page shifted more than expected, or a variant moved the header — turn on Search whole page if not found. Scanix then widens the hunt across the entire page. Wider matches are penalised, so a strong, close-by match still wins over a faint one far away; you get the safety net without it hijacking a good local result. This is the option to reach for when a template works on most documents but occasionally comes back empty on an outlier.

More anchors mean better correction

A single anchor can do one thing: shift the zone so it follows its landmark from side to side and up and down. That handles plain drift — the page fed in high or the field moved down a couple of lines.

The moment two or more anchors resolve, Scanix does something stronger. It compares where the landmarks sit now against where they sat on your sample and works out the page's overall transformation — not just how far it moved, but how much it grew or shrank and how far it rotated. It then applies that same correction to the zone. So a page that came through the feeder at an angle, or scanned at a slightly different size, still lands the zone exactly on the value.

This is why adding anchors is always at least as good as having fewer: with one, you get position correction; with several, you get position, scale, and rotation, triangulated automatically from whichever ones resolve. You don't choose between these modes — Scanix picks the strongest correction the resolved anchors support.

You add extra references with Add anchor under Additional anchors. A practical pattern is one landmark near the top of the page and one near the bottom, so the correction is anchored across the whole sheet rather than one corner of it.

Zones whose height changes

Some values don't sit at a fixed spot because what's above them grows or shrinks — a line-items table that runs long on one invoice and short on the next, pushing the total further down. For these, Dual anchoring (Top + Bottom) lets a zone hold an independent top anchor and a bottom anchor, so its top and bottom edges each track their own landmark and the zone stretches or contracts to fit. Reach for it when a field's vertical position depends on how much content precedes it.

AI-suggested anchors versus your own

When you build a template with Smart Template, the AI proposes an anchor for each zone alongside the zones themselves. On the Anchoring tab these show up under an AI-suggested anchor banner with Accept and Reject:

• Accept keeps the AI's suggestion as-is.
• Reject discards anchoring for that zone, which then falls back to its fixed box.
• Editing any part of a suggested anchor automatically confirms it as yours — the anchor's source flips from AI-suggested to operator, your stamp that you've reviewed it.

Treat the AI's suggestions as a strong first draft. Preview each one on your sample, and Accept or adjust rather than assuming it's right — a landmark that resolves on the sample but is too generic (a word that repeats elsewhere on the page) is exactly the kind of thing a quick preview catches.

What happens when an anchor can't be found

Even a good anchor occasionally fails to resolve — a torn corner, a missing barcode, a label that didn't print. You decide in advance what the zone should do in that case, with Failure behavior:

• Get value from default area — fall back to the zone's original fixed box. This is the default, and the safest choice for high-volume runs where you'd rather get a best-effort read than a gap.
• Leave unpopulated — capture nothing, leaving the field empty.
• Surface error for operator review — flag the document so a person checks it before it exports.

The Preview anchor on sample result is colour-coded so you can judge confidence at a glance: green for a high-confidence match, amber for a medium one, and red for a match that falls below the threshold. A green preview before you save is the best sign an anchor will hold up at run time.

How anchoring fits the bigger picture

Anchoring is one of several ways a template stays accurate on messy input. Cleaning up the page first — deskewing, cropping, and so on in the Designer — reduces how much drift there is to correct. Zone settings decide how each captured value is read and validated once the zone is in the right place. And when pages come out wrong at the source — feeding crooked, scanning at the wrong size — scanning issues is where to start, since a steadier scan makes every anchor's job easier.

Next steps

Related

• The Designer
• Zone settings
• Scanning issues