Split a Filled Package into Sub-Forms
Cuts a completed package into one piece per form, following an assignment of source pages to forms, so each prior form can be pointed at the matching section of the new package.
Overview
Almost every team that produces multi-form deliverables has a drawer full of finished ones. Last quarter's appraisal report. The bid that won. Those documents are the clearest statement of house style — how your people phrase a narrative, which optional fields they actually complete, what they leave blank on purpose.
The problem is that a finished package arrives as one flat PDF: forty pages of main form, addenda, schedules and exhibits, with no marker saying where one form ends and the next begins. That matters because a section fills better from a small, relevant set of inputs than from the whole package at once.
This tool cuts the package apart. Given an assignment of source pages to forms, it copies each stretch of pages into a piece of its own and labels that piece with the form it belongs to, so the piece can be used as the filling reference for the matching section of the new package. Photo pages and the sketch page produce a second, picture-only piece as well.
What It Does
| Capability | What it means for you |
|---|---|
| Page-level slicing | The pages assigned to a form are copied into a piece of their own, and the source page numbers are kept with it |
| Per-form pieces | Each piece is labelled with the form it came from, so it can be routed to that form's section instead of to everything |
| Section-level references | A section is filled against its counterpart from the prior package rather than against all forty pages |
| Derived picture pieces | Photo pages and the sketch page yield a picture-only piece alongside the sliced form |
| Independent of who made the PDF | Slicing reads page numbers, not content, so a package produced by hand or by other software is cut the same way |
When the Agent Chooses This Tool
Only the appraisal workflow can call this. A pre-bid run has a cutting step of its own, which breaks an uploaded invitation into page-level pieces; a coating-inspection run has no cutting step at all. Neither produces per-form pieces.
Within an appraisal run, nothing about a document's contents triggers the split. A file becomes a "completed prior package" at upload, when the router that classifies incoming files reads its name; a file whose name the router does not recognise is classified from the text of its first pages instead. By the time the agent starts, that decision has already been made, and the workflow's rule is simply: for every attachment carrying that classification, split it.
The rules that follow from it:
- Read before cutting. The workflow inspects the upload and maps each page of the source to a form from the registered list. What it has to work with is a short sample of text taken from each page.
- Once per completed package. The workflow issues one split per such upload.
- Split first, then build. Splitting runs before the new package's sections are created, and the pieces it produces are what the later steps route.
- The original is barred from everything downstream. The workflow is told the completed package itself must never be used as a filling source and never as an image source. Only the pieces cut out of it may be used — and those are filling sources for text only, never image sources.
- Pictures take the other route. The picture-only pieces are the ones that go into image positions, via Place an Image in a Slot.
- The pieces are used on the run that made them. On later runs of the same package the workflow is told not to feed them to any section again.
Once the pieces exist, each section is filled against its counterpart rather than against the whole prior package — see Fill Sections from Source Documents. Where a form has fields the prior package is authoritative for, filling can be restricted to exactly those fields.
Inputs and Outputs
Takes: a completed package that is already an attachment on the package being built, plus an assignment of source pages to forms — which pages belong to the main form, which to the first addendum, and so on. Pages are numbered from 1, and a page number outside the document is dropped.
Returns: one identifier per form that produced a piece. The picture-only pieces are created at the same time but are not part of what comes back.
Example
An appraiser starts a new report for a property their office appraised three years ago, and uploads the completed report from that job alongside this run's files.
The workflow reads a per-page sample of the old report's text and works out the structure: the first pages are the main report form, then a run of photo pages, then the sketch, then correspondence. It assigns those page ranges to forms and issues one split.
Each form named in the assignment comes back as its own piece, labelled with that form. The photo pages also produce a picture-only piece, and the sketch page produces one of its own. The new package's sections are then created and filled: the new main form is filled against the old main form, not against the whole old report, and the picture pieces are placed into image positions rather than used as text.
The correspondence at the back was never assigned to a form, so it is not cut out at all — it stays part of the original upload and plays no part in the split.
Limits
- Splitting the same package twice produces a second full set of pieces. Nothing is replaced and nothing is matched against what already exists, so a re-run leaves duplicates behind.
- A piece that cannot be produced is left out and the split still reports success. As long as one piece was built, the run continues as though the whole split had worked.
- The page-to-form assignment rests on text the pages actually carry. A package with no text layer — a scan, a photographed set, a fax — gives the step that reads it nothing to work from.
- Individual photographs are pulled out of one kind of page only, and at most three of them, matched to positions by the order they are stored in the file rather than by what they show. Every picture stored in those pages counts, including a logo or a rule repeated on each page, so repeated furniture can take the place of real photographs.
- The photo-addendum pages and the sketch page are not separated into individual pictures — each page crosses over as one image. On the addendum pages that image is a fixed region of the page, sized for one specific form layout. On the sketch page it is the region enclosed by the outline detected on that page.
- Splitting produces references, not filled sections. Nothing is written into the new package until filling runs.
- Pages that are not assigned to a form are not cut out. Only the pages named in the assignment become pieces.
- What comes out reflects what went in. A prior package that was filled inconsistently, or filled wrongly, is still the example the new one is written against. A prior example is a style guide, not a fact-checker.
FAQ
Can I use a report we already finished as a template for new ones?
Yes — that is what this is for. The completed package is cut into per-form pieces, and each new section is filled against its counterpart rather than against the whole prior document.
Does the prior package have to come from Instafill?
No. The cutting itself works on page numbers, so a package filled by hand, produced by another vendor or exported from software you no longer run is sliced the same way. What does depend on the file is the step before: it is treated as a completed prior package because of how the upload was named, or failing that because of what its first pages say.
What happens to photos inside the old package?
It depends on which page they are on. From one kind of photo page, up to three pictures are pulled out as separate image files, in the order they are stored. The photo-addendum pages and the sketch page are handled differently: each page becomes a single image rather than being separated into individual pictures — a fixed region of the page for the addendum pages, and the region enclosed by the detected outline for the sketch page. Either way, the result goes into the new package through Place an Image in a Slot.
How does it know which pages belong to which form?
It doesn't work it out on its own — the assignment is an input, and the workflow builds it before calling the split. What the workflow has to build it from is a short sample of text taken from every page of the upload, together with the list of forms registered for that package. The split itself applies the assignment as given.
Will it copy mistakes from the old report?
It can. The prior package is being used as the example to write against, and nothing in the process knows that a phrase in it was wrong. A confidently-wrong prior example is a real risk worth knowing about.
Is this the same as splitting a PDF by page range?
The mechanic is similar; the point is different. Each resulting piece is labelled with the form it represents, which is what lets a new section be filled against the matching old one. A generic page-range split gives you loose files with no such relationship.