Scoring

A score template defines what to calculate; its engine decides how. Pick the engine that matches where your logic already lives: a rules matrix, an Excel model, a SQL query, or a document to read.

Score templates

Setting	Property	Purpose
Engine	ScoreEngine	The plugin that calculates the score. See Scoring engines.
Applies to	Object	The kind of record scored (for example Contact, Document, or Message).
Method	Method	Where the engine’s input comes from. See Data sources.
Value type	ValueType	Type of the overall score value.
Engine parameters	EngineParameters	Engine-specific options, in query-string form.
Render	RenderID	A file the engine uses: the Excel model or the XSLT transform.
Repeatable	Repeatable	One score per record, one active, or many. See Templates.

Score items

Each item is one data point:

Setting	Purpose
Name	The item’s name. Engines match it by its safe name (spaces and special characters removed: Due Date becomes DueDate).
Value type	Money, Int, Float, Boolean, String, Date, or Enumerated. The value is stored in a typed column.
Prompt	For text parsing, the pattern to find; for AI, the instruction describing the value.
Enumeration	Allowed values for an Enumerated item.
Weight	How much the item counts toward the score’s overall confidence (default 1; 0 excludes it).
Label, Sequence	How the item is labeled and ordered on screen and in audits.
Mask, Unmask permission	Mask a sensitive value, and the permission needed to see it unmasked.

Data sources

The template’s Method decides what the engine receives:

• A method signature, such as Contact/Summary?ID={ObjectID} or a custom statement. Its results (rows or XML) are passed to the engine.
• Parameters: the values passed on the calculate call are used directly. This turns a score into a calculation service.
• Empty: the engine works from the scored record itself (for example, the document being read).

Scoring engines

Engine	Name	Input	How it scores
Matrix	Matrix	Parameters, query row, or record	Looks up the rules matrix with the same name as the template, and saves each matrix output to the matching item.
Excel	Aspose Cells integration	Record summary, query results, or parameters	Merges data into the template’s Excel model, recalculates it, reads results from named cells, and saves the completed workbook with the score.
API	ApiEngine	Method results	Saves any returned column whose name matches the score or an item. Ideal for caching expensive statistics.
XSLT	Xslt	Method returning XML	Transforms the XML with the template’s XSLT into elements named after items.
Text parsing	TextParsing	A document or message	Finds each item’s prompt pattern in the text, running OCR on documents first if needed.
AI (text)	AiScoreEngine	A document or message	Converts the document to text and asks an AI model to extract every item, with confidence.
AI (multimodal)	AI integration	A document	Sends the original file (PDF, image, audio) to the model, with no text conversion.
Amazon Textract	TextractEngine	A document	Extracts each item by name with Amazon Textract and saves its confidence.
Composite	Scoring integration	Any	Runs several engines or templates and keeps the most confident value for each item.

Excel models

In the workbook, give a cell a name and add a comment starting with ?= to inject data:

• ?=FieldName places a single value.
• ?= followed by a method signature places a table of results.
• ?=Matrix:{Matrix}:{Output} places the best-matching rule output.

To extract results, name cells ScoreValue (the overall value) and after each item’s safe name. A cell named {Item}Confidence sets that item’s confidence.

Text parsing prompts

Prompt	Finds
Invoice Total: {Value}	The value after Invoice Total:, parsed as the item’s type.
Due{Value}{Newline}	The value between Due and the end of the line.
Signature on file (Boolean item)	Yes if the text appears, No if not.

One match gives confidence 100. No match gives 0. Several matches lower confidence by 25 per match.

AI engines

Set AgentName in Engine parameters to choose the configured AI agent (it defaults to the engine’s name without Engine). UseConfidence=false skips per-item confidence. The multimodal engine also accepts FileExpirationMinutes, to reuse an uploaded file when the same document is scored repeatedly.

AI · built in

Write each item’s prompt as a plain description of the value you want (“Total amount due, including fees”). The prompt is what the model reads.

Composite scoring

Set Engine parameters to choose what to combine:

Parameter	Default	Effect
Engines	AiScoreEngine,TextractEngine	Engines to run, in order, against this template.
Templates		Other score templates to run instead (their item names must match).
Label	{Score} - {Name}	Name of each child score.
Confidence	0.8	Stop once a child score’s confidence exceeds this. Set 0 to always run everything.

Each child is saved as its own score and linked to the composite, so you can compare models side by side.

Audits

To audit completed work:

1. On the worklist template being audited, create an audit matrix. Its outputs choose which completed items to sample: Template (the audit worklist and score template name), Percent, Minimum, and Maximum.
2. Create a score template with the same name as the matrix’s Template output. Each item becomes an audit question.
3. Run Audit on the worklist. QBO samples eligible items into the audit worklist; auditors answer the questions on the original page.

Start from a matrix

If your logic is already a rules matrix, the CreateFromMatrix statement builds the score template for you: one template per selected matrix (named after it, using the Matrix engine), with an item for each matrix output and a matching value type. Matrices that already have a score template are skipped.