FileMaker AI Grammar banner
andykear andykear

FileMaker AI Grammar

Testing community

Description

171 engine-measured test vectors and 29 verified behaviour rules for FileMaker's calculation engine, round-trip tested against real FileMaker Pro rather than assumed or transcribed from docs. Includes a loadable Claude skill covering the traps AI models get confidently wrong. Companion to FileMaker AI Vocabulary.

Installation

This entry records only its repository, not the path inside it, so there is no exact command to give. Open the source below and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

README

[![Stars](https://img.shields.io/github/stars/andykear/FileMaker-AI-Grammar?style=social)](https://github.com/andykear/FileMaker-AI-Grammar) [![License](https://img.shields.io/badge/license-CC%20BY%204.0-green)](https://creativecommons.org/licenses/by/4.0/)

FileMaker AI Grammar

171 test vectors covering FileMaker's calculation engine, each one measured against real FileMaker Pro rather than assumed. 29 rules pulled out of that corpus and a Claude skill built from the seven that matter most.

Developed by Andrew Kear of Clockwork Creative Technology and shared openly with the FileMaker/Claris community.

The problem

FileMaker's calculation language has no published formal grammar. Claris maintains a full operator order of evaluation table, at `operators-in-formulas.html`, it just isn't cross linked from the four separate operator category pages that describe the individual operators.

So this repo tests instead of describing. Every vector here is a calculation expression, round trip tested against a real FileMaker Pro file, with the value it actually returned logged as the expected output.

The reason that matters: FileMaker's calculation engine has real, specific behavior that doesn't match general purpose language convention, or plain intuition. Unary minus binds tighter than exponentiation. `Trim` strips only spaces. `Substitute` is case sensitive while `Position` isn't. Some of this is stated in Claris's own help pages, it's just easy to miss, or to contradict with intuition carried over from another language. A model reasoning from general programming convention gets a predictable slice of this confidently wrong, with nothing to signal that anything is off.

Tested against ChatGPT

All eight predicted traps below were run against ChatGPT, logged honestly rather than cherry picked. Four produced a confirmed wrong answer. Four the model already had right.

Prompt Predicted wrong answer Actual Result
"abc" = "ABC" False