Real-world findings

Real form-logic failures, shown rule by rule.

In a retained July 2026 convenience review of 64 public-source artifacts across SurveyJS, REDCap, and FHIR Questionnaire, FormPath recorded 4 supported defect shapes affecting 7 files. The examples below show one inspectable case for each shape.

Each case shows the rule that failed, the values available in the form, and what users or data collectors could miss. Working SurveyJS answer combinations are replayed in SurveyJS; supported impossible rules use finite proofs whose encoding is sample-checked against it for that form. Supported generated rule shapes are exhaustively cross-checked in release tests. REDCap and FHIR conclusions concern logic translated from the supplied definitions, not replay in their native runtimes, and are labeled accordingly.

This deliberately assembled set exercises varied definitions and adapters; it is not a random sample and does not estimate how common these defects are. Public source files are summarized rather than republished, and the analyzer uses safe, made-up reproductions.

64retained artifacts across three form ecosystems

7affected files across 4 published defect shapes

4 defect shapes · July 2026

This convenience set was collected for validation, not prevalence measurement. The public ledger preserves all 64 artifact IDs, hashes, assignments, and the evidence boundary for each ecosystem without republishing third-party definitions.

Inspect the retained review

FP–001FHIR Questionnaire

Translated-model proof

Hand-authored FHIR transcription of a public CDC STEADI referral form

A follow-up waited for an answer that did not exist

In the retained transcription, the rule expects “vision-20-40,” but the checklist offers “vision.” The follow-up cannot be enabled by any offered answer.

Impact: In the retained transcription, selecting the vision referral reason does not enable the side follow-up.
How it was checked: Checked against the listed answers after translating the hand-authored FHIR Questionnaire logic; no original CDC Questionnaire or native renderer was tested
Affected files: 1 file

Read the related explanation

Controlling rulereasons for referral = 'vision-20-40'

Available answer codes'foot-abnormalities', 'vision', 'balance-difficulties', …

TRANSLATED-MODEL PROOF

FP–002REDCap

Translated-model proof

Four RADAR-CNS study dictionaries

Four site copies waited for an answer code none of them offered

All four branching rules wait for code 17, but none of the four answer lists includes it.

Impact: The “describe other ethnicity” question never appears. In the two copies with codes 0–6, that hidden field is also marked required.
How it was checked: Checked against the listed choices after translating the REDCap branching logic
Affected files: 4 files · 2 required follow-ups

Read the complete case

Controlling rule[ethnicity_1] = '17'

Available answer codesTwo copies offer '0'–'6' (“Other” = '6'); two offer '18'–'22' (“Other” = '22')

TRANSLATED-MODEL PROOF

FP–003SurveyJS

Confirmed root cause

Mortgage application example

Missing brackets turned an answer check into text that never matched

The second clause compares two fixed text literals instead of reading {property-use}; it also misspells “residential.” The expression is false for every respondent.

Impact: The mortgage-status question never appears, including for a house purchase on a residential property.
How it was checked: Proved impossible from the listed answer values; the rule encoding was cross-checked against SurveyJS
Affected files: 1 file

Read the related explanation

Controlling rule({loan-purpose} = 'house-purchase') and ('property-use' == 'residental')

Definition values{property-use} stores 'residential', 'buy-to-let', or 'holiday-let'

PROVEN

FP–004SurveyJS

Confirmed root cause

Public-sector evaluation example

A final score referred to three names that did not exist

The grand-total expression references three names absent from the definition. The form instead contains differently named section totals, including I_total_aspectos_administrativos and II_total.

Impact: The supplied form loads without a JSON error and displays 0% and “Crítico” even as its defined section totals change.
How it was checked: Checked for references to fields missing from the SurveyJS definition
Affected files: 1 file · 3 affected outputs

Read the related explanation

Controlling ruleround({A_total} + {B_total} + {C_total}, 2)

Defined inputsNo fields or calculated values named A_total, B_total, or C_total

PROVEN

Try a safe reproduction

Try the same kind of defect in a synthetic form.

Open the analyzer, then choose one of FormPath’s compact, made-up examples. Public source files are summarized rather than republished.

Open the analyzer