This document explains what the Service is, what it is not, what kinds of errors it may make, and what your responsibilities are when using its output.
What the Service is
Redline accepts an uploaded construction drawing set, of one or many pages up to the per-analysis page limit published in our FAQ, and produces a structured report. You choose the kind of review to run, and the emphasis of the report varies with that choice: general review, detail review, clash detection, code compliance, annotation review, or coordination check. The report typically includes:
- A list of potential defects, organized by severity (critical, major, minor)
- References to codes, standards, or industry practices the AI believes are relevant
- Suggested corrective actions for each defect
- An overall summary and recommended coordination checks
The report is produced by large language models analyzing the drawing along with prompts crafted to focus the model on engineering concerns. Reports are typically produced in seconds to minutes.
What the Service is not
Known categories of error
You should expect the following kinds of errors and review accordingly:
False positives
The Service may flag items as defects when no defect exists. Common causes include misreading drawing notes, applying general guidance to a project that has a specific deviation, and identifying patterns in the drawing that are not actually present.
False negatives
The Service may fail to flag defects that are present. Common causes include defects involving multiple sheets that the Service cannot cross-reference, defects requiring deep code knowledge the model lacks, and visual conditions the model cannot perceive clearly.
Incorrect code citations
Section numbers, edition years, or scope of cited codes may be wrong. Treat every code reference as something to verify independently before relying on it.
Location ambiguity
The Service may identify the wrong page or area of the drawing for a flagged defect, or may describe a location in a way that is hard to verify on the sheet.
Confident wrongness
AI systems can produce output that sounds authoritative but is factually incorrect. The presence of code citations, professional vocabulary, or precise-sounding measurements in the output does not mean those details are accurate.
Your responsibility as the engineer of record
The Service's output may inform your review. It may speed up your review. It does not constitute your review.
You must independently verify every item from the Service before acting on it, communicating it, or incorporating it into deliverables. You must not represent the Service's output as constituting professional engineering review or opinion.
The stamp and the decision record
When you confirm, dismiss, or dispute a finding, the Service records that decision against your account together with the time you made it, and can include that record in an exported report. It is an attribution record: it shows who decided what, and when.
It is not a tamper-evident audit trail, it is not a legal instrument, and it does not by itself establish that a review was complete. Applying your seal image in the Service does not constitute a sealed engineering review, and the Service does not verify that you have considered every finding before it produces an export. Whether a review is complete, and whether a drawing may be sealed, remain matters of your professional judgment.
Restricted uses
The Service is not designed or validated for use in:
- Life-safety-critical determinations (fire alarm, emergency egress, fall protection) made without independent licensed review
- Medical facility code compliance (NFPA 99, FGI guidelines) made without independent licensed review
- Nuclear, defense, or aviation engineering work
- Direct submission to authorities-having-jurisdiction without engineer review
- Forensic or legal proceedings, including expert opinion or testimony
- Any application where an error in the output could foreseeably cause death, serious bodily injury, or significant property damage and where no independent licensed engineering review will occur before action is taken
None of these are absolute prohibitions on the underlying engineering work, only on relying on the Service as more than an early-stage aid in those contexts.
User acknowledgments
By using the Service, you acknowledge each of the following:
- You are using the Service as a professional review aid only, in connection with your own licensed engineering judgment or under the supervision of a licensed engineer.
- You will not rely on the Service's output as the sole basis for any engineering opinion, code-compliance determination, or signed deliverable.
- You will independently verify every defect, citation, and recommended action before acting on it.
- You understand that the Service generates both false positives and false negatives and that the rate of these errors is not guaranteed.
- You will not represent the Service's output as a substitute for a peer review, an AHJ review, an owner's review, or your own sealed engineering review.
- You will not use the Service in the restricted contexts described in Section 5 without separate independent engineering review.
- You assume sole professional responsibility for any work product you produce that draws on the Service's output.
Relationship to other terms
This Disclaimer is incorporated by reference into our Terms of Service. If there is any conflict between this Disclaimer and our Terms of Service on the subject of AI output and engineer responsibility, the more restrictive provision governs.
Nothing in this Disclaimer creates an engineer-client relationship between you and us, between you and any contributor to the Service, or between us and any third party who relies on output you have shared with them.