How rmForge works: fixed rules, AI drafts, your team decides

rmForge splits the work of a Hazard Analysis three ways. Fixed rules do the risk math, check the rows and lay out the workbook, the same way on every run. AI, meaning artificial intelligence models, reads your documents, drafts the rows and makes the judgment calls listed below, inside limits the software checks. Your team makes the decisions. This page lists which is which, as the software runs today.

Done by fixed rules: same input, same answer

  • Risk math. Po, the probability that harm occurs, comes from P1 and P2 (the probability of the hazardous situation, and of harm once it exists) through a lookup table, and the Risk Level from Severity and Po through a second table. Nothing is multiplied and no AI is asked. The Risk Region tab of your workbook shows both tables.
  • The software rule. When a row's cause is software and its Severity is 4 or 5, P1 is set to 5, the highest level.
  • Which hazards a row may name. Your setup answers decide which hazards from the list in Annex C of ISO 14971, the international standard for medical device risk management, apply to your device. A row that names any other hazard is sent back to be rewritten.
  • Harm wording. A harm taken from the rmForge harm library must match the library word for word. A row that does not is sent back to be rewritten.
  • Writing checks. Each row's hazardous situation, harm and sequence of events are checked against written grammar rules. A field that breaks one is rewritten once by AI, and the rewrite is kept only if it breaks fewer; a row that still breaks one carries a note for your reviewer.
  • The workbook layout. Its 32 columns, its Po and Risk Level formulas, its reference tabs and its cover page come from code and are the same on every run.
  • Word, Excel and comma separated values files are turned into text by code on the rmForge server, not by AI. The AI steps then work from that text.

Done by AI, and its limits

  • AI reads the text of your Instructions for Use and design inputs and answers the 48-question Hazard Questionnaire.
  • AI drafts each hazard row: the sequence of events, the hazardous situation, the harm, the cause, the Severity, P1 and P2, and at least one proposed risk control.
  • A row that names a hazard your setup answers do not allow, words a library harm differently, or puts a fault under misuse without saying why that is foreseeable is rewritten by AI up to three times. If it still breaks the rule, it is kept, with a note on the cell for your reviewer.
  • Three kinds of row are left out rather than noted: a row with a required field missing or invalid, a row the software reads as being about workers who make, sterilize, pack or ship the device, and a row written under the wrong hazard category.
  • For medical electrical equipment, AI decides Essential Performance for each row above Low; Low rows read N/A (not applicable).
  • When a harm's name is not found in the published list of the International Medical Device Regulators Forum (IMDRF), AI picks the published term for the same injury, or the closest broader one. Check those codes against your own coding.
  • When you bring a product specification or a requirements matrix, AI decides which of its entries each risk control matches, so the workbook can show your own requirement number, design output and verification.
  • When you bring your own harm list, AI decides whether each of your harms is the same injury as a harm in the rmForge library.
  • A PDF (Portable Document Format) file the rmForge server cannot read is read by AI.
  • Each AI step starts on one model, and backup models take over when it fails. The models that answered for your analysis are listed on the workbook's Disclaimer sheet.

What your team decides

  • Whether each row is right. Every row is a draft for your engineers to review, edit and approve. The workbook says so on its cover and its Disclaimer sheet.
  • What risk is acceptable. rmForge reports a Risk Level from its tables. What is acceptable for your device is yours to set.
  • What goes in. You bring your documents and your answers to the setup questions, and the rmForge operator enters them with you.
  • Approval, and the FMEA baseline. Your team approves the Hazard Analysis on the sign lines of its Cover Page. A Design or Process FMEA (Failure Mode and Effects Analysis) is built only on a Hazard Analysis you accepted as its baseline: the rmForge operator records, with the date, that you accepted it. That record is not a signature; your approval stays on the Cover Page and in your own quality system.

rmForge does not claim that a file built with it meets any regulation or will be accepted by a regulator. See what a Hazard Analysis is and the questions and answers.

See a sample rmForge workbook, built for a test device.

See a sample workbook

Or create an account and book a working session.