Boeing’s redesign of MCAS in March 2016 had 2 approvers, a date and a purpose. On the public record it had no list of what depended on it.
In March 2016 the general manager of Boeing’s 737 MAX programme and its chief project engineer approved a redesign of the flight-control software called the Maneuvering Characteristics Augmentation System, or MCAS. According to the final report of the US House Committee on Transportation and Infrastructure, the change gave MCAS more authority to move the horizontal stabiliser at low speed, so that the aircraft could meet stall requirements needed for certification.
Flight testing had begun 2 months earlier. The Department of Transportation’s Office of Inspector General dates the engineering record, Revision D, to 30 March and calls it a significant change.
Before Revision D, MCAS could activate only above Mach 0.67, and afterwards anywhere from Mach 0.2 to 0.84. Its maximum movement of the stabiliser rose to 2.5 degrees. The inspector general puts the earlier limit at 0.55 degrees, an increase of over 300 percent, while the committee gives 0.6, citing a Boeing coordination sheet from July 2015.
By 15 August 2016 the 2.5 degree version sat in the flight-control software Boeing intended as final. In October 2018 and March 2019 repeated activations of MCAS pushed Lion Air Flight 610 and Ethiopian Airlines Flight 302 into dives neither crew could recover from, and 346 people died.
The chief project engineer later told committee staff what he had not known when he approved the change. MCAS took its reading from a single angle-of-attack sensor, and it could fire repeatedly.
Boeing also held internal simulator data in which one of its own test pilots took more than 10 seconds to respond to an uncommanded activation, a result described as catastrophic. The approval came from people with authority to give it, and it went ahead without facts the new behaviour depended on.
On 19 January 2016, 10 weeks before the redesign, Boeing had completed a single and multiple failure analysis of 75 failure cases. A system safety assessment and the certification plans described MCAS, and FAA engineers and test pilots worked on the programme. After the redesign, the committee found, Boeing “failed to reevaluate the system or perform single- or multiple-failure analyses of MCAS”.
In a presentation to the FAA in December 2018, Boeing described its own failure analysis as completed before the design change and “not reevaluated”. The inspector general found the hazard table in the safety assessment left as it was, and the FAA engineer who recommended delegating that assessment back to Boeing said he had not known of the increase to 2.5 degrees.
After the crashes the FAA convened the Joint Authorities Technical Review, whose members came from the FAA itself, NASA and 9 civil aviation authorities. The review found that certification deliverables were not updated during the programme to reflect the changes to MCAS, and that the certification plans and the preliminary system safety assessment did not describe its expansion into the low-speed part of the flight envelope. Recommendation R4 asked the FAA for feedback paths that keep compliance, system safety and human factors in view throughout development and certification.
The mechanisms for reconsidering the design were present in 2016, and no reassessment of the whole aircraft followed. My reading of the record is that the redesign entered it as a revision prompted by flight test, and the reports reviewed here do not list what rested on it.
Kettei calls this authority leakage. A programme approves its founding decision with care, and then the authority to commit the endeavour drains into execution, where people with real authority make choices to keep the work moving. Those choices bind the programme as firmly as the founding decision did, yet the choices go unrecorded as decisions. The March 2016 redesign was one.
It changed the Scope of what the aircraft would do on its own, at speeds where MCAS had not acted before, and within 5 months it was in the software meant for delivery. The record names 2 approvers and no Owner for whether 2.5 degrees stayed safe, and shows no Coupling tying the change to the failure analysis or safety assessment, both built on the earlier design, or the pilot training commitment.
In Kettei a commitment of that weight, made during execution, is a decision in its own right. It begins as Proposed and moves to Active only when its 7 properties are written down.
The set includes a Reversibility classification, a Validity Boundary naming the assumptions it stands on, and the Coupling that declares what depends on it. When a decision with declared dependents changes, each dependent comes back for review by its own owner inside a response window, and a change to Scope goes through a governed transition, recorded and dated with its reasons.
Three gaps the chief project engineer described, the single sensor, the repeat activation and the slow test response, are the kind of assumption a written boundary puts in front of an approver before the signature. Written properties alone promise no different answer, since the people in the room still decide what to do with a review.
The properties give the review an object, a named owner and a date. For a programme past its approval gate, the useful audit starts from the product as built and works back to the record. Which commitment made inside execution has changed what an earlier decision assumed, and who has checked that assumption since?
Kettei: The Discipline of Governed Intent is on Leanpub: https://leanpub.com/kettei-dogi
More on Kettei, the free Boeing case and the Kettei Workshop: https://ishi.co.nz/kettei/
