M45
Method

How M45 reconstructs intent

Engineering programs produce artifacts. The reasoning behind them is supposed to survive — and rarely does. Here is the shape of the work.

01 · Today

Engineering programs produce artifacts.

RequirementsHazard analysisDesign rationaleSubsystem spec

Requirements, designs, hazard analyses, rationale notes. Each lives in its own system, in its own format. Each one is a partial view of the program.

02 · The shape of the problem

Each artifact projects from the same underlying system. The system itself is nowhere written down.

the underlying system(implicit, unwritten)each artifact is a partial view of the system

The artifacts aren't the system. They are projections of it. The system exists in the relationships between them, and in the reasoning that connects them. That reasoning is what gets lost.

03 · What senior engineers already do

Senior engineers reconstruct the system from the artifacts to check whether it adds up.

SYS-041SYS-042SYS-043reconstructevery reviewPilothuman controllerAutopilotautomated controllerAircraftcontrolled processnot preserved between reviews

Reading SYS-041, SYS-042, SYS-043 about autopilot disengagement, the engineer assembles a picture: a control loop, the pilot, the autopilot's authority, the sensor feeding it. That picture isn't preserved. It has to be reconstructed every review cycle, by someone with enough context to do it.

04 · What M45 does

M45 reconstructs the model from the artifacts. Each artifact gets mapped to the structural role it plays.

SYS-041Disengage oninvalid airspeedSYS-042Alert pilot ondisengagementSYS-043Manual controlalways availablePilothuman controllerAutopilotautomated controllerAircraftcontrolled processAirspeed sensorfeedbackcontrol actionfeedbackcontrol actionfeedbackdataAll passindividually

A control structure built from the artifacts. SYS-041 constrains the autopilot's disengagement behavior. SYS-042 constrains the autopilot-to-pilot feedback channel. SYS-043 constrains the aircraft's controllability after disengagement. Each requirement now has a structural role.

05 · The gap

The model has slots that requirements could fill. Some have requirements. Some don't.

Pilothuman controllerAutopilotautomated controllerAircraftcontrolled processAirspeed sensorfeedbackPilot's process modelwhat the pilot believesabout airspeed validityGap detectedNo requirement covers what the pilot needs to believe about airspeed validityafter disengagement. Pilot is left flying on the same bad data.

The pilot's process model has a slot: what they need to believe to control the aircraft. After disengagement on bad airspeed, no requirement tells the pilot the airspeed is unreliable. The artifacts pass review individually. The model says they don't add up.

06 · AI infers. Engineers decide.

M45 surfaces the gap. The engineer decides what to do.

M45 inference · Hypothesis ready for reviewCoverage gap: pilot process model after autopilot disengagementInferred from: SYS-041, SYS-042, SYS-043, hazard H-12,control structure analysis. Confidence: surfaceable.Acceptadd requirementRefinescope the gapDeferpark with rationaleRejectnot a gapDecision recordedwith rationale, carried forward

M45 shows the inferred role of each artifact, the slot in the model that's uncovered, and why the inference fired. The engineer accepts, refines, defers, or rejects. Every decision is recorded with its rationale and carried into the next iteration.

07 · Across iterations

The engineering context grows richer. Reasoning compounds instead of resetting.

Iteration 18 accepted4 deferred2 rejectedIteration 211 accepted5 deferred0 rejectedIteration 314 accepted3 deferred1 rejecteddecisions, rationale, and inferences carry forward

Each iteration starts where the last one left off. Decisions carry forward, deferred items stay open, the model grows, and what was once an inference becomes part of the project's working record.

If this is the work you're trying to do, we'd like to understand your case.

We're talking with a small number of design partners: aerospace programs where the cost of misalignment is real. Not to demo, but to understand how reasoning gets lost in your specific program, and whether the way we're thinking about it would help.

Get in touch