space-rha.comShort courseChapter 11 · The assurance decision

Chapter 11 · Lesson 1 of 3 · about 14 minutes

The assurance argument and its closure

An assurance case connects the required mission behavior to the evidence through stated reasoning. A stack of test reports supports the claim only when the tested parts, conditions and endpoints match it. The path from environment and part response to function should be visible enough that another engineer can challenge it.

What you need to know

An assurance case is an argument. It states a claim about a mission function, gives the reasoning that connects the claim to evidence, and presents the evidence, with the context, the assumptions and the open issues attached where they apply. A stack of test reports on its own is not that argument. A report says what was observed under its conditions; the argument has to say why those conditions cover the flight. The case is reviewable when another engineer can follow the path from the environment (chapter 1), through transport (chapter 2), the requirements (chapter 3), the part responses (chapters 4 to 8), the rates (chapter 9) and the mitigations (chapter 10) to the function, and can point at the link they doubt. Notations such as the goal structuring notation record goals, strategies, solutions, contexts, assumptions and justifications as a graph. The notation shows the structure of the argument. Whether each inference holds is a separate question.

Claims are written at the scope of a function and kept separate by effect. Cumulative performance (U3 stays within its error budget at end of life), destructive-event protection (U1 survives a high-current event, or its loads do) and interruption availability (the recorder is available for its windows) are three claims that call for three kinds of evidence. One part generally appears under all three, and one test generally supports only one of them. Because several parts and operating modes together satisfy one requirement, the argument is organized by function rather than by the parts list. One test cited under several branches of the graph is still one piece of evidence; citing it several times does not make it independent corroboration.

A requirement closes when the evidence answers its question for the flight hardware, the flight conditions, the function and the criterion, and the record says so. The verification matrix lists, for each requirement, the method, the evidence, the reason that evidence applies to the flight configuration, the result and its limitations, the owner, and the disposition. The disposition is one of four: verified; verified with conditions (an operating restriction, a monitoring need); open with an action; or a documented deviation accepted by the responsible authority. Technical findings, engineering recommendations and project dispositions are kept as three separate records. A waiver is a controlled decision about a requirement; it adds no evidence. A point estimate below the criterion can hide a wide bound (chapter 8 lesson 4) or a gap, and a status color cannot show either, so the matrix records the bound and the gap next to the status.

Worked example 11.1. The argument for the recorder function, written out

LevelStatementSupported by
GoalThe recorder delivers its science data with at most one lost recording window per month over three years in the sun-synchronous orbitthe three strategies below
Strategy AArgue over each effect separately: cumulative damage, destructive events, interruptionschapters 3 and 8 lesson 4
Goal A1 (cumulative)U3 and U4 remain within their circuit budgets at end of life; U1 and U2 regulation and timing parameters remain within specificationTID trajectories and the tolerance bound (8.3); U4 CTR curve against the five-year fluence (10.4); the low-dose-rate basis for U3
Context A1Mission dose 15 krad(Si) at the parts, design margin 2, end-of-life circuit analysis at 70 °Cchapter 1 package; chapter 3 flowdown
Open issue A1The 90/90 tolerance bound of the U3 lot is 25 krad against 30 krad required; the LDR offset error is 12 mV against a 10 mV budgetdisposition pending: calibration in flight, or a replacement part (11.2)
Goal A2 (destructive)A high-current event on U1 does not damage U1 or its loads; its probability over three years is below 1 % (worst case 0.50 %)U1 curve and IRPP rate (8.2, 9.1); the bench demonstration with the output capacitor present (10.3); no SEB or SEGR to 112 V derated (7.4)
Assumption A2The flight lot shares the die revision of the tested lotprocurement record; reopened by any die or process change
Goal A3 (interruptions)Interruptions of the recorder are detected and cleared within 88 s in at least 88 % of cases, and the remainder by ground command within one orbitSEFI curves and proton term (9.2); watchdog coverage by injection (10.2); the reset ladder timing (7.3)
Goal A3.1Uncorrectable words in U2 are rare against the data loss allocationinterleaving against the measured cluster distribution; scrub interval sized for the anomaly pass (10.1)
JustificationRates are best estimates, with the X/100 worst case reported alongside; proton SEFI in the flight mode is a listed gap, with a proton run scheduled8.4 budget

Written out this way, the argument shows which two links the goal depends on: the U3 lot bound (A1) and the watchdog coverage (A3). Both are questions of evidence, and both are the items whose cost and value lesson 11.2 weighs.

The four teaching parts in this chapter

Four constructed parts run through chapters 6 to 12 so that each chapter's work can be followed on the same hardware: U1, a shared regulator that powers sensing, logic and the recovery controller; U2, a 1 Gbit recorder memory under SEC-DED error correction and periodic scrubbing; U3, a precision analog chain with an accuracy allocation; and U4, an optical interface that must keep its signal margin. Their response curves are stated in the build (constructed from a declared true curve with Poisson sampling), so that every number in these chapters can be traced; they are teaching cases, not commercial devices.

PartFunctionThis chapter's step
U1 · shared powerA shared regulator powers the sensing, the logic and the recovery controller.Tie the functional claim to the configuration that was tested, the recovery path that was demonstrated and the assumptions about the shared rail. A change to the rail or to the firmware can reopen the argument.
U2 · recorder memoryA memory stores science data under ECC and periodic scrubbing.Put the physical-to-logical mapping, the ECC configuration, the response data and the recovery path into the controlled baseline. A change to the firmware or to the memory's construction calls for a review.
U3 · precision analogA precision analog chain has to hold an accuracy and timing allocation for the application.Tie the accuracy claim to the end-of-life circuit analysis and to the degradation envelope it was based on. A change in load or bias can remove a margin that looked comfortable.
U4 · optical interfaceAn optical interface has to deliver enough signal margin through the whole mission.Record the emitter and detector construction, the test endpoint, the transfer model and the circuit limit. A change of supplier or of internal construction can reopen the question of whether the DD data still apply.

Exercise

Start from one bounded claim about a mission function and trace each supporting inference to the evidence and the assumptions it depends on in the controlled record.

Keep a requirement-to-evidence matrix that states, for each requirement, any conditions on the finding, the owner, the remaining actions and who has the authority to disposition it.

Going deeper

A sound structure does not make the inferences true

Whether each inference holds depends on whether the evidence applies to the flight case, whether every credible mechanism has been considered, and whether the assumptions about how the parts combine are right. One test or model cited under several branches is still one piece of evidence. Review how strong the reasoning is and what could defeat it, including evidence that would overturn a central assumption.

A fully linked graph can still contain a false premise or a system failure path that nobody tested.

Do not collapse uncertainty into a status color

A point estimate below the criterion can hide a wide statistical bound, a disagreement with the model or a mode that was never tested. Look at the confidence statement and at whether the requirement itself is complete. A more severe consequence justifies a heavier burden of evidence, but no fixed threshold or traffic-light scheme replaces the project's technical review and its decision authority.

A waiver is a controlled decision about a requirement. It adds no evidence.

The common mistake

GSN or another notation records the reasoning. The reviewers still have to judge whether the reasoning is sufficient.

Takeaway

  • Writing the argument out shows which assumptions and which evidence the mission-level claim depends on.
  • A requirement closes against the question it originally asked, and the record keeps the limits of the evidence that closed it.

Check yourself

A review package contains twelve test reports and a summary table of pass marks. What is missing for it to be an assurance case?

The reasoning. The package needs to say which claim each report supports, why the test conditions cover the flight configuration and environment, which assumptions the argument depends on, and what remains open. A table of pass marks hides the bounds and the gaps.

The same U1 heavy-ion test appears under the destructive claim and under the interruption claim. How many pieces of evidence is that?

One, used twice. The two claims draw on different quantities from the same run (the bound on destructive outcomes and the high-current cross section), and a flaw in that run (range at the wells, chapter 6 lesson 3) affects both branches at once.

When does the U3 requirement close?

When the evidence answers the question for the flight lot at the flight dose rate and conditions, or when the responsible authority records a deviation with the numbers in front of it. A tolerance bound below the requirement with a pending action leaves it open.

The verification matrix marks the U1 destructive claim green because the worst-case probability of 0.50 % over three years is below the 1 % criterion. What does the color hide?

The bound and the gaps. The 0.50 % is the X/100 worst case from one heavy-ion curve, and the status says nothing about the assumptions the figure rests on: that the flight lot shares the die revision of the tested lot, that the bench demonstration included the output capacitor, and that solar heavy ions are not in the figure (lesson 11.2). A point estimate below the criterion can hide a wide bound (chapter 8 lesson 4) or an untested mode, so the matrix records the bound and the gap next to the status; a status color cannot show either.

Why is the recorder argument organized by function (A1, A2, A3) rather than by the parts list U1 to U4?

Because several parts and operating modes together satisfy one requirement, and one part generally appears under all three effects. U1 appears under cumulative performance through its regulation parameters, under destructive-event protection and under interruptions, and each claim calls for a different kind of evidence; one test generally supports only one of them. Organized by function, the argument shows which links the mission-level goal depends on, in the worked example the U3 lot bound (A1) and the watchdog coverage (A3), so that a reviewer can point at the link they doubt.

The U2 availability claim (A3) holds only if the scrubber runs at its interval and the watchdog coverage measured by injection holds in flight. Which of the four dispositions applies, and what follows from it?

Verified with conditions. The disposition records the operating restriction (the scrub interval is part of the assessed configuration) and the monitoring need (the telemetry has to show that the scrubber runs, and the flight record of interrupts updates the coverage). Those conditions go into the operations handoff of lesson 11.3 and the telemetry design of lesson 12.1. A plain "verified" would drop them, and a firmware change that alters the interval reopens the claim.

Further reading

  • Assurance Case Working Group, "Goal Structuring Notation community standard," version 3, 2021.
  • T. Kelly and R. Weaver, "The goal structuring notation: a safety argument notation," Proc. DSN 2004 Workshop on Assurance Cases, 2004.
  • ECSS-Q-ST-60-15C, "Space product assurance: radiation hardness assurance, EEE components," 2012.
  • NASA, "Avionics Radiation Hardness Assurance (RHA) Guidelines," NASA/TM-20210018053, 2021. NTRS 20210018053.
  • K. A. LaBel et al., "Emerging radiation hardness assurance (RHA) issues: a NASA approach for space flight programs," IEEE Trans. Nucl. Sci., vol. 45, no. 6, pp. 2727-2736, 1998. doi:10.1109/23.736521.