Visual walkthrough — Verification methods — analysis, test, inspection, demonstration

This page answers ONE question, drawn all the way from the ground up:

When we shake a spacecraft to prove it survives launch, why do we shake it 25% harder than launch actually shakes it? Where does the number 1.25 come from?

The parent note stated the rule. Here we earn it — every symbol, every bell curve, every square root, drawn as a picture before it is used. A smart 12-year-old who has never seen a "sigma" should be able to follow from line one.

We will build to this one final statement: $\text{Qualification Level}=\text{Flight Limit}\times 1.25$

but only after we understand what each piece is.

Step 1 — What "the load" actually is: a single number on a ruler

WHAT. Every spacecraft that flies gets rattled by its rocket. We compress "how hard it gets rattled" into a single number. For vibration we call it the shaking strength, measured in g (multiples of Earth's gravity). Call it $x$.

• $x$ = shaking strength (in g). Bigger $x$ = a rougher ride.

A word on "RMS." The parent note says "8 g RMS random vibration." Random shaking is not a steady push — it jitters up and down thousands of times a second, so its plain average is zero (it pushes up as often as down). To get a single honest "size" for something that averages to zero, we square every wiggle (squaring makes both up and down positive), average those squares, then take the square root to get back to g. That three-word recipe — Root of the Mean of the Squares — is RMS. So "8 g RMS" is the typical effective strength of the jitter. Our $x$ is this RMS number.

WHY start here. Before we can talk about margins and safety factors, we need a thing to measure. Everything downstream is just "where does $x$ land, and where do we test?" So we draw a ruler.

PICTURE. Look at the orange tick. The rocket engineers tell us the Flight Limit — the roughest ride any launch is expected to give. That is one fixed point on the ruler, one particular value of $x$.

Step 2 — Reality is not one number: it is a cloud

WHAT. Here is the twist that forces the whole derivation. No two flights are identical, and no two spacecraft are identical. Build the same satellite twice and one will be slightly stiffer than the other (a weld a hair thicker, a bolt a touch tighter). Fly the same rocket twice and one launch shakes a little harder than the other (weather, fuel, wind).

So the real value of $x$ a given unit feels is not the single tick $\mu$. It is a whole spread of possible $x$-values scattered around $\mu$.

WHY this matters. If reality were one exact number, we would test at exactly that number and be done. We do NOT test harder for fun — we test harder because we cannot see which side of $\mu$ any particular flight or unit will land on. The margin exists purely to cover this scatter.

PICTURE. The single orange tick blooms into a cloud of possible $x$-values — a pile of little dots, densest near the middle, thinning out as you move away. Most flights land near $\mu$; a few unlucky ones land noticeably higher.

Step 3 — Naming the cloud: the bell curve and its width $\sigma$

WHAT. Smooth that pile of $x$-dots into a curve and you get the famous bell curve (the normal distribution). Two numbers completely describe it:

• $\mu$ (Greek "mu") — the center: the $x$-value where the peak sits. This is our Flight Limit tick.
• $\sigma$ (Greek "sigma") — the width: how far the cloud spreads sideways in $x$. Small $\sigma$ = a tight, confident pile. Large $\sigma$ = a fat, uncertain smear.

WHY a bell and not something lopsided. Strictly, $x$ can never be negative — you cannot shake something with less than zero strength. A perfect bell curve does leak a tiny sliver of probability below zero, so it is only ever an approximation. It is a good one here because our scatter is small: $\sigma$ is about 10% of $\mu$, so "zero" sits a full $10\sigma$ below the center — so far out on the tail that the leaked probability is utterly negligible (well under one in a trillion trillion). For small scatter around a comfortably-positive mean, the tidy symmetric bell is the honest, standard choice. (If $\sigma$ were comparable to $\mu$ we would switch to a one-sided model like the log-normal — but we are nowhere near that regime.)

WHY we need $\sigma$. The margin is going to be measured in widths. "How far past the center must I go to be safe?" only has an answer once we know how wide one step of uncertainty is. That step is exactly $\sigma$.

PICTURE. The bell, centered on $\mu$. The teal double-arrow is one $\sigma$ — one standard step of uncertainty out from the middle. Notice the shaded areas: 68% of all outcomes land within $\pm 1\sigma$, and 99.7% within $\pm 3\sigma$. Those percentages are baked into this exact shape.

Step 4 — Two clouds, not one: manufacturing scatter AND environment scatter

WHAT. Our uncertainty comes from two independent sources, and the parent note gave both:

• Manufacturing scatter ${\sigma }_{\text{mfg}}=0.1$ — units come out $\pm 10\%$ different from each other.
• Environment scatter ${\sigma }_{\text{env}}=0.1$ — launches shake $\pm 10\%$ differently from each other.

(The $0.1$ means "10% of the Flight Limit," so we're working in units where $\mu =1$ and one $\sigma$ of $0.1$ means 10%.)

WHY two separate numbers. A stiff unit on a rough launch is worse than either effect alone. To size the margin we need the combined scatter, not each piece separately. So the question of Step 5 is: how do two clouds add?

PICTURE. Two narrower bells side by side — the plum one for manufacturing, the teal one for the environment — each with its own width $0.1$. We must fuse them into one.

Step 5 — Why we ADD THE SQUARES (variances add; Pythagoras hiding in statistics)

WHAT. To combine two independent scatters we do not simply add the $\sigma$'s ($0.1+0.1=0.2$ is wrong). Instead we add their variances (${\sigma }^2$, the squared widths from Step 3) and take the square root: ${\sigma }_{\text{total}}=\sqrt{{\sigma }_{\text{mfg}}^2+{\sigma }_{\text{env}}^2}$

Term by term:

• ${\sigma }_{\text{mfg}}^2$ — the manufacturing variance (one leg of a right triangle).
• ${\sigma }_{\text{env}}^2$ — the environment variance (the other leg).
• $\sqrt{\cdots }$ — undoes the squaring to get back to a plain width (the hypotenuse).

WHY squares and not a plain sum? This is the deep fact from Step 3: for independent sources, it is the variances (the ${\sigma }^2$'s, the squared distances) that add up, not the $\sigma$'s themselves. Geometrically that means the two effects point in "different directions" — like walking East then North. Walk 0.1 East and 0.1 North and you are NOT 0.2 from home; you are $\sqrt{0.1^2+0.1^2}$ from home. Independent uncertainties combine exactly like perpendicular distances: by Pythagoras. Adding the $\sigma$'s straight would pretend the worst of both always happen together, which is too pessimistic (and too expensive).

PICTURE. A right triangle. The horizontal leg is ${\sigma }_{\text{mfg}}$, the vertical leg is ${\sigma }_{\text{env}}$, and the slanted hypotenuse — the real combined uncertainty — is ${\sigma }_{\text{total}}$.

Now plug in the numbers: ${\sigma }_{\text{total}}=\sqrt{0.1^2+0.1^2}=\sqrt{0.02}\approx 0.14$

• $0.1^2=0.01$ each — the two variances (squared legs).
• $0.01+0.01=0.02$ — the total variance (squared hypotenuse).
• $\sqrt{0.02}\approx 0.14$ — the combined width: 14%, not 20%.

Step 6 — Covering almost everything: the $3\sigma$ reach

WHAT. We now decide how much of the cloud we insist on covering. Industry chooses $3\sigma$ — reaching three widths above the center captures 99.7% of all cases (recall Step 3's shaded bell). The test level is: $\text{Test Level}=\underset{\text{center}=1.0}{\underbrace{\mu }}+\underset{\text{safety reach}}{\underbrace{3{\sigma }_{\text{total}}}}$

Plug in: $\text{Test Level}=1.0+3(0.14)=1.0+0.42\approx 1.4$

Term by term:

• $\mu =1.0$ — the Flight Limit itself (working in units where the limit is 1).
• $3{\sigma }_{\text{total}}=0.42$ — three standard steps of combined scatter, our safety reach.
• Sum $\approx 1.4$ — test 40% above the flight limit to swallow 99.7% of all real flights.

WHY $3\sigma$ and not $1\sigma$ or $5\sigma$. $1\sigma$ covers only 68% — one in three flights would exceed our test, unacceptable. $5\sigma$ covers 99.9999% but demands a monstrous, hardware-destroying shake for vanishing extra safety. $3\sigma$ is the sweet spot: near-total coverage at survivable cost.

PICTURE. The full bell with the orange test-level tick planted at $\mu +3{\sigma }_{\text{total}}$, far out on the right tail. Everything left of that tick — 99.7% of reality — is covered by our test.

Step 7 — From 1.4 down to the real-world 1.25 (the degenerate & practical cases)

WHAT. The clean statistics, fed our illustrative $\pm 10\%$ inputs, give $\approx 1.4$. The published industry standard (e.g. NASA-STD-7001 / GEVS and MIL-STD-1540 heritage) is 1.25 — historically expressed as a +3 dB qualification margin on the vibration spectrum, which multiplies the RMS level by $\sqrt{2}\approx 1.41$... in energy, but the flat-level factor customarily applied and quoted is $1.25$. The gap from our 1.4 to 1.25 is not a fudge: it reflects that the $0.1/0.1$ figures were worst-case teaching numbers, and that decades of flight data show mature, well-characterised hardware and launch environments scatter less than that.

WHY 1.25 is chosen. Three edge/degenerate cases pin it down:

1. Less scatter than assumed — the $0.1/0.1$ pair is deliberately conservative. When flight-heritage data give a tighter, empirically-measured scatter (roughly ${\sigma }_{\text{mfg}}\approx {\sigma }_{\text{env}}\approx 0.06$ for characterised hardware), then ${\sigma }_{\text{total}}=\sqrt{0.06^2+0.06^2}\approx 0.085$, and $1+3(0.085)\approx 1.25$. So the standard's 1.25 factor is the $3\sigma$ answer once you plug in measured rather than assumed-worst scatter — which is exactly why standards bodies settled there.
2. Zero scatter (perfect world) — if ${\sigma }_{\text{mfg}}={\sigma }_{\text{env}}=0$, then ${\sigma }_{\text{total}}=0$ and Test Level $=\mu =1.0$. No uncertainty ⇒ no margin. This is the sanity check: the whole margin exists only because of scatter.
3. Acceptance vs qualification. The design is proven once at $1.25\mu$ on a qualification unit. Each flight unit then gets an acceptance test at $1.0\mu$ — just enough to catch a manufacturing defect, without overstressing a good unit you actually intend to fly.

PICTURE. A number line from $1.0$ to $1.4$ with the three landmarks: $1.0$ (acceptance / zero-scatter), $1.25$ (qualification, the standard, from measured scatter), and $1.4$ (the $3\sigma$ result for the pessimistic $0.1/0.1$ teaching inputs). The chosen band sits at 1.25.

The one-picture summary

Everything above, in a single frame: the Flight Limit tick blooms into a cloud of $x$-values, the cloud gets a width $\sigma$, two variances combine by Pythagoras into ${\sigma }_{\text{total}}\approx 0.14$, three of those widths reach out to cover 99.7%, and using measured rather than worst-case scatter pulls the standard factor to 1.25.

See also: Verification Methods — Analysis, Test, Inspection, Demonstration · Vibration Testing · Acceptance Testing · Margin Philosophy · Requirements Development · Traceability Matrix · Model Validation