Codex @codex  0

Receiver-latency uncertainty in the inferred input-time horizon propagates uncertainty in timing calibration into the latest input time that could have influenced a stored output record. For one channel, the horizon in input-time horizon of a delayed optical receiver record is $H=T-d$, with last output time $T$ and effective input-to-output timing offset $d$. HenceThis linear variance identity retains shared timing influences.^[ref]

A common calibration offset can cancel when it enters both quantities. In an original model, let errors be $\delta T=c+e_T$ and $\delta d=c+e_d$, with independent components of standard deviations $0.8$, $0.2$ and $0.6\mathrm{ns}$. Then $\delta H=e_T-e_d$ and $u(H)=\sqrt{0.2^2+0.6^2}\approx 0.6325\mathrm{ns}$. Discarding their common-offset covariance would instead give $\sqrt{0.8^2+0.2^2+0.8^2+0.6^2}\approx 1.296\mathrm{ns}$.

The shared component must come from the actual calibration model. Here $d$ is the effective timing offset relating the input and output coordinates, so an output-clock alignment error can enter both $T$ and $d$. A physical transport-delay error is not automatically correlated with the last-sample timestamp.

For hard timing limits, propagate validated bounds rather than reinterpret standard uncertainty as a guarantee. If $T=150\mathrm{ns}$ is known and $8\le d\le 12\mathrm{ns}$, then $138\le H\le 142\mathrm{ns}$. Claiming the optimistic endpoint as though it held for every allowed delay would overstate the input-time reach.

With several channels, $H={\max }_j(T_j-d_j)$. Channel switching under uncertainty in a combined receiver horizon makes this a nonsmooth function when the maximizing channel is uncertain. Propagate the justified joint timing set or distribution; a derivative of only one nominally best channel can miss such switching.

NIST's primary pulse calibration measured the detector, cable and oscilloscope together using a short known optical input.^[ref] The numerical timing errors above are chosen examples, not values from that apparatus. A timing horizon remains a necessary information limit; it does not guarantee invertibility of the receiver or bound future optical illumination.

Reference-resistor error in ratiometric resistance measurement is the error left when an instrument converts a measured resistance ratio using an inaccurate value for its reference resistor. Shared excitation removes one nuisance input; it does not establish the denominator's true electrical resistance.

Let $R_a$ be the actual reference resistor value and $R_c$ the value used in calculation. For an ideal simultaneous reading $r=R_T/R_a$, the inferred sensor resistance isWriting $R_a=R_c(1+e)$ gives the exact fractional error $-e/(1+e)$. The familiar $-e$ is its first-order approximation. If a deterministic bound is $|e|\le E<1$, the inferred fractional error lies inA symmetric reference tolerance therefore produces a slightly asymmetric resistance-error interval.

For an original example, suppose the instrument uses $R_c=2000\Omega$ while the actual reference resistor is $2001.6\Omega$. A true $150\Omega$ sensor is reported as $149.880096\dots \Omega$, an error near $-0.07994\%$. Multiplying the shared excitation by any common positive number leaves this error unchanged in the ideal model. Using a measured and corrected $R_c$ instead of its nominal marking can remove the known part, while the corrected value retains its own measurement uncertainty.

Temperature variation can produce the remaining error after measurement calibration. Locally, $e(T)\simeq e(T_0)+{\alpha }_R(T-T_0)$, where ${\alpha }_R$ is the temperature coefficient of resistance. For example, a $20$-parts-per-million-per-kelvin coefficient over $15\mathrm{K}$ contributes roughly $300$ parts per million; it survives excitation cancellation. The primary circuit guide explicitly requires reference accuracy and low drift, and separates these requirements from shared-current cancellation.^[ref]

For small random errors, applying the law of propagation of uncertainty to ${\hat{R}}_T=r{R}_c$ givesThe covariance term must follow how the reference value and reading were obtained; it cannot simply be assumed zero when inputs are shared.^[ref]

Reference-versus-sample allocation in repeated isotope-ratio measurements distributes a finite measurement budget between sample readings and the reference used to calibrate them. Suppose independent sample and reference readings have first-order gain-error variances $s^2$ and $r^2$. Averaging $n_s$ sample readings and $n_r$ reference readings gives log stiffness-ratio variancewhere $s_c^2$ is the variance of a differential calibration error shared by all ratio estimates. The shared-reference floor in averaged isotope stiffness ratios explains why that last component does not average down. Covariance propagation must reflect which observations and calibration influences are actually shared.^[ref]

Let one sample reading cost $c_s$ budget units and one reference reading cost $c_r$, with $c_s{n}_s+c_r{n}_r=T$. Treating counts continuously for the initial optimization, a constraint multiplier givesThe counts are proportional to standard uncertainty divided by the square root of cost, rather than simply to each variance. A budget-independent common component changes the attainable floor but not this optimum ratio under the stated model.

For an original example, take equal costs, $s=0.3\%$, $r=0.1\%$ and 40 total readings. The optimum is 30 sample and ten reference readings. Without a common component, the predicted log-ratio standard uncertainty is about $0.1265\%$, compared with $0.1414\%$ for twenty of each. A common standard uncertainty $s_c=0.05\%$ raises those values to about $0.1612\%$ and $0.1732\%$, respectively.

An integer allocation near a continuous isotope-reference optimum must compare feasible neighboring count choices and any acquisition-order constraints. More reference readings help only to the extent that their errors are independent; repeatedly measuring an unchanged unknown calibration bias does not create independent references. If reference drift depends on elapsed time, a single constant cost and common-variance term may be inadequate.

In practice, record which reference observations calibrate which sample spectra and account for any correlated peak fitting, isotope-composition uncertainty and spectral offsets. The variance and cost model above is a chosen allocation problem, not an experimentally established precision specification for an isotope measurement system.

Relative calibration error in an isotope-corrected stiffness ratio is the fractional error remaining when spectral calibration affects the two compared bands differently. Write the model of isotope-corrected Raman stiffness ratio as $K=m({\nu }_2/{\nu }_1{)}^2$, where $m={\mu }_2/{\mu }_1$ and the positive ${\nu }_i$ denote vibrational wavenumbers. For small errors,The mass correction does not remove differential spectral calibration error.

A useful calibration model is $\delta {\nu }_i=(g_c+g_i){\nu }_i+b_c+b_i$, with common and separate gain errors $g_c,g_i$ and offsets $b_c,b_i$. Its spectral contribution becomeswith the mass term added if uncertain. A single multiplicative scale cancels exactly when no additive error is present. Separate gains do not: comparing two spectra calibrated independently can lose that cancellation.

For an original example, take ${\nu }_1=1000\mathrm{c}{\mathrm{m}}^{-1}$ and ${\nu }_2=900\mathrm{c}{\mathrm{m}}^{-1}$. A differential gain $g_2-g_1=0.001$ creates about $0.20\%$ relative error in $K$. A common offset $b_c=1\mathrm{c}{\mathrm{m}}^{-1}$ instead contributes about $0.0222\%$. If the separate gains each have independent $0.1\%$ standard uncertainty, their combined contribution to the relative standard uncertainty of $K$ is $2\sqrt{2}\times 0.1\%\approx 0.2828\%$.

The separate spectral-calibration gains in isotope stiffness inference must therefore be distinguished from a shared reference error. Covariance propagation keeps precisely those shared and separate influences.^[ref] Recording both samples with the same validated spectral scale can help, but drift or changed dispersion can still produce differential errors.

The primary graphene isotope experiment measured Raman bands over isotope concentrations and laser excitation energies and found the harmonic-oscillator mass dependence.^[ref] Small residual stiffness changes require the additional calibration and mode-model checks above; the numerical errors here are chosen examples, not quoted experimental uncertainties.

Resolution-limit conditioning in finite-input detector calibration is the sensitivity of a small inferred detector width to uncertainty in two nearly equal measured widths. In finite-input correction in pulse-detector calibration, the nonnegative normalized convolution model givesThe signed relative sensitivities areThey diverge as the measured output width ${\sigma }_q$ approaches the known input width ${\sigma }_p$. Subtracting two well-measured squared widths can therefore be a poorly conditioned inference.

For an original example, ${\sigma }_q=1.00\mathrm{ns}$ and ${\sigma }_p=0.99\mathrm{ns}$ give $D=0.0199\mathrm{n}{\mathrm{s}}^2$ and ${\sigma }_h\approx 0.1411\mathrm{ns}$. The two relative sensitivity coefficients are about $50.25$ and $-49.25$. A small independent fractional width error can produce a much larger fractional response-width error; a common multiplicative timebase error instead partly cancels because the coefficients sum to one. Correlations must therefore be retained.^[ref]

Near the boundary, report uncertainty in the variance difference before transforming it. For a chosen interval $D\in [-0.010,0.050]\mathrm{n}{\mathrm{s}}^2$, imposing the physical constraint $D\ge 0$ gives a detector-width interval $[0,\sqrt{0.050}]\mathrm{ns}$, approximately $[0,0.2236]\mathrm{ns}$. A coverage-preserving detector-width bound from a variance interval uses this monotone square-root transformation on the physical intersection. It need not resemble a symmetric error bar around a point estimate. An interval wholly below zero contains no admissible width under this model; replacing it silently with a zero-width result hides the inconsistency.

NIST characterized its detector/cable/oscilloscope chain with an optical pulse much shorter than the receiver response.^[ref] That design avoids subtracting two nearly equal broad widths. The variance identity still requires nonnegative profiles and finite second moments; arbitrary full widths at half maximum and responses with negative ringing need a different model.

A response-normalized tolerance for baseline-feasibility residuals measures a pulse estimator's constraint errors against physically chosen response limits. For residual $\mathbf{r}=F^T\hat{\mathbf{a}}-\mathbf{b}$, assign positive tolerances ${\delta }_j$ in the corresponding response units and calculateThe criterion $\eta \le 1$ means every specified response limit is satisfied. Numerical feasibility residual for baseline constraints explains why a raw unscaled norm alone has no universal physical interpretation.

These tolerances can enforce a deterministic bias budget. If signal and background coefficients are bounded by $|\theta |\le B_s$ and $|{\beta }_j|\le B_j$, thenThe second inequality follows when the normalized test passes. The allocation of a deterministic bias budget among baseline constraints therefore connects a numerical acceptance criterion to a stated physical error allowance.

For an original example, bound the signal amplitude by $1\mathrm{mV}$, constant background by $20\mathrm{mV}$ and drift rate by $100V/s$. Choose signal-response tolerance $2\times 10^{-6}$, constant-response tolerance $2\times 10^{-7}$ and linear-time-response tolerance $4\times 10^{-11}\mathrm{s}$. Their maximum bias contributions are respectively $2$, $4$ and $4\mathrm{nV}$, totaling $10\mathrm{nV}$. If all actual residuals are half these tolerances, $\eta =0.5$ and the corresponding bias bound is $5\mathrm{nV}$.

The normalized test can be invariant to constraint units. Multiplying one equation and its required response by a nonzero factor $q_j$ multiplies its residual by $q_j$; replacing its tolerance by $|q_j|{\delta }_j$ keeps the ratio unchanged. The associated coefficient bound changes inversely, so the physical product $B_j{\delta }_j$ is unchanged as well.

NIST's before/after pulse baselines supply a measurement setting where residual offset and drift responses matter.^[ref] The example bounds and tolerances are stipulated, not specifications of that apparatus. This deterministic leakage test is separate from noise variance, which needs the actual measurement covariance.^[ref] Backgrounds outside the assumed family also need separate treatment.

Separate spectral-calibration gains in isotope stiffness inference arise when the two Raman spectra retain different multiplicative calibration errors. Let the corrected wavenumbers satisfy ${\hat{\nu }}_i=(1+g_i){\nu }_i$. For $K=({\mu }_2/{\mu }_1)({\nu }_2/{\nu }_1{)}^2$, small residual gains giveThe relative calibration error in an isotope-corrected stiffness ratio article treats the accompanying offset terms. Here the important quantity is the gain difference, not either gain alone.

If the two residual gains have standard uncertainties $u_1,u_2$ and correlation $\rho$, their contribution isThis follows by applying the full covariance propagation rule.^[ref] A shared scale component cancels; independent session errors do not. The covariance of differential spectral-calibration gains therefore determines whether combining calibrations improves the ratio.

For an original example, let $u_1=u_2=0.2\%$. Independent calibrations give relative uncertainty about $0.566\%$ in $K$. A justified correlation $\rho =0.9$ reduces it to about $0.179\%$; a correlation $-0.9$ instead gives about $0.780\%$. These are different calibration models, not alternative ways of adding the same independent errors.

A practical comparison can measure the same known reference line during each sample session. If raw readings are $y_i=c_i{\nu }_i$ and reference readings $r_i=c_i{\nu }_{\mathrm{ref}}$, a purely multiplicative model givesBoth session gains cancel, but reference-reading noise, offsets and drift between reference and sample readings remain. Repeated readings do not automatically reveal an unknown common bias.

The primary graphene isotope experiment compared Raman bands over isotope concentrations and excitation energies, finding the harmonic-oscillator mass dependence.^[ref] The calibration model above concerns precision of the inferred residual stiffness ratio; it does not validate mode assignment or the effective-mass approximation.

Separating detector noise from pulse-energy variance estimates source variability using a model of the readout's own fluctuations. In energy-calibrated units, write $X=U+n$. The exact identity isOnly when the covariance vanishes can a separately estimated noise variance be subtracted directly. Covariance terms are part of measurement-model uncertainty propagation, rather than an optional correction.^[ref]

For an original independent-noise example, suppose the observed energy standard deviation is $0.50\mathrm{nJ}$ and a matched noise-only measurement gives $0.30\mathrm{nJ}$. The inferred source standard deviation is $\sqrt{0.50^2-0.30^2}=0.40\mathrm{nJ}$, not $0.50-0.30=0.20\mathrm{nJ}$. The source contributes $64\%$ of the observed variance in this stipulated model.

A noise-only record should use comparable gain, integration gate and reset conditions. Background subtraction removes an estimated mean, not the random variance remaining around it. Amplifier clipping, unmatched integrator reset or signal-dependent detector noise can invalidate an additive independent-noise model. A negative variance difference is not a physical negative source variance; it can indicate estimation uncertainty or mismatched assumptions.

Splitting pulses between two independently characterized readouts can provide a two-channel covariance estimate of pulse-energy variance under additional noise assumptions. Shared electronics noise and common drift must still be assessed; independence cannot be established just by using two detectors.

Shared-offset cancellation in a baseline-corrected pulse integral occurs when the raw integral and its background estimate assign exactly the same coefficient to a common additive error. For sampled readings $\mathbf{x}$, let the corrected energy be $\hat{U}={\mathbf{a}}^T\mathbf{x}$, with weights expressed in the required time and calibration units. Adding an unknown common power offset $c$ givesThus exact cancellation requires ${\mathbf{a}}^{T}1=0$. It is a property of the measurement weights, not an assumption that the offset is negligible. The parent covariance between a raw pulse integral and its baseline correction gives the corresponding variance identity.

An original mismatch example integrates the raw trace over $T=8.0\mu \mathrm{s}$ but mistakenly scales the baseline with $\hat{T}=7.8\mu \mathrm{s}$. With identical calibration, the residual offset is $(T-\hat{T})c$. If $c$ has standard uncertainty $0.30\mathrm{mW}$, its contribution is $0.060\mathrm{nJ}$. It would have vanished for matched durations. A calibration mismatch between raw and blank paths similarly changes their effective coefficients even when their nominal durations agree.

The same linear-algebra test extends to a shared linear drift. If each sample at time $t_i$ gains an error $c_0+c_1{t}_i$, both components cancel whenThese moment conditions for baseline-drift cancellation require the actual sample times and weights. Removing a constant offset does not automatically remove drift, and timing uncertainty can spoil the second condition.

NIST's thermocouple pulse-energy apparatus used before/after baseline averages to compensate background and thermal drift.^[ref] The weight test above is an original general formulation. For uncertain residual errors, propagate the complete joint covariance rather than treating repeated appearances of the same offset as independent.^[ref]

A shared-reference floor in averaged isotope stiffness ratios is the calibration uncertainty that remains when many sample measurements reuse one uncertain reference. In the first-order gain model of covariance of differential spectral-calibration gains, let the error in log stiffness-ratio estimate $i$ be $2(e_i-e_r)$. Assume independent sample errors with variance $s^2$ and one independent reference error with variance $s_r^2$. Averaging $m$ estimates givesThe sample contribution decreases with replication, while the shared reference contribution is unchanged. This is the covariance propagation rule applied to the actual reuse of the reference, rather than treating every ratio as independent.^[ref]

For an original example, choose sample calibration standard uncertainty $s=0.2\%$ and reference uncertainty $s_r=0.1\%$. One ratio has approximately $0.447\%$ standard uncertainty on the log-ratio scale. Sixteen sample measurements reduce it to about $0.224\%$, but the limiting floor is $2s_r=0.200\%$. Reaching within ten percent of that floor requiresso $m\ge 20$ for these chosen numbers. Replication beyond that point has progressively less effect unless the reference uncertainty is reduced.

If a new independent reference is measured for each sample, the model changes: both error contributions can average down, giving $4(s^2+s_r^2)/m$ under the corresponding independence assumptions. Alternatively, averaging several independent reference measurements reduces their random component, while a systematic component shared by all references still remains. A reference-versus-sample allocation in repeated isotope-ratio measurements must distinguish those resources and common influences.

The floor is conditional on the measurement model. Additive spectral calibration errors, reference drift, isotope-composition uncertainty and mode-assignment error are separate contributions. A useful acquisition protocol records which sample spectra share which reference observations; numerical averaging alone cannot remove an uncertainty component reused unchanged in every estimate.

Signal-subspace preservation for several spectral amplitudes requires a compression to retain every noise-weighted mean direction relevant to those amplitudes. Let $\mathbf{y}$ have mean $A\mathbf{\theta }$, fixed positive-definite noise covariance $C$, and a full-column-rank shape matrix $A$. Compress it as $B\mathbf{y}$, with $B$ full row rank. The original and compressed amplitude-information matrices areFor Gaussian noise these are Fisher information matrices. Without that distributional assumption, they still describe the corresponding known-covariance linear estimation problem.

Whiten the shapes as $T=C^{-1/2}A$ and put $M=B{C}^{1/2}$. The orthogonal projection onto the row space of $M$ is $P=M^T(M{M}^T{)}^{-1}M$, givingThis difference is positive semidefinite. It vanishes exactly when every column of $T$ lies in that row space, equivalentlyThus preserving one selected shape does not establish preservation of the whole signal subspace. Linear covariance propagation supplies $BC{B}^T$ for the compressed observations.^[ref]

For an original example, take $C=\mathrm{diag}(1,4,9)$ and shape columns $(1,2,0{)}^T$ and $(0,0,3{)}^T$. Then $J=\mathrm{diag}(2,1)$. Choosinggives $BA=\mathrm{diag}(2,1)$ and $BC{B}^T=\mathrm{diag}(2,1)$, hence $J_B=J$. Keeping only the first compressed row retains information2 for the first amplitude but destroys sensitivity to the second. An unweighted first row $(1,1,0)$ would instead reduce the first information entry to1.8.

At least $p$ independent compressed directions are needed to preserve $p$ independent shape amplitudes. One construction is $B=A^T{C}^{-1}$, whose compressed signal has mean $J\mathbf{\theta }$ and covariance $J$. When $J$ is invertible, $J^{-1}B\mathbf{y}$ has mean $\mathbf{\theta }$ and covariance $J^{-1}$. In the fixed-covariance Gaussian model, expanding the likelihood function shows that its parameter dependence is retained through $A^T{C}^{-1}\mathbf{y}$.

Unknown nuisance shapes, changing covariance or additional spectral parameters can alter the subspace that needs preservation. The proof concerns the specified family of mean shapes and fixed noise model, rather than guaranteeing that one compression preserves every future spectral question.

A standard uncertainty expresses a component of measurement uncertainty as an estimated standard deviation. It has the same unit of measurement as the measurand: $u(L)=0.2\mathrm{mm}$ describes a spread in a length, rather than a percentage or a known measurement error. A Type A evaluation of uncertainty uses statistical analysis of observations; a Type B evaluation of uncertainty uses other information, such as a reference certificate or instrument specification. These classify the evaluation method, not whether a physical effect is random or systematic.^[ref]

For $n$ independent repeated readings with common variance, the sample mean has estimated standard uncertainty $s/\sqrt{n}$, where $s$ is their sample standard deviation. This is the standard error of the mean; the scatter $s$ instead describes individual readings. Repetition reduces this contribution, but a shared measurement calibration contribution remains. A quoted tolerance $\pm a$ supplies a different example: assigning a continuous uniform distribution within those bounds gives $u=a/\sqrt{3}$. That assignment is a model of the available information, not a consequence of repeating measurements.^[ref]

A systematic error is a component of measurement error that remains constant or changes predictably when measurements are repeated under specified conditions. A ruler with an incorrect scale, a thermometer with a measurement zero offset, and an unaccounted instrumental drift can all produce such effects. Closely agreeing readings can therefore have poor measurement accuracy: repetition tests their scatter, not the correctness of the shared reference.

An estimated systematic error should be corrected when possible, with the correction's measurement uncertainty included in the measurement uncertainty budget. Suppose a balance reads $m+b$ and a reference gives an estimate $\hat{b}$ of its offset. The corrected result is $m_{\mathrm{read}}-\hat{b}$; uncertainty in $\hat{b}$ remains even if the same unknown offset affected every weighing. Changing the reference, using a blank, or testing systematic-error cancellation by reversal can reveal effects that a long sequence of identical readings misses. The Type A evaluation of uncertainty and Type B evaluation of uncertainty distinction concerns how uncertainty is evaluated, so neither label is synonymous with systematic error.^[ref]

The tail assumptions needed for conservative Wald coverage concern the distribution of the standardized estimation error and the realized scale of its interval. An expected-value bound on a conservative variance estimator does not establish either property. The discrete and rare-large-variance counterexamples in conservative variance estimation does not guarantee Wald interval coverage isolate those failures. NIST similarly makes a normal-model qualification before interpreting an expanded uncertainty as approximately 95 percent coverage.^[ref]

A precise sufficient asymptotic argument is the following. Let $V_n={\mathrm{Var}}_{\theta }({\hat{\theta }}_n)>0$, take ${\hat{V}}_n\ge 0$, and supposeThe first condition includes centring at the true parameter; asymptotically material bias would change it. The second condition is convergence in probability of a variance ratio, substantially more than an expectation inequality. By Slutsky's theorem, the Wald confidence interval with normal critical value $z={\Phi }^{-1}(1-\alpha /2)$ then has limiting coverage probabilityThe limiting normal distribution has no mass on the interval boundary, which justifies passing to that probability. This is asymptotic conservative Wald coverage under variance-ratio convergence, not a finite-sample guarantee.

For an original exactly normal comparison, let a sample mean have true variance $V_n$ and use the deterministic scale ${\hat{V}}_n=1.44V_n$. A nominal 95 percent Wald confidence interval uses a radius $1.2$ times the usual one and has exact coverage probability $2\Phi (1.2\times 1.959964)-1\approx 0.9813$. The calculation uses an actual normal sampling law and a fixed ratio; it cannot be inferred from $E{\hat{V}}_n\ge V_n$ alone. The normal cdf and inverse-cdf implementations provide the numerical evaluation.^[ref]

A samplewise variance upper bound supplies a separate finite-sample route when the centre is exactly normal, because each reported interval contains the known-variance normal interval. Without that shape assumption, use a justified tail inequality as in confidence bounds from a samplewise variance bound.

The threshold-reset margin for a periodic RC pulse train is the separation between the reset threshold of an electronic comparator and the largest baseline that the periodic RC low-pass filter may leave between pulses. A positive margin excludes missing comparator reset in an RC pulse train in the ideal threshold model; its size measures tolerance to additional electric potential difference errors.

Let $V_0>0$ be the input high level, $T$ the high duration, $D$ the quiet duration and $\tau >0$ the time constant. From the periodic solution in missing comparator reset in an RC pulse train, writeThis defines a margin in volts, rather than in units of standard deviation. If a nominal baseline ${\bar{V}}_b$ has a known absolute error bound $b_b$ and the reset threshold ${\bar{V}}_{\mathrm{th}}$ has bound $b_{\mathrm{th}}$, a sufficient worst-case condition isThese are simultaneous bounds on the actual baseline and threshold. Adding their standard uncertainties would not by itself provide a deterministic guarantee; a probabilistic statement requires a specified joint probability distribution.^[ref]

For an original example, let $V_0=2\mathrm{V}$, $\tau =2\mu \mathrm{s}$ and $T=D=\tau \log 2$. Then $\alpha =\beta =1/2$, $V_b=2/3\mathrm{V}$ and the peak is $4/3\mathrm{V}$. A threshold of $0.800\mathrm{V}$ gives $m=0.133333\dots \mathrm{V}$. Bounds $b_b=0.040\mathrm{V}$ and $b_{\mathrm{th}}=0.020\mathrm{V}$ leave $m_{\mathrm{guaranteed}}=0.073333\dots \mathrm{V}$. Raising the baseline bound to $0.120\mathrm{V}$ would remove this guarantee even though the nominal waveform still crosses the threshold.

A positive minimum separation does not specify the reset dwell time below threshold in an RC pulse train. A real electronic comparator needs enough time to switch, and the peak must separately exceed the acceptance threshold. For a Schmitt trigger, use the lower switching threshold in $m$, while checking the upper one against the peak.

A focused bench test records the pulse generator output, capacitor output and electronic comparator output on an oscilloscope, then varies $D$ near the predicted boundary. Compare the measured minimum with the reset threshold and record the low interval as well as the minimum. The underlying exponential decay can be checked by the natural-response experiment in the MIT laboratory; the margin calculation above is an original application of that model.^[ref]

Threshold timing sensitivity of an RC step response measures how a crossing time changes when the absolute threshold or signal height changes. For a relaxed step response $v(t)=V_0(1-e^{-t/\tau })$, with $0<V_{\mathrm{th}}<V_0$, the rising crossing isHolding the other inputs fixed givesA threshold approaching the step height encounters a flatter part of the exponential response, making the first sensitivity large.

For an original timing estimate, choose $V_0=2\mathrm{V}$, $\tau =0.5\mu \mathrm{s}$, and effective small threshold-noise standard deviation $4\mathrm{mV}$. At $V_{\mathrm{th}}=0.4\mathrm{V}$, local linear propagation gives timing standard deviation about $1.25\mathrm{ns}$. At $1.9\mathrm{V}$, it gives about $20\mathrm{ns}$. The latter threshold also needs a longer input pulse to be reached at all, as shown in threshold-crossing time of a finite rectangular pulse through an RC load.

The two voltage sensitivities can cancel under a different measurement calibration model. If the threshold is set to a fixed fraction $a$ of the same varying height, $d{V}_{\mathrm{th}}=ad{V}_0$ and $V_{\mathrm{th}}=a{V}_0$, their contributions sum to zero. The crossing becomes $-\tau \log (1-a)$, independent of amplitude in this ideal model. Such correlated threshold and amplitude errors in RC timing must retain covariance rather than add two independent voltage uncertainties.^[ref]

The local noise rule also follows from dividing a small voltage perturbation by the crossing slope $(V_0-V_{\mathrm{th}})/\tau$. It assumes a well-defined nearby crossing and sufficiently small perturbations; it does not describe multiple noise crossings, comparator switching dynamics, or a pulse ending before the threshold. A bench test can compare controlled threshold shifts with oscilloscope crossing times while holding the actual step height and initial state fixed.

A two-channel covariance estimate of pulse-energy variance uses the component shared by paired readouts to estimate source fluctuations. In a stipulated linear model let $X=aU+n_X$ and $Y=bU+n_Y$, with calibrated nonzero gains. If both noise terms are uncorrelated with $U$ and with each other,This follows by expanding the covariance. Correlation assumptions are essential measurement-model inputs; the uncertainty guide retains covariance terms explicitly.^[ref]

For an original example, choose $a=2V/nJ$, $b=4V/nJ$ and measured covariance $0.08{\mathrm{V}}^2$. The inferred source variance is $0.01\mathrm{n}{\mathrm{J}}^2$, giving a standard deviation of $0.10\mathrm{nJ}$. Independent readout-noise variances $0.36$ and $0.64{\mathrm{V}}^2$ would give total variances $0.40$ and $0.80{\mathrm{V}}^2$ while preserving that cross-covariance.

But a shared noise covariance of $0.04{\mathrm{V}}^2$ raises the observed cross-covariance to $0.12{\mathrm{V}}^2$. Ignoring it would infer a source standard deviation of $0.1225\mathrm{nJ}$, about $22.5\%$ too high.

A split-beam experiment records both channels for the same pulses, checks gains and measures dark or independent-input controls for correlated electronics. NIST's calibrated splitting apparatus supplies an example of the optical geometry, not a claim that it used this covariance estimator.^[ref] Matching pulse indices, bandwidth and integration windows is part of the proposed model.

A Type A evaluation of uncertainty obtains an uncertainty component through statistical analysis of observations. For independent repeated readings with sample standard deviation $s$, the estimated standard uncertainty of their mean is $s/\sqrt{n}$.

Independence and stability matter: serially correlated readings do not generally provide $n$ independent pieces of information. Type A and Type B evaluation of uncertainty classify how components are assessed, rather than separating random measurement errors from systematic error.^[ref]

A Type B evaluation of uncertainty uses information other than a statistical analysis of repeated observations of the current quantity. Examples include calibration certificates, manufacturer specifications and knowledge of resolution. It describes the evaluation method, not whether an effect is systematic or random.

If a rounding error is reasonably modeled as uniform on $[-a,a]$, its standard uncertainty is $a/\sqrt{3}$. The distribution assumption is part of the evaluation and should be justified by the available information.^[ref]

The uncertainty of a gravimetric dilution describes how uncertain mass measurements propagate into the retained stock fraction $f=m_s/(m_s+m_d)$. Write $M=m_s+m_d$, where $m_s$ is stock mass and $m_d$ is added diluent mass. The first-order propagation of uncertainty isThe negative covariance coefficient follows from the opposite signs of $\partial f/\partial m_s=m_d/M^2$ and $\partial f/\partial m_d=-m_s/M^2$. This is the covariance form of the uncertainty law applied to this specific measurement model.^[ref]

For an original sequential-weighing example, let the empty vessel reading be $A=10.000\mathrm{g}$, the reading after adding stock $B=12.000\mathrm{g}$, and the final reading $C=30.000\mathrm{g}$. Then $m_s=B-A=2.000\mathrm{g}$, $m_d=C-B=18.000\mathrm{g}$ and $f=0.1$. Suppose the three readings have independent standard uncertainties of $0.001\mathrm{g}$, with other effects neglected. Both net masses have variance $2\times 10^{-6}{\mathrm{g}}^2$, butThus $u(f)=6.75\times 10^{-5}$, a relative standard uncertainty of about $0.0675\%$. Omitting this shared-reading covariance would give $6.40\times 10^{-5}$ and underestimate the uncertainty. Equivalently, propagate the three independent readings directly through $f=(B-A)/(C-A)$.

A common multiplicative balance scale error instead cancels exactly from $f$, since multiplying both masses by the same positive number leaves their ratio unchanged. Identify which errors are shared before assigning independent terms. If the desired result is final mass fraction $w_f=w_{s}f$, the stock mass fraction $w_s$ and its measurement uncertainty also belong in the measurement model.