Documentation sections
API Changelog
Every change to the public API that a caller could notice, newest first. Each entry is dated the day it was announced here; a change that breaks existing callers is announced no later than the release that makes it, and is marked Breaking. 29 of the 81 entries below are breaking.
rf-link-budget takes 0.5 dB of atmospheric loss when atmosphericLoss is left out
POST /api/py/v1/calculatePOST /api/py/v1/calculate/batch
A call that left out
atmosphericLosswas computed with 0 dB while itsprovenance.inputsreported the documented default, 0.5 dB. It is now computed with 0.5 dB, as the calculator page and the provenance say, sorxPower,linkMarginandmaxRangeKmmove by that half decibel for such a call. A call that sends the input is unchanged.Every other input of every calculator was checked the same way: left out, it is either refused (400, nothing charged) or computed with its documented default.
GET /v1/usage can return calls per day
GET /api/py/v1/usage
Pass
?days=N, from 1 to 90, and the response gainsdaily: one{date, calls}entry for each of the last N UTC days, oldest first and ending today. A day'scallsare summed over every key of the account, net of calls given back, and a day with none reads 0. Days before 2026-09-24, when per-hour counts began, read 0.Without
daysthe response is unchanged. Reading usage is still never metered.A Touchstone upload's port count is read from its rows before its name
rf_cascadesparam_pipelineimpedance_matcheye_diagram
Where a file's count of values admits more than one port count, its rows decide before its extension: three one-port points in a file named
.s2pare one port, not one two-port point, and a name the data contradicts is reported (port_count_mismatchin rf_cascade). A[Number of Ports]keyword still decides first.The eye diagram no longer fails on a file whose extension names more ports than its data holds.
The filter Monte Carlo runs up to 1,000 trials free and 100,000 on Pro and API, within a work budget
POST /api/py/v1/jobs (filter_monte_carlo)POST /api/py/v1/jobs/estimateGET /api/py/v1/jobs/typesMCP simulate_filter_monte_carlorftools-io SDK
monteCarloIterationsgoes to 1,000 on the free tier (was 500) and 100,000 on Pro and API (was 10,000). The job's time ceiling is 180 s (was 120 s).A run's cost is its trials times the ladder evaluations each takes, set by its parts and
passCriteria. A run predicted over its lane's budget (120 s on the paid lane, 60 s on the free) is refused at submission and by the estimate witherrorKind: "too_large", before anything is charged. The detail namesmonteCarloIterationsand gives the cost, the limit and, inallowed.max, the most trials that fit. 100,000 trials of a 5-part ladder on three criteria fit; a 20-part ladder on eight criteria fits up to 53,304.The estimate reports
withinBudget, abudgetof kindladderEvaluations, andworkastrialsandladderEvaluationsPerTrial, which a job'scomputeCost.workrecords too.An RF cascade's stage list is checked at submission
POST /api/py/v1/jobsPOST /api/py/v1/jobs/estimaterf_cascade
A malformed
stages(not JSON, empty, a stage that is not an object, a value that is not a finite number) is refused with a 400 naming the stage and key, before anything is charged. It used to be accepted and then fail as a fault.Stage values are bounded:
gain_db±1000 dB,nf_db0 to 1000 dB,iip3_dbmandp1db_dbm±1000 dBm, and the chain's gain after every stage within ±1000 dB.NaN,Infinityandtrue/falseare refused; numbers written as text are still read.A stage key that differs from a known one only in case (
gain_dB,IIP3_dbm,Name) is refused, naming the stage and the key it should be; it used to be ignored and the stage ran on the default. Any other key a stage does not take (notes,part) is ignored with anunknown_stage_keywarning.A stage whose
sparamFilewas not uploaded with the job is refused, not computed with its typed gain.No intercept is null, not 99 dBm
POST /api/py/v1/jobsrf_cascade
A passive stage's IIP3, and a chain's IIP3, OIP3, P1dB and SFDR where no stage limits them, are
null(they were 99 dBm, and 99 + gain). The Monte Carlo IIP3 percentiles and histogram arenullthere and the IIP3 yield is 100%.A passive stage no longer adds a 99 dBm intercept: a 150 dB pad before a 0 dBm-IIP3 amplifier now reads 150 dBm, not 99.
RF cascade: passive noise from files, Monte Carlo sign, exact intercepts, warnings
rf_cascade
A passive stage whose gain comes from a file takes its noise figure from that gain.
The Monte Carlo moves a passive stage's noise figure with its loss; the sign was reversed.
The input IIP3 and P1dB are computed exactly however low they fall (three 300 dB amplifiers: −600 dBm and −610 dBm, which read −270 dBm before); a −270 dBm floor remains only as a guard where the sum behind a point is not a positive, finite number, which the stage bounds keep out of reach, so a P1dB no longer fails the job.
Warnings
unknown_stage_type,unknown_stage_keyandfile_not_uploaded.Impedance matching snaps on a log scale and sweeps the power-wave reflection
POST /api/py/v1/jobsimpedance_match
E12, E24 and E96 parts are the standard value nearest on a log scale, the way the series are spaced; exactly halfway, the smaller one. 2.099 nH is now 2.2 nH, not 2.0 nH. Where the two rules differ, the parts, the return loss and the branch
autopicks may change.|S11|, the Smith trajectory, the worst-case return loss and the bandwidth use the power-wave reflection (Z − Zs*)/(Z + Zs), as the figures at f0 do. A real source is unchanged; a complex source, typed or read from a file, reads a different |S11|.
Impedance matching states its design frequency, its load's frequency model and its figures at f0
POST /api/py/v1/jobsPOST /api/py/v1/jobs/estimateimpedance_match
Each new parameter is optional; a request without them gets the answer it got before.
designFreq(Hz): f0, stated apart fromfreqStart–freqStopand inside them; absent, the band's centre.loadModel:constant(absent),series_rcorseries_rl(fitted at f0; a reactance of the wrong sign is refused) orfile(the uploaded load's own frequency points).branchPreference:any,lowpassorhighpass, with abranch_preference_unmetwarning where no such branch exists. A.s1pload file is accepted.New results:
designFreqHz,loadModel,unmatchedLoad,performance.idealand.standard(return loss, VSWR, mismatch loss removed and the RL ≥ 10 dB bandwidth with its edges and share of f0, resolved by bisection),branchandqDefinitions.The eye diagram takes a launch swing, a CTLE and a mask
POST /api/py/v1/jobs (eye_diagram)MCP simulate_eye_diagram
Each new parameter is optional, and a request without them gets the answer it got before.
launchSwing(mV peak to peak, default 2000, the ±1 V every earlier result assumed) scales every voltage;eyeHeightPctOfSwinggives the eye height as a share of it.equalization: ctleapplies the IEEE 802.3 Annex 93A continuous-time filter.ctleDcGain(−12 to 0 dB) fixes its gain; blank keeps the best of −12 to 0 dB in 1 dB steps.ctlereports the gain, the peaking at Nyquist, the sweep and the eye before it.maskType(hexagon or rectangle),maskHalfHeight(mV) andmaskWidth(UI) givemask.pass,margin_mVand the violations.samplingPhase(UI after the crossing) andlossFrequency(Hz) addsamplingandlossAtFrequency.Every result carries
channelFile,uiMeasuredandtraceCount.Radar detection takes a random seed
POST /api/py/v1/jobs (radar_detection)provenance.seed
randomSeed(a whole number, 0 or more) replaces the fixed Monte Carlo seed 42 and is recorded asprovenance.seed, so a run can be repeated and a what-if resubmitted on the same draws. Without it every result is unchanged.The filter Monte Carlo runs a ladder as built, against criteria you set, and says what to tighten
POST /api/py/v1/jobs (filter_monte_carlo)MCP simulate_filter_monte_carlorftools-io SDK
Each new parameter is optional, and a request without them gets the answer it got before, with new result fields beside it.
parts(1 to 20, source to load:ref,kindL or C,positionseries or shunt,valuein H or F, optionaltolerancein %) simulates the ladder as given, with no E24 rounding.passCriteria(1 to 8:insertionLossoverfStart–fStopat mostmaxDb,cutoff3dBinsidefMin–fMax,attenuationatfreqor overfStart–fStopat leastminDb) replacesrippleSpec_dbandrejectionSpec_db.yieldTarget(%, 95 when blank).New results:
circuit,yieldTargetwith a verdict against the 95 % interval,criteriawith statistics, histograms and each part'svarianceShare(r²),failingTrials(the first 200),sampleTrialsand, on entered criteria,whatIfs: the parts that drive the most-failed criterion, one tolerance step tighter, re-run on the same trials. Resubmitting with the run'srandomSeedand those tolerances gives the same yield.The SMPS control loop takes its compensator by parts, solves for a crossover and gives a verdict
POST /api/py/v1/jobsPOST /api/py/v1/jobs/estimatesmps_control_loop result
Each new input is optional, and a request without it gets the answer it got before.
compensatorType: type2_otawithcompRc,compCc,compChf,compGmandcompVrefdescribes a transconductance Type II by its parts, mapped exactly onto K, f_z and f_p.compVrefaboveVoutis refused.solveFor: crossoverwithtargetCrossover(Hz, default Fsw/10),resistorSeries(E96) andcapacitorSeries(E12) places R_c, C_c and C_hf in peak-current mode and reports what they achieve.minPhaseMargin(45°) andminGainMargin(10 dB) set the verdict and the Monte Carlo yield.New results:
verdict,polesZeros,nominal.phaseCrossoverHz,nominal.loopGainAt100HzDb,compensatorParts,solve,runSummaryandbodeAboveHalfFsw, the responses to 2·Fsw, where no margin is read.Sixteen timing calculators judge their timing against a budget
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batchprovenance.validRange
Each new input is optional, and a request without it gets the answer it got before, except where the entries below say otherwise.
uart-baud-rate:oversampling(16 or 8),divisorFraction(0 a whole divisor, 1 a fraction in steps of 1/oversampling) anderrorBudget(%, 2, which the baud-error warning follows);divisorSet,baudError_pct(signed) andstopSampleOffset_bits.can-bus-timing:sjw(tq, 1);bitRateActual,bitRateError_pct,samplePointActualandoscTolerance_pct, the clock tolerance ISO 11898-1 allows each node.lin-bus-timing:slotTime(ms, 10);frameTimeMax, 1.4 × the nominal frame (LIN 2.x).modbus-frame-timing:charGap(ms, 0), which the frame time now counts;t15,charGapCharsandt15Chars.spi-timing:riseTime(2.2τ) andedgeShare_pct.i2s-timing:mclk(MHz, 16.9344) andrateBudget(%, 0.1);mclkDivider,bclkActual,sampleRateActualandsampleRateError_pct.pwm-duty-cycle:timerClock(MHz, 16) andrequiredBits(8);periodCounts,onCounts,dutyActual,frequencyActualanddutyResolution_bits.555-timer:targetTime(ms, 2),errorBudget(%, 5),rToleranceandcTolerance(%, 0);timeMin,timeMaxandtimingError_pct.rc-time-constant:rTolerance(%, 1),cTolerance(%, 5) andtauBudget(%, 10), with a warning beyond it;tauMinUs,tauMaxUsandtauErrorPct.inductor-energy:requiredTime(µs, 20).battery-charge-time:requiredTime(h, 3).supercapacitor-backup:holdUpTime(s, 60).motor-inrush-current:dipLimit(%, 10).via-step-response:slowdownBudget(%, 10), which the slowdown warning now follows. Each is refused at or below 0.inrush-current-limiter:ntcR25(Ω, 0 for the required value) andntcTolerance(%, 20);peakInrushMax. WithntcR25set,peakInrush,timeConstantandlimitingResistorfollow the fitted thermistor.pid-tuning:overshootBudget(%, 25);overshootPctandsettlingTimefrom a simulated closed-loop step.Fourteen frequency calculators count their products against a band
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batchprovenance.validRange
Each new input is optional, and a request without it gets the answer it got before, except where the entries below say otherwise.
mixer-spur-calculator:rfBandLow/rfBandHigh(MHz, 902/928),ifBandwidth(MHz, 10) andnearMargin(MHz, 50);spursInBand,spursNearBand,nearestSpur,imageFreq,imageToBandandloToBand, from every RF input that converts to the IF, f_RF = (m·f_LO ± f_IF)/n (m the LO harmonic, n the RF harmonic, as the older count has them), m + n up tomaxOrder.intermodulation-distortion:bandLow/bandHigh(MHz, 88/108);productsInBand.vibration-phase-noise:offsetBandLow/offsetBandHigh(Hz, 1000/10000);sidebandsInBand.phase-noise-to-jitter:pointsIntegrated.ism-coexistence:channelWidth1_MHz/channelWidth2_MHz(20/2);overlappingChannels, with the channels spread evenly across the band.channels1andchannels2are limited to 100.via-stub-resonance:dataRate(Gb/s, 10);nyquistFreq,resonancesInBandandresonancesNearBand.chassis-resonance:bandLow/bandHigh(MHz, 30/1000);f011MHzandmodesInBand.room-modes:bandLow/bandHigh(Hz, 20/200);modesInBand,modesNearBandandmodalBandwidth.sampling-nyquist:harmonicOrder(5);aliasesInBand(the tone itself aliased into the signal band),harmonicAliasesInBand(its harmonics folded into it, which matter as far as the converter's SFDR) andlowestAliasingInput.fft-bin-resolution:toneFreq(Hz, 1000) andsidebandSpacing(Hz, 100);toneBin,toneLevelError,resolutionBandwidthandsidebandsUnresolved(null for a tone above f_s/2).am-modulation-index:channelWidth(kHz, 10);powerInChannel,sidebandsInChannelandsidebandsOutside.fm-modulation-index:channelLow/channelHigh(kHz, 98000/98200);occupiedBandwidth,powerInChannel,sidebandsInChannelandsidebandsOutside(null above β = 1000, with a warning).oversampling-snr:signalBandwidth(Hz, 24000);modulatorRate,outOfBandNoiseandfoldBands.digital-filter-order:toneFreq(Hz, 1000) andmaxHarmonic(5);harmonicsInPassband,harmonicsInTransition,nearestHarmonic,butterworthAttenuationandchebyshevAttenuation.can-bus-timing: the prescaler and quanta closest to the bit rate
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
Of every bit of 8 to 25 time quanta (CAN 2.0), with the prescaler on either side of the one it calls for, the closest to the bit rate; then the sample point nearest the one asked for (CiA 601-3); then 16 quanta; then the smallest prescaler. It took the prescaler nearest 16 quanta whatever rate that gave: 36 MHz at 500 kbps was 5 × 14 quanta, 2.86 % fast, and is 3 × 24, exact.
prescaler,nominalBitTime,tqand the segments change where an exact or closer setting exists; every default and preset is unchanged.The sample point lies at most a quantum before the end of the bit, so the segments add up to it (8 quanta at 95 % were 9).
modbus-frame-timing: fixed timers above 19 200 Bd
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
Above 19 200 Bd
t35is a fixed 1.75 ms (Modbus over Serial Line V1.02, section 2.5.1.1), no longer 3.5 characters (0.334 ms at 115 200 Bd 8E1), andmaxFrameRatefollows; the newt15is a fixed 750 µs there.intermodulation-distortion: im3High is a magnitude
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
im3Highis |2f₂ − f₁|, asim3Lowalready was: below f₂ = f₁/2 it was negative (100 and 40 MHz: −20, now 20 MHz).chassis-resonance: exact c and the (0,1,1) mode
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
The speed of light is exact, 29 979 245 800 cm/s, not 3 × 10¹⁰:
fLowest,f101MHzandf110MHzfall 0.07 % (901.39 → 900.76 MHz at the defaults).fLowestincludes (0,1,1), the lowest mode when the length is the shortest side (Pozar, Microwave Engineering, 4th ed., section 6.3): a 10 × 30 × 20 cm box was 1581 MHz and is 900.76 MHz;wavelengthAtResonancefollows.phase-noise-to-jitter: the profile integrated as power laws
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
With
profileMode1 the profile is integrated as power laws between its points, straight lines on log-log axes (Drakhlis, Microwaves & RF, January 2001), and the band ends at the profile's own level there. A linear trapezoid overstated a falling profile, and L₄ was held to the band's end even where it lies beyond it: the default profile at 100 MHz was 0.166 ps and is 0.084 ps.rmsJitter,cycleToCycleJitter,phaseNoisePower,rmsPhaseErrorandadcSnrLimitchange in that mode; the single-level mode is unchanged.sampling-nyquist: the alias folds to the nearest multiple of f_s
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
aliasFreq_Hzis |f − k·f_s| with k the integer nearest f/f_s (Kester, MT-002). It was |f_s − f|, right only up to 1.5·f_s: 10 kHz at 4 kSa/s was 6000 Hz and is 2000 Hz.fm-modulation-index: the Bessel bandwidth from the Bessel functions
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
besselBandwidthspans the sidebands above 1 % of the unmodulated carrier, 2·n·f_m (Haykin, Communication Systems): it was 2(β + 1)·f_m, Carson's rule again. At β = 5 it was 180 kHz and is 240 kHz. Above β = 1000 it is null, with a warning.digital-filter-order: the exact elliptic order
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
ellipticOrdersolves the degree equation K(k)·K′(k₁)/(K′(k)·K(k₁)), with k = f_p/f_s and k₁ = ε_p/ε_s. It was the Chebyshev order over 1.5, too high in most designs and too low in some: the defaults need 3, not 4; a 2:1 transition at 1 and 20 dB needs 3, not 2.via-stub-resonance: no stub, no resonance
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
With the exit on the last layer there is no stub:
resonantFreqandbackdrillBenefitare null (an unbounded resonance), no longer 0.room-modes and inductor-energy: refusals that name what is wrong
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
room-modesrefuses a speed of sound or a reverberation time at or below 0 with its own message, and returns no values beside a refusal; it returned zeros and named the room's dimensions.inductor-energyrefuses a series resistance at or below 0, as the site always did; the API failed on it.Eight more converter and driver calculators take an operating range, choose their parts and add a loss model
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batchprovenance.validRange
As
buck-converterdoes: each input that varies in service is a pair<input>Minand<input>Max, where 0 (the default) means the nominal input. Every worst case is an output, with the input values it occurs at as outputs of their own (rippleWorstAatrippleWorstVin). Parts are chosen at or above what the worst corner needs, an inductor in E12 and a voltage rating the next standard one at or above 1.5 times the worst voltage (the flyback's switch at or above its clamped drain peak). The loss model's inputs have typical defaults; its losses are at the nominal point, in W, and add tolossTotal. A request without the new inputs gets the answer it got before, except where the entries below say otherwise.boost-converter:inputVoltageMin/inputVoltageMax;coutEach, and the loss model'srdsOn,tSwitch,qg,vGate,vDiode,qrr,dcr,coreLoss,esrOut. New outputs includedutyMin/dutyMax,inductorMinWorstUHandinductorChosenUH,ilPeakWorst,inputCurrentWorst,q1RmsA,coutRmsA,outputRippleMV,vdsRating, eight losses andefficiencyLossModel; theefficiencyinput still sets the currents.flyback-converter:vinMin/vinMax, run on the nominal transformer at each corner; the loss model'srdsOn,tFall,qg,vGate,vDiode,rPri,rSec,coreLoss,leakageInductance(µH, default 2.6) andclampRatio(default 2, from 1.1), with the RCD clamp's loss ½·L_lk·I_pk²·f·k/(k − 1). New outputs include the peak and RMS currents of both windings,reflectedVoltage,clampVoltage,vdsPeakWorst,vdsRating,diodeVrWorst,diodeVrRating, seven losses andefficiencyLossModel.charge-pump-voltage:inputVoltageMin/inputVoltageMaxandpumpCurrentMin/pumpCurrentMax;switchResistance(Ω, default 1) andesr(mΩ, default 10), added to the charge-sharing resistance as the ICL7660 data sheet's R_O is. New outputs includeloadedVoltageMin/loadedVoltageMax,outputResistance,loadedVoltageWithLosses, the capacitors' and switches' ratings, three losses andefficiencyLossModel(equal toefficiencywith ideal parts).switching-regulator-ripple:outputCurrent(A, default 2),inputVoltageMin/inputVoltageMax, and the buck converter's loss model on its own parts (itsesris the output capacitor's). New outputs include the worst-case ripples and RMS currents,vdsRating, eight losses andefficiency.mosfet-power-dissipation:drainCurrentMin/drainCurrentMaxandtAmbientMin/tAmbientMax(any value but 0, which is the nominal);rdsonTempco(%/°C, default 0), solved to the junction in closed form,coss(pF, default 0),qgandvGate. New outputs includejunctionTempWorstwith its current and ambient,totalLossWorst,efficiencyWorst,vdsRating, four losses andefficiencyLossModel.h-bridge-selectionandmotor-driver-power:motorCurrentMin/motorCurrentMaxand the supply's range (motorVoltageMin/motorVoltageMax,supplyVoltageMin/supplyVoltageMax); dead timetDeadand the body diode'svBody(TI SLVA504A), and on the H-bridgedutyCycle(%, default 75),tSwitch,qgandvGate. New outputs include the worst bridge loss, the hottest FET,vdsRating, the losses andefficiencyLossModel.class-d-efficiency:outputPowerMin/outputPowerMaxandsupplyVoltageMin/supplyVoltageMax;switchingFreq(kHz, default 400),filterCutoff(kHz, default 40),tSwitch,coss,qg,vGate, and the filter inductor'sdcrandcoreLoss. New outputs include the output filter (TI SLAA701B) as computed and as chosen, five losses,loadPower,efficiencyLossModel, the worst efficiency and loss with their corners,headroomWorst(negative where the bridge cannot swing the output), the filter's ripple and peak currents, and the ratings.boost-converter: the duty cycle is TI SLVA372D's
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
The duty cycle was 1 − V_in/(V_out·η), so losses shortened it, at odds with the calculator's own input current. It is now TI SLVA372D eq. 1, D = 1 − V_in·η/V_out. At the defaults (5 V to 12 V, 0.5 A, 100 kHz, 88 %)
dutyCycle52.65 % becomes 63.33 %,inductorValue64.35 µH becomes 77.41 µH andoutputCapacitor4.388 µF becomes 5.278 µF.The refusal and the warning move with it: an input between η·V_out and V_out, refused before, is now computed (only V_out ≤ V_in is refused), and the 95 % duty-cycle warning comes at V_in·η/V_out ≤ 0.05.
switching-regulator-ripple: an output at or above the input is refused
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
The calculator is a buck. With
outputVoltageat or aboveinputVoltageit returned a 99 % duty cycle and a negative ripple; it now refuses with "Output voltage must be less than input voltage" and returns no values.buck-converter: an input range, the chosen parts and their stresses, and a loss model with the efficiency
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batchprovenance.validRange
vinMinandvinMaxgive the input as a range (0, the default, is the nominalvin). The inductor is sized where its ripple is largest and chosen at or above that in E12 (inductorMinWorstUH,inductorChosenUH); each stress is reported where it is largest, with the input voltage it occurs at:rippleWorstA/rippleWorstVin,ilPeakWorst/ilPeakWorstVin,outputRippleMV/outputRippleVin,cinMinWorstUF,cinRmsA,q1RmsA,q2RmsA(each with its…Vin), anddutyMin,dutyMax. The output bank iscoutCountcapacitors ofcoutEachµF; the parts' voltage ratings arevdsRating,cinVoltageRating,coutVoltageRating(the rating sold at or above 1.5 times the voltage).A loss model, every input optional:
rdsOnHigh,rdsOnLow,tSwitch,qg,vGate,tDead,vBody,dcr,coreLoss,esrOut,esrIn. At the nominal input and load the losses arelossDcr,lossCore,lossHsConduction,lossLsConduction,lossHsSwitching,lossGateDrive,lossDeadTime,lossCapEsrandlossTotal(W), withoutputPowerandefficiency(%). A request without the new inputs gets the five outputs it got before, exceptcinUF(see buck-converter-input-capacitance).buck-converter: the input capacitor holds the input's ripple
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
cinUFused to divide by the output's ripple voltage (deltaVoutofvout). It now takes the input's own ripple, the new inputdeltaVin(% ofvin, default 1), as TI SLVA477B eq. 12 has it: C_in = I_out·D·(1 − D)/(f_sw·ΔV_in). At the defaults (12 V to 5 V, 1 A, 500 kHz) 9.72 µF becomes 4.05 µF.Twenty-three more calculators take a part series and report the design as built
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batchprovenance.validRange
eSeries(0 for ideal values, or 12, 24, 48, 96) rounds each resistor, capacitor or inductor the calculator designs or reads to the nearest value of that series, and every output is then the rounded design's. Its default leaves a request without it with the answer it got before: 24 onattenuator-designer,led-resistorandzener-diode, whose E24 outputs (piR1NearesttotR2Nearest,rNearest,iActual,pNearest,e24Resistor) now follow it, and 0 onvoltage-divider,opamp-gain,comparator-hysteresis,schmitt-trigger,transistor-switch,current-mirror,crystal-load-capacitance,photodiode-transimpedance,lm317-resistors,rs485-termination,pll-loop-filter,wilkinson-power-divider,rc-snubber,rcd-clamp-snubber,emi-filter-lc,power-supply-ripple-filter,conducted-emissions-filter,differential-mode-filter,speaker-crossoverandtweeter-capacitor.directional-coupler's parts are lines and take none.New outputs describe the rounded design:
attenuator-designerpiAttenuationBuilt(dB) andpiInputImpedanceBuilt(Ω);wilkinson-power-dividerisolationBuilt(dB at f₀, null with the exact resistor, where it is unbounded);rcd-clamp-snubberclampVoltageBuilt,rippleVoltageBuilt,clampPowerBuiltandpeakDrainVoltageBuilt;pll-loop-filtercrossoverBuilt(kHz, the loop's 0 dB crossing;pmActualis the margin there);rs485-terminationfailsafeVoltage(mV, the idle bus's A–B voltage);emi-filter-lcinsertionLossBuilt(dB at 10·f_c);conducted-emissions-filtercornerBuilt,attenuationBuiltandoutputLevelBuilt;speaker-crossoverwooferCutoffBuiltandtweeterCutoffBuilt;tweeter-capacitorcrossoverBuilt;zener-diodezenerCurrentBuilt(mA) andzenerPowerBuilt(mW) at full load through the fitted resistor, the current negative where the resistor cannot carry the load.Labels, keys unchanged: the attenuator's rounded resistors read "(rounded)" for "(E24)";
rNearest"Nearest Series Value",pNearest"Power (series R)",e24Resistor"Nearest Series Resistor";current-mirror'serror"Output Current Error", the shortfall against N·I_ref (with a series it includes the rounding, and a Widlar output is solved from V_T·ln(I_ref/I) = I·R_E with the rounded R_E).So that a part can be solved for a target output, these inputs now state a maximum (a value above it is still computed, with the provenance warning):
comparator-hysteresisr1andr210 000 kΩ;photodiode-transimpedancefeedbackResistance1e7 kΩ andfeedbackCapacitance100 000 pF;differential-mode-filterinductance100 000 µH andcapacitance1 000 µF;power-supply-ripple-filterinductance100 000 µH andcapacitance100 000 µF.schmitt-trigger: thresholds by superposition at the + input
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
The band was V_cc·R2/(R2 + R3), which leaves out R1, set symmetrically about the divider's voltage. The thresholds are now superposition at the + node (TI SNOA997A): V_TH+ = V_cc·R1/(R1 + R2∥R3) and V_TH− = V_cc·(R1∥R3)/(R2 + R1∥R3);
centerVoltageis the band's middle. At the defaults (5 V, 10, 10 and 100 kΩ)vth_high,vth_lowandhysteresis2.727, 2.273 and 0.4545 V become 2.6190, 2.3810 and 0.2381 V. The circuit is an inverting trigger; the description called it non-inverting.conducted-emissions-filter: L and C match the load
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
L and C were split as √(LC·Z) and √(LC/Z), which set √(L/C) to √Z (7.07 Ω into 50 Ω; the filter peaked 17 dB at 47 kHz from a voltage source). They are now L = Z/ω₀ and C = 1/(ω₀·Z). At the defaults (80 dBµV at 150 kHz, 20 dB, 50 Ω)
lValue23.7 µH becomes 167.8 µH andcValue0.475 µF becomes 0.0671 µF. The corner, and so the 40 dB/decade attenuation, is unchanged.attenuator-designer, led-resistor, zener-diode: E24 values below 10 Ω
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
Their E24 searches began at 10 Ω, so a part under 10 Ω was fitted as 10 Ω: a 1 dB, 50 Ω T pad's 2.875 Ω series resistors came out as 10 Ω (the pad built 1.336 dB and 54 Ω), and a 5 Ω LED resistor as 10 Ω. The searches now begin at 0.10 Ω (3.0 Ω and 5.1 Ω). Above 10 Ω nothing changes.
pll-loop-filter and class-d-efficiency: a refusal returns no values
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
pll-loop-filter(every refusal) andclass-d-efficiency(in the API) returned zeros beside the error. They return no values, as other calculators do.filter-designer: the response of the ladder as built, an exact cutoff shift, and cutoffs to 1 GHz
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batchprovenance.validRange
Six new outputs describe the ladder the part values build, solved as a cascade of ABCD matrices between Z₀ and its load:
loadImpedance(Z₀, or an even-order Chebyshev's Z₀/g_{N+1}, g_{N+1}·Z₀ for high-pass),cutoffBuilt_Hz(the −3 dB frequency, low- and high-pass),fLowBuilt_HzandfHighBuilt_Hz(a band-pass's −3 dB edges),passbandLoss_dB(the worst loss to 0.8·f_c, above 1.25·f_c, or across the middle 80 % of a band) andstopbandBuilt_dB(the loss where the prototype reaches twice its cutoff). WitheSeriesset they are the rounded design's; with ideal parts the stopband equalsattenuation_at_2fc.fcShiftPercentwas half the largest part error. It is now how far rounding moved the built ladder's −3 dB frequency (a band-pass: the centre of its −3 dB band) from the ideal ladder's: the 100 MHz 5th-order Butterworth with E24 parts gives +0.997 %, where the estimate gave +1.31 %. WitheSeries0 it is 0, as before.fctakes up to 1 GHz (it stopped at 10 MHz).voltage-drop: every gauge computes, and aluminium is IEC 60889's
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
A
wireGaugethe table does not list (an odd gauge, or a fractional one) was computed with 12 AWG's resistance. It now takes the same formula as the table, IEC 60028 copper (1/58 Ω·mm²/m) over the gauge's ASTM B258 area; the listed gauges are unchanged.conductorType1 (aluminium) took 1.61 times copper's resistance. It now takes IEC 60889's 0.028264 Ω·mm²/m, 1.6393 times copper, so every aluminium drop is 1.8 % larger: 10 AWG at 15 A over 30 m at 120 V drops 4.03 %, where it read 3.96 %.coax-loss: attenuation from the manufacturers' data; eight more cables, input power and power out
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batchprovenance.validRange
The cable table ran about 1.7 times the datasheets' dB/100 ft, so every cable's loss was too low. Attenuation per 100 m is now k₁√f + k₂f with f in MHz: Times Microwave's published coefficients (25 °C) for the LMR cables, and a least-squares fit to the Belden and CommScope tables (20 °C) for the others. Old → new
lossPer100mat 100, 435 and 2400 MHz: LMR-400 2.15 → 4.10, 4.62 → 8.74, 11.50 → 21.70; LMR-240 3.60 → 8.05, 7.69 → 17.04, 19.40 → 41.51; LMR-200 4.80 → 10.64, 10.35 → 22.43, 26.00 → 54.18; LMR-100A 10.80 → 23.84, 23.45 → 51.01, 61.00 → 127.68; RG-58/U 10.00 → 12.45, 22.65 → 28.77, 64.00 → 84.61; RG-8X 6.10 → 10.10, 13.50 → 22.74, 38.00 → 63.62; RG-213/U 4.70 → 6.30, 10.35 → 15.21, 28.50 → 48.34; RG-174/U 19.00 → 28.12, 43.33 → 65.75, 120.00 → 197.64; Belden 9913 2.50 → 4.34, 5.42 → 9.23, 13.70 → 22.75; Andrew LDF4-50A 0.70 → 2.17, 1.53 → 4.66, 3.95 → 11.81.attenuation_dBand the power and voltage shares follow. Andrew LDF4-50A is the 1/2 in cable; it was labelled 7/8 in.cableIndex0–9 keep their cables and now runs to 17: LMR-600, LMR-195, LMR-400-UltraFlex, LMR-400-LLPL (plenum), RG-316, RG-6 (75 Ω), Andrew LDF5-50A (7/8 in) and LMR-900. New optional inputinputPower(dBm, default 30); new outputspowerOut_dBm,powerOut_WandlossPerMeter. A frequency above a cable's rated maximum, past the last point of a fitted table, or below the frequency from which a fit holds within 10 % gets a warning, not a refusal.Eighteen calculators take lists: stages, losses, terms, layers, loads, parts and surfaces
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batchprovenance.validRange
Each list is a count and indexed inputs (
numLoadswithload1Current,load2Current, …), so a request stays a flat object. Every count defaults to the list the calculator had before, so a request without the new inputs gets the answer it got before (checked on 201 requests per calculator, refusals included). An indexed input past the count is ignored.noise-figure-cascade:numStagesnow runs to 10 (it ran to 4), withstage5Gain,stage5NF,stage5IIP3tostage10IIP3, and an optionalbandwidth(MHz, default 1); new outputscascadedGain,noiseTemperature(290 K × (F − 1)) andnoiseFloor(kT₀B plus the cascaded noise figure).lora-link-budget,ble-link-budget,5g-nr-link-budget:extraLossCount(0–6) ofextraLoss1toextraLoss6(dB), andtotalLoss.clock-jitter:jitterTerms(0–6) ofterm{n}Jitter_ps, added in quadrature.emi-margin-budget:marginTerms(0–6) ofterm{n}Allowance, andtotalAllowance.sensor-accuracy-budget:errorTerms(0–6) ofterm{n}Error, in quadrature.4-20ma-transmitter:loopElements(0–6) ofelement{n}Resistance, andtotalLoopResistance.heatsink-calculator,heatsink-selection,thermal-resistance-network,junction-temperature:numLayers(0–6) oflayer{n}Theta(°C/W), in series between case and heatsink; the two heatsink calculators also returnthetaJSandheatsinkTemp.battery-life:numLoads(0–7) ofload{n}Currentandload{n}Duty, andselfDischargeCurrent.battery-runtime-motor:motorDuty(%, default 100) andnumLoads(0–5) ofload{n}Current,load{n}Dutyandload{n}Efficiency, andbatteryCurrent.solar-panel-sizing:loadHours(h a day, default 24) andnumLoads(0–9) ofload{n}Powerandload{n}Hours.gear-ratio:extraStages(0–5) ofstage{n}Driver,stage{n}Drivenandstage{n}Efficiency, andtotalEfficiency.series-parallel-resistor:numParts(1–8, default 4) withr5tor8, andsumOfValues;r1may be 0, a part not fitted, as the other positions already could.rt60-reverberation:numSurfaces(0–8) ofsurface{n}Areaandsurface{n}Alpha, andotherAreaandmeanAbsorption.Link budgets answer a distance of 0 with no path loss
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
lora-link-budget,ble-link-budgetand5g-nr-link-budgetrefused a distance of 0. They now answer it with 0 dB of path loss: the link with no path, which the pages use to show the budget before the path. A negative distance is still refused.clock-jitter: no timing budget left is an unbounded share, not 0 %
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
Where setup and hold time take the whole clock period,
budgetUsed_pctwas 0, which reads as no jitter at all. It is now null (an unbounded share), with a warning that no timing budget is left when there is no setup violation to report.series-parallel-resistor: the divider ratio is the two fitted parts
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
With exactly two parts fitted,
voltDividerRatiowas R2/(R1 + R2) whichever two they were, so it was 0 with R1 and R3 fitted. It is now the second fitted part over the sum of both, which is unchanged when they are R1 and R2.optical-sensor-range: a refusal returns no values
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
A safety factor of 0 or less was refused with
maxRangeandnormalizedRangeof 0 beside the error. The refusal now returns no values, as every other refusal does.broadside-coupled-pair: no common-mode, uncoupled impedance or plane height without planes
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
An unshielded pair (
shielded0) has no reference planes, socommonImpedance,uncoupledImpedanceandheightToPlaneare undefined, as the warning already said. They were 0 and are now null.rise-time-bandwidth: no harmonic number without a clock
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
With
clockFrequency0, which the input allows to leave the clock out,harmonicNumberwas 0, a value no clock gives. It is now null, as other outputs are where a value does not exist.Thirteen calculators judge their result against a published standard
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
Each takes the standard (and, where one applies, the class) as an optional input, and returns the limit, the result as a percentage of it (above 100 fails), and the value of an input that would comply. The limits are computed in the calculation, so the API gives the page's verdict. Every new input is optional, and at its default every existing output is unchanged, except where an entry below says otherwise.
power-density:standard(0 FCC 47 CFR §1.1310, 1 ICNIRP 2020),exposureClass,groundReflection; returnslimitDensity,limitPercent,minDistance,maxEirp.trace-width-current:standard(it reads the drawntraceWidththe API already takes);limitCurrent,limitPercent,minTraceWidth.microvia-current-capacity:standard,conductorClass;limitCurrent,limitPercent,minVias,minViaDiameter.pcb-trace-temp:standard,laminate;chartRise,limitRise,limitPercent,minTraceWidth,maxCurrent.minimum-conductor-spacing:conductorSpacing;spacingPercent,maxVoltage(the limit isspacing).padstack-annular-ring:standard,padDiameter,producibilityLevel;minLandDiameter,landPercent,maxDrillDiameter.solder-paste-volume:releaseCriterion;maxStencilThickness,stencilPercent,maxApertureReduction.bga-land-pad:ballType,landDiameter;ipcLand,landDeviation,landTolerance,landPercent,compliantLand.i2c-pullup:pullup;riseTime,riseTimePercent,maxBusCap(the limit isriseTimeTarget_ns, the mode isspeedMode).usb-termination:impedanceError,impedanceTolerance,impedancePercent,compliantImpedance(the standard isusbVersion).esd-tvs-diode:standard,testLevel,esdRating;testVoltage,limitPercent.esd-clamp-selection:standard,testLevel;testVoltage,limitPercent.radiated-emission-estimate:standard,emissionClass;eFieldUvm,limitField,limitPercent,maxCurrent,maxLoopArea.Where a standard sets no limit (ICNIRP below 30 MHz, CISPR 32 outside 30 MHz–6 GHz, IPC-6012 Class 2 annular rings, an A6 spacing above 3,050 m), the new outputs are NaN, returned as null, with a warning that says why.
Eleven transmission-line calculators return the guided wavelength
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
microstrip-impedance,controlled-impedance,asymmetric-stripline,dual-stripline,differential-pair,edge-coupled-internal-symmetric,edge-coupled-internal-asymmetric,edge-coupled-embedded,broadside-coupled-pair,coplanar-waveguideandcoax-impedancetake an optionalfrequency(GHz, default 2.4) and returnguidedWavelengthandquarterWavelength(mm), λg = c/(f·√εeff), the coupled lines from their odd-mode εeff. A request withoutfrequencyreturns every existing output unchanged, plus the two new ones at 2.4 GHz.radiated-emission-estimate: the margin follows the chosen standard and class
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
marginis the limit of the chosen standard and class at the request's frequency and distance, minus the field:standard0 is CISPR 32 and 1 is FCC Part 15, section 15.109,emissionClass0 is Class B and 1 is Class A. A request without them gets CISPR 32 Class B, as before, so its margin is unchanged. FCC limits apply from 30 MHz to 40 GHz, and the margin is null outside its standard's range. The output's label is now "Margin to Limit".minimum-conductor-spacing adds the component-lead columns A6 and A7, and a B4 output
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
condition4 is A6 (external component lead or termination, uncoated, sea level to 3,050 m) and 5 is A7 (external component lead or termination, conformal coating), both from IPC-2221B Table 6-1. Above 3,050 m an A6 request returns null spacing with a warning, since the table gives A6 only to 3,050 m. The new outputspacingB4is column B4 (external conductors, permanent polymer coating), besidespacingB1,spacingB2andspacingB3. Conditions 0 to 3 and the existing outputs are unchanged.bga-land-pad sizes the land by ball diameter, as IPC-7351B does
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
The land was the ball diameter × 1.0, 0.9 or 0.8 by density level, which no IPC table gives. It now follows IPC-7351B: for a collapsing ball, 25, 20 or 15 % smaller than the ball by ball size. At the defaults
padDiagoes from 0.55 to 0.45 mm, and the outputs derived from it follow. ThedensityLevelinput is retired, since IPC-7351B does not apply density levels to a BGA land: a request that sends it is answered, with a warning that the value was not read.bldc-winding: a coil spans at least one slot
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batch
Where slots divided by poles rounds to 0 (6 slots under 14 poles, for example), the coil span was 0, the winding factor 0, and
actualKvanddeltaEquivKvwere null. A concentrated coil is wound round one tooth, so the span is now at least one slot and those outputs are numbers.emi_radiated jobs apply the tighter limit at a band edge
POST /api/py/v1/jobs
At exactly 88, 216 and 960 MHz (FCC Part 15) and 230 MHz (CISPR 32), the looser of the two adjoining limits was used. 47 CFR 15.109(c) says the tighter limit applies at the band edges, and CISPR 32 applies the lower limit at a transition frequency; both are now applied.
Input ranges on 31 calculators
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}POST /api/py/v1/calculate/batchprovenance.validRange
Inputs on 30 calculators gained a maximum, most of them geometry (lines, pads, antennas, waveguides), so the page can solve for them, and
i2c-pullup's bus capacitance now runs to 550 pF, Fast-mode Plus's limit. The API does not refuse a value beyond an input's range;provenance.validRangereports it as outside.Twelve sensor calculators gain a front end: its voltage, the ADC code and the resolution
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}
ntc-thermistor,rtd-temperature,pt100-resistance,thermocouple-voltage,wheatstone-bridge,strain-gauge-bridge,pressure-bridge-output,load-cell-amplifier,hall-effect-sensor,current-shunt,accelerometer-sensitivityandlvdt-sensitivitytake optional front-end inputs and return three new outputs. The inputs are a divider (fixedResistor,sensorPositionorntcPosition) or an amplifier (frontGain), and an ADC (adcRef,adcBits), reusing an input a calculator already had (gain,amplifierGain,adcVoltage,supplyVoltage). The outputs are the front end's voltage (dividerVoltageoramplifiedVoltage),adcCode(a unipolar N-bit code, held between 0 and 2^N − 1) andresolution(the physical quantity per ADC step at that point).load-cell-amplifiertakes the load on the cell (appliedLoad) and returns its output (bridgeOutputMv);accelerometer-sensitivitytakesaccelerationand returnsoutputVoltage.Every new input is optional, and at its default every existing output is unchanged: a request written before this change returns the values it did, plus the new outputs.
current-shunt: the amplifier gain defaults to 20
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}
When a request omits
amplifierGain,current-shuntnow uses 20 where it used 50. At the other defaults (10 A through 10 mΩ, a 3.3 V ADC), gain 50 put 5 V on the ADC, beyond its reference. Gain 20 gives 2 V.amplifiedVoltageandadcResolutionCurrentchange for such requests; a request that namesamplifierGainis unaffected.hall-effect-sensor: a field of zero or below is refused
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}
hall-effect-sensornow returns an error forBfieldat or below 0 mT, as the site already did, where the API used to return values computed from it.frequency-ppm-tolerance: warnings cite the clock tolerances that apply
POST /api/py/v1/calculatePOST /api/py/v1/calculate/{slug}
The warnings of
frequency-ppm-tolerancecited a ±50 ppm UART budget (a UART tolerates about 2 %) and called ±100 ppm too loose for USB, which allows ±2500 ppm at full speed. Above ±50 ppm the warning now cites 100BASE-TX Ethernet and the Bluetooth LE radio clock. Above ±100 ppm it cites every Ethernet clock, and above ±500 ppm USB 2.0 high speed. The thresholds at 50 and 100 ppm are unchanged; the one at 500 ppm is new.Board-specific trace width and temperature rise from rftools' own thermal model
POST /api/py/v1/calculate (trace-width-current, pcb-trace-temp)POST /api/py/v1/calculate/batchPOST /api/py/v1/calculate/solveMCP run_calculationrftools-io SDK
trace-width-currentgainswidthBoardMm, the minimum width for this board, andpcb-trace-tempgainstempRiseBoardandmaxTempBoard, the rise and maximum temperature on it; both gaininModelRange. They come from rftools' own thermal model, never labelled IPC-2152: a closed-form thermal network fitted to a test-only 2-D finite-volume solver of the trace, laminate, plane and still air (convection and radiation), which is checked against published measurements (IPC-TM-650 2.5.4.1A conditions, Brooks & Adam's fits to IPC-2152's charts, and others) and carries a 15 % margin on the rise from the worst of them. Inside its validated range (trace width 0.25–5.1 mm, board 1.4–3.2 mm, an internal trace from 1.27 mm wide, a rise of 5–100 °C, a plane at 0.1 mm or more) the fit is within 2.81 % of the solver.New inputs, appended with defaults so an existing request computes unchanged:
boardThickness(mm, default 1.6),planeDistance(mm to a copper plane, 0 for none) andplaneCopperWeight(oz, default 0.5) on both calculators, andisExternalonpcb-trace-temp. Every existing output is unchanged. At the defaultswidthBoardMmis 0.30212 mm, beside Figure 5-1's 0.78144 mm, andtempRiseBoard10.2818 °C.Outside the validated range the result is still reported, with
inModelRange0 and a warning naming the bound; a board under 1.4 mm points to Figure 5-1, where the model and IPC-2152's published thickness data disagree. A thermal runaway returns null for the board-specific rise.provenance.validRangegainsmodelson these two calculators, each fitted model with its owninside: IPC-2152 Figure 5-1's fit first, then the thermal model, andstatusisoutsideif either is. A request for these calculators with a zero or negative current, rise, copper weight or length is refused in the browser as by the API, and a negative current is refused rather than answered 500.Seven outputs the calculators could not compute are removed
POST /api/py/v1/calculate (differential-via, usb-termination, parabolic-dish-antenna)POST /api/py/v1/calculate/batchMCP run_calculationrftools-io SDK
differential-viano longer returnsdiffImpedance,commonImpedance,oddImpedance,evenImpedanceorsingleEndedImpedance. A field solution (a 3-D quasi-static solver licensed on four closed forms, cross-checked full-wave with openEMS to 0.4 %) showed that a via pair's impedance depends on the pitch of the planes it passes through, which is not an input: at the defaults it is 85.6 Ω at a 0.1 mm plane pitch and 96.0 Ω at 1.0 mm, where the calculator read 44.10 Ω. No relation in the calculator's inputs comes within 10 % of every case, and the common-mode impedance also depends on where the return current flows. The stub resonance and the losses are still returned; see the fixed entry.usb-terminationno longer returnseyeOpening, a heuristic no USB specification defines.parabolic-dish-antennano longer returnsnoiseTemperature, which was 290·(1 − η): an antenna's noise temperature depends on its elevation, its spillover and the sky, none of which the calculator takes.A caller that reads one of these keys must stop: the key is absent, not null. Every other output of the three calculators is still returned.
Values that do not exist are absent or null, not zero or a sentinel
POST /api/py/v1/calculatePOST /api/py/v1/calculate/batchPOST /api/py/v1/calculate/solveMCP run_calculationrftools-io SDK
These outputs are now null where the quantity is undefined, where they used to be a number:
usb-termination'ssignalRiseTimefor USB 3.0, which specifies no transmitter rise time (USB 2.0 reads its 500 ps minimum, T_HSR);pcb-crosstalk-emc's coupling outputs above the weak-coupling limit, (x_C + x_L)/2 = 0.1, where the lumped model does not hold;subwoofer-box'svbSealedandf3Sealedfor Qts ≥ 0.707, where no sealed alignment exists (they read 2·Vas);planar-spiral-inductor'sqFactorat or above the self-resonance;aperture-diagonal's covering drill and router bit when no stocked tool fits (they read 0);lc-resonance'sqFactorwith no resistance, a lossless circuit's unbounded Q (it read 9999, and the bandwidth f₀/9999; the bandwidth is now 0); anddifferential-via'sstubResonancewhen there is no stub (it read 0 GHz). Each null comes with a warning that says why.A refused input now returns no values, with its error, where these calculators returned zeros or partial values:
power-density,thermal-via-array,bga-breakout-widthwith no channel,dual-striplinewith h₂ ≤ 0,fresnel-zone, and every refusal ofdoppler-shift,reactance-calculator,broadside-coupled-pair,fusing-current,rise-time-bandwidth,current-divider,delta-wye-conversion,voltage-drop,rt60-reverberationandvfd-motor-speed. A zero beside an error read as a result.A caller that treated these values as numbers must check for null, or for an absent key beside an error.
/calculate/solvealready skips points without the output.Calculators corrected against their references, with three new inputs and new outputs
POST /api/py/v1/calculatePOST /api/py/v1/calculate/batchPOST /api/py/v1/calculate/solvePOST /api/py/v1/jobs (emi_radiated)MCP run_calculationMCP simulate_emi_radiatedrftools-io SDK
Constants. Copper resistivity is IEC 60028's 1/58 Ω·mm²/m everywhere (trace-resistance 100.217 → 100.458 mΩ at its defaults; the other copper calculators move 0.008 %, skin depth 0.004 %). The microstrip engine takes η₀/2π where it took 60 and the exact speed of light, so every microstrip-based impedance moves −0.069 % (microstrip-impedance 80.954 → 80.898 Ω, differential-pair 99.983 → 99.914 Ω, and solved widths with them). Every calculator now shares one η₀ = µ₀·c (CODATA 2018).
power-density's wave impedance is µ₀·c, 376.730 Ω (it was 120π).controlled-impedance's delay is √εeff/0.299792458 ps/mm (6.52995 → 6.53447). Theemi_radiatedjob uses π·µ₀/c and µ₀ for its fields, so it now agrees withradiated-emission-estimate(it read 0.021 dB and 0.023 dB high).Models.
planar-spiral-inductortakes Jow & Ghovanloo's parasitics and its shape's trace length: self-resonance 68.0 → 601.7 MHz and Q at 100 MHz 814.5 → 116.8 at the defaults, with the new optional inputfrequency(MHz, default 100) for Q.pcb-crosstalk-emcis Paul's two-end model: the existing keys are per end (half the old figures),farEndCrosstalkis new, and the defaults are a 3.3 V, 25 MHz clock.subwoofer-boxtunes the port by Margolis & Small (20.246 → 40.273 Hz) and addsf3Vented.pcb-crosstalkruns on the coupled-microstrip engine with the new inputdielectricConst(default 4.5; coupling 6.77 → 5.72 %).patch-antenna's edge resistance is Balanis' radiating-slot model with the exact η₀, unclamped (306.80 → 321.22 Ω; the feed resistance and the 50 Ω inset follow).lc-resonancetakes its series resistance as a series loss in either topology, Q = Z₀/R (parallel mode computed R/Z₀), and addsimpedanceAtResonance_ohm: R in series, L/(C·R) in parallel.differential-via's stub resonance is labelled the unloaded upper bound, and its conductor loss is the dense-plane upper bound (0.030 → 0.011 dB).Rules. A microvia array is rated by its summed cross-section, IPC-2221 note 5 (ten vias 2.762 → 1.466 A). Via self-heating is the barrel's peak, P·L/(8kAN), in
via-voltage-drop(0.2506 → 0.0313 °C) andmicrovia-current-capacity.bga-breakout-widthcounts the outer row (escapeRows1 → 2).padstack-annular-ringjudges aspect ratio against IPC-2222 Table 9-4, 8:1 at every class, with a minimum-drill check by Table 9-6, and Class 3's internal ring is 0.025 mm.thermal-via-arraycounts whole vias.i2c-pullupandrs485-terminationreturn their resistances unrounded (the display still rounds).aperture-diagonaltakes the new inputrouterBit.A request that omits one of the new inputs gets its default.
Four outputs that held one value at every input are removed
POST /api/py/v1/calculate (balun-transformer, opamp-bandwidth, capacitor-code)POST /api/py/v1/calculate/batchMCP run_calculationrftools-io SDK
Each of these outputs returned the same value whatever the inputs, and none was a result:
balun-transformer'svswr(1.0, "VSWR (ideal)") andinsertionLoss(0.3 dB, "Typical Insertion Loss");opamp-bandwidth'sslewLimited(0, a placeholder for a check that needs slew-rate and amplitude inputs the calculator does not have); andcapacitor-code'svoltageCode(0 V, a rating a 3-digit code does not encode). They are no longer returned, not even beside a refusal.A caller that reads one of these keys must stop: the key is absent, not null. What each value stood for is stated in
provenance.assumptionsinstead:ideal-transformeronbalun-transformer(matched by construction; core and winding losses are not modelled, and a ferrite balun typically loses 0.1-1 dB) andcapacitance-onlyoncapacitor-code. Every other output of the three calculators is unchanged.loop-antenna's gain includes its radiation efficiency
POST /api/py/v1/calculate (loop-antenna, capacitor-code)POST /api/py/v1/calculate/batchPOST /api/py/v1/calculate/solveMCP run_calculationrftools-io SDK
loop-antenna'sgainwas 1.76 dBi at every input: a small loop's directivity, 3/2, with none of its loss. It is now that directivity times the radiation efficiency e = R_rad/(R_rad + R_loss), from the two resistances the calculator already returns: G = 10·log₁₀(1.5·e) (Balanis, Antenna Theory, 4th ed., Chapter 2 and section 5.2). At the defaults, a 1 m loop of 10 mm copper at 7 MHz, it reads −9.4043 dBi, 11.2 dB below the old figure; a loop whose loss is small against its radiation resistance still reads close to 1.76 dBi. The efficiency is the new outputradiationEfficiency(%), 7.6467 % at the defaults.capacitor-codeon the API rounded a fractional code half to even where the page rounds half up, so 104.5 decoded as 100 nF on the API and 1 µF on the page. Both now give 1 µF.Eight inputs that changed no result now set one; eleven new outputs
POST /api/py/v1/calculatePOST /api/py/v1/calculate/batchPOST /api/py/v1/calculate/solveMCP run_calculationrftools-io SDK
Each input below was declared and read by no output. It now sets a new output, appended after the existing ones and computed from a relation the calculator already makes; no existing output changes at any input.
patch-antenna:feedOffsetsetsfeedResistance_ohm, R_edge·cos²(π·y₀/L) (Balanis, Antenna Theory, 4th ed., section 14.2.1, Eq. 14-20a), 306.795 Ω at the default edge feed and 65.501 Ω at 10 mm, andinsetFeedPosition_mmis the inset that gives 50 Ω, 10.594 mm.flyback-converter:fswsetsprimaryInductance(µH), Vin·D/(fsw·Ipk) for the calculator's own boundary-conduction peak current, 130.56 µH.solar-panel-sizing:batteryCapacitysetsbatteryAutonomyDays, C·V·0.5/(P·24 h) at the 50 % depth of discharge the required capacity assumes, 0.25 days.rs485-termination:baudRatesetsbaudRateUtilisation(%), the rate as a share of the 10⁸ bit·m maximum, 11.52 %.adc-snr:samplingFreqsetsnyquistZone, ⌊f_in/(f_s/2)⌋ + 1, andaliasFrequency(MHz), |f_in − f_s·round(f_in/f_s)| (Kester, Analog Devices MT-002): zone 1 and 1 MHz at the defaults, zone 2 and 30 MHz for 70 MHz at 100 MSPS.bldc-thermal:continuousCurrentsetsratedCurrentShare(%), the operating current against the rating, 75 %.bldc-winding:supplyVoltagesetsnoLoadSpeed(rpm), Kv·V, 11,406.44 rpm. Figures are at each calculator's defaults.Two more new outputs:
assumedBoomLength(m) onyagi-antenna, the boom its 0.2λ element spacing implies, (N − 1)·0.2λ, 1.6655 m at the defaults, which replaces its retired boom-length input (see that entry); andradiationEfficiency(%) onloop-antenna(see its gain entry). On the site only,ism-coexistence's live regulatory widget now shows the band itsbandIndexselects; its result is unchanged.A request that omits one of these inputs gets its default.
patch-antennawithoutfeedOffsetandsolar-panel-sizingwithoutbatteryCapacitywere answered 400 until now; both are answered.Twelve inputs that never changed a result are retired; a request that sends one is answered with a warning
POST /api/py/v1/calculatePOST /api/py/v1/calculate/batchPOST /api/py/v1/calculate/solveMCP run_calculationMCP get_calculator_schemarftools-io SDK
These inputs were declared, but no value of them changed any output, warning or refusal:
smith-chartfrequency,via-calculatorsignalLayer,stackup-builderlayerCount(the model is a fixed four-layer stack, which its title now says),i2c-pullupnumDevices,fm-modulation-indexcarrierFreq,quantization-noisefrequency,555-timervcc,bldc-efficiencypolePairs,opamp-bandwidthphase_margin,battery-internal-resistancetemperature,rs485-terminationnumNodesandyagi-antennaboomLength. They leave the catalogue, the provenance bounds and the MCP schema; each calculator lists them underretiredInputs, with the date.A request that still sends one is answered with the same values as before and the existing warning "Input '<key>' is not read by this calculator and was ignored." It is never refused.
POST /calculate/solverefuses to solve for one, as for any undeclared input; no grid of it could have moved an output. Ayagi-antennarequest withoutboomLength, answered 400 until now, is answered.The next
rftools-iorelease keeps each retired keyword: passing one raises aDeprecationWarningnaming the key and its retirement date, and the key is not sent.trace-width-current reads its IPC-2152 width on IPC-2152 Figure 5-1; pcb-trace-temp reports the IPC-2152 rise
POST /api/py/v1/calculatePOST /api/py/v1/calculate/batchPOST /api/py/v1/calculate/solveMCP run_calculationrftools-io SDK
trace-width-current'swidth2152mm, labelled IPC-2152, was the IPC-2221 width × 0.75 on an external layer and × 0.85 on an internal one, a factor IPC-2152 does not contain and narrower than every IPC-2152 reading from 2 A upward. It is now IPC-2152 (2009) Figure 5-1, the chart IPC-2152 recommends for internal and external conductors in all environments, through the published fit I = 0.024·ΔT^0.44·A^0.725, the same on either layer: 0.7814 mm at the defaults, was 0.2266 mm; at 5 A, a 10 °C rise and 1 oz on an outer layer 7.194 mm, was 2.086 mm. It reproduces Olson's (2010) Figure 5-1 readings of 300 and 680 sq mil within 5 %. The key is unchanged; its label is now "Min Width (IPC-2152 Fig. 5-1)".Both calculators take copper as 35 µm per ounce, so
width2221mmmoves −0.58 % (0.3004 mm at the defaults, was 0.3021 mm).trace-width-current'sresistance,voltageDropandpowerDissare computed for a new input, the drawn trace width, where they used the IPC-2221 minimum, with copper at IEC 60028's 1/58 Ω·mm²/m: 0.10433 Ω at the defaults, was 0.17326 Ω.New optional input, appended; a request that omits it gets its default:
traceWidthontrace-width-current(mm, 0.5). New outputs:inChartRangeon both calculators (1 inside the range IPC-2152 Figure 5-1 draws, 0 outside, carried intoprovenance.validRangewith the fit's worst residual), andtempRise2152andmaxTemp2152onpcb-trace-temp, the rise on Figure 5-1 (k = 0.024, any layer) beside the IPC-2221 external rise: 25.815 °C at the defaults, beside 5.342 °C.pcb-trace-temp's existing outputs keep their keys and values.Results outside the charts are reported and flagged, never refused: a warning names the quantity and the chart bound, for Figure 5-1's 10 to 45 °C curves and 700 sq mil, IPC-2221 chart A's 10 to 100 °C and 700 sq mil on an outer layer, and chart B's ½ to 3 oz; beyond Figure 5-1,
provenance.validRange.statusisoutside.Citations:
trace-width-current,pcb-trace-temp,microvia-current-capacityandvia-calculatorcited IPC-2221B for the current chart, which that edition (2012) does not contain. They now cite IPC-2221A (2003) Figure 6-4, the chart of IPC-2221 (1998), and IPC-2152 (2009) Figure 5-1; no value of the last two changes.Twelve calculators take the model their page and their reference describe
POST /api/py/v1/calculatePOST /api/py/v1/calculate/batchPOST /api/py/v1/calculate/solveMCP run_calculationrftools-io SDK
Each calculator below computed a model that stopped short of what its page and its reference describe, the same way in the browser, the API and the MCP server. Each now computes the published model and is tested against a published case, and the parity sweep holds the browser and the API to the same numbers. Saved calculations store inputs, so they recompute.
bldc-winding, part of the same change, has its own entry of 2026-10-11.Values change for:
cable-shield-effectiveness(transfer impedance R_T + jωM_T, and shielding effectiveness against the line impedance, Hähner & Mund 1998: 64.95 dB at the defaults, was 30.97 dB against 1 Ω, with no 0 dB floor),ferrite-bead(Analog Devices AN-1368's R_DC + (R_AC ∥ L ∥ C), anchored to the entered 100 MHz impedance, and the complex insertion loss between source and load: the default 600 Ω bead reads 94.9 Ω at 1 GHz, was 1897 Ω),mosfet-operating-point(the operating point on the drain load line, Sedra & Smith 7e section 5.3; see the breaking entry below),bldc-motor(thrust at the speed where the motor's and the propeller's torques balance, Drela 2007, with C_P 0.047: 140.91 g at the defaults, was 80.43 g at a fixed 75 % of no-load speed),ethernet-cable(ANSI/TIA-568-C.2 cable insertion loss at each PHY's signalling frequency, 31.25, 62.5 or 400 MHz: 7.69 dB at the defaults, was 16.5 dB; Category 8 is a 30 m channel at every speed),radar-range-equation(the range at the SNR Albersheim's equation requires for P_d and P_fa: 35.23 km at the defaults, was 74.96 km at SNR = 1, andsnrAtMaxRangereports that SNR, where it was 0),bldc-efficiency(copper loss I²·R with the terminal resistance, Drela 2007 and the maxon Formulae Handbook: 4.500 W at the defaults, was 6.750 W),bldc-thermal(the winding resistance at the temperature it reaches, maxon Formulae Handbook section 9.2, with the same copper loss: 207.61 °C at the defaults, was 184.47 °C; see the breaking entry below),class-d-efficiency(conduction through the FETs that carry the load at once, I²·R_DS(on)·N/2: 0.625 W at the defaults, was 1.250 W),motor-driver-power(two conducting FETs and one hard-switched, TI SLVA504A: a 0.1440 W bridge at the defaults, was 0.2088 W), andcoax-impedanceandvia-calculator(Z₀ from η₀/2π = 59.9585 Ω with the CODATA 2018 constants, where they used 60: 47.990 Ω and 35.2243 Ω at the defaults, were 48.023 Ω and 35.2487 Ω).New optional inputs, appended; a request that omits one gets its default:
transferInductance(nH/m, 0.159155, which reproduces the previous transfer impedance at 10 mΩ/m) andlineImpedance(Ω, 50) oncable-shield-effectiveness;beadInductance(µH, 1.208),parasiticCapacitance(pF, 1.678) andsourceImpedance(Ω, 0) onferrite-bead;biasMode(0 solves the load line, 1 takes V_DS as entered; default 0) onmosfet-operating-point;pd(0.9),pfa(1×10⁻⁶) andpulses(1, integrated noncoherently) onradar-range-equation;transitionTime(ns, 50) onmotor-driver-power.New outputs:
vdsOperatingonmosfet-operating-point,loadedRpmandloadedCurrentonbldc-motor, andgateDriveLoss(2·Q_g·V·f, dissipated in the gate driver) onmotor-driver-power.Keys kept, meanings now matching their labels:
motor-driver-power'sconductionLossis the loss of one conducting position (I²·R),switchingLossthat of the one hard-switched FET,totalLossPerFetthe hottest FET, andefficiencyis taken against V·I·D and counts the gate drive. ThephaseResistanceinput ofbldc-efficiencyandbldc-thermalis labelled as the terminal resistance, measured between two leads, it always was.Category 8 insertion loss uses the draft coefficients the IEEE 802.3bq channel model quotes for TIA Category 8 (draft d0.7: 1.8·√f + 0.005·f + 0.25/√f dB per 100 m), and every Category 8 result says so, since the published TIA-568.2-D table is not at hand.
New refusals and warnings:
ferrite-beadrefuses a 100 MHz impedance its L and C cannot reach, naming the largest they can;radar-range-equationrefuses a P_d no detector meets at that P_fa and warns outside Albersheim's stated range;cable-shield-effectivenesswarns when the cable is longer than a sixth of a wavelength or the coupling is no longer weak;ethernet-cablewarns when a category is evaluated beyond the frequency it is specified to.mosfet-operating-point solves the load line by default; bldc-thermal returns null where the winding runs away
POST /api/py/v1/calculate (mosfet-operating-point, bldc-thermal)POST /api/py/v1/calculate/batchPOST /api/py/v1/calculate/solveMCP run_calculationrftools-io SDK
mosfet-operating-pointnow solves the square law together with the drain load line V_DS = V_DD − I_D·R_D by default (Sedra & Smith, Microelectronic Circuits, 7th ed., section 5.3, Example 5.5), sovddandrdLoadset the operating point and an enteredvdsno longer does. At the defaults it reads 11.96 mA in the triode region at V_DS 0.0401 V, where it read 450 mA, which the 1 kΩ load cannot pass from 12 V. A caller that relies on the enteredvdsmust sendbiasMode: 1, which keeps the previous computation and warns when V_DS + I_D·R_D is more than 1 % from V_DD. The new outputvdsOperatinggives the operating point's V_DS in either mode.bldc-thermalnow includes the winding resistance's rise with temperature (maxon Formulae Handbook, 2012, section 9.2). Where α·Rθ·R₂₅·I² ≥ 1 there is no steady state: the copper loss grows with temperature faster than the thermal path removes it. TheretempRise,steadyStateTemp,thermalMarginandhotResistanceare null, with a warning naming the current below which the winding settles; they used to be numbers (467.97 °C at 25 A on the default motor, which now runs away above 23.21 A). A caller that reads them as numbers must handle null.deratedCurrentandtimeToLimitare always numbers.bldc-thermal'stimeToLimitis null, with a warning, when the peak current settles the winding below its limit, so the limit is never reached. It used to be the sentinel 9999 s, which read as a real time.bldc-winding: fill counts both coil sides, and the star-of-slots winding factor
POST /api/py/v1/calculate (bldc-winding)POST /api/py/v1/calculate/batchPOST /api/py/v1/calculate/solveMCP run_calculationrftools-io SDK
fillFactornow counts both coil sides in each slot, 2·N·A_wire/A_slot (Pyrhönen, Jokinen & Hrabovcová, Design of Rotating Electrical Machines, 2008, section 2.9): one coil on every tooth puts two coil sides in every slot. It used to count one, so every fill doubles. It is no longer capped at 100 %: a winding that cannot fit returns its real figure, with the 75 % and 90 % warnings as before.windingFactortakes its distribution factor from the star-of-slots method, k_d = sin(π/6)/(z·sin(π/(6z))) for q = S/(3P) = z/n in lowest terms (Bianchi & Dai Pré, 2006). It was 1 for every q that rounded to 1, so fractional-slot windings read high: 12 slots and 14 poles now give 0.93301 (was 0.96593), 18 slots and 16 poles 0.94521 (was 0.98481). Integral-slot windings are unchanged.turnsPerCoil,actualKv,deltaEquivKvandphaseResistancefollow: the default 935 Kv winding stays at 4 turns and achieves 1027.61 Kv (was 992.59), and 920 Kv on a 23 × 10 mm stator takes 6 turns (was 5).The calculator's default
maxCurrentis 5 A (was 10 A), so its default EMAX MT2213 winding builds: 4 turns of AWG 18 at 47.24 % fill. At 10 A it needs AWG 14 and 119.44 %. The API applies no defaults, so a request that sendsmaxCurrentis computed at the value it sends.Thirty calculators corrected to the relation they cite
POST /api/py/v1/calculatePOST /api/py/v1/calculate/batchPOST /api/py/v1/calculate/solveMCP run_calculationrftools-io SDK
Each calculator below computed something other than its cited reference, the same way in the browser, the API and the MCP server. Each now computes the reference relation and is tested against a published value. Saved calculations store inputs, so they recompute.
Values change for:
equalizer-q-factor(constant-Q octave bandwidth),capacitor-code(a multiplier digit 8 means ×0.01, 9 means ×0.1),opamp-gain(bandwidth is GBW over the noise gain),coax-impedance(TE11 cutoff on the mean circumference),h-bridge-selection(two FETs conduct at a time),bldc-winding(the √3 factor belongs to wye; new defaults from a real 12N14P motor),speaker-crossover(2nd-order inductors √2·R/ω_c, twice the old value, in the KiCad export too),esd-tvs-diode(the IEC 61000-4-2 level takes the standard's 3.75 A/kV first peak),common-mode-choke(a complex series impedance in a 50 Ω/50 Ω system),uart-baud-rate(mark and space parity count one bit),bldc-motor(static thrust C_T·ρ·n²·D⁴ with C_T 0.11),resistor-color-code(orange ±0.05 %, yellow ±0.02 %; a white tolerance band is refused),pt100-resistance(IEC 60751's C below 0 °C, and the full slope),torque-units,energy-unitsandtorque-unit-converter(exact NIST SP 811 factors),via-thermal-resistance(current capacity from the IPC-2221 fit),thermal-via-array(the plated barrel is copper whatever the fill),stackup-builder(L3 solved as a stripline, L1 with copper thickness, copper counted in the total),coplanar-waveguide(grounded CPW stays finite for wide strips),i2c-pullup(V_OL is 0.2·V_DD at or below 2 V),vibration-phase-noise(the random term falls with offset; the deviation is the peak),inrush-current-limiter(the cold-start peak is V/R_cold),shielding-effectiveness(reflection loss carries η₀; multiple reflections added),ethernet-cable(Category 5e at 10 Gb/s fails),ground-plane-impedance(inductance µ₀·h·l/w; one-sided AC resistance) andminimum-conductor-spacing(IPC-2221B Table 6-1 by band, with its per-volt rates above 500 V).New optional inputs:
separationonground-plane-impedance(mm, default 1) andtempRiseonvia-thermal-resistance(°C, default 10). A request that omits them gets the default, which reproduces the published worked examples.New outputs:
hotRestartInrushoninrush-current-limiterandmultipleReflectionLossonshielding-effectiveness.Outputs that have no value now come back null with a warning, where they used to be a number:
waveguide-cutoff's guide wavelength and phase and group velocities below cutoff, andstackup-builder's width for a target impedance the layer cannot reach.comparator-hysteresisandpcb-trace-tempchange only their labels and stated formula: no number changes.POST /api/py/v1/calculate/solverefuses agridfiner than 1×10⁻¹² of the range's magnitude with a 400gridrefusal, before any evaluation and without a charge. Such a grid could hold a request until it timed out.radiated-emission-estimate: margin against CISPR 32 Class B, null where it sets no limit
POST /api/py/v1/calculate (radiated-emission-estimate)POST /api/py/v1/calculate/batchMCP run_calculationrftools-io SDK
The margin is now taken against the CISPR 32 Class B limit at the measurement distance: Tables A.4 and A.5, the 3 m row below √30 m and the 10 m row above it up to 1 GHz, the 3 m average rows from 1 to 6 GHz, 20 dB/decade from the tabulated distance, and the lower limit at a band edge. It used to be a flat 40 dBµV/m at every frequency and distance.
Below 30 MHz and above 6 GHz CISPR 32 sets no radiated limit, so
marginis null there, with a warning. It used to be a number. A caller that readsmarginas a number must handle null.The field is the free-space small-loop estimate, now with the exact constant π·µ₀/c. A measurement over a ground plane can read up to 6 dB higher, as the page says.
filter-designer and six RF calculators corrected
POST /api/py/v1/calculateMCP run_calculationrftools-io SDK
filter-designer: the Chebyshev prototype values were wrong from order 4 (g3 1.5733, now 2.3661), a first-order Butterworth gave g1 = 1 (now 2), and the band-pass element values were ten times too small at Q = 10. The values are now computed, not tabulated.waveguide-cutoff,free-space-path-loss,rf-link-budget,fresnel-zone,link-marginandradar-range-equationuse the exact speed of light, not 3×10⁸: lengths move by about 0.07 % and path loss by 0.006 dB.radar-range-equationalso uses the exact Boltzmann constant.planar-spiral-inductorrefuses an inner diameter at or above the outer one instead of returning NaN.Fourteen calculators were added on 2026-10-05 and 2026-10-06, including the new
iotcategory;GETthe catalogue orlist_calculatorsfor them.PDN impedance: a converged plane model, a derived port and two new parameters
POST /api/py/v1/jobs (pdn_impedance)POST /api/py/v1/jobs/estimateMCP simulate_pdn_impedancerftools-io SDK
An omitted
portX_mmandportY_mmnow place the port at the board's one-third point (a/3, b/3) rather than its centre. The centre is a null of every cavity mode with an odd index, so the first resonances never showed on a default run. A request that states the port gets the port it asked for.Two optional parameters:
portSize_mm(default 1 mm, 0.25–20), the side of the square through which the IC's current enters the planes, andboardInductance_nh(default 1 nH, 0–20), the inductance between the regulator and the planes, which was fixed at 1 nH. A port that overhangs a board edge is accepted and modelled with its image; a port whose centre is off the board, or that is larger than the board, is refused.Results change. The plane's modal sum was truncated relative to
freqMax_hzand, with a point port, never converged, so results moved with the sweep. It now converges and does not depend on the sweep below the first mode. At the default port the plane's spreading inductance is about 1.5 nH where it read about 0.2 nH, so the bare plane's series resonance falls from about 550 to about 210 MHz, and recommended capacitor sets change. Past results are not recomputed.The verdict and
summary.maxZPdnDbnow cover the whole decoupling band, which ends at 0.8 × the board's first cavity mode, with the worst case refined between grid points. New result fields:summary.band,summary.worstHz,summary.unreachable,resonances[].portCoupling, and the port, port size and board inductance used, underparams. Every existing field is kept.A target-solve call: search a calculator input for a target output
POST /api/py/v1/calculate/solve
POST /api/py/v1/calculate/solvefinds the value of one calculator input that brings one of its outputs to a target — the trace width that gives 50 Ω, the gap that gives a target differential impedance — and returns it with the forward result at exactly that value:values,warnings,errorsandprovenance, exactly what/calculatereturns for those inputs. One metered call replaces the several forward calls a client would otherwise spend searching by hand.The search runs over the input's stated bounds, narrowed by an optional
range; an input with no stated bound needs one. An optionalgridrounds the returned value to its nearest multiple inside the range, alongside the unrounded solution. A target the search cannot reach still succeeds, withreached: falseand the nearest value found, and is still charged. Refusals — an unknown calculator, an input or output name the calculator does not declare, a non-finite value, a missing range on an unbounded input — cost nothing, as for every other calculator route.The Python SDK's
Client.solve()(rftools-io0.4.0) and the MCP server'ssolve_calculationtool (2.2.0) expose the same call.A spent allowance on the calculator routes is 402, not 401
POST /api/py/v1/calculatePOST /api/py/v1/calculate/batch
The calculator routes answered 401 "Invalid or quota-exceeded API key" both for a key that was not valid and for an allowance that was spent, so a script could not tell a bad key from a busy month. A spent allowance is now 402, as the jobs route already answered, with
errorKind: "rate_limited",reason,limit,used,resetAt,upgradeUrl,overageUrland aRetry-Afterheader.401 now means only that the key is missing, malformed, unknown or revoked, and never mentions an allowance. A request with no key at all is 401 with a
keyUrlwhere a free key can be had.A call is checked before it is counted: an unknown calculator (404) or an input that is not a finite number (400) costs nothing, and a calculation that fails is given back. In a batch, only the items that pass those checks are counted, and items whose calculation fails are given back. Branch on the status and
errorKind, never on message text.The monthly allowance belongs to the account, shared by all its keys
POST /api/py/v1/calculatePOST /api/py/v1/calculate/batchPOST /api/py/v1/jobs
Each key used to carry its own monthly allowance, so an account holding several keys could spend its allowance once per key. The allowance is now the account's: the calls of every key it holds, revoked ones included, count against one monthly figure, which resets at the start of each calendar month in UTC.
An account with one key sees no difference. Calls are still attributed to the key that made them, and the dashboard and
GET /api/py/v1/usagelist them per key.A usage endpoint, and usage headers on every metered response
GET /api/py/v1/usageX-Usage-AllowanceX-Usage-UsedX-Usage-ResetX-Usage-Overage
GET /api/py/v1/usageanswers a key (X-API-KeyorAuthorization: Bearer rfc_…) or a signed-in session with the tier, the allowance, what is used and what is left, when it resets, the overage state and the calls attributed to each of the account's keys. Reading it is never counted.Every metered response carries
X-Usage-Allowance,X-Usage-Used(including that call) andX-Usage-Reset, andX-Usage-Overagewhen overage was counted. Browsers can read them.Every result says how it was computed
POST /api/py/v1/calculatePOST /api/py/v1/calculate/batchGET /api/py/v1/jobs/{jobId}
Calculator results and each batch item carry a
provenanceobject with nine members:method,version,formulaRef,assumptions,validRange,computedAt,inputs,seedandelapsedSeconds. Job results already carriedprovenance; it gainsformulaRef,assumptions,validRangeandcomputedAtrather than a second object, and itsversionis writtenworker@<revision>rather than the bare revision.An input outside the range a calculator is stated for is still computed:
validRangenames it and a warning is added. A result stored before this change keeps the members it had; treat an absent one as unknown.Programs can get a key through the device authorization grant
POST /api/py/v1/oauth/device/codePOST /api/py/v1/oauth/device/token/device/
A program with no browser in it can get a key without its user pasting one (OAuth 2.0 device authorization grant, RFC 8628). It asks
/oauth/device/codefor a code, shows its user the code and the address https://rftools.io/device/, and polls/oauth/device/tokenat the interval it was given. Once the user signs in and approves the code, the next poll returns an ordinary API key, once.The key is metered, listed and revocable like any other, and counts toward the account's key limit.
Key links for clients, and labels on keys
POST /api/py/v1/apikeys/dashboard/?newKey=1
POST /api/py/v1/apikeysaccepts an optional{"label", "client"}body and returns both; leaving the body out works as before. A client can send its user tohttps://rftools.io/dashboard/?newKey=1&client=<clientId>to create a key labelled for it, shown once.Refusals name the limit and the way past it
429 responses402 from POST /api/py/v1/jobsAuthorization: Bearer rfc_…
The per-minute 429 carries
errorKind: "rate_limited",limit,windowSecondsandupgradeUrlbesidedetail, as every other refusal carries itserrorKind. The jobs route's 402 for a spent allowance gainsreason,limit,used,upgradeUrlandoverageUrl, and the 402 for a spentfdtd_sparamsolve-mode allowance carriesreason: "solve_mode". Existing fields are unchanged.The calculator routes accept a key as
Authorization: Bearer rfc_…as well asX-API-Key, as the jobs routes already did.
The full reference is the API documentation; the machine-readable description is the OpenAPI playground and the OpenAPI document.