A balance does not become more accurate because you asked it for less. Repeatability, nonlinearity, sensitivity tolerance and the temperature coefficient of that sensitivity are specifications of the instrument, and together with air buoyancy they make up the combined uncertainty of a mass determination [1]. Under favourable conditions the relative uncertainty of weighing data is often in the 10-5 to 10-4 range, depending on the net and gross weights and on the densities involved [1]. That figure carries a large caveat, stated in the same work: it assumes weighing goods free of disturbances such as electrostatic charges, air draughts, evaporation and water adsorption [1]. When you are weighing two milligrams of a dry peptide powder on an open bench, those four disturbances are the whole problem.
Repeatability is an absolute scatter, so small samples pay for it
Weigh the same object ten times, lifting it off and replacing it each time, and the standard deviation of those readings is your balance's repeatability under your conditions: your bench, your room, your hands. That standard deviation is a quantity in milligrams, and the relative uncertainty it contributes therefore depends on the net weight [1]. The same 0.02 mg of scatter is 0.01% of a 200 mg sample and 1% of a 2 mg sample.
A minimum sample weight is simply the smallest net mass at which that relative contribution still meets a tolerance you have committed to. The algebra is worth writing out. If your tolerance is a relative uncertainty of 0.1% and you cover the repeatability standard deviation s with a factor of 2, the smallest defensible net mass is 2s/0.001, that is 2000s. Measure s = 0.015 mg on your own instrument, in your own room, and the minimum comes out at 30 mg — a number most people find uncomfortably large the first time they compute it. The coverage factor and the tolerance are yours to justify against whichever requirement you work to; the structure of the calculation is not negotiable.
Be precise about what those ten weighings license you to conclude. They characterise scatter, and scatter only. They say nothing about whether the balance reads high or low, because a bias repeats perfectly and so contributes nothing to a standard deviation. And weighing below your minimum does not simply add noise you can average away across replicates. A single weighing error fixes a single wrong number for that stock solution, and that number then propagates identically into every dilution drawn from it. The scatter is visible in your data; the offset is not.
The mechanism under the pan
A high-resolution balance is not a spring. It is a compliant mechanism with flexure hinges and a lever-based linear guide, in which an electromagnetic force compensates the load. The stiffness of that mechanism and its sensitivity to quasi-static ground tilt are design parameters that can be deliberately adjusted, for example by attaching trim weights to the levers of the linear guide; parametric study of a planar compliant mechanism with semi-circular flexure hinges showed that stiffness and tilt sensitivity can both be reduced significantly relative to the non-adjusted cell, and the effect of the trim weights was verified experimentally on a commercially available weighing cell [2]. That work concerns cells designed for 1 kg mass standards rather than milligram samples, but the structural point transfers: the indication depends on the geometry and adjustment of a jointed elastic mechanism.
Two bench consequences follow. Levelling is not cosmetic, because a cell with residual tilt sensitivity reads differently when the bench is out of level or has been knocked, which is why the level is checked and the balance re-adjusted after any move. And because the load is transmitted through a guide, placing the sample off-centre loads that guide differently: corner-load, or eccentricity, error is conventionally tested on its own rather than being folded into repeatability, because ten replicate weighings of a centred object cannot reveal it. Nonlinearity is a third and separate contribution [1], which is why a balance adjusted near full scale can still misread at the bottom of its range.
Buoyancy is a bias, not a wobble
Air buoyancy has been described as the largest systematic influence in mass determination, and correcting for it by the classical route requires the air density, which in turn demands simultaneous traceable measurement of air temperature, air pressure and relative humidity [3]. An artefact-based alternative was developed precisely because a field laboratory may not hold fully traceable instrumentation for those three room-climate parameters; it uses only a mass artefact, at the cost of a higher calculated uncertainty than the classical method [3]. Neglecting the systematic effect entirely, as proposed in the EURACHEM/CITAC guide against which that work compares, is the third option evaluated [3].
Air density itself comes from an internationally agreed equation of state, and that equation is tightly controlled: the CIPM-2007 revision, which adopted a redetermined argon mole fraction and an updated molar gas constant, resolved a relative deviation of 6.4 × 10-5 between gravimetrically determined air density and the earlier CIPM-81/91 formula [4]. That 6.4 × 10-5 is the residual disagreement in the air density value itself, not the size of the buoyancy correction; the correction scales with the volume of the object sitting on the pan, and none of the work cited here fixes its magnitude for a particular sample. It is systematic, and it does not change appreciably within a single weighing. If your reading is wandering by a percent, buoyancy is not the explanation and a buoyancy correction will not rescue it.
What static charge does to a dry powder
Dry organic powders acquire charge by contact and separation. Triboelectrification of α-lactose monohydrate in a cyclone charger fitted with interchangeable steel and polymer contact surfaces showed charge acquisition inversely related to particle size where contamination of the contact surface was negligible, and a more complex picture once powder adhered to that surface, since particles then collided with both clean and contaminated regions [5]. In a separate study, micronised salbutamol sulphate or ipratropium bromide monohydrate mixed with 63–90 µm α-lactose monohydrate in cylindrical vessels of stainless steel, polypropylene and acetal charged in a way that depended on both the powder and the vessel surface; the materials could be arranged in a triboelectric series with ipratropium at the positive end and polypropylene at the negative end, and coating a vessel with drug or excipient removed the influence of the vessel material on charge generation [6]. Raising relative humidity across the range 0–86% produced a corresponding decrease in charge and adhesion for each drug, for lactose and for the formulations [6].
Charge also leaves, and for anyone watching a display the decay rate matters more than the peak magnitude. After corona charging, salbutamol sulphate, ipratropium bromide monohydrate and α-lactose monohydrate all showed tri-phasic charge decay, at different rates ranked ipratropium bromide > lactose > salbutamol sulphate, and every material decayed faster as relative humidity was raised up to 86% [7]. This is consistent with the failure mode you meet at the balance: an indication that creeps in one direction over tens of seconds and then slows, in a dry room, with a sample that behaved perfectly in humid weather. What the cell is responding to is an electrostatic disturbance rather than a change in mass, which is why electrostatic charge is listed among the conditions that must be absent before low weighing uncertainties are achievable [1]. Note the analyte class before you generalise: these are micronised small-molecule drug substances and a disaccharide carrier, not peptides. The charging mechanism is a property of insulating particulate solids in contact with steel and polymer surfaces, and a lyophilised peptide cake ground to a powder is such a solid, but no charge magnitude from those experiments transfers to your material.
The remedy is to remove the charge rather than to average over it. Ionisers work by corona discharge: in a passive design characterised for roll-to-roll web handling, a grounded needle electrode releases ions once the field at its tip exceeds the corona inception condition, and an air flow ejected beneath the needle transports those ions to the charged surface, with performance quantified as a charge elimination current and a charge elimination rate [8]. That characterisation was performed on a moving charged roll, not on a weighing pan, so read it as evidence for the mechanism rather than as a specification for your instrument. The vessel-material result [6] has the more immediate bench consequence: the container is part of the experiment, and moving from a polypropylene boat to earthed glass or metal changes the charging behaviour rather than merely changing what holds the powder.
Water that arrives while you weigh
Freeze-dried solids are not dry in the sense of containing no water. They hold an equilibrium water content set by the water activity around them. In freeze-dried lysozyme–sucrose systems characterised by differential scanning calorimetry, thermogravimetric analysis, sorption calorimetry and small-angle X-ray scattering, residual moisture content increased with added lysozyme, and this followed from an increase in equilibrium water content with lysozyme concentration at constant water activity, with the glass transition temperature of the mixtures rising alongside [9]. That is a protein–disaccharide lyophile rather than a synthetic peptide, and the numbers do not carry across, but the principle does: the solid on the pan is in exchange with the room.
The effect is large enough to be a controlled variable in regulated gravimetry. Comparing manual and automated weighing of particulate-matter-loaded filters across several humidities, the two methods could in most cases be treated as referential in the range 30–55% RH, with very slightly better device stability at 30% RH, and the authors note that water vapour condensed on a filter can affect the particulate concentrations derived from the weighing [10]. Filters loaded with deposited particulate are not peptide powders, and the transferable claim is narrow: ambient humidity is a term in the mass you record, and a hygroscopic solid taking up water during the weighing produces an upward drift that re-tareing does not remove. It is worth separating this from a static drift. Moisture absorption and electrostatic charge appear as distinct entries in the list of disturbances that prevent low weighing uncertainty [1], and they call for different fixes: enclosure and humidity control for one, charge elimination for the other. Letting a cold vial reach room temperature before opening it belongs to the first category, not the second.
When the quantity is too small to weigh
Two routes exist when the mass you need falls below your minimum sample weight, and both trade the weighing problem for a different and better-characterised one.
Weighing by difference — weigh the closed vessel and its contents, dispense, reweigh the vessel — avoids taring a container and avoids depositing loose powder on a pan. It does not halve anything. Two independent weighings contribute, so their variances add, and the quantity you report is the difference of two much larger numbers. Its real advantage is procedural: the vessel stays closed and can stay earthed between readings, which addresses the electrostatic and hygroscopic failure modes above rather than the repeatability floor.
The stronger route is to weigh a mass you can defend, dissolve it completely, and deliver a fraction of the resulting solution gravimetrically. Gravimetric dispensing of microlitre drops has been characterised in exactly this spirit: a micropipettor modified to take removable microcapillaries was used to prepare liquid scintillation sources from a well-characterised 241Am reference solution, giving an independent radiometric check on the dispensed masses; controlled dispensing onto gold foils showed losses of no more than 0.34(4)% of the total drop activity, and a detailed measurement equation for the weighing technique was presented, including corrections for evaporation, with a full accounting of the associated uncertainties [11]. The analyte there was a radionuclide solution, not a peptide, and the transferable lesson is the structure rather than the figure: evaporation appears explicitly in that measurement equation, which is the warning. Solution transfer trades electrostatic and hygroscopic error for evaporative loss, and evaporative loss has to be measured, not assumed away.
What the weighing licenses about a molarity
Converting a weighed mass into a molar concentration multiplies three things you have to believe: the mass, the fraction of that mass which is actually the peptide, and the molar mass you divide by. The balance governs only the first.
The second is where a synthetic peptide differs from a small-molecule standard. Quantitative 1H NMR of two commercially sourced custom synthetic peptides, DR and DRVYI, uncovered mannitol as an undeclared constituent at 20 and 43% w/w, quantifiable even on a 60 MHz benchtop instrument [12]. Peptides made by solid-phase synthesis are additionally obtained as trifluoroacetate salts, because TFA serves both as cleavage agent and as ion-pairing reagent; a systematic study quantified residual TFA- by FT-IR, 19F-NMR and HPLC with evaporative light-scattering detection, exchanged it for Cl- by freeze-drying from HCl with 10 mM found optimal, and used the detected counterion contents to understand the impact of sequence differences, especially positive charges, on the amount and potential localization of counterions [13]. The weighed mass includes counterion and any excipient; the molar mass you calculate from the sequence does not.
This is why value assignment for peptide reference standards runs through a mass balance approach applied to a bulk material, which is then used as a standard to assign a value to the vialed material [14]. It is also why direct methods are attractive. In a multi-laboratory study on the nonapeptide oxytocin, comparing an HPLC assay, quantitative NMR and amino acid analysis, the HPLC assay using the same peptide bulk material as the standard showed the lowest inter-laboratory variability — with the authors' own caveat that the coefficient of variation was calculated without counting the uncertainty associated with the mass balance purity assignment of that standard — while qNMR, a direct measurement against an internal standard and not difficult to perform under common laboratory conditions, was judged to deserve further exploration as a primary method for value assignment [15].
The propagation arithmetic is unforgiving in one specific way. A weighing error is not noise on a concentration; it is a scale factor. Record 2.0 mg when the pan in fact held 2.1 mg and every concentration computed from that stock understates the truth by about 5%, in the stock, in each working dilution, and in every value referenced to them. The error cancels nowhere. Relative uncertainties in mass, in peptide content and in volume combine in quadrature, so it pays to know which term dominates before optimising any of them. A combined ISO-GUM and Monte Carlo evaluation of an HPLC-UV assay for a small-molecule active in a pharmaceutical preparation found the accuracy of the sample volume and of the calibration standard to be the dominant contributors, at 39.9% and 36.2% of the total uncertainty and 76.1% combined, with an expanded uncertainty at k = 2 of (99.41 ± 0.69)% [16]. That budget belongs to one assay of one compound and does not transfer, but the exercise does: build the budget for your own method and the term you have been polishing is often not the one that matters.
Record what makes the number reconstructable later: the balance and the repeatability standard deviation you measured on it, the net mass and whether it cleared your minimum, the vessel material, whether charge elimination was used, and the ambient humidity. A concentration reported without the weighed mass and its uncertainty cannot be re-derived by anyone, including you.