Hand-considered hypotheses, 5001–5100

Written one at a time, not composed. The hundred entries here are restricted to the physical and chemical mechanisms of pharmaceutical manufacturing: wet and dry granulation, drying, blending, compaction and tabletting, film coating, freeze-drying, and the transfer of these operations between scales. No label is repeated, so the variety is structural as well as topical; each entry states a claim, a mechanism, and a test that could separate it from the obvious alternative. None has been tested. No data are reported and nothing is cited; where a magnitude appears it is an estimate and is marked as one.

Wet granulation: nucleation, growth, and the liquid balance

Misattribution
H5001. Granule growth in a high-shear mixer is generally charged to binder viscosity, but once the binder is thick enough to immobilise the liquid bridges the growth rate is governed by pore saturation, and viscosity mainly sets how fast that saturation is approached. Two binder grades added at the same liquid-to-solid ratio but differing several-fold in solution viscosity should give the same median size after the same number of impeller revolutions, while matched viscosity at two saturations should not. The predicted viscosity-insensitive region begins above roughly 50 mPa·s for a typical polymeric binder solution (estimate).
Hidden additive term
H5002. Impeller torque is read as a proxy for wet-mass consistency, yet the signal also carries the powder's weight in the bowl and the bowl's own drag, both of which change with fill level. Dividing torque by the mass charged should collapse measurements made at different fill fractions onto one curve, and any residual fill dependence is the geometry term. Running one formulation at three fill levels, with impeller speed and binder addition fixed, separates the two contributions without a new sensor.
Ceiling set by breakage
H5003. The largest granule a high-shear mixer sustains is set by the balance between coalescence and breakage at the impeller tip, and at a fixed tip speed that ceiling belongs to the wet mass rather than to the blade pattern. Changing blade angle should alter how quickly the ceiling is reached, not where it lies. Running two impellers of different blade angle at the same tip speed, to the same revolution count, and sieving at the same elapsed time tests the distinction.
Threshold from a stress ratio
H5004. Below a minimum impeller speed, nuclei are wetted and coated by binder but never consolidated, so granule size stops responding to time however long the batch runs. The threshold is set by whether collision stress in the bed exceeds the wet granule's yield stress, which means it should move with binder concentration but not with spray rate. A speed ladder at fixed spray rate, sampled at fixed revolution counts, locates the threshold where the size-versus-time curve flattens.
Identity of two complaints
H5005. Overgranulation and poor flow are tracked separately, yet both may be one state: a saturated pore network consolidated into a dense, low-porosity granule. If that is so, the two should co-occur along a saturation ladder and no formulation change should move one without the other. Measuring flow function and granule porosity on the same samples across the ladder would confirm the identity.
Scale argument
H5006. Geometric scale-up at constant impeller tip speed preserves collision velocity but not spray flux, because liquid delivery scales with batch mass while powder surface scales with mass to the two-thirds power. The quantity to hold constant is liquid volume per unit powder surface per unit time, which forces a tenfold batch to run about 2.2 times longer at the same relative fill rather than ten times longer (estimate). Two batch sizes run at matched spray flux and at matched tip speed should differ in granule porosity only in the first case.
Drying artefact in the ranking
H5007. Size distributions compared after drying rank formulations partly by how much each binder shrinks as it loses water, so a film-forming binder can look coarser than a sugar binder when wet and finer after drying. Measuring the same batch before and after drying, by wet sieving or by in-process imaging, should show the ranking invert. The shrinkage term is computable from the binder's solids volume fraction, which makes the correction testable.
Conservation argument
H5008. The binder concentration delivered to the powder is lower than the concentration prepared, because a fraction adheres to the bowl, the impeller and the spray line, and because water evaporates during addition. Closing a liquid balance by weighing the bowl before and after and by trapping the exhaust should account for the difference, which can reach a fifth of the charge in small batches (estimate). If the balance does not close, the missing liquid is inside the granule and the prepared concentration is not the granulation concentration.
Non-monotone response
H5009. Raising impeller speed first increases granule size by promoting coalescence and then decreases it once breakage dominates. The turning point should move to higher speed as binder concentration rises, because a thicker binder film raises the granule's yield stress. A speed ladder at two binder concentrations, with size measured at fixed revolutions, tests whether the curves shift or cross.
Count
H5010. A wet-granulation regime map has at most three independent coordinates — pore saturation, a collision-intensity measure, and a binder-to-solid ratio — so a map drawn with more axes is projecting the same space twice. Dimensional analysis supports the count, because the remaining candidate variables enter only through these groups. Fitting a growth model to data from a two-dimensional grid should show the third axis adding no independent variance.
Definition trap
H5011. Spray flux is defined with a droplet diameter, and models variously use the volume-median or the Sauter mean, which differ by a factor of two or more for a real spray (estimate). Because flux enters regime predictions linearly, one process can be classified as nucleation-dominated or layering-dominated depending on which diameter was chosen. Recomputing a reported regime boundary with both diameters, from one droplet measurement, exposes the sensitivity.
Latent variable
H5012. Feed powder water content is logged, but the variable that governs nucleation is the water activity of the particle surface, which depends on the formulation's sorbed water and on the humidity of the room where the powder was stored. Two feeds matched by loss on drying yet stored at different humidity can therefore nucleate differently, and conditioning both to matched water activity should remove the difference. Splitting one batch into three humidity conditions and drying all back to the same loss on drying tests which variable predicts the nucleation fraction.

Dry granulation and roller compaction

Hidden machine compliance
H5013. Specific compaction force is computed as force divided by ribbon width, yet part of the applied force is absorbed by elastic closure of the roll gap, so the powder never sees the quoted value. Machine stiffness, measurable by running an empty gap against a hardened shim, supplies the correction term. Two machines with the same nominal force but different frame stiffness should make ribbons of different density, and the difference should vanish once force is corrected for gap deflection.
Misattribution to force
H5014. Ribbon density variation along a compacted strand is usually blamed on roll force, but at constant force the dominant cause can be feeder slip: an uneven screw delivery makes the nip fill fluctuate and the ribbon thickness oscillate. Recording screw torque and ribbon thickness together should show the thickness oscillation lagging torque by a fixed angle. A run at constant screw speed with a deliberately pulsed feed rate would reproduce the banding without any change in roll force.
Biting condition
H5015. Roll compaction bites only when powder-to-roll friction exceeds the tangent of the nip angle, so below a critical combination of gap and roll speed the material slips and the ribbon starves. The condition is geometric and depends on internal friction rather than on particle size directly. A speed and gap ladder on one blend, recording ribbon continuity, locates the slip threshold, which should be independent of the roll force setting.
Lower bound on porosity
H5016. The lowest porosity a ribbon reaches at a given force is bounded by the nip's release geometry, because material that has not relaxed when the rolls separate retains elastic strain and springs back. The floor should move with the material's elastic recovery, measurable on a compacted slug in a press, not with the mill setting. Compacting one blend at a force series and measuring ribbon porosity immediately and after a day separates instantaneous from recovered porosity.
Identity of two readouts
H5017. Ribbon hardness and the size of granules milled from it are optimised in separate experiments, yet both may be readouts of one densification state, hardness measuring the bond area that milling then has to break. If so, granule median size against ribbon hardness should collapse across formulations and roll forces onto a single curve. Testing several formulations at several forces, with milling held constant, shows whether the two responses are one.
Rate similarity failure
H5018. Roll diameter is scaled to preserve nip angle, but strain rate in the nip rises as a larger circumference turns at the same surface speed, so ribbon strength need not be preserved. The governing group is the Deborah number, the material's relaxation time over its residence time in the nip, and it changes with diameter at matched surface speed. Compacting one blend on two roll diameters at matched nip angle and surface speed, then measuring ribbon tensile strength, isolates the rate effect.
Volume-definition artefact
H5019. Envelope density includes closed pores that a helium pycnometer would exclude, so a ribbon that looks a fifth porous by geometry can be nearly fully dense to the skeleton (estimate). The two measures diverge most for materials that fragment and trap air during compaction. Measuring both on the same ribbon segments gives the closed-pore fraction and shows which number a model needs.
Hysteresis
H5020. Re-compacting a ribbon at the same gap and force gives a denser ribbon than the first pass, because the first pass closed pores that do not reopen on unloading. The second pass therefore compacts a different material, which is why recycle ratio changes ribbon properties non-linearly. Running one, two and three passes at fixed settings and measuring porosity after each separates the irreversible part from the reversible elastic part.

Drying: air, vapour, and the moisture that is actually water

Proxy substitution
H5021. Fluid-bed drying endpoints are often called on product temperature, but the temperature rise at the end of drying tracks the inlet air's humidity as much as the granule's moisture, so a humid day makes a dry batch look wet. The drying potential is the humidity ratio of the inlet air, not its temperature. Two runs at the same inlet temperature but different dew points, both stopped at the same product temperature, should leave different residual moistures.
Enthalpy bookkeeping
H5022. The heat carried by inlet air includes the latent heat of its water vapour, and at high humidity that term is a large fraction of total enthalpy, so inlet temperature understates the difference between two air streams. Computing enthalpy from dry-bulb and dew point should collapse drying curves that temperature alone keeps separate. A pair of runs matched in temperature and differing in dew point by a few grams per kilogram tests the collapse (estimate).
Diffusion ceiling
H5023. Once the surface stops being saturated, the drying rate is bounded by vapour diffusion through the partially dry shell, and no increase in air velocity can raise it because the resistance has moved inside the particle. The bound is set by shell thickness and effective diffusivity, both fittable from a single falling-rate curve. Drying the same granules at two air velocities and finding identical rates after the critical point confirms that external resistance no longer matters.
Critical-point threshold
H5024. A drying curve has a critical moisture at which the surface stops being fully wetted, and that point is a property of the pore structure rather than of the air. Below it the rate falls, above it the rate is set by the external mass transfer coefficient. Measuring the curve at two air velocities should shift the constant-rate branch but leave the critical moisture unchanged, which separates a pore threshold from an equipment effect.
Synonymy claim
H5025. Case hardening and skin formation are treated as different defects, yet both are a low-porosity surface layer whose diffusivity is below the core's, and both should arise at the same drying intensity. If they are one phenomenon, the two names should map onto a single dimensionless drying rate, and no formulation should show one without the other. Drying slabs at a series of rates while measuring surface porosity and internal moisture profiles tests the mapping.
Square-law scaling
H5026. Tray drying of a deep bed is limited by diffusion, so drying time scales with the square of bed depth rather than linearly, and the square law holds only while a moisture gradient persists. Halving the depth should cut the time by roughly four, not two, in that regime (estimate). A bed-depth ladder at one air condition, recording time to target moisture, shows where the square law breaks and surface resistance takes over.
Measurand substitution
H5027. Loss on drying is read as water, yet it also removes residual solvent and volatile binder components, so a formulation change that alters the binder can move the number without changing the water. Thermogravimetry with a moisture-specific sensor, or Karl Fischer titration on the same sample, separates the contributions. A drying specification written against loss on drying alone therefore specifies total volatiles, not water.
Energy closure
H5028. An energy balance across a fluid-bed dryer, inlet enthalpy minus outlet enthalpy against the enthalpy of evaporation from the measured moisture loss, should close within instrument accuracy. A persistent gap points to heat lost through the column walls or to water released by a reaction, and the sign of the gap says which. Running the balance on an empty column and on a loaded one at the same conditions isolates the wall term.
Inversion of a monotone expectation
H5029. Raising inlet temperature usually shortens drying, but above a point it can lengthen it by forming a dense surface skin that blocks vapour, so the time-temperature curve has a minimum. The minimum should sit at lower temperature for formulations with a film-forming binder, because the skin forms earlier. Drying one batch at four inlet temperatures and recording the time needed to reach the target moisture locates the minimum and tests the mechanism.
Equipment crossover
H5030. As bed depth increases, fluid-bed and tray drying of the same granules cross over: the fluid bed wins at shallow beds through external mass transfer and loses at deep beds through back-mixing and bypass. The crossover depth should depend on air flow per unit mass rather than on equipment type. Drying identical batches at three depths in both configurations at matched inlet conditions should show the curves cross once.
Aliasing
H5031. Outlet humidity sampled more slowly than the bubble eruption frequency folds the eruption cycle into a slow oscillation that looks like a real drying wave. The apparent period should change when the sampling interval changes, which is the signature of aliasing rather than of the process. Recording at ten times the bubble frequency for one batch and decimating the record afterwards separates the two.
Depth masking
H5032. Near-infrared moisture measurement penetrates only the outer few hundred micrometres, so a granule with a dry skin over a wet core reads dry, and the discrepancy grows with drying intensity. The failure is a sampling-depth mismatch rather than an instrument error. Comparing near-infrared values with loss on drying on the same granules, before and after milling, should show the gap collapse when the core is exposed.

Blending and segregation

Sampling artefact
H5033. Blend uniformity failures are usually charged to the mixer, but a thief sample drawn from a bed of free-flowing granules preferentially collects the fine fraction that flows into the cavity, so the sample is not the blend. The bias is a function of sampler geometry and bed permeability, not of mixing time. Comparing thief samples with whole-container assays, or with samples from a frozen bed, should show a systematic offset that no mixing change removes.
Hidden charge term
H5034. Powder cohesion in a blender contains an electrostatic term that depends on the particles' charge history, and that term can rival van der Waals attraction at low humidity. Because charge is generated by the same tumbling that mixes, the blender changes the interaction while it mixes. Grounding the vessel and ionising the headspace, with everything else fixed, changes blend RSD only if the electrostatic term matters.
Counting limit
H5035. The best uniformity attainable in a blend is bounded by the number of minor-component particles in a tablet-weight of blend and by the size ratio of the components, so a blend cannot be more uniform than counting statistics allow. The bound is a counting limit, not a mixing limit. Measuring RSD at two tablet weights from the same blend should show the smaller failing at a uniformity the larger achieves.
Percolation threshold
H5036. Lubricant blending has a percolation threshold: below a critical surface coverage the lubricant sits in isolated patches and compaction is unaffected, and above it a continuous film forms and strength falls sharply. The threshold should appear as a break in the strength-versus-blending-time curve rather than as a gradual decline. Compacting blends at a series of lubricant loadings and times, with strength measured on one press, tests for the break.
Identity of mechanisms
H5037. Segregation and demixing are treated as separate failure modes, but both are differential motion of particles under a driving field, so a blend that segregates by one mechanism should segregate by the others at a different rate. If so, formulations should rank in the same order whether tested by vibration, by pouring or by tumbling. Testing a set of blends in three segregation assays and comparing the orderings would show whether the mechanisms are one.
Similarity mismatch
H5038. Blender scale-up by Froude number preserves the centrifugal-to-gravitational ratio but not the number of revolutions needed for a given uniformity, because circulation time grows with vessel diameter. The relevant scaling is revolutions, and revolutions and time coincide only at constant Froude number. Running pilot and production blenders at matched Froude number, sampling uniformity against revolution count, shows whether the curves collapse.
Sample-size artefact
H5039. Blend RSD depends on sample mass, so two studies of the same blend can disagree twofold because one took 1 g samples and the other 10 g (estimate). Sample size enters through counting statistics, not through the mixer's performance. Re-assaying archived samples at two masses, or splitting each sample, separates sampling variance from blend variance.
Optimum with over-blending
H5040. Longer blending does not always improve uniformity: beyond a point, fine cohesive particles form agglomerates that shear cannot break, and RSD rises again. The reversal should be more pronounced for blends carrying a small fraction of fines below about ten micrometres (estimate). Blending one batch and sampling at a series of revolution counts, rather than stopping at a fixed time, exposes the minimum.
Fill-dependent crossover
H5041. V-blenders and bin blenders cross over as fill volume rises: the V works well at partial fill because it splits and recombines the bed and poorly when overfilled, while the bin's shear-driven motion is less sensitive to fill. The crossover should occur at a fill fraction set by vessel geometry rather than by formulation. Running one blend in both at three fill heights, with revolutions matched, locates the crossing.
Transport delay
H5042. In continuous blending, a feeder's response to a setpoint change reaches the outlet after a transport delay independent of the blend's mixing behaviour, so a loop tuned on the blender's response will oscillate. The delay is set by volumetric holdup and throughput, both measurable with a tracer. A step change in one feeder's rate, with outlet composition recorded, gives the delay directly and separates it from the mixing time constant.

Compaction and tabletting: force, time, and the die

Misattribution of fracture mode
H5043. Capping is usually attributed to air entrapped during compression, but at production speeds the commoner cause is elastic recovery exceeding the tablet's tensile strength on decompression, and the two leave different fracture patterns. Entrapped air opens a horizontal crack near the band, while recovery separates a cap at the edge. Compressing one formulation with and without a precompression stage should distinguish them, since the stage lets air escape.
Composite force
H5044. Ejection force is read as die-wall friction, but it also contains the elastic recovery that presses the tablet against the die as the punch withdraws, so a formulation with high recovery shows a high ejection force even at low friction. Ejecting tablets of two thicknesses at one diameter and recording the force at the moment of movement separates the recovery term, which scales with axial strain, from friction, which does not. The correction matters because lubricant studies otherwise credit the lubricant with a recovery effect.
Saturation ceiling in strength
H5045. Tablet tensile strength is bounded by interparticle bond density, which rises with relative density and saturates once the contacts are fully formed, so no formulation can be strengthened indefinitely by pressure. The bound should appear as a plateau in strength against porosity, with the plateau level set by bond area per unit volume. Compressing a series of tablets to porosities between 5% and 40% (estimate) and locating the plateau tests it.
Transition threshold
H5046. The brittle-to-ductile transition in a compact occurs where the particles' yield stress is reached before fracture, and it should show as a change in slope of a Heckel plot rather than as a change in tablet strength. The transition pressure moves with temperature and moisture, both of which lower the yield stress. Compressing the same powder at two temperatures and locating the slope change tests whether the transition is a material threshold rather than a press artefact.
Single-cause identity
H5047. A lubricant level that lowers tablet strength almost always lowers ejection force too, which suggests both are readouts of one boundary film of magnesium stearate rather than two independent effects, since that film lowers interparticle bonding and die-wall friction together. If so, no lubricant level should lower strength without also lowering ejection force, and the two effects should scale together across lubricant grades. Compacting a lubricant ladder and measuring both responses on the same tablets tests the coupling.
Parameter count
H5048. A porosity-strength relationship for a brittle compact has two independent parameters — an intrinsic strength at zero porosity and a sensitivity constant — so a three-constant model is absorbing batch variation. The same two constants should describe tablets made at different speeds if the mechanism is unchanged. Fitting one- and two-parameter forms to a speed ladder shows whether the extra constant earns its place.
Time-scale mismatch
H5049. Compaction on a rotary press differs from a simulator in dwell time even when peak pressure is matched, and the difference matters for materials whose strength depends on time under load. What governs is the ratio of the powder's relaxation time to the dwell time. Compressing one blend on a simulator programmed to the rotary dwell time, and on the rotary press itself, should match strength only if that ratio is preserved.
Formula artefact
H5050. Tablet hardness from diametral compression is converted to tensile strength through a formula that assumes a Poisson ratio and a plane-stress state, and both assumptions fail for very thin or very thick tablets. The reported strength therefore carries a thickness term that is not a material property. Measuring one formulation at three thicknesses and comparing with a three-point bend on the same tablets exposes the dependence.
Sign reversal with speed
H5051. For brittle materials, faster compaction can give stronger tablets, because particles fracture rather than deform and fresh surfaces bond before the load is removed. The reversal should appear as a positive slope of strength against compression speed up to some speed, turning negative once plastic deformation dominates. Compressing a brittle and a plastic grade over the same speed ladder and comparing slopes tests the sign.
Cancellation point
H5052. The sensitivity of tablet strength to compression speed changes sign between plastic and brittle materials, so a blend of the two can be speed-insensitive at the composition where the contributions cancel. That composition should be predictable from the pure components' speed sensitivities. Compressing blends across a composition range at two speeds, and finding where strength is unchanged, tests the cancellation.
Conversion assumption
H5053. Tensile strength computed from a hardness test depends on the tablet's Poisson ratio, which is rarely measured, so two formulations with equal hardness can have different true tensile strengths. The error is largest for materials whose Poisson ratio lies far from the assumed 0.2 to 0.3 (estimate). Measuring Poisson ratio by resonant ultrasound on the same tablets and recomputing shows which comparisons survive.
Unit trap
H5054. Hardness is reported in kiloponds, newtons or arbitrary units depending on the instrument, and conversion is not a single factor when test geometry differs. A specification written in one unit and verified in another can pass a tablet that fails the intended mechanical criterion. One audit re-measuring a lot on two instrument types, converting through force rather than through the displayed number, catches the discrepancy.
Path dependence
H5055. Precompression changes the main compression: a tablet precompressed at low force and then taken to the same peak force is denser than one compressed in a single step, and the difference survives ejection. The mechanism is that precompression lets air escape and permits particle rearrangement before the load path stiffens. Compressing one formulation at two precompression forces while holding the main force fixed, then measuring density and ejection force, quantifies the path dependence.
Coverage saturation
H5056. Die-wall friction falls with lubricant loading up to a surface coverage at which the film is complete, and beyond that point extra lubricant adds no further reduction while continuing to lower strength. The saturation coverage is set by lubricant particle size and die surface roughness. Compacting a loading ladder and plotting ejection force against loading locates the plateau, which should move with lubricant grade if coverage is the mechanism.
State versus amount
H5057. Tablet strength varies between granule lots at equal moisture content, and the hidden variable may be where the water sits rather than how much there is: water at interparticle contacts plasticises them, while water in closed pores does not. Two lots equilibrated to equal loss on drying yet differing in sorption history should therefore differ in strength. Dynamic vapour sorption and compaction on the same samples tests the location hypothesis.
Delayed expression
H5058. Viscoelastic relaxation during ejection can mask a capping tendency, because a tablet that relaxes slowly is still under radial stress when it leaves the die and relieves that stress later as a crack. The crack therefore appears after the press, which is why capping is often blamed on downstream handling. Measuring tablet strain immediately after ejection and again an hour later, with acoustic emission, shows whether the crack forms in the die or later.
Null result
H5059. If polishing a punch beyond a certain smoothness produces no further reduction in sticking, then sticking is not controlled by punch surface finish but by a film of formulation adhering to the punch. The null result is informative because it bounds the contribution of tooling finish. Compacting one blend on punches at three roughness levels and measuring the force needed to detach the tablet tests the bound.
Irreversibility
H5060. Work hardening of punch tips is irreversible, so a press that has run for years at production speed has different tooling from a new one even at the same dimensions. The change should appear as a gradual rise in the force needed to reach a given tablet thickness rather than as a dimension change. Measuring punch tip hardness and the press's force-thickness curve at intervals over a long run shows whether the two drift together.
Volume balance
H5061. Die fill is a volumetric transaction: the volume delivered by the feeder must equal tablet volume plus loss, and most fill problems are failures of that balance rather than of the feeder's setting. Weighing tablets and measuring dimensions over a run, then comparing implied fill volume with the die's nominal volume, locates the loss. The balance also separates a worn die, which changes volume, from a feeder that under-delivers.
Symmetry assumption
H5062. Rotary presses are assumed to apply equal and opposite forces through upper and lower punches, but the force path is not symmetric because each punch bears on a different compression roller. Asymmetry in measured forces should therefore be normal, and a symmetric reading indicates a faulty load cell. Instrumenting both punches at one station and comparing forces across a cycle tests the assumption.
Common-node inference
H5063. When two lots of tablets show the same strength loss, a shared raw-material supplier or a shared granulation line is a likelier explanation than a shared formulation, because the formulation did not change. Comparing the lots' granule porosity and binder distribution, rather than their tablet strength, should reveal the common node. The coincidence is a clue to an upstream cause, not a material property.
Survivor selection
H5064. Friability testing discards tablets that break during the test, so the reported value describes the survivors and a lot with a few very weak tablets can appear acceptable. Counting the tablets that fail, not only the mass lost, changes the ranking of formulations. Running the test on a lot spiked with a known fraction of weak tablets, reporting both mass loss and breakage count, quantifies the selection.

Film coating: droplets, passes, and the barrier

Rate-versus-flux confusion
H5065. Coating defects in a perforated pan are often attributed to spray rate, but at fixed rate the defect count tracks residence time per pass, which sets how long a tablet stays in the spray zone and how much liquid it receives per pass. What matters is the liquid deposited per unit of tablet area in each pass, not the total rate. Two runs delivering the same total rate at different pan speeds should differ in defect rate, and the difference should follow residence time.
Hidden driving force
H5066. Evaporation capacity in a pan coater is computed from inlet temperature and flow, but inlet humidity is a hidden term that can remove much of the available driving force on a humid day (estimate). Drying rate is set by humidity ratio and air velocity over the bed, not by temperature alone. Two runs matched in temperature but differing in dew point should give different film formation, with the humid run showing more orange peel.
Percolation bound on barrier
H5067. The minimum film thickness that resists acid penetration is bounded by the percolation threshold of the coating's pores: below a critical thickness, pores connect through the film and the barrier fails whatever the polymer chemistry. The threshold should scale with pigment volume concentration, because pigment particles create the pore network. Coating tablets at a series of thicknesses and measuring both acid uptake and film porosity locates it.
Droplet-drying threshold
H5068. A sprayed droplet forms a film if it reaches the substrate before it dries and a spray-dried particle if it dries first, so the transition is a threshold in the ratio of droplet drying time to flight time. The threshold moves with droplet size squared, which is why finer sprays form films more easily at the same solids content. Coating at two spray pressures, with droplet size by laser diffraction and film morphology by microscopy, tests the ratio.
Morphology identity
H5069. Orange peel and spray-dried particles are catalogued as separate defects, yet both reduce to one event: a droplet that dried before it coalesced with the film. If so, the two should appear at the same spray conditions and their surface feature size should match the droplet size distribution. Coating at a series of spray rates and comparing feature size with measured droplet size tests the identity.
Area scaling
H5070. Coating scale-up by pan speed preserves Froude number but not spray flux per unit tablet surface, because bed surface area grows more slowly than mass. Holding liquid volume per unit tablet area per unit time constant forces a longer run at larger scale rather than a higher spray rate. Two pan sizes run at matched Froude number and matched flux should give the same thickness distribution, while matched tip speed should not.
Gravimetric artefact
H5071. Film thickness is usually inferred from weight gain, but the gain includes residual moisture and, for aqueous coatings, water still in the film when tablets are weighed. The discrepancy is greatest for freshly coated tablets and shrinks over hours. Weighing the same tablets immediately, after equilibration and after vacuum drying separates the water term and shows whether a thickness specification is really a moisture specification.
Mass closure
H5072. Polymer sprayed must end up on tablets, on pan walls or in the exhaust filter, and a mass balance across a run usually shows wall loss that the nominal weight gain ignores. Measuring polymer in the pan wash and in the filter against the mass sprayed closes the balance. If it does not close, the difference is overspray that missed the filter or a solids content different from nominal.
Optimum with attrition
H5073. Increasing pan speed improves uniformity by increasing passes through the spray zone, but beyond a point it damages the film by attrition and impact. The optimum should move to lower speed for friable cores, whose attrition threshold is lower. Coating a friable and a robust core over a speed ladder, with uniformity and defect count measured separately, tests the reversal.
Process crossover
H5074. Wurster and pan coating cross over in efficiency as batch size grows: the Wurster's per-pass residence time is short and uniform, while pan mixing becomes less efficient at large fill. The crossover should depend on the ratio of batch mass to air flow rather than on the machine's make. Coating the same cores at two batch sizes in both configurations, with thickness distribution by image analysis, locates the crossing.
Phase locking
H5075. Tablet samples taken once per pan revolution are phase-locked to rotation, so the same tablets are sampled every time and measured uniformity is better than true uniformity. Sampling at an interval that is not a multiple of the revolution period removes the lock. Taking samples at two non-commensurate intervals from one batch should give different RSDs if the lock is real.
Substrate variable
H5076. Core surface porosity is rarely measured but may control film adhesion more than formulation does, because a rough porous surface gives mechanical keying while a smooth one relies on interfacial bonding. Two cores of the same composition but different porosity should adhere differently at equal coating thickness. Measuring adhesion by peel or by a fatigue test on cores of graded porosity tests the hypothesis.
Optical masking
H5077. Opacifying pigments hide cracks, so a coating that looks uniform can be mechanically discontinuous and the defect appears only when the tablet is stressed. The masking is optical rather than physical. Inspecting the same tablets under transmitted light, and after a controlled strain, should reveal cracks that reflected-light inspection misses.
State dependence of curing
H5078. Curing a film is not reversible: a coated tablet held above the polymer's glass transition coalesces further, and the gain in barrier performance persists after cooling. A second cure therefore starts from a different state than the first. Curing one batch in two stages, with a barrier measurement between them, tests whether the second stage adds anything beyond the first.
Irreversibility of coalescence
H5079. Once a film has coalesced, redissolving it does not restore the original latex particle structure, so a reworked batch has different barrier properties from a virgin one at the same composition. What is lost is the particle-boundary network that slowed diffusion. Comparing virgin and redissolved films at equal thickness, by permeability and by microscopy, quantifies the irreversible part.
Two-regime discrimination
H5080. If two coating polymers of very different oxygen permeability give the same barrier at equal thickness, the barrier is set by pinholes rather than by the polymer, and the comparison is measuring defect density. A thickness series should separate the regimes, with pinhole-limited barrier falling steeply and polymer-limited barrier falling slowly. The null result would redirect the work from formulation to defect control.
Facial asymmetry
H5081. A tablet is not symmetric top to bottom, yet coating uniformity is usually reported as one mean, which assumes the faces are equivalent. The band and the faces have different curvature and spray exposure, so the assumption should fail for biconvex tablets. Measuring thickness separately on top face, bottom face and band at two pan speeds tests the assumption.
Mode count
H5082. Defects in a coating operation sort into at most three independent modes — incomplete coverage, mechanical damage and optical non-uniformity — because each maps to a distinct process variable. A classification with more categories is usually subdividing one mode by appearance. Coating at a factorial set of spray rate, pan speed and air flow, then testing whether defect counts collapse onto three factors, checks the count.

Freeze-drying: nucleation, sublimation, and the cake

Thermodynamic misattribution
H5083. Collapse temperature is usually identified with the glass transition of the maximally freeze-concentrated phase, but collapse is kinetic: the cake sags when viscosity falls below a threshold on the drying time scale. The two temperatures therefore differ, and the difference grows with drying rate. Drying one formulation at two shelf temperatures on either side of the transition, measuring cake height loss against time, separates the thermodynamic from the kinetic reading.
Radiative term
H5084. Vial heat transfer coefficients fitted from a run contain radiation from chamber walls and shelf edges, not only conduction through the vial base, so a value measured at one shelf position does not apply at another. The radiative term grows with the vial's view of the door, which is why edge vials dry faster. Mapping the coefficient across the shelf, with centre and edge vials, quantifies the term.
Transport ceiling
H5085. The maximum sublimation rate is bounded by choked flow through the dried cake: once vapour velocity in the pores reaches the speed of sound, lowering chamber pressure no longer raises the flux. Pore size and cake thickness set the bound, not shelf temperature. Drying at a series of chamber pressures at constant shelf temperature, and finding the flux plateau, tests the bound.
Nucleation control
H5086. Ice crystal size is set at nucleation: the number of nuclei formed in the first few degrees of supercooling fixes how many crystals grow, and the subsequent freezing rate matters less than that number. The threshold is the supercooling at which nucleation becomes probable, which varies from vial to vial. Nucleating a batch at a controlled temperature, by ice fog or pressure drop, and comparing with spontaneous nucleation, tests whether that moment fixes crystal size.
Failure-mode identity
H5087. Meltback and collapse are logged as different failures, yet both are loss of the cake's mechanical support when the frozen matrix softens, so they should occur at one product temperature. If so, the two are interchangeable labels for one event, meltback being the advanced case. Drying slabs over a temperature series and recording the temperature at which each fails would settle whether they are one.
Area-to-mass scaling
H5088. Shelf contact governs vial heat transfer, and contact area per unit mass changes with vial size, so a cycle scaled by shelf loading alone will not transfer. The controlling ratio is heat-transfer area over product mass, which falls as vials get larger. Drying the same fill mass in two vial sizes at one shelf temperature, and comparing drying times, tests whether that ratio predicts the difference.
Observer effect
H5089. Thermocouples inserted into product vials act as nucleation sites and change the freezing event they are meant to measure, so a monitored vial is not representative of the batch. The artefact should be largest for vials that would otherwise nucleate at high supercooling. Comparing product traces from instrumented and non-instrumented vials, the latter measured remotely, quantifies the bias.
Water closure
H5090. Water removed in primary drying must equal water frozen plus water desorbed, and weighing the batch before and after, plus integrating condenser accumulation, closes the balance. A gap points to water that bypassed the condenser or to a leak. Running the balance on a batch of known fill mass, with the condenser isolated, tests closure.
Reversal with crystallisation
H5091. Slower freezing usually coarsens ice, but in a formulation whose excipient crystallises during freezing, slower cooling gives more complete crystallisation and a different pore network, which can yield a finer cake rather than a coarser one. The direction of the effect therefore depends on the solute's crystallisation kinetics. Freezing at three rates, with X-ray diffraction to confirm crystallinity, tests it.
Limiting-mechanism crossover
H5092. Primary drying rate against shelf temperature crosses over from heat-limited to vapour-limited behaviour, and beyond the crossover extra heat adds no flux because transport through the cake limits it. The crossover should move with cake thickness. Drying cakes of two thicknesses over a temperature ladder, finding where flux stops responding, tests the shift.
Fitted coefficient
H5093. Dividing the sublimation flux by the difference between shelf and product temperature yields a vial heat transfer coefficient that inherits the errors of both and is not a property of the vial alone. It depends on chamber pressure because gas conductance between shelf and vial base changes with pressure. Measuring the coefficient at three pressures with the same vials shows whether one number is adequate.
Measurement location
H5094. Chamber pressure is usually measured at the wall, but pressure at the product is higher during sublimation because gas flows through the cake, so the wall reading understates local pressure. The difference is the pressure drop across the cake and depends on flux. Comparing a wall gauge with a gauge near the product at two fluxes quantifies the drop.

Scale-up and transfer between equipment

Similarity failure
H5095. Pharmaceutical scale-up cannot match every dimensionless group that governs a process, because geometric similarity fixes some ratios while heat and mass transfer fix others, so a choice must be made about which group to preserve. That choice is often left implicit, which produces a process that looks similar and behaves differently. Running pilot and production units at matched values of each candidate group in turn, and comparing the response that matters, identifies the controlling group.
Bottleneck migration
H5096. The rate-limiting step in a manufacturing train changes with scale, so a parameter critical in the pilot plant can be irrelevant in production. At small scale wall heat transfer is often limiting, at large scale mixing or mass transfer becomes limiting. Measuring the same response at three scales, and finding where a change in the pilot-critical variable stops mattering, locates the shift.
Geometric hidden term
H5097. Scale-up by volume assumes heat and mass transfer areas scale with volume, but they scale with area, so the area-to-volume ratio falls as the two-thirds power of the scale factor. That hidden term is why drying and cooling take longer at large scale even when every intensive variable is matched. Comparing measured heat-up times at two scales with the ratio predicted from geometry alone tests the term.
Regime crossing
H5098. A process can cross a regime boundary during scale-up even when every dimensionless group is held constant, because the boundary depends on a ratio that is not among the groups being matched. The crossing should appear as a discontinuity in a response that was smooth at the smaller scale. Running a scale series through the suspected boundary, with the response measured continuously, shows whether the discontinuity is real.
Error cancellation
H5099. Two scale-dependent errors can cancel at one scale and not another, so a process validated at pilot scale can pass for the wrong reason. The classic pair is a longer drying time that over-dries and a lower heat transfer coefficient that under-dries. Measuring both terms separately at two scales shows whether agreement was cancellation, and perturbing one term at pilot scale should break it.
Invariance criterion
H5100. The only scale-up rule that survives a change of equipment is written in terms of a material response rather than a machine setting, because machine settings do not transfer between geometries. A target granule porosity or a target ribbon density is invariant, while an impeller speed or a roll force is not. Applying the same material target on two equipment types, and checking whether the product response matches, distinguishes the two kinds of rule.

Status

100 of the standing list. The numbering is continuous with the neighbouring blocks, and the subjects were chosen to cover the unit operations rather than to follow a grid: granulation, drying, blending, compaction, coating, freeze-drying, and scale-up. Each was written with its own logical form, and no form label is used twice.

What a list like this is worth depends on the tests, not on the entries. None has been tested, and no result is claimed.

If one of these is worth an experiment, it is H5006 — that matched spray flux, not matched tip speed, is what transfers between granulation scales, and that the correction is a factor of the cube root of the batch ratio. It is checkable with two batch sizes and one sieve.

Incoprea Research · hand-considered hypotheses 5001–5100 · 3 November 2026 · ~/Desktop/Nov3/