Hand-considered hypotheses, 3801–3900
Surface cleaning, finishing and preparation, written one at a time. Degreasing and the aqueous bath, rinsing and drying, pickling and passivation, blasting, polishing and lapping, plasma and laser cleaning, contamination control, adhesion preparation, and the verification that is supposed to prove any of it. Each entry carries the logical form chosen for it, printed above the claim, so the variety of argument can be audited against the text instead of taken on trust. None has been tested; no citation is given because none is invoked, and every quantity offered as an estimate is marked as one.
Degreasing, solvents and the aqueous bath
Misattribution
H3801. Vapour degreasing is credited to the solvent's solvency, yet in a working plant the cleaning is done by fresh condensate refluxing onto a cold part, so the removal rate collapses as the part warms toward the vapour temperature while the solvent itself is unchanged. The transport term is condensation flux, not dissolving power, and holding part temperature fixed while changing solvent separates them. A contaminated coupon run at three preheat temperatures in one vapour zone, with removal measured gravimetrically, decides it.
Hidden term
H3802. An alkaline cleaner's working strength is quoted as total concentration, but the part of it that removes oil is the free alkalinity left after the builder has been consumed by saponification, and that fraction falls as the bath loads with oil. The bath therefore weakens at a constant titration setpoint, which is why cleaning failures cluster late in a shift. Titrating free alkalinity against oil load at fixed total concentration, with cleaning rate measured on a standard oiled coupon at each point, shows whether the drop accounts for the failures.
Threshold
H3803. Ultrasonic cleaning has an amplitude below which the field holds only stable, oscillating bubbles that stream without cleaning, and the threshold is set by the liquid's dissolved gas and surface tension rather than by the transducer's rated power. Above the threshold the same field begins to erode soft metals, so the useful window is narrow and material-dependent. Mapping aluminium-foil erosion against amplitude at two dissolved-gas levels locates both edges of that window in one experiment.
Inversion
H3804. Adding surfactant past its critical micelle concentration can slow the removal of a tenacious oil, because micelles buffer the monomer concentration and flatten the interfacial-tension gradient that drives the oil to roll up. More surfactant then means less cleaning, which contradicts the dose-response habit behind bath top-ups. A concentration series across the CMC, on one oil at one temperature, shows the rate maximum sitting before the CMC rather than after it.
Trade-off
H3805. Temperature works both ways in an aqueous ultrasonic tank: it accelerates saponification and lowers viscosity, but it also raises vapour pressure and softens cavitation collapse. There is therefore a temperature of best cleaning that belongs to the bath and the part together, not to the chemistry alone. A temperature series at fixed chemistry, with soil removal and cavitation intensity measured in the same run, either finds that optimum or shows one effect is negligible.
Identity claim
H3806. Roll-up and emulsification are described as two mechanisms of oily-soil removal, but both are detachment of a contact line pinned until the local contact angle exceeds a threshold, so they differ in the angle reached rather than in kind. If that holds, one surface and one oil cannot show one mechanism and not the other at the same temperature. A series of surfaces with graded contact angles, cleaned identically and watched in situ, should show a single transition rather than two.
Scale argument
H3807. Rinse efficiency is usually argued from rinse-water concentration, but what a part carries in is a film whose thickness is set by drainage, so the controlling ratio is drag-out volume per part to rinse volume, not tank concentration. Below a certain ratio one well-agitated rinse does what three still rinses cannot. Adding stages while holding that ratio constant should change nothing, and the experiment is cheap enough to run before any capital decision.
Causal ordering
H3808. In solvent cleaning the residue left behind is decided by the last solvent that evaporates, not by the total volume used, so the order of the steps carries more information than their number. A pure final solvent that evaporates cleanly erases the history of the earlier ones; a contaminated final solvent writes that history back. Two sequences with identical solvent consumption but reversed final stages give different residue, and that is the whole test.
Hysteresis
H3809. Cleaning the same surface twice does not return it to the same state, because the first pass opens pores and fissures that then fill with the second pass's contaminated liquid, so the residue after drying differs although the chemistry applied was identical. Cleanliness is therefore a function of history, not of the last treatment. Repeating a standard clean on one coupon with an ellipsometric measurement between passes shows whether the second result converges or moves away.
Null claim
H3810. For an oil that is fully saponifiable on a hot alkaline surface, agitation should make no difference once the chemical step has become rate-limiting, because transport is then faster than reaction and the reaction sets the clock. The prediction is a flat line where the process window assumes a strong agitation dependence. Cleaning the same oiled coupon at four agitation speeds, at a temperature high enough to saponify quickly, should give that flat line or refute the claim outright.
Unit trap
H3811. Oil-removal efficiency is commonly reported as a percentage of the oil applied, but the quantity that matters to the next process is the residual film mass per area, and the two are not interchangeable because a percentage hides the loading. A cleaner that removes 99 per cent of a heavy film can leave more per square metre than one that removes 90 per cent of a light film. Reporting mg/m² alongside the percentage, on two loadings, shows whether the ranking survives the change of unit.
Count
H3812. A counterflow rinse line with n stages carries n−1 independent concentration ratios, so a control scheme that trims each stage separately is over-parameterized and its extra handles will fight one another. The physical content is the flow ratio and the drag-out, and every other setpoint follows from the mass balance. Perturbing one stage and counting how many others must move to hold the endpoint is a direct count of the real degrees of freedom.
Rinsing, drying and what the water leaves
Bound
H3813. Final cleanliness is bounded by the purity of the last liquid to touch the part, not by the number of rinses before it, because each stage equilibrates the surface with its own bath and the last one leaves the last film. A line with six stages and a contaminated final rinse is worse than two stages and a clean one. Deliberately spoiling the final rinse at graded conductivity, and measuring residue, gives the bound and its slope.
Artefact mimicking effect
H3814. The rings and spots left after drying are read as contamination, but their geometry is a map of where droplets sat and how the air moved, not of what was on the surface before the rinse. Solute concentrates at the pinned contact line as the droplet recedes, so a clean part can spot and a dirty one can dry clear. Drying the same cleaned part under three airflow patterns, with the pre-dry residue measured separately, separates the pattern from the contamination.
Crossover
H3815. Whether a droplet leaves by being blown off or by evaporating depends on the ratio of drag to evaporation at its own size, and that ratio crosses unity at a diameter set by air velocity and humidity. Below the crossover the droplet dries in place and leaves its solute; above it the droplet leaves first. A blotter of graded droplet sizes dried at two air speeds shows the crossover as a boundary in spot density.
Reversal
H3816. Hot air is used to dry parts faster, yet faster evaporation concentrates solute at the receding contact line and can leave a heavier ring than slow drying in still air, so the residue rises while the water falls. The mechanism is the coffee-ring transport that makes one stain out of one drop, merely hurried. Drying identical cleaned coupons in still air, warm air and warm filtered air, then counting rings, puts a sign on the effect.
Conservation claim
H3817. Solute leaving on the part equals what the flow carries away in a counterflow rinse, so the stage that limits the line is the one with the largest product of local flow ratio and local drag-out, not the dirtiest tank. Operators clean the dirtiest tank and leave the true bottleneck alone. Injecting a tracer at each stage and measuring the outlet concentration identifies the limiting stage without dismantling anything.
Lag
H3818. Inline conductivity responds to the tank's mixed volume rather than to the film on the part, so during a short rinse it reports a cleanliness the surface has not reached, and the lag grows with the tank's turnover time. Endpointing on it stops the rinse early in exactly the cases where the film is thickest. A step change in feed conductivity, timed against a simultaneous surface-residue measurement, gives the lag directly.
Selection effect
H3819. Spot-free drying is qualified on mirror-polished coupons because spots are easy to see there, but those coupons also drain and dry most evenly, so the qualification selects the easiest surface in the plant. A process that passes on the mirror can fail on a blasted part whose pits hold liquid. Running the same qualification on one mirror coupon and one production-surface coupon, with identical drying, exposes the selection.
Threshold
H3820. Condensation on a part moved from a rinse to a drier has a threshold in dew-point margin rather than in relative humidity, so a room that is comfortable by its hygrometer can still fog cold work. The margin is the difference between surface temperature and the air's dew point, and it is the only number that predicts fogging. Transferring parts across a range of margin, with the surface watched for fog, locates the threshold and its dependence on transfer time.
Hidden variable
H3821. Deionised water specified by conductivity can still leave residue, because the species that survive deionisation are often neutral or weakly ionised and carry no current, so the meter is blind to them by construction. Silica and dissolved organics are the usual examples, and both concentrate on drying. Evaporating a fixed volume of the same water onto a clean wafer, and comparing the residue with the conductivity, shows whether the specification tracks what it claims.
Pickling, descaling and passivation
Transport control
H3822. Pickling rate is usually credited to acid strength, but in a still bath the rate is often set by how fast dissolved metal diffuses away from the surface, so the acid is present in excess and the bath is transport-bound. That is why stirring changes a pickling time that a titration-based model says is fixed. Pickling the same steel at fixed acid concentration and temperature but at three agitation levels, with thickness loss measured, puts the transport term where it belongs.
Optimal point
H3823. Over-pickling roughens the surface and loads it with hydrogen at the same time, and both degrade a subsequent coating, so there is an exposure with the best adhesion rather than a monotonic rule that cleaner is better. The optimum sits where oxide removal is complete but base-metal pitting has not begun. An exposure series on one alloy, with roughness and coating adhesion measured at every point, should show a maximum that the usual 'until water-break-free' rule does not predict.
Hidden term
H3824. The smut left on aluminium after alkaline or acid attack is enriched in the alloy's minor elements far beyond their bulk fraction, so its composition measures segregation at the surface rather than fingerprinting the acid used. Two acids giving the same etch rate can leave different smuts because they expose different phases. Analysing smut from one alloy in two baths at matched metal removal, and comparing against the bulk analysis, tests the enrichment claim.
Threshold
H3825. Hydrogen embrittlement from pickling has a threshold in absorbed hydrogen rather than in immersion time, and that threshold is set by the steel's hardness and trap density, so a mild steel can be pickled for hours while a high-strength one fails after minutes. Time-based specifications therefore transfer badly between alloys. Charging at a fixed current density to graded hydrogen contents, then testing notched tensile strength, gives the threshold independently of how the hydrogen arrived.
Inversion
H3826. A pickling inhibitor can raise a bath's total acid consumption while lowering its metal loss, because it slows general attack but concentrates what remains at inclusions and grain boundaries, and those pits consume acid locally and seed roughness. Inhibitor efficiency measured by weight loss then reads as improvement while surface quality falls. Comparing inhibited and uninhibited baths at equal metal loss, with pit density counted, separates the two effects.
Identity claim
H3827. Desmutting and pickling are treated as separate steps, but both are dissolution of a surface layer at a potential set by the bath's oxidising power, and the boundary between them is a convention about how much metal is removed. A bath that desmuts is a weak pickle, and the two names describe doses rather than mechanisms. Running one bath at graded oxidising power and measuring both smut removal and metal loss should give one continuous curve, not two regimes.
Causal ordering
H3828. Acid passivation is credited with creating the passive film on stainless steel, yet the film forms within seconds in air or water on a clean surface; what the acid removes is exogenous iron and other contamination that would otherwise sit beneath it. If that ordering is right, a part passivated in an oxygen-free glovebox should come out worse than one passivated and then exposed to air. Passivating under nitrogen and exposing to air for graded times tests which step made the film.
Equivalence claim
H3829. Citric and nitric passivation are held interchangeable for stainless steel, but on free-machining grades the sulfide inclusions need an oxidiser to be removed, and citric acid alone does not supply one, so a residual sulfide film survives and later releases hydrogen sulfide at the interface. Equivalence then holds for one alloy class and fails for another, which no general statement can express. Passivating two grades in both acids and measuring surface sulfur is a direct test.
Coupling claim
H3830. Pickling and plating are optimised separately, but the pickle sets the substrate's hydrogen content and that content changes how the plate nucleates and how it adheres, so the two steps are coupled through a variable neither specification tracks. An intermediate bake would break the coupling without changing topography. Plating after pickling with and without that bake, at matched surface roughness, isolates the hydrogen term.
Limit case
H3831. At long enough pickling time the surface approaches an etch texture whose scale is set by grain orientation and second-phase spacing, so roughness converges on a value that further acid exposure cannot change. That floor belongs to the alloy's microstructure, not to the bath, and it is why two acids reach the same endpoint slowly. Pickling one coupon through a long time series with roughness measured at each point shows whether the curve plateaus and at what scale.
Bookkeeping deficit
H3832. Descaling acid is consumed by two reactions, dissolving scale and dissolving base metal, and most bath-life accounting counts only the first, so the acid looks longer-lived than it is while the base-metal term quietly dominates at long dwell. The error is small per part and large per campaign. Measuring iron in solution and separately measuring scale removal over a full bath life splits the two consumptions.
Blasting and mechanical removal
Misattribution
H3833. Blasting is credited with removing oxide by impact energy, but on a brittle scale over a ductile substrate most of the removal happens when the substrate deflects and the scale spalls, so a compliant substrate cleans at lower particle energy than a stiff one. Impact energy alone therefore predicts the wrong ranking between materials. Blasting the same scale on two substrates of different stiffness at matched particle energy, and measuring removal, shows the deflection term.
Lower limit
H3834. There is a particle velocity below which an oxide is not removed however long the blast runs, because the crack a particle starts does not reach the scale-metal interface, and the threshold depends on scale thickness and toughness. Coverage, the variable operators control, cannot substitute for velocity at all. A velocity series at fixed mass flow and fixed time, with removal measured rather than coverage, finds a floor that coverage-based specifications cannot express.
Saturation
H3835. Surface roughness rises with blasting and then saturates while the surface is still not clean, because roughness responds to the first few overlapping craters whereas cleanliness requires the whole area to be struck. The roughness average therefore stops moving at a point unrelated to the process endpoint. Measuring roughness and residual oxide together across a coverage series shows the two curves diverging.
Hidden variable
H3836. After blasting a soft metal the dominant contaminant is often embedded media that no roughness measurement and no solvent wipe detects, because the particles sit beneath the surface in deformed metal. It is invisible to the methods used to release the part and appears later as coating defects or corrosion nuclei. Dissolving a thin layer of the surface and analysing the solution for the media's elements quantifies it directly.
Trade-off
H3837. Finer blast media improve finish and reduce compressive residual stress at the same time, so a specification that tightens roughness and demands fatigue life is asking for two things that move in opposite directions under one handle. The acceptable window is a band rather than a point, and its width is the real process capability. Blasting a fatigue-grade alloy with a media-size series, measuring both roughness and the residual stress profile, locates the band's edges.
Crossover
H3838. The angle at which a blast stream removes material best shifts from cutting at high incidence to peening at low incidence, and the crossover angle is set by media shape rather than media hardness, because angular grit cuts while round shot deforms. That is why two suppliers' media of the same size give different removal maps. Measuring removal and residual stress against angle for one angular and one round medium shows the crossover moving with shape.
Count
H3839. A pressure blast system has at most four independent settings that change the surface: particle velocity, mass flow, incidence angle and coverage, and every other knob is a way of arriving at one of those four. Claims of five- or six-parameter optimisation therefore contain a hidden redundancy, usually the nozzle's distance folded into velocity. Sweeping a claimed fifth parameter at fixed settings for the other four, and finding no effect, counts the real handles.
Artefact mimicking effect
H3840. The Almen strip is used to set peening intensity, but the strip curves as it is peened, so the incidence angle drifts through the pass in a way the flat production part never sees, and the quoted intensity carries the fixture's geometry. Two machines can then agree on Almen intensity and disagree on the part. Peening a flat strip and a curved holder instrumented for angle, at otherwise matched conditions, shows how much of the intensity is geometry.
Non-monotonicity
H3841. Reusing blast media first improves cleaning as fresh grit's sharp edges wear to a stable cutting shape, then degrades it as the edges round and fines load the stream, so media life has a best point inside it rather than an end. Suppliers specify a replacement fraction that assumes monotonic decay. Cleaning rate against media cycles, measured on identical coupons, maps the hump and locates the real replacement point.
Symmetry claim
H3842. Blasting a surface and the surface eroding the nozzle are the same momentum exchange seen from two sides, but they are not the same wear problem, because the particle's spin decides which side takes the damage. That asymmetry is why nozzle life follows media shape more closely than media hardness. Firing angular and round media of equal hardness at identical velocity, and measuring nozzle mass loss against particle spin, separates the two.
Memory effect
H3843. Blast-induced residual stress relaxes with time at temperature, so a coupon characterised a week after blasting is not the surface that enters service, and the relaxation is faster in the first hours than a shelf-life specification implies. The measured stress is a snapshot of a decaying quantity. Measuring the same coupon at intervals after blasting, and again after a mild anneal, gives the decay the qualification misses.
Scale argument
H3844. Coverage by a peening stream approaches completeness exponentially, so the last few per cent of area cost about as much again as the first ninety-five (estimate), and the cost curve is why full coverage is a convention rather than a measurement. Past that point the process adds time and damage without adding area. Timing a coverage series against a fluorescent tracer, rather than by eye, prices the last few per cent.
Polishing, lapping and the topography that results
Hidden mechanism
H3845. The final mirror in mechanical polishing is credited to the fineness of the last abrasive, but at that stage removal is often hydrodynamic: pad, slurry and part form a viscous film, and the finish follows the film's thickness rather than the grit's size. Pad hardness then matters more than the last grade of paper, which is a nuisance for anyone buying paper. Polishing one metal with a single final grit on two pads of different hardness, at matched load and speed, tests which variable the finish follows.
Measurement artefact
H3846. A roughness average is not a property of the surface alone: it depends on the evaluation length and the cutoff the instrument applied, so two laboratories reporting the same parameter for one coupon can both be right and still disagree. The number is a filtered quantity whose filter is rarely quoted beside it. Measuring one coupon on two instruments at two cutoffs, and keeping the unfiltered profile, shows how much of the difference is the filter.
Bound
H3847. The best roughness a mechanical polish can reach is bounded by the coarse tail of the abrasive's size distribution rather than by its mean, because the largest particle in the slurry sets the deepest scratch that survives the pass. Buying a tighter mean without tightening the tail therefore buys nothing measurable. Polishing with two abrasives of equal mean but different tail widths, and counting deep scratches, tests the bound directly.
Inversion
H3848. Electropolishing can raise a surface's roughness average while lowering its contamination, because it removes the disturbed layer and exposes grain facets whose height variation exceeds that of the smeared layer it replaced. A rougher electropolished part can then outperform a smoother buffed one downstream, which inverts the usual reading of the number. Measuring roughness and surface carbon on buffed and electropolished coupons of one alloy separates the two directions of change.
Depth identity
H3849. Buffing-compound residue and the smeared metal layer beneath it are the same contaminant at two depths: both are material the last operation moved rather than removed, and a solvent wipe takes the top one while leaving the other. The test that passes or fails the part therefore depends on which one is present. Wiping, then etching a controlled depth and re-measuring carbon, shows whether the wipe removed the layer or only its surface.
Optimal point
H3850. Lapping pressure has an optimum because removal rate rises with pressure while slurry transport into the gap falls, and past the optimum the rate drops while subsurface damage deepens. The optimum is set by slurry viscosity and plate grooving rather than by abrasive grade, so it moves when the fixture does. A pressure series with removal rate and damage depth measured together gives the peak and shows which side a production process sits on.
Aliasing
H3851. A periodic polish mark whose spacing is close to the profilometer's sampling interval appears in the profile as random roughness, so a surface with a regular defect is recorded as a stochastic one. The instrument then reports a texture the eye would call a pattern, and the two descriptions lead to different corrections. Sampling one deliberately ruled surface at several intervals, against a fine reference scan, shows the mark folding into the noise.
Necessary but not sufficient
H3852. A low roughness average is necessary for a mirror finish but not sufficient, because specular reflection is degraded by waviness at wavelengths far longer than the cutoff, which the average excludes by construction. Two surfaces with identical roughness can differ visibly in gloss, and the gloss is what the customer sees. Measuring gloss and roughness on a set of polished coupons, with a long-wavelength scan added, shows the case where the number fails to predict the appearance.
Symmetry breaking
H3853. Polishing imposes a direction on the surface that survives cleaning, because the abrasive moves material along the stroke and leaves a shallow anisotropic texture aligned with it. That texture biases wetting and later coating flow along the stroke, so an isotropic specification is not met by an anisotropic process. Cleaning and then measuring contact angle along and across the stroke, on the same coupon, tests whether the direction is still legible afterwards.
Criterion claim
H3854. Scratch visibility has a threshold in depth set by the observer's contrast sensitivity and the illumination, not by a physical transition in the material, so a specification of no visible scratches names a viewing condition rather than a surface. Change the light and the same part passes or fails. Showing one scratched coupon to a panel under two illuminations, and measuring the scratch depths, calibrates the wording to something physical.
Rate-limiting step
H3855. In vibratory finishing, cycle time is usually attributed to abrasive wear, but the rate-limiting step is often media renewal at the part's surface, so media size and shape control the result more than media grade. A finer grade in the same bowl then changes little, which is not what the purchasing decision assumes. Running one part in three media shapes at matched grade, logging removal against time, shows whether the slope follows the shape.
Conservation claim
H3856. Mechanical removal in polishing obeys a work relation in which rate is proportional to the product of pressure and relative speed, so neither variable alone can be optimised and two processes with the same product should remove the same amount. This is why speed can be traded for pressure only within the damage limit. A pressure-speed grid at fixed slurry, comparing removal at equal products, tests the relation and finds where it breaks.
Plasma, laser and dry cleaning
Misattribution
H3857. Plasma cleaning is credited to oxidation of hydrocarbons, but at low pressure much of the removal is physical sputtering and vacuum-ultraviolet scission, so an argon plasma with no oxygen at all can clean as effectively as an oxygen one. The oxidation story then mispredicts which gas to use for which residue. Cleaning identical contaminated coupons in argon, oxygen and argon-oxygen at matched ion dose compares the three without the ambiguity.
Hidden variable
H3858. The dose a plasma delivers to a part is the ion flux at that surface, not the generator's power multiplied by time, and the flux depends on the part's position relative to the sheath and on the chamber's loading. Identical recipes therefore deliver different doses on two racks, which is why plasma processes transfer poorly between machines. Mapping a dose-sensitive indicator over the chamber volume at fixed settings shows the variation the recipe hides.
Saturation
H3859. Plasma treatment's effect on wettability saturates at a monolayer while its effect on adhesion keeps changing, because wetting reads only the outermost chemistry whereas adhesion depends on the crosslinked and roughened layer beneath it. A process tuned to a contact angle therefore stops improving at the point where the adhesion story is still developing. Measuring both as functions of dose on one polymer shows the two curves separating.
Reversal
H3860. An atmospheric plasma jet can make a surface dirtier as the treatment lengthens, because the electrode erodes and deposits metal oxide downstream, so beyond a dose the added contamination outweighs the added activation. The effect grows with power and with the jet's age, so a well-used jet is a different tool from a new one. Treating a clean polymer over a time series, with surface metals and contact angle both measured, shows the turnover the recipe does not mention.
Diffusion scaling
H3861. Hydrophobic recovery after plasma treatment is the diffusion of low-molecular-weight material from the bulk to the surface, so the contact angle returns as the square root of time and the rate is set by molecular weight and free volume rather than by storage humidity alone. Recovery is therefore fastest in the first hours and slows without stopping. Measuring angle against time on a series of molecular weights collapses the curves onto one slope if the claim holds.
Null claim
H3862. On a gold surface, whose only contamination is adsorbed hydrocarbon, ultraviolet-ozone treatment should do nothing that a clean argon sputter does not, because gold grows no oxide and the ozone's action is confined to carbon removal. The claim is that the treatment is generic carbon removal and nothing else. Running both treatments on one gold film and comparing carbon by spectroscopy tests the null and, if it holds, simplifies the recipe.
Artefact mimicking effect
H3863. A contact angle that falls after plasma treatment is often read as chemical activation, but ion etching also roughens the surface, and roughness alone lowers the apparent angle on a wetting system. The measurement then confirms a chemistry it never tested, and the process is tuned to the artefact. Treating a series of polymers of known etch resistance, measuring angle and roughness together with an independent chemical probe, separates the contributions.
Causal ordering
H3864. In a plasma chamber the residual water desorbing from the walls is a stronger oxidiser than the process gas, so pump-down time and wall conditioning set the result more than the gas mixture does. Runs that differ only in pump-down then differ in chemistry, and the difference is attributed to the gas. Varying pump-down at fixed gas and power, with water partial pressure measured on the residual gas analyser, orders the causes properly.
Window claim
H3865. The fluence window for laser cleaning widens as the pulse shortens, because heat diffusion is confined to a depth smaller than the contamination layer, so a short pulse removes the film before the substrate can be heated. Long-pulse recipes then work with a window narrower than the physics allows, and their yield losses look like material variability. Measuring the ablation threshold and the substrate damage threshold at three pulse durations, on one film, maps the window against duration.
Hidden term
H3866. Dry-ice cleaning is modelled as momentum transfer from the sublimating pellet, but the pellet also cools the surface sharply, and the resulting thermal stress contributes to film detachment in a way a momentum-only model cannot predict. That is why the process works on some films at energies that should have been too low. Comparing removal at matched momentum but different pellet temperatures isolates the thermal term.
Crossover
H3867. Laser removal of a thin film crosses from explosive ablation to thermal vaporisation as the film thickens past the absorption length, so a fluence that cleans a thin film can bake a thick one. The crossover is a property of the film's optical depth rather than of its chemistry, which is why one recipe cannot serve a thickness range. A thickness series at one fluence, with removal and residue chemistry measured, shows the change of regime.
Regime identity
H3868. Corona treatment and dielectric-barrier discharge are names for one discharge regime at two current levels, and the differences reported between them are differences of gas chemistry and dwell time rather than of discharge physics. If so, the two should be interchangeable once gas and dose are matched, and the literature's disagreement dissolves into bookkeeping. Treating one polymer by both methods at matched dose and gas, and comparing the surfaces, tests the identity.
Observability claim
H3869. A plasma's cleaning endpoint cannot be read from its electrical parameters, because the discharge changes mode as the surface's secondary electron yield falls, so voltage and current move for reasons unrelated to cleanliness. Inline electrical monitoring then reports the wrong thing confidently, and the endpoint is set by recipe time. Running a cycle to completion while logging electrical parameters against an independent downstream gas signal shows which of them tracks the endpoint.
No-go claim
H3870. No low-pressure plasma can clean the bottom of a high-aspect-ratio blind hole, because the sheath excludes ions from the interior and the neutral flux is choked by the geometry, so the bottom sees a different process from the mouth. The criterion is aspect ratio against sheath thickness, and it is a hard limit rather than a rate that can be improved by patience. Cleaning a series of holes of increasing ratio and sectioning them afterwards shows where the interior stops changing.
Contamination control and handling
Misattribution
H3871. Particles found on a part after cleaning are usually debited to the cleaning process, but most are deposited after the last wet step, during drying, transfer and handling, so the count measures the room more than the bath. Cleaning harder then changes nothing, and the effort is spent in the wrong place. Sampling at three points along the line with the same method at each attributes the count to the stage that added it.
Hidden term
H3872. A cleanroom's particle count includes the contribution of the sampling probe's own tubing and of the disturbance the probe creates, so the class number is partly a property of the monitor and its placement. Two rooms can be certified differently because they were measured differently, and neither measurement is wrong. Measuring one room with two probe lengths and two sampling positions quantifies the instrument's share of the number.
Threshold
H3873. Particle adhesion changes character above a relative humidity at which capillary bridges form at the contact, and beyond that threshold dry removal methods lose most of their efficiency because the holding force rises sharply (estimate). The threshold belongs to the particle-surface pair, not to the room, so one humidity specification serves two surfaces unequally. Challenging a standard deposit at graded humidity, with removal by one dry method, locates it.
Inversion
H3874. Adding a final filtration stage can increase the particles delivered to a part, because a filter near the end of its life sheds more than it captures and the shedding arrives undiluted. Air that is nominally cleaner then deposits more, and the extra stage is blamed on the room. Measuring particle concentration downstream of fresh and loaded filters at the same upstream challenge shows the non-monotonicity and sets a change-out criterion.
Count
H3875. A contamination budget resolved from one surface measurement can distinguish at most three independent sources, because the measurement returns a total and each source contributes through two ratios. Budgets that name five sources from one sample are accounting rather than measurement. Tagging one suspected source with a tracer and refitting shows how many the data can actually support.
Selection effect
H3876. Witness coupons are placed where they can be placed, which is usually where the flow is unobstructed and the surface is accessible, so they measure the cleanest part of the chamber and certify it for the dirtiest. The bias is systematic and always optimistic, which is the worst combination. Mounting coupons at one nominal location but three heights, and comparing, shows the vertical structure that standard placement misses.
Lag
H3877. Contact angle on a porous surface responds to the last liquid in the pores rather than to the chemistry of the outer layer, so it lags the actual organic loading and can report a surface that no longer exists. The lag grows with pore volume, which is why castings and sintered parts read inconsistently. Wetting a contaminated porous coupon and a non-porous one identically, and tracking angle against time, shows the porous one arriving late at the same value.
Ratchet
H3878. Contamination accumulates in the cleaning equipment faster than on the product it cleans, so a bath or rinse tower becomes a source as it ages: its cleaning rate falls while its shedding rises, and both trends run the same way between services. The process therefore degrades on a one-way path that maintenance schedules do not model. Logging removal efficiency and effluent particles across a service interval shows the two curves crossing.
Thickness identity
H3879. A fingerprint and an adsorbed organic film are the same contaminant at different thicknesses, the fingerprint being a thick film with entrained salt, so the physics of removal is shared and only the dose differs. Methods tuned to one should work on the other with an adjusted time, and the apparent need for special fingerprint removers is a dosing problem. Cleaning a graded series from monolayer film to fingerprint with one chemistry shows a single removal curve.
Bound
H3880. Removal of a particle from a surface by fluid flow is bounded by the boundary layer: a particle much smaller than the layer sits in nearly stagnant liquid and never sees the free-stream velocity, so no increase in flow rate dislodges it. The bound is set by particle size against layer thickness, not by pressure. Measuring removal against flow rate for three particle sizes shows the size at which the curve goes flat.
Hysteresis
H3881. A surface that has been contaminated and then cleaned passes and fails the same test at two different thresholds, because the pores retain enough of the first contamination to change the wetting path. History therefore decides the verdict at a fixed contamination level, and a supplier and a customer can disagree while both are honest. Testing virgin and previously contaminated coupons across a graded series, with one method, gives two curves and the width between them.
Confounder
H3882. The frequently reported correlation between particle count and coating defects can be produced by handling alone, since handling adds both particles and mechanical damage, and the defects then follow the handling rather than the particles. The correlation would survive even if particles caused nothing at all. Comparing defects on parts handled identically but cleaned differently, and on parts cleaned identically but handled differently, breaks the confound.
Adhesion preparation and its interface
Trade-off
H3883. Roughening a surface for mechanical keying helps adhesion only until the adhesive can no longer penetrate the texture, so there is an optimum roughness set by adhesive viscosity and texture scale, and finer texture is not automatically better. Beyond the optimum the bond line contains voids that start the failure, and the strength curve turns down while the roughness number keeps improving. Bonding one adhesive to a roughness series, with void content and strength both measured, finds the peak.
Causal ordering
H3884. A conversion coating continues to grow after the primer is applied, drawing water through the primer, so the interface that is tested is not the one that was qualified and the growth after priming is what decides durability. The ordering error is invisible at time zero and shows up only in long exposure. Measuring coating thickness and interfacial water before and after a primer cure, on one coupon, shows whether growth continued and at whose expense.
Hidden variable
H3885. The open time of a prepared surface is set by the adsorption of airborne organics, so its variability between plants is dominated by air-change rate and occupancy rather than by the preparation chemistry. Specifications that quote a time without a room condition are therefore not portable, and each plant's number is locally true. Exposing prepared coupons in two rooms at measured air-change rates and occupancy, and bonding at intervals, attributes the variance.
Reversal
H3886. A silane coupling layer thicker than a monolayer can weaken a bond, because the excess hydrolyses and condenses into a loose network that fails cohesively beneath the primer. More coupling agent then means less strength, contrary to the dose-response assumed by topping up the bath. Depositing silane at graded coverage, from sub-monolayer to multilayer, and measuring strength and failure locus shows the maximum sitting before the multilayer.
Identifiability claim
H3887. Splitting a surface energy into dispersion and acid-base components from two probe liquids is underdetermined, because two measurements cannot fix three unknowns without an assumption, so a reported change in the polar fraction may be a change in the assumption rather than in the surface. The number is then not comparable between papers, though both are computed correctly. Measuring the same surfaces with three probe liquids and checking internal consistency shows how wide the admissible range is.
Necessary but not sufficient
H3888. A high surface energy is necessary for wetting but not sufficient for adhesion, because a weak boundary layer can wet perfectly and still fail at a fraction of the bulk strength. Specifications that stop at the contact angle therefore certify the wrong thing, and they certify it well. Bonding surfaces of matched high energy but different subsurface damage, and comparing failure loci, makes the distinction visible.
Threshold
H3889. Contamination below roughly a monolayer changes bond strength very little, so a cleanliness specification tighter than that threshold costs money without buying strength, and the threshold is what a specification should be written against. Above it strength falls steeply and the cost of contamination appears suddenly. Bonding a graded series that spans sub-monolayer to multilayer, with strength measured, locates the knee and gives the specification a physical basis.
Null claim
H3890. For one alloy and one adhesive, anodising and chromate conversion may give the same durable strength after water exposure, because both present the same oxide chemistry to the primer and the reported difference is oxide thickness rather than surface species. If so, the choice between them is a manufacturing one and not an interfacial one. Bonding both treatments at matched oxide chemistry, and testing after identical water exposure, tests the null.
Crossover
H3891. The failure locus of a bonded joint moves from cohesive in the adhesive to interfacial as temperature rises, and the crossover is where the interphase's modulus falls below the adhesive's, so a joint tested cold says nothing about the same joint hot. Strength rankings can invert across the crossover, which is why supplier data taken at one temperature misleads at another. Testing one joint over a temperature series, recording the locus and not only the load, finds the crossover and the ranking change.
Identity claim
H3892. A weak boundary layer and a contaminated surface are the same failure mode described from two ends, one from the fracture path and one from the chemistry, so either diagnosis should predict the other. Where they do not, one of the two measurements is wrong, and the disagreement is more informative than either result. Taking joints that fail interfacially and analysing their fracture surfaces for the contaminant the weak-boundary-layer model requires tests the identity.
Scaling claim
H3893. Water-driven durability of a prepared interface follows diffusion through the interphase, so a joint that fails in a week at an elevated temperature should fail in months at room temperature with a shift set by the activation energy, and the shift should be the same for every joint failing by the same path. A shift that differs between adherends then means the path differs. Testing two temperatures and two adherends, and checking the shift's constancy, tests the mechanism rather than the number.
Hidden term
H3894. The oxide on a metal is in residual stress before any load is applied, because it grew at a different temperature or in a different volume from the metal beneath, and that stress adds to whatever the adhesion test applies. An intrinsic bond strength measured on a thick oxide therefore includes a term the thin-oxide case does not have, and the thickness comparison is confounded. Comparing oxides of two thicknesses on one metal, with curvature measured, separates the term.
Verification, specification and what a test can see
Proxy failure
H3895. Qualification by witness coupon certifies the process rather than the part, because the coupon's thermal and flow history differs from the part's, so a line can pass qualification and ship parts that never saw the qualified condition. The proxy is honest only where the histories match, and that condition is rarely checked. Instrumenting one part and one coupon in the same run, at the same locations, shows the gap in a single shift.
Traceability claim
H3896. A cleanliness specification written as visually clean is not traceable to a physical quantity, so two suppliers can both comply and deliver surfaces differing by orders of magnitude in residue, and the disagreement sits in the wording rather than in the work. The fix is to name the method, the instrument and the limit, which also makes the specification falsifiable. Measuring residue on parts that all pass the visual test shows the range the wording admits.
Specification defect
H3897. A holding time between cleaning and coating is written as a time, but the quantity that matters is the accumulated organic dose, which depends on concentration and air movement as well as time, so the same time is safe in one plant and unsafe in another. The specification then fails without anyone violating it, which is the most expensive kind of failure. Exposing coupons for equal times in two rooms of different organic loading, and bonding them, shows the time is the wrong variable.
Resolution limit
H3898. Destructive sampling of a few parts per lot cannot resolve a rare adhesion failure, because the number of samples needed scales inversely with the failure rate, and a lot with a one-in-a-thousand rate will pass almost every plan that a plant can afford. The test then certifies a rate it cannot see, and the certificate is sincere. Running a deliberately degraded process to a known failure rate, and checking what the plan detects, calibrates the plan rather than the product.
Falsifiability claim
H3899. A claim that a cleaned surface stays clean indefinitely cannot be tested, because any observed failure is attributed to handling, so the claim absorbs its own counterexamples. Restating it as a dose to failure under a specified exposure makes it testable and makes the number comparable between suppliers. Exposing coupons to a controlled organic atmosphere and measuring the dose at which adhesion falls converts the claim into a measurement.
Bound
H3900. The cleanliness a process can deliver is bounded by the cleanliness of the water or gas that last touches the part, so all upstream effort is capped by a downstream utility, and the cap can be measured once instead of rediscovered at every failure. Past the cap, more chemistry and more stages buy nothing at all. Measuring residue after a final contact with utility streams of graded purity, changing nothing else, gives the cap and shows how far the process sits below it.
Status and plan
One hundred entries, numbered in sequence. They were composed individually, each with its logical form chosen before its subject, so that no two share an argument shape; the labels are printed for that reason and can be checked against the sentences that follow them.
The subject stays inside the preparation of a surface: what removes oil, what removes oxide, what removes metal, what removes nothing but rearranges it, and what the next layer finds when the part arrives. No claim here concerns a product, a manufacturer, or a person.
If one of these is worth an afternoon, it is the first: the claim that a vapour degreaser cleans by condensation rather than by solvency is testable with three preheated coupons, one tank and a balance, and it would explain why a bath that tests clean still fails to clean when the work is warm.
Incoprea Research · hand-considered hypotheses 3801–3900 · 3 November 2026 · ~/Desktop/Nov3/