Bivalve Larval Counts, Subsamples and Survival Explained

The number in the cup is not the number in the tank

A hatchery technician counts 402 larvae in a small microscope chamber. The record beside it says 0.5 mL and a prepared concentrate of 2,000 mL. Someone copies “402 larvae” into a daily population report. That translation is wrong because the observation belongs to a sub-sample, not the whole container. The correct question is how many larvae the measured volume represents after the parent suspension has been mixed, and whether the concentrate itself represents the complete recovered culture.

Bivalve larval enumeration has three levels. The microscope count concerns the chamber; the chamber volume converts that count to individuals per millilitre; and the total resuspended volume converts density to an estimated number of individuals in the batch. If the original tank contained more water, its full water volume is irrelevant after all larvae were retained and concentrated, provided transfer recovery was complete. Losing animals on a sieve invalidates that condition. Accurate reporting therefore starts with a map of physical containers and transfers, not a calculator.

This guide explains the measurement chain, offers checked hypothetical calculations, and distinguishes enumeration, viability, developmental stage and cohort survival. Examples are teaching cases, not measurements from a real hatchery and not universal husbandry limits. The FAO bivalve hatchery manual describes the practical use of mixed samples, counting chambers, sieves and size-frequency measurements; its methods provide a foundation, but a local laboratory must validate its own tools and procedures.

Decide which population the report actually names

A word such as “stock,” “larvae,” or “yield” can silently change its denominator. The fertilised embryos stocked on day zero may be compared with normally formed D-larvae at a later inspection, all larvae alive during a water change, or competent pediveligers at a later stage. These are different populations measured at different times. A figure of 1.2 million may be the number put into culture, transferred into a new tank, or recovered after grading. The verb accompanying the number is evidence.

Read each log as an event: collection, screening, concentration, mixing, sub-sampling, counting, grading, return, loss and discard. Note whether containers were pooled; keep tank IDs separate if treatment effects matter. A sample labelled “Tank A, after 80 µm sieve, before discard” cannot automatically represent larvae that went through that mesh or died before the sample. If the log lacks the event boundary, a correct arithmetical extrapolation can still answer the wrong biological question.

Build a measurement chain before multiplying

Write the chain in plain words: initial culture was drained through a designated mesh, retained larvae were rinsed into a measured concentrate, concentrate was kept gently mixed while several aliquots were pipetted, the aliquots were counted in a calibrated chamber, and counts were expanded back to the concentrate. Every arrow is an assumption about recovery, mixing or volume. People often skip the middle arrows because the last multiplication looks straightforward. The middle steps decide whether it has meaning.

A complete entry records original tank or pool ID, nominal and measured concentrate volume, dilution step if used, aliquot volume, replicate counts, counting rules, counted stage, time, equipment calibration, and relevant exclusions. If an aliquot has been diluted before loading, the expansion must include the inverse dilution fraction exactly once. If technicians count only an outlined subset of a chamber, its effective volume also matters. A reproducible estimate is one whose path another competent person can reconstruct.

Dimensional analysis exposes the common error

Suppose a chamber contains 402 larvae in 0.5 mL. The concentration is 402 divided by 0.5, or 804 larvae/mL. Multiply by a 2,000 mL concentrate to obtain 1,608,000 larvae. The units show the logic: larvae per mL multiplied by mL leaves larvae. The 2,000 mL is not an extra biological growth factor; it is the physical volume of the suspension for which the aliquot is intended to be representative.

If a worksheet instead reports 402 × 2,000 = 804,000, it has silently treated a half-millilitre count as though it came from one millilitre. Conversely, using the original 1,000 L tank volume after concentrating everything into 2 L could inflate the estimate five hundredfold. Before doing arithmetic, label each input. The FAO manual gives an illustrative triplicate count of 414, 389 and 402 in 0.5 mL aliquots from 2 L, leading to about 1.608 million larvae.

Averaging replicate chambers is not pooling unknown volumes

Imagine three equal 0.5 mL aliquots taken during continuous gentle agitation yielding 414, 389 and 402 larvae. Their mean is 401.67 larvae per half-millilitre, which equals 803.33 larvae/mL. Multiply by 2,000 mL for approximately 1,606,667 larvae. Reporting 1.61 million is more honest than presenting a string of meaningless final digits, since biological mixing and the chamber count introduce uncertainty beyond arithmetic rounding.

This mean method assumes all aliquots have the same effective volume. If the first is 0.25 mL, the second 0.5 mL and the third 1.0 mL, do not simply average raw counts. One may combine total counted individuals divided by total sampled volume when samples share a coherent sampling frame, or compute densities first with a suitable weighting scheme. State whether aliquots were independent draws or repeated views of the same chamber. Those choices change how sampling precision should be represented.

Dilution has direction: a worked tenfold example

A dense larval concentrate is prepared in a 1,500 mL vessel. A technician mixes 1 mL of concentrate with 9 mL of appropriate diluent, making 10 mL total. That is a tenfold dilution of original concentration, provided volume additivity is a defensible approximation for the procedure. A 1 mL counted aliquot of the diluted mixture holds 120 larvae. The diluted sample has 120 larvae/mL, so the original concentrate has 1,200 larvae/mL and approximately 1,800,000 larvae in 1,500 mL.

The trap is to divide by ten at the last step or to multiply by ten twice. Write an origin arrow on the worksheet: original → one part in ten → observed chamber. To return to the original, multiply density by ten once. Distinguish dilution factor from the original volume fraction: a 1:10 dilution can be ambiguously written in laboratory shorthand, so the robust record says “1 mL stock plus 9 mL diluent, final 10 mL.” It is also important to verify that the animals were not lost or stuck during dilution.

Two serial dilutions need a full audit trail

Consider a 1:5 preparation followed by a 1:4 preparation, each defined by final-volume fraction. The combined dilution is twentyfold, not ninefold and not a difference of four. If the final solution gives 25 larvae in 0.5 mL, its density is 50/mL, the original density is 1,000/mL, and a 3,000 mL concentrate would hold an estimated 3 million larvae. Each step can be recomputed separately to catch accidental inversion.

Serial dilution labels should specify sample taken, diluent added and final total at every stage. The example assumes both steps are volumetrically accurate, fully mixed and representative. If intermediate mixing is incomplete, multiplication preserves the wrong measurement with impressive precision. For some samples, dilution may change organism distribution through aggregation, settling, adhesion or handling stress; that is a sampling concern rather than an algebraic one. A defensible translation flags uncertainty rather than concealing it inside a tidy factor.

Mixing is the quiet foundation of the estimate

Larvae are not dissolved salt molecules. They may swim, sink, rise, cluster around edges or redistribute when aeration stops. A sample taken from the surface after a long pause need not represent the full volume. The FAO manual illustrates sampling while suspensions are kept agitated with a perforated plunger. The key is repeatable, gentle redistribution that does not damage organisms and a consistent timing relationship between mixing and pipetting.

A practical test is to compare aliquots taken from different accessible positions under a deliberately specified mixing procedure. If top and bottom counts systematically differ, more replicates from the same biased point will not repair the estimate. Engineers call this an issue of representativeness. Keep track of whether any organism-rich residue remained on the sieve, bucket walls or transfer tubing. A large sample count can reduce random counting variation while still missing a systematic recovery failure.

Replicate variability and counting precision

A hypothetical sequence of counts of 98, 102, 101 and 99 per equal aliquot gives a mean of 100 and modest observed variation. A sequence of 35, 170, 48 and 147 also averages 100 but warns of poor mixing, nonuniform concentrations, inconsistent chamber filling or identification disagreement. A report that prints only the same average for both series hides crucial information about how dependable its extrapolation is.

For a basic random-counting approximation, relative Poisson counting uncertainty for one count n is about one divided by the square root of n. A count of 100 has an approximate ten-percent relative counting component, while a count of 400 has about five percent. This is not a guarantee of total uncertainty, because larvae may be clumped, aliquots are affected by mixing and pipettes have volumetric errors. It describes only a limited statistical model. Replication and independent controls help discover when the model is inappropriate.

A simple interval is not a licence to ignore sampling design

Suppose six well-mixed equal aliquots give a total of 2,400 counted larvae across 6 mL. Their pooled estimate is 400 larvae/mL. Under a simplistic independent count model, square-root-of-count variation is about 49 individuals across the six millilitres, or roughly 8.2 larvae/mL. Extrapolated over a 2,000 mL concentrate, the point estimate is 800,000 and this counting-only standard-error component corresponds to roughly 16,300 larvae.

This is not a complete confidence interval for the tank. It does not account for uncertainty in the total concentrate volume, lost animals, between-aliquot heterogeneity, incorrect species classification or handling. Reporting “800,000 ± 16,300” without a clear qualifier would overstate knowledge. An honest record might say “estimated 0.80 million; counting-only standard error about 0.016 million under an independent model; further mixing and recovery uncertainty not quantified.” The difference is not cosmetic: it governs decisions about culture density and survival.

Empty shells, live larvae and abnormal development

Counting chambers can contain empty shell material, dead or moribund larvae, normally swimming animals and individuals with abnormal development. An electronic particle counter may tally objects with similar dimensions without reliably distinguishing these biological categories. The FAO hatchery manual explicitly notes the need for visual examination of culture quality despite the convenience of electronic counters. An accurate translated summary should therefore name what the instrument actually detects.

If a report contains 200 counted objects but only 170 are assessed as live larvae, the live fraction among counted objects is 85 percent; it is not automatically culture survival from stocking. If 155 of those 170 are normally formed at the relevant stage, normal morphology among live larvae is about 91.2 percent. Each fraction uses a different denominator. When numbers are inferred from separate sub-samples instead of the same individuals, that distinction should be clear and uncertainty may be wider.

What a sieve does to the population frame

Mesh aperture influences which organisms are retained, passed through or trapped against the screen. A figure such as “80 µm mesh” specifies the screen opening, not a perfect universal boundary for larval shell length. The FAO manual’s retention guidance varies by larval shape and species, and it treats mesh-size relationships as indicative. If a report says “greater than 80 µm,” ask whether that means aperture, a measured shell dimension or the laboratory’s operational retained fraction.

If larger larvae were retained on 120 µm and smaller ones on 80 µm, the two sieve fractions must be combined appropriately to estimate the full surviving population. Taking only the 120 µm count and calling it total survival undercounts smaller larvae. If unusually small larvae passed through the smallest mesh, even adding retained fractions misses them. Report what was recovered, what was discarded and which sizes were not sampled. This is particularly important when growth is uneven and small individuals remain biologically important.

Developmental stage is more informative than a day label

An embryo, D-larva, veliger, pediveliger and settled spat are not interchangeable biological units. Temperature, species, food and culture conditions affect the speed of development, so “day four” is not a robust substitute for observing morphology. The FAO manual distinguishes stages and documents differences among oysters, clams, mussels and scallops. A useful translation retains the stage name, the evidence used to classify it and whether the count includes abnormal or delayed individuals.

A report saying “75 percent D-stage” is incomplete unless the population is defined. Does that mean normally formed D-larvae among all originally stocked embryos, among currently recovered organisms, or among the subset judged alive? These values can vary dramatically. If 800,000 embryos were stocked and 540,000 normal D-larvae recovered, yield relative to stocking is 67.5 percent. If 600,000 total D-stage objects were present, the normal proportion in that D-stage collection is 90 percent. Both numbers can be true.

Survival rate and recovery yield are different claims

Suppose a culture starts with 2 million larvae and an end-of-period enumeration finds 1.4 million live larvae after a complete transfer. The apparent surviving proportion is 70 percent. But if 100,000 larvae were deliberately removed earlier for another purpose and 80,000 were transferred to a separate study vessel, the observed count is not a closed-system survival endpoint. The denominator and disposition of removals must be reconciled before saying that 30 percent died.

One useful accounting identity is beginning population plus introductions minus known removals minus known losses equals expected ending population, with an explicit unaccounted balance. For living cohorts, population estimates are noisy and metamorphosis may change what is counted, so the balance is an audit rather than proof of causation. Avoid turning an unexplained shortfall into a specific mortality mechanism without direct evidence. The report should label its numerator and denominator in words as well as symbols.

Specific example: survival across two transfers

A synthetic batch has 1.5 million live larvae at the first water change. After one interval, a new count estimates 1.2 million, and 50,000 healthy larvae have been sampled and removed for a documented study. If the question is how many remain in the production tank relative to the original 1.5 million, the answer is 80 percent. If the question is the minimum number still accounted for as alive across both destinations, the count is 1.25 million, or about 83.3 percent.

Neither figure automatically identifies deaths because sieving, aggregation and imperfect counts could contribute to discrepancies. The second figure is an accounted-for survivor share under the assumption that study animals remained alive at that measurement. A spreadsheet must not silently add removed animals back into the production tank. The lesson is that numerator definitions reflect a question: inventory remaining, recovered live animals, or biological survival across multiple paths.

Batch-level aggregation versus mean of percentages

Two tanks each begin with different larval counts. Tank A starts with 100,000 and ends with 90,000; Tank B starts with 900,000 and ends with 450,000. Their individual proportions are 90 percent and 50 percent. Averaging these gives 70 percent, which describes the unweighted average tank-level survival, not the fraction of all original larvae recovered. At batch level, 540,000 of 1 million remain, or 54 percent.

Both the unweighted tank average and the pooled count-based proportion can be defensible for different analytical questions. If the purpose is forecasting surviving units across the facility, 54 percent is often the relevant pooled proportion. If each tank represents an independent experimental replicate and the question concerns average treatment response, tank-level analysis matters. Do not change the unit of replication merely because a larva is easy to count. Explain whether the statistical unit is an individual, a chamber, a tank or a hatchery run.

Counting chamber capacity and image recognition

A Sedgwick-Rafter style chamber or equivalent counting slide has a known observation geometry, but instruments and chambers differ. Verify loaded volume and whether the observer counted the whole chamber, selected fields or a gridded subset. If half of a chamber was examined, extrapolating to its full volume requires an explicit spatial-sampling assumption. Larvae accumulated along an edge may violate that assumption, especially after settling or uneven loading.

Automated imaging can speed measurements but requires checks for image focus, overlap, debris, organism stage and species ambiguity. A classifier’s “detected objects” is not automatically a validated “live larvae” count. Save the raw images where permissible, human review rules and a confusion check on representative samples. A transparent report keeps algorithm threshold changes traceable; otherwise two dates may show a fictitious trend because a software setting changed rather than the culture.

When counting zeros changes the calculation

A zero count in one small aliquot could occur by chance at low densities. It does not prove that the parent tank contains no larvae. If five 1 mL chambers give zero counts from a 2 L concentrate, a simple Poisson model suggests an upper approximate 95-percent density boundary of about three organisms divided by the 5 mL sampled, or 0.6 larvae/mL; that would correspond to about 1,200 larvae in the 2 L concentrate. This is a counting-model illustration, not a complete detection limit.

A real limit of detection depends on classification sensitivity, chamber geometry, sample aggregation and transfer recovery. At very low densities, increasing total counted volume can be more informative than repeating tiny aliquots. State “none observed in 5 mL” rather than “zero in the tank.” The distinction matters when interpreting residual contamination, broodstock crossings or whether a near-empty culture has genuinely been cleared.

Larval length and count must be measured independently

A count tells how many organisms were recovered; shell length describes a dimensional trait. A culture with 600,000 larvae averaging 150 µm is not automatically more successful than one with 500,000 averaging 160 µm. Decisions depend on animal health, development, size distribution, desired stage and future performance. The FAO manual discusses measuring samples of roughly 50–100 larvae for mean shell length during routine husbandry, while keeping full counts and morphology as separate observations.

If 50 larvae are measured and one unusually large individual biases the mean, the median and a histogram may reveal the situation better. Report instrument calibration, orientation of shell length, stage and whether measurements were chosen randomly or only from actively swimming larvae. A statement like “grew by 20 µm” must specify mean change over an interval, not imply that every organism gained exactly twenty micrometres. Selective mortality can also change a population mean without an equal amount of individual growth.

Numerical example: growth and changing size composition

At day one, a mixed culture includes half its members at 100 µm and half at 140 µm, yielding a mean of 120 µm. At day three, suppose all 100 µm individuals die while the original 140 µm group remains unchanged. The observed mean becomes 140 µm, a rise of 20 µm, but no surviving individual has grown. This exaggerated synthetic example illustrates a real interpretive danger: mean size is not a tracked longitudinal record.

In practice, larval growth, size-selective mortality, sorting and sieve selection can happen together. Where feasible, compare consistent size-frequency distributions and counts through time and describe any grading. If the same individuals cannot be followed, call the result a cross-sectional mean shift rather than a precisely observed individual growth rate. The quality of the translated claim depends as much on sampling design as on micrometre accuracy.

How to decide whether the estimator is trustworthy

Ask five physical questions: were the organisms fully recovered; was the concentrate volume measured; was the suspension mixed when aliquots were drawn; did the counter distinguish the intended biological category; and were there enough independent aliquots to evaluate heterogeneity? If one answer is unknown, the estimate should carry that limitation. Extra significant figures cannot substitute for an omitted transfer-recovery check.

The most valuable repair sometimes involves repeating the count with a validated mixing procedure, not fitting a sophisticated model to flawed samples. A blank chamber can reveal contamination from equipment or previous specimens; a known reference suspension can test a counter; duplicate observers can expose classification disagreement. Each check addresses a different error source. A report should state which checks were performed and avoid representing an unperformed control as passed.

Temperature, salinity and timing are contextual variables

Water temperature and salinity affect larval development and activity, but they are not mathematical substitutes for the actual number counted. Report them as measured conditions with dates and equipment, not as generic proof that animals survived. A count performed immediately after vigorous aeration may differ from one after a long static period because of redistribution. A record should distinguish biological change from timing effects around water changes and feeding.

A careful daily summary connects environmental conditions and the numerical endpoint without implying causation from one snapshot. If oxygen declined and a later count was lower, those observations warrant investigation; they do not alone establish that oxygen caused every disappearance. Species, stage, density, pathogens, mechanical handling, feed and measurement error may all matter. Scientific translation means keeping correlation, hypothesis and demonstrated cause in separate sentences.

Contamination and cross-batch identity

Shared pipettes, sieves or vessels can move organisms between batches. Even without a serious biosecurity event, a handful of stray larvae can distort a very low-density enumeration. Mark containers, rinse and validate equipment according to a local standard operating procedure, and document whether pooled samples contained multiple species or lines. When a mixed sample cannot be separated reliably, report a total count with a stated species-identification limitation.

An accidental cross-batch transfer can have two consequences: it biases population estimates and potentially compromises the lineage record. A calculation cannot reveal an identity error by itself. Traceability fields should include the batch source, receiving vessel, batch code, date and operator. If the provenance cannot be established, the honest action is to flag the ambiguity before the result is used for breeding, shipment, experiments or survival claims.

The difference between estimate and inventory

A hatchery may display “estimated 2.4 million larvae,” then remove ten thousand from a tank for microscopy. The numeral is a statistical extrapolation, not a perfect ledger in which each individual has been physically counted. When subtracting removals, keep the uncertainty of the starting estimate in mind. A thousand-animal difference might be smaller than ordinary sampling variation and should not be marketed as a meaningful survival trend.

Use an inventory ledger for intentional transfers and a measurement table for sampled population estimates. Reconcile them periodically, but do not merge them without a clear explanation. Where the objective is treatment comparison, repeated measurements from several independently assigned tanks are usually more informative than one extremely precise chamber count from a single tank. Physical independence of experimental units matters more than the number of organisms visible under a microscope.

An end-to-end worked example with a conservation check

A 750 L tank is drained through a validated collection system. All retained larvae are rinsed into 2.4 L of concentrate. Three 0.4 mL aliquots yield 252, 264 and 249 live larvae. Their mean is 255 per 0.4 mL, or 637.5 larvae/mL. Multiplying by 2,400 mL gives an estimated 1,530,000 live larvae. Notice that the original 750 L never appears in this final multiplication: it can be used to describe initial culture density only after a full recovery estimate exists.

Now imagine that 150 mL of concentrate containing the same measured density is deliberately diverted for a viability trial. The estimated diverted number is 95,625 larvae. The remaining concentrate volume, assuming no other losses, is 2,250 mL, corresponding to 1,434,375 estimated larvae. These figures should be rounded according to the measurement quality; the example preserves intermediate digits to show arithmetic. If the measured concentrate volume was only approximate, precise subtraction would make the final totals look more certain than the evidence permits.

As an independent sense check, density expressed in the original tank is 1,530,000 divided by 750 L, or 2,040 larvae/L, if recovery was complete. This low-looking number should not be silently swapped with the concentrate density of 637.5 larvae/mL. They describe the same estimated organisms across different volumes. A unit table showing 750 L tank, 2,400 mL concentrate and 0.4 mL chamber prevents the most damaging interpretation errors.

Exercise one: a dilution chain with a provided answer

A 1,250 mL concentrate is diluted by placing 2 mL of it into 18 mL of diluent. From the resulting 20 mL mixture, a 0.5 mL counting chamber holds 47 live larvae. Determine the count in the original concentrate, assuming the chamber is representative and there is no handling loss. Start by writing the sample fraction and units; avoid multiplying all visible numbers without a plan.

Answer: the diluted sample concentration is 47 ÷ 0.5 = 94 larvae/mL. The original liquid is ten times as concentrated because 2 mL stock became 20 mL total. Original density is 940 larvae/mL. Multiply by 1,250 mL for an estimated 1,175,000 larvae. A common incorrect answer of 117,500 reverses the dilution. A common incorrect answer of 11,750,000 multiplies by ten twice. Both can be diagnosed by asking whether dilution should make the counted chamber contain fewer organisms than the original.

Exercise two: distinguish a mean from a survival statistic

Three independent tanks are started with 200,000, 400,000 and 800,000 larvae. At the next inventory, estimated live counts are 150,000, 200,000 and 640,000. Compute each tank’s retained proportion, the simple average of the proportions and the pooled retained proportion. Describe what each result means, without claiming the differences represent experimental treatment effects from a single pair of measurements.

Answer: tank proportions are 75 percent, 50 percent and 80 percent. Their unweighted average is about 68.33 percent. Total initial count is 1.4 million and total recovered count is 990,000, so pooled retention is about 70.71 percent. These numbers answer different questions and neither alone establishes causes. The pooled figure answers “what share of all initially counted larvae is now recovered?” while the unweighted average answers “what was the average tank proportion?”

Exercise three: a failed mixing test

Aliquots from the top of an unmixed vessel yield densities of 50 and 60 larvae/mL. Aliquots near the bottom yield 210 and 220 larvae/mL. A clerk proposes using their average, 135 larvae/mL, as a population estimate. What should the report say? Assume the vessel geometry, sampling positions and interval have not been validated for volume-weighted representation.

Answer: a simple four-aliquot average is not a verified population density. The strong spatial pattern indicates heterogeneity, and the sampling points may not represent equal volumes. The technician should apply a documented gentle mixing procedure and repeat representative sampling, or use a validated stratified method with known layer volumes. The report can retain the raw observations but should not quietly label 135 larvae/mL a complete inventory estimate. Precision requires a defensible selection mechanism.

Exercise four: a changing denominator

A culture begins with 1 million embryos. At the D-stage count, 700,000 total D-stage objects are collected, 600,000 of which are assessed as normal, and 550,000 are assessed as alive. If the two classifications were obtained from the same individually evaluated set, which percentages are meaningful? What cannot be inferred from the totals alone?

Answer: the fraction classified as normally formed among collected D-stage objects is 600,000 ÷ 700,000 = 85.71 percent. The living fraction among collected objects is 550,000 ÷ 700,000 = 78.57 percent. The living count relative to embryos stocked is 55 percent apparent recovery. Without a joint classification table, one cannot know how many of the 550,000 live individuals are also in the 600,000 normal category. It is unsafe to multiply percentages or claim an exact “live and normal” yield without more information.

An editorial translation checklist

Before publishing a sentence that contains a larval number, verify the organism and life stage, sample origin, sampling time, chamber volume, counted objects, dilution steps, concentrate volume, exclusions and whether values are observed or inferred. Name uncertainty sources close to the result. Write “estimated live larvae recovered from the 2.4 L concentrate” rather than “the hatchery produced exactly 1,530,000 healthy oysters.” The latter adds precision, species identification and a health claim that an ordinary counting calculation did not establish.

Explain necessary symbols in running language: mL means millilitre; µm means micrometre; density has units of organisms per volume; percentage is a fraction multiplied by one hundred. Maintain original decimals when comparing methods, but round reported totals honestly. If source notes conflict, keep both the literal observation and the unresolved difference, and request an internal technical clarification rather than silently fixing the record. This approach also works in translation between natural languages, where the meaning of a denominator is easily lost.

Frequently asked question: may one high count replace replicate sampling?

A very large chamber count can reduce one kind of random counting variation, but it does not prove that the sampled portion is representative. An unrepresentative aliquot can contain thousands of animals and still misstate the tank. Replicate aliquots and mixed sampling reveal variation that one chamber cannot. Where the operator must minimise handling, a defensible reduced sampling plan should be documented and validated rather than presented as equivalent to a fuller design.

Counting high-density specimens can also increase overlaps or visual fatigue and may reduce classification accuracy. Diluting appropriately can improve viewing, but the dilution must be incorporated in the inverse calculation, and all transfers must be checked for losses. The best procedure balances count quality, organism welfare, handling risk and precision instead of maximising the number of visible objects at any cost.

Frequently asked question: does a counter detect viability?

Not automatically. An electronic particle counter can count particles by their instrument-defined physical properties; an image classifier may distinguish some shapes; and a microscopic observer may apply separate behavioural or morphological criteria. None of these should be translated as a universal proof that an individual will survive subsequent culture. The term “viable” requires a validated operational definition appropriate to the biological stage and objective.

If observers decide whether larvae are normal, alive or competent to settle, preserve the exact criteria and any inter-observer disagreement. For a practical hatchery, competence may be assessed by stage, morphology or behaviour under a defined protocol. A generic instrument-generated object count cannot replace that context. Separate the enumeration instrument from the biological decision it was asked to inform.

Frequently asked question: should every difference be called mortality?

No. Changes in measured count can reflect real deaths, transfers, intended culling, sieving losses, sampling variability, escaping the measurement frame, or an identification change. A report of fewer organisms at the next inspection is evidence for a reduced recovered count under a defined method. Mortality becomes a stronger claim only when relevant alternative paths are accounted for or direct death observations support it.

This is particularly important after grading. A tank may contain fewer larvae because smaller individuals were moved elsewhere; the biological cohort may remain mostly intact. Conversely, a constant estimated count could mask changing health or replacement by contamination. Good interpretation pairs numbers with dated operations and observed quality, rather than elevating a single end-point difference into a complete causal story.

How to Translate Easily | How to Read Bivalve Hatchery Larval Counts, Subsamples and Survival Estimates