One batch of eggs can produce four completely different percentages
A hatchery reports 90 percent fertilisation, 80 percent hatch, 60 percent early survival and 30 percent nursery survival. Those figures sound like a chain of four successes. They are also easy to misread as a 30 percent overall yield. That conclusion is only justified if the final percentage was explicitly computed against the original egg count. More often each figure describes a transition from one developmental stage to the next. The denominators change as embryos hatch, larvae start feeding and juvenile fish are moved into a nursery. The first job is to recover those denominators.
This guide shows how to translate aquaculture hatchery records from fertilisation observations into real cohort accounting. It explains the difference between viable eggs, fertilised eggs, hatched larvae, feeding larvae, fry and saleable fingerlings, then works through checked numbers and uncertainty. The numbers in our exercises are invented for teaching, not measurements from a hatchery. Species, temperature, salinity, breeding system and disease history matter enormously; no example here is a treatment plan or a universal production benchmark.
A hatchery manager may use a percentage to answer whether incubation equipment worked; a customer may use the same number to estimate delivery; a biologist may need to separate failed fertilisation from embryo loss. Each reader is translating a different question from one table. Good reporting carries the cohort identity, counting method, observation time and stage definition alongside every rate. Without those details, multiplying printed percentages gives an illusion of precision instead of a verified account of animals actually alive at the final stage.
Start with a stage ledger rather than a headline survival rate
Draw six columns: eggs allocated to the batch, eggs assessed as fertilised, larvae observed to hatch, larvae alive at first feeding, fry alive at transfer, and fingerlings meeting a declared size and health criterion. One column must never silently replace another. Eggs removed for quality checks, dead embryos, larvae transferred out, sampled animals and escapees need separate entries. If a cohort is split between tanks, record the split and its uncertainty before adding the counts back together.
A source document should distinguish a snapshot count from a cumulative total. A tank may contain 12,000 larvae at 09:00 and 11,000 at 17:00, but the difference cannot automatically be called 1,000 deaths. Animals may have been removed for research, escaped through an outlet or moved into another compartment. A complete ledger reconciles beginning stock plus known additions minus documented transfers, mortalities and endings. An unexplained residual is a measurement or record discrepancy until investigated.
The ledger also separates biological status from saleable status. An animal may be alive but too small for a contract, excluded during grading or held for another week. Calling this mortality inflates biological loss, whereas calling all living animals finished product inflates saleable yield. A precise sentence might say that 8,700 live juveniles remained, 7,900 met the dispatch grade, and 800 were retained for further rearing. The three numbers answer different commercial and biological questions.
Define fertilisation observations without pretending the microscope sees everything
A fertilisation rate is usually an estimate derived from a defined inspection of eggs or embryos, not a direct census of every egg in a large batch. The observer may count normal cleavage, visible developmental structures or another species-specific marker. The exact criterion and time after spawning determine what the reported numerator means. Abnormal development, damaged eggs and ambiguous specimens must not be silently classified as either fully viable or unfertilised simply to simplify a worksheet.
Suppose 500 eggs are randomly sampled and 430 meet a specified fertilisation criterion. The observed rate is 430 divided by 500, or 0.86, meaning 86 percent of inspected eggs. If a separately estimated cohort contains 40,000 eggs, a naive projection suggests 34,400 fertilised eggs. It is a projection, not 34,400 individually confirmed cases. If collection favoured floating eggs and the full cohort includes sinking eggs, the sample may be biased even when the arithmetic is perfect.
Do not confuse fertilisation with viability. A fertilised embryo can die during incubation, and an embryo showing early cleavage is not guaranteed to produce a normal larva. A hatchery may report a percentage of buoyant eggs, percentage developing normally, or percentage fertilised. Buoyancy is an operational screening measure, not the same definition as confirmed fertilisation. Translate each phrase into the actual observational rule, identify the sample frame, and explain why the categories can overlap without being identical.
Hatch success uses a different numerator and may use a different denominator
A hatch rate can be reported as hatched larvae divided by fertilised eggs placed in incubators, or as hatched larvae divided by all eggs originally collected. Neither convention is intrinsically wrong if labelled. They answer different questions. The first measures conditional success during incubation among eggs accepted as fertilised. The second combines fertilisation and hatching losses into one overall result. Comparing hatch rates across laboratories requires confirming which convention was used.
Work through an illustrative cohort of 50,000 eggs. Suppose an estimated 40,000 were fertilised and 32,000 larvae hatched. Conditional hatch success is 32,000 divided by 40,000, or 80 percent. Hatches per collected egg are 32,000 divided by 50,000, or 64 percent. A report saying merely “80 percent hatch” is incomplete if a buyer believes 40,000 larvae are available from the original 50,000 eggs. The numerators are the same, but the denominators are not.
Counting a hatch is itself a biological decision. A partly emerged embryo, dead hatchling or malformed larva may or may not meet the protocol definition. Different sampling intervals may miss rapid mortality. For a defensible comparison, include hatch-window start and end, temperature history, representative sampling, counting method and the rule for live versus total hatch. The reader should be able to reproduce which animals entered the numerator, not merely copy the displayed percentage.
Translate stage survival as conditional and cumulative probabilities
Let F be fertilised proportion, H the conditional hatch proportion, L the fraction of hatched larvae alive at first feeding, and N the fraction of those larvae surviving to a nursery milestone. When a single closed cohort is followed and every stage is correctly nested, expected cumulative yield is F times H times L times N. The multiplication is a chain of conditional proportions. It is not a formula for independent biological events, and it is invalid when the stages refer to mismatched tanks or cohorts.
Take an illustrative batch with F=0.90, H=0.80, L=0.60 and N=0.50. The cumulative fraction is 0.90 × 0.80 × 0.60 × 0.50 = 0.216, or 21.6 percent of original eggs. Out of 100,000 eggs, that means 90,000 estimated fertilised, 72,000 hatched, 43,200 at first feeding and 21,600 at the later milestone. The 50 percent nursery survival is not a 50 percent yield from the egg stage. Calling 21.6 percent a 50 percent outcome hides almost four-fifths of total cohort loss.
Conversely, if the final 21,600 were counted directly, the overall output proportion is 21.6 percent even if intermediate survival estimates are incomplete. Do not multiply a measured final output by conditional rates a second time. A direct end count and a product of stage estimates are alternative estimates to reconcile, not two separate filters. Differences between them can reveal sampling error, omitted transfers or changing definitions rather than unexpected biology.
What an egg-count estimate means before a percentage is applied
Large egg batches are often estimated rather than counted one by one. A volume method multiplies sampled egg density by a measured water volume; a gravimetric method estimates eggs per unit egg mass. Neither method is free of assumptions. Clumping, settling, stratification and species-specific egg swelling may change the relationship between the small sample and the full container. A report should state the method, replicate counts and whether it estimates total eggs, viable eggs or eggs of a particular size class.
Imagine a vessel containing 200 litres of egg suspension, with a homogenised sample estimating 250 eggs per litre. Multiplying gives an estimated 50,000 eggs. But if half the vessel is incompletely mixed and counts range from 150 to 350 eggs per litre across replicate samples, a single 250 figure conceals substantial spatial variation. The repair is a defined mixing and sampling protocol with multiple aliquots, not rounding the final total to five significant figures.
In a gravimetric estimate, suppose three independent subsamples show 900, 1,020 and 980 eggs per gram. Their arithmetic mean is about 967 eggs per gram, with variation that should inform uncertainty. If 40 grams of suitably comparable egg material are weighed, the point estimate is roughly 38,700 eggs. This estimate is only as defensible as the sample representativeness and mass definition. Wet, swollen eggs do not automatically share the same conversion as dried eggs from another protocol.
A small sample creates uncertainty even if the count is unbiased
Counting 43 fertilised eggs among 50 sampled eggs gives the same point proportion, 86 percent, as counting 430 among 500. It does not give the same confidence in that estimate. With independent representative Bernoulli observations, a simple standard-error approximation for a sample proportion p is the square root of p(1−p)/n. At p=0.86 and n=500 the estimate is about 0.0155, or 1.55 percentage points; at n=50 it is about 0.0491, or 4.91 percentage points.
A rough normal approximation of plus or minus 1.96 standard errors would be about three percentage points in the larger illustrative sample and nearly ten in the smaller one. Real hatchery sampling may be clustered, spatially structured and dependent, making that simple model optimistic. For small samples and proportions near zero or one, a Wilson interval or a suitable model is generally preferable. Precision calculated under an invalid sampling design is not a substitute for representative collection.
If 10 eggs are taken from each of 10 different parents, those 100 outcomes are not necessarily 100 independent biological replicates. Parental identity, fertilisation method and incubation environment may affect all eggs in a cluster. Record batch and broodstock identifiers. An uncertainty interval should reflect relevant clustering or replicate batch variation. Translating 100 eggs as a hundred independent production units can make a very narrow interval look more credible than the experiment deserves.
The denominator changes again when larvae begin feeding
Newly hatched larvae may depend on yolk reserves before beginning external feeding. A first-feeding survival rate therefore captures a particular transition, not a generic survival claim from fertilisation onwards. To calculate it, first define the stage at which larvae are counted and when first-feeding competence is assessed. A protocol that counts living larvae at 24 hours cannot be compared automatically with another that counts actively feeding larvae at 96 hours.
Suppose 32,000 larvae hatch, 24,000 remain alive near mouth opening, and 18,000 demonstrate the specified feeding criterion. Survival from hatch to life at mouth opening is 24,000/32,000=75 percent. The fraction meeting the feeding criterion among the 24,000 alive at that inspection is 18,000/24,000=75 percent. The fraction from hatch to observed feeding competence is 18,000/32,000=56.25 percent. All three figures are true about this hypothetical cohort, but none can replace the others.
A feeding criterion may depend on gut inspection, observed strike behaviour, ingestion of tracer prey or a different method. A single feeding observation does not show that every larva will continue feeding, assimilate nutrients or grow normally. Avoid translating “first feeding achieved” into “successful weaning” or “healthy juveniles.” Weaning from live prey to manufactured feed is a separate process with its own timing, survival and growth consequences.
Follow mortality and removals separately across tanks
An incubator starts with an estimated 40,000 viable eggs. An operator transfers 25,000 larvae to Tank A, 6,000 to Tank B and retains 1,000 for observation. Adding gives 32,000 total hatched larvae allocated across destinations. If Tank A later records 19,000 alive and Tank B 4,800 alive, an overall survival calculation cannot omit the observation group without stating whether the denominator was restricted to Tank A and B. A partial sampling frame changes the question.
Within Tank A, 25,000 starting larvae minus 19,000 ending larvae equals 6,000 unaccounted for before documented transfers and sampling. Suppose 1,200 were moved to a different tank and 200 were removed for diagnostics. The residual is 4,600, but it is not automatically a confirmed mortality count if escapement or counting error remains possible. A transparent record labels it an estimated unexplained loss until evidence supports a more specific category.
Combining tanks by averaging their percentage survival can also be wrong. Tank A surviving at 76 percent and Tank B at 80 percent does not justify treating total survival as their simple mean of 78 percent when they started with different numbers. The pooled survival of the two tracked tanks is (19,000+4,800)/(25,000+6,000)=76.77 percent. A weighted total protects the denominator. Averaging tank percentages equally instead answers the different question, “What is the average tank-specific percentage?”
Why time windows are part of the meaning
Two reports may both say “larval survival 70 percent,” but one refers to day 1 through day 7 and the other to day 1 through day 30. These are different exposures. Survival probability over a longer interval will generally be no higher than survival over a shorter nested interval in a closed cohort, though counting noise can obscure that relationship. Always preserve age or calendar time at the start and end of a reported survival window, and the basis used to assign age.
A cohort of 10,000 at day 0 and 8,000 at day 5 has five-day survival of 80 percent. If 6,000 remain at day 10, ten-day survival relative to day 0 is 60 percent, whereas conditional survival during days 5–10 is 6,000/8,000=75 percent. Multiplying 0.80 by 0.75 reproduces 0.60. A chart may plot daily retention or cumulative retention; the accompanying text should state which. Otherwise identical curve shapes can be attached to contradictory claims.
A hazard rate is not the same thing as a percentage dying per day. Under a simplified constant continuous mortality hazard m, survival over time t can be expressed as exp(−mt). If 80 percent survives five days, estimated m=−ln(0.8)/5, about 0.0446 per day. This is a mathematical summary under a strong assumption, not proof that each day’s physiological risk was constant. Hatching, first feeding and transfer can produce sharply changing risks.
Distinguish biological loss from measurement and handling loss
Larvae may be too small to count reliably by eye, may collect in corners, or may be transported with nonuniform water flow. Sampling with a net can injure or exclude the smallest animals. The measurement itself can create stress. A survival estimate after transfer therefore bundles biology, handling, detection and the count method unless the study was designed to separate them. A reduction in reported animals is not always direct evidence of disease or poor nutrition.
Consider a pre-transfer sample estimate of 12,000 larvae with a method-specific error range and a post-transfer count of 10,500. Even if those two numbers are exact, they do not identify the cause of a 1,500 difference. Some animals could remain in the source tank, be caught in tubing, pass unnoticed through a screen or die. The next task is to reconcile transfer volumes and equipment recovery, repeat independent counts and inspect the physical pathway. An emotionally compelling narrative is not a substitute for an inventory.
When comparing methods, record whether counters use manual aliquots, imaging systems, optical beams or mass-based conversion. An image analysis model can misclassify debris as larvae or count overlapping animals as one. A claimed gain in hatch rate after switching counting systems may be a detection change. Validate the old and new methods on shared specimens before writing that a new treatment improved fertilisation or incubation performance.
Keep species and environmental context attached to the numbers
Egg development, hatch timing and larval feeding requirements vary between fish and crustacean species and across temperature and salinity conditions. Historical Food and Agriculture Organization hatchery accounts of seabass and other species illustrate how hatch rates and stage losses can vary under defined methods. Such reports are evidence for those experimental or operational conditions, not modern universal targets. A number copied from one species to another may be easy to translate and biologically indefensible.
Temperature affects developmental speed. If a warmer incubator hatches earlier, counting both incubators at the same clock hour may make the cooler one appear unsuccessful simply because its embryos have not completed development. Use a biologically relevant hatch window and recorded temperature history. The relationship is not an unlimited benefit from warmth; temperatures beyond a species’ tolerance can damage embryos. A careful explanation avoids prescribing a single incubation temperature without species and life-stage evidence.
Salinity, oxygen, microbial exposure, egg handling and parental condition may all influence observed outcomes, but a correlation in hatchery logs does not isolate causality. A batch with better hatch success and higher dissolved oxygen might also have higher-quality broodstock or a newer incubator. Compare batches using appropriate replication and controls before crediting one operating variable. An explanatory article should show what a rate measures, then distinguish observed association from a justified intervention claim.
Worked case: the apparently excellent ninety-percent hatchery
A fictional hatchery collects an estimated 120,000 eggs, samples 600 eggs and finds 540 passing a fertilisation criterion. The projected fertilised number is 120,000×540/600=108,000. Next the incubator count estimates 91,800 hatchlings. Conditional hatch success is 91,800/108,000=85 percent, while hatchlings relative to collected eggs are 91,800/120,000=76.5 percent. Printing only “90 percent fertilised” and “85 percent hatched” can tempt a hurried reader to report 85 percent of total eggs hatched.
Suppose 64,260 larvae survive to a defined first-feeding inspection. Conditional survival from hatch is 64,260/91,800=70 percent. If 38,556 reach a subsequent nursery checkpoint, conditional survival over that interval is 38,556/64,260=60 percent. Overall retention from collected eggs is 38,556/120,000=32.13 percent. The product 0.90×0.85×0.70×0.60 also equals 0.3213, confirming the accounting chain. This agreement is a mathematical consistency check, not evidence the original observations were unbiased.
Now suppose 3,000 of the final 38,556 are alive but held back from a customer’s minimum size grade. Dispatchable output is 35,556, or 29.63 percent of the original eggs. The final biological output and dispatchable output should both be reported. The retained animals have not died, and moving them into “mortality” simply to fit a sales dashboard would misrepresent the biology. Conversely, claiming delivery of 38,556 animals when only 35,556 meet the specification misrepresents the contract.
Worked case: a comparison that reverses when denominators change
Batch A begins with 20,000 eggs, has an estimated 95 percent fertilisation and 70 percent conditional hatch. It produces 13,300 hatchlings. Batch B begins with 20,000 eggs, has 80 percent fertilisation and 80 percent conditional hatch. It produces 12,800 hatchlings. Which hatchery is better? Batch B has the higher conditional hatch percentage, but Batch A has the higher number of hatchlings per collected egg. Neither statement alone describes final performance.
If the buyer is considering incubation efficiency among eggs already confirmed fertilised, Batch B’s 80 percent may be relevant. If the buyer needs larvae per egg collected, Batch A’s 66.5 percent compared with Batch B’s 64 percent is relevant. If Batch A uses expensive broodstock conditioning or suffers a later larval loss, either ranking could reverse again. Ask which decision is being made, not which percentage is largest. Good translation preserves the denominator before presenting a comparison.
Suppose the two batches are run in tanks of different sizes, with different broodstock, different salinity and different counters. A ranking then becomes more uncertain still. Any claim that Batch B’s incubation process is superior should address plausible confounding, sampling variance and method comparability. A worked numerical difference is not a causal experiment. It teaches what the arithmetic says and what remains unknown.
Worked case: time-to-hatch can make a chart look wrong
Imagine two egg incubators each receive 10,000 fertilised eggs. At 24 hours, 8,000 in Tank A and 4,000 in Tank B have hatched. A report limited to that time says A achieved 80 percent and B 40 percent. At 36 hours, B reaches 8,500 hatched larvae while A remains at 8,000. The earlier ranking is not false for the snapshot, but it is not the final hatch success ranking. The discrepancy is a time-window mismatch.
To interpret the chart, ask whether hatch is recorded as cumulative newly hatched individuals, living hatchlings present at each time, or a rate over each interval. A hatchling dying between timepoints might remain in a cumulative hatch count but disappear from a live-count chart. Those two series can move differently. Some studies also remove hatchlings at each inspection, altering subsequent counts. The legend should define the numerator and fate of animals after each sample.
A robust report could say “cumulative hatch by 24 hours” and “cumulative hatch by 36 hours,” with the number of fertilised eggs entered at time zero stated once and verified. It could also present survival at each inspection separately. These labels protect the reader from interpreting developmental timing as a biological failure or falsely assuming that earlier hatching always means better eventual juvenile quality.
Worked case: a tank split must be reconstructed
A batch has 30,000 hatchlings. Operators allocate 10,000 to Tank 1, 15,000 to Tank 2 and 5,000 to Tank 3. Later they count 7,000, 12,000 and 4,500 alive respectively. The tank-specific survival rates are 70, 80 and 90 percent. The simple average is 80 percent, but the weighted combined survival is 23,500/30,000=78.33 percent. Neither is an arithmetic mistake; they answer different questions and should be named appropriately.
Suppose Tank 3 contains the most robust size grade selected at the split. Its 90 percent survival does not demonstrate that Tank 3’s environment is better. The fish populations were different before treatment. To test a tank intervention, assignment and baseline health should be comparable or adjustment should account for the difference. A survival table without a cohort-selection column risks turning a grading decision into a false causal conclusion.
The next practical step is to record each destination’s starting count and selection rule, plus any transfer losses or losses to observation samples. If a manager combines the tanks again, their new shared starting count is 23,500, not 30,000. A later 90 percent survival measured after recombination should be applied to that current baseline, producing an expected 21,150 animals under the simplified assumption. Resetting the denominator without reporting it obscures attrition history.
Worked case: uncertainty compounds through the chain
Suppose the original egg count is estimated at 100,000 with a plausible uncertainty of several percent, fertilisation is estimated from 200 inspected eggs, and larval survival is measured by subsampling a crowded tank. It would be misleading to report exactly 21,638 final animals merely because a spreadsheet multiplied proportions to three decimal places. Every estimated stage has measurement error, and some errors may be correlated because a biased count method is reused.
One transparent approach is to report observed counts with their sampling procedures and give a reasonable range for the total rather than manufacturing exact animal-level precision. If errors are modelled statistically, the distribution of each proportion and its dependence structure matter. A Monte Carlo calculation may propagate uncertainty by drawing stage counts or proportions under defensible models, but no simulation can repair an unrepresentative sample frame or unknown undocumented removals.
The uncertainty of cumulative yield is not necessarily the sum of four percentage-point margins. Proportions multiply, errors may be asymmetric, and sample counts from nested stages are dependent. A fully specified binomial or hierarchical model can address those details where warranted. For routine communication, an explicit estimate with method, approximate interval and caveat is usually more honest than several decimals and a vague statement that the figures are “accurate.”
Errors to repair before the report reaches a reader
One frequent error is adding percentages from consecutive stages. If 80 percent of eggs hatch and 75 percent of those hatchlings survive a later stage, cumulative retention is 0.80×0.75=60 percent, not 155 percent and not the arithmetic mean of 77.5 percent. Another error is interpreting a percentage decrease in the rate as a percentage-point decrease. A decline from 80 to 70 percent is ten percentage points but 12.5 percent relative to the original rate.
A second error is silently treating every missing animal as dead. Sampling removal, shipment, intentional grading and equipment escape must be documented separately. A third is transferring a rate from a sample to a cohort without telling the reader that the total is modelled. A fourth is comparing hatch rates without ensuring that both use fertilised eggs, or both use all collected eggs, as denominator. Each can change operational conclusions more than an impressive decimal precision ever could.
A practical repair workflow begins with an auditable map: source record, batch ID, start time, finish time, stage boundary, numerator, denominator, counting instrument and any exclusion. Recalculate each percentage from the raw counts. Reconcile flow between tanks. Inspect whether a metric is directly observed or projected. Record outstanding ambiguity as an explicit question rather than hiding it. Only then choose plain-language terms such as fertilised, hatched, alive, feeding, transferred, retained or dispatchable.
Transfer problems with worked answers
Problem one: A cohort contains an estimated 75,000 eggs; 60,000 are judged fertilised, 48,000 hatch and 30,000 survive to first feeding. What are the fertilisation, conditional hatch and hatch-to-feeding survival rates? The answers are 60,000/75,000=80 percent, 48,000/60,000=80 percent, and 30,000/48,000=62.5 percent. Overall first-feeding retention relative to collected eggs is 30,000/75,000=40 percent. The three stage rates must not be given a shared denominator after the fact.
Problem two: Tank A starts with 2,000 larvae and ends with 1,600; Tank B starts with 8,000 and ends with 6,000. The individual survival percentages are 80 and 75 percent. What is pooled survival? Total alive is 7,600 out of 10,000, so 76 percent. The unweighted average of percentages is 77.5 percent and describes a different statistic. If the sizes of the tanks reflect preferentially allocating large larvae to one tank, even the pooled value should be presented as a descriptive output, not causal evidence about husbandry.
Problem three: At noon, 6,000 of 10,000 fertilised eggs have hatched. By midnight, 8,000 have hatched cumulatively, but only 7,500 hatchlings remain alive. What is the 24-hour hatch success and the live hatchling proportion at that point? They are 80 percent and 75 percent respectively, assuming the full cohort is observed. The difference represents 500 larvae no longer alive at the final inspection, under the simplified closed-cohort account. If animals were removed, that interpretation must be revised.
Problem four: A report calls an estimate of 18,000 eggs “exact” after using 30 aliquots. What must be checked? Look for sampling representativeness, mixing, aliquot volumes, microscope counting uncertainty, replicate variance and the original bulk volume or egg mass. The presence of 30 samples does not guarantee absence of systematic bias. A valid answer is not an invented confidence interval; it is a plan to reconstruct how the estimate was obtained and identify which errors the samples can and cannot detect.
Frequently asked questions
Does a 95 percent fertilisation rate mean that 95 percent of eggs will become fry? No. Fertilisation is an early stage and does not include embryo mortality, hatch failures, first-feeding losses, nursery mortality or later grading. To estimate later production, follow the same cohort through the relevant stage transitions and multiply correctly defined conditional retention proportions, or count the final animals directly. Calling early fertilisation “fry yield” confuses potential with realised output.
Can I compare hatchery A and hatchery B from hatch rate alone? Only for a tightly defined comparison. Confirm species, starting egg condition, denominator, incubation conditions, hatch observation window and counting protocol. If one hatchery uses all spawned eggs and another uses screened fertilised eggs, their printed rates describe different things. Differences in broodstock or handling may further confound claims about operator skill or equipment quality.
Why can reported survival exceed 100 percent? In a true closed cohort with complete counts, it cannot. Apparent values above 100 percent arise from measurement error, underestimated initial counts, new animals entering the system or mismatched observation frames. Do not clip such a value to 100 percent without checking the ledger; the discrepancy is diagnostic information. State the unexplained count difference and correct the source if an audit finds a defensible correction.
Is a higher early hatch rate always the best decision? No. A decision may prioritise healthy feeding larvae, long-term juvenile survival, animal welfare, genetic diversity, disease control or sustainable broodstock use. A high early rate can coexist with later poor outcomes. Do not assume an intervention is beneficial without examining the whole pathway and appropriate comparator evidence. The purpose of a stage ledger is to expose rather than erase the trade-offs.
What if counts are based on subsamples? Report the sampling method, number of replicates and uncertainty when possible. An estimated cohort count should be labelled estimated. If samples come from different tanks or times, investigate whether they genuinely represent the same population. Precision from a formula is not independent verification of the biology or the counting process. If a count is too uncertain for an operational decision, the honest result is that more measurement is needed.
Primary evidence, limitations and related guides
Food and Agriculture Organization historical hatchery manuals document stage-specific hatch and survival data for seabass and other cultivated species, demonstrating why developmental milestones and denominators matter. Their numerical results belong to their recorded species and conditions. This guide’s worked examples are intentionally hypothetical. Readers should consult species-specific current technical guidance, local veterinary and aquaculture standards, and actual operational measurements before making animal health or stocking decisions. No fabricated measurements or unverified treatment claims have been used.
A useful reader can now look at a hatchery spreadsheet and ask four separate questions: What is the biological stage? Which cohort did it belong to? What was the denominator at that stage? Was the count observed, estimated or transferred from another calculation? If those answers can be recovered, the difference between fertilisation, hatch, live first-feeding larvae and finished juveniles becomes clear. If they cannot, the correct translation is a qualified statement that preserves the unresolved gap rather than an invented success rate.
References and connected reading
FAO seabass propagation and stage-specific hatch/survival. FAO chapter on fish propagation and hatchery measures. FAO shrimp hatchery production design assumptions. eduKateSG aquaculture RAS circulation interpretation. eduKateSG aquaculture feed conversion and live biomass