Bycatch

Prime #

677

Origin domain

Marine Science Oceanography

Subdomain

fisheries management → Marine Science Oceanography

Core Idea

A selection or capture process aimed at one target class also captures non-target classes because of the selector's finite specificity; the captured non-targets are real outputs — often the dominant fraction of real impact — yet the harm persists because the success metric counts only target capture and never enters the process's own ledger.

How would you explain it like I'm…

The Wrong-Catch Net

Imagine you throw a big net in the water to catch one kind of fish you want. But the net also scoops up turtles and crabs and other animals you didn't want, because the net can't tell them apart. Those extra animals are caught even though you weren't trying to catch them. That's bycatch: catching things you weren't aiming for, just because your net isn't picky enough.

Caught By Accident

Bycatch happens when a process tries to grab one kind of thing but also grabs other kinds, because it can't tell them apart at the moment it grabs. Think of a fishing net meant for tuna that also traps dolphins, or an email spam filter that throws out a real letter along with the junk. The thing you wanted is the 'target,' and the stuff you didn't want is the 'non-target.' What makes bycatch sneaky is that nobody counts the non-target stuff: the scoreboard only tracks the tuna, so the trapped dolphins don't show up anywhere, and the problem keeps happening because it stays hidden.

Off-Target Capture

Bycatch is the pattern where a selecting process aimed at one target class also captures members of other classes, simply because the selector can't distinguish them at the point of capture. The captured non-targets aren't the goal, aren't measured by the success metric, and usually aren't budgeted for, but they're real outputs of the process and are often the biggest part of its real impact. Two features make this persistent: the selector has limited specificity (it can't tell target from non-target), and the success metric is blind to the non-target burden (it counts only target captures). Because the cost never shows up in the system's own ledger, there's no pressure to fix it, so it doesn't self-correct. It's different from a 'side effect' because the same act does both the wanted and unwanted work at once, not as a separate downstream consequence.

 

Bycatch names a structural pattern by which a selection or capture process aimed at one target class also harms or processes members of non-target classes as a consequence of the selector's finite specificity. It carries five structural commitments: (1) a selective process — a sorting, capture, classification, or enforcement mechanism — with limited discrimination; (2) a target class it is designed to act on; (3) one or more non-target classes it also acts on because it cannot distinguish them at the point of action; (4) an asymmetry of magnitudes, where target value is high per unit but non-target volume is large in absolute terms; and (5) metric invisibility, since the success metric counts only target capture. The load-bearing combination is the third and fifth commitments together: capture happens because of limited specificity, and the harm persists because of invisibility in the success metric. Bycatch is thus an internalized cost the process does not internalize to itself — an output it produces but does not see, which is exactly why it is persistent rather than self-correcting. It is distinct from 'spillover' or 'side effect' because the same act does both target and non-target work, and the burden is invisible in the process's own ledger. Many domains have local names for it — false positives, off-target effects, collateral damage, collateral consequences — but the structural identity is the same across them.

Broad Use

• Fisheries (origin): trawl and gillnet gear capture target species plus non-target sharks, turtles, dolphins, and juveniles, often larger by tonnage than the landed target.
• Law enforcement: a sweep or stop-and-frisk aimed at a target population also detains non-target people who match the selector imprecisely, uncounted in the arrest metric.
• Spam filtering: a filter removing prohibited content also removes legitimate messages that resemble the selector — false positives that rarely reach the dashboard.
• Pharmacology & agriculture: a drug acts on similar receptors (off-target binding); a pesticide kills pollinators and soil microbes alongside the pest.
• Search & surveillance: a query reaches adjacent unwanted results; bulk collection captures non-targets on the same wire.
• Medical screening: broad tests produce false positives leading to non-target interventions.

Clarity

Separates the impact the process intends from the impact it actually has, locating the divergence in finite selector specificity rather than operator carelessness — which matters because the two diagnoses point to different remedies (training versus selector redesign).

Manages Complexity

Compresses false positives, off-target effects, and collateral damage into one pattern with a small parameter set, and organizes the response into five moves: refine the selector, retime or relocate the selection, enable release, internalise the cost by metering, or substitute a cleaner method.

Abstract Reasoning

Installs a standing audit posture — ask of any selective process not only "what does it catch that it wants?" but "what does it catch that it doesn't want, and who pays?" — treating collateral harm as a budgeted cost rather than an invisible byproduct.

Knowledge Transfer

• Fisheries → pharmacology: refine-the-selector and enable-release map onto receptor-specificity engineering and antidote co-administration.
• Fisheries → policing: the bycatch audit becomes "collateral consequences" and hit-rate auditing in criminology.
• Statistics → fisheries: the precision/recall and ROC vocabulary inherits directly into gear-selectivity design.

Example

A binary classifier scored by precision and recall is bycatch on a confusion matrix: the decision boundary is the finite-specificity selector, false positives are non-target capture by the same thresholding act, low base rates make non-targets outnumber targets, and recall-only reporting is the metric invisibility — fixed by also reporting precision and adding a low-friction appeal.

Relationships to Other Primes

Parents (1) — more general patterns this builds on

• Bycatch presupposes, typical Type I & Type II Errors — Bycatch is what a false-positive RATE becomes when it acts on the world — the real-world non-target capture, with a magnitude-asymmetry + ledger-invisibility dimension the bare error taxonomy lacks. Built on the Type-I/false-positive category; presupposes it. The file: 'Bycatch is the impact-and-visibility structure built on top of the bare error category.'

Path to root: Bycatch → Type I & Type II Errors → Trade-offs → Constraint

Not to Be Confused With

• Bycatch is not an Externality because it is non-target capture by the same selective act at the point of selection from finite specificity, whereas an externality is a cost on a third party through a separate downstream mechanism.
• Bycatch is not a Type I Error because it adds the magnitude asymmetry and ledger invisibility of the real-world harm a false-positive rate produces, whereas the error taxonomy names only the bare false-positive category.
• Bycatch is not Selection Bias because it harms the captured non-targets themselves (real turtles, real bystanders), whereas selection bias distorts an inference about a population.