Skip to content

Additive Bias

Prime #
619
Origin domain
Cognition Judgment Decision Making
Subdomain
problem solving biases → Cognition Judgment Decision Making

Core Idea

Additive bias is the pattern that, when an agent — a person, team, organization, evolving lineage, or regulator — is asked to improve a system, the agent reliably reaches first for adding a component (a feature, rule, step, piece, person, layer) and reliably underweights the alternative of removing an existing one. The asymmetry holds even when subtractive transformations are demonstrably better, and it recurs at the organizational, legal, technical, and biological scale, where the ratio of additions to removals over a system's lifespan is strikingly skewed regardless of which would optimize the system's stated objective.

The structural commitment is a direction-asymmetric search distribution: when generating candidate transformations of an existing system, agents draw far more heavily from the "add-a-component" region than from the "remove-a-component" region of transformation-space, even when the latter holds better solutions. The asymmetry is not a one-off error but a bias in the search distribution itself, persisting across substrates because its generative factors recur — the visibility of additions versus the invisibility of removals, the credit-assignment asymmetry between adding and removing, and the cognitive difficulty of reasoning about counterfactual absence versus counterfactual presence. The mechanism produces a characteristic accretion signature in long-lived systems: components accumulate over the system's history at a rate that outpaces removal, the system grows in size and complexity along a trajectory tied more to iteration count than to purpose, and removals, when they happen, are infrequent, deliberate, and require special machinery — sunset clauses, refactoring sprints, deliberate purges.

How would you explain it like I'm…

Add-A-Block Habit

When something is wrong, most people try to ADD something to fix it instead of taking something away. If a tower wobbles, we want to glue on more blocks, when really we should pull out the crooked one. Taking things away is just as good a fix, but our brains forget to try it. Additive Bias is reaching to add a piece when removing one would work better.

Pile-It-On Bias

When people are asked to improve something, they almost always think of adding a part, a rule, a step, or a feature, and they almost never think of removing one. This happens even when removing would clearly make things better. It shows up everywhere: in companies, in laws, in machines, even in nature. Over a long time, things just keep piling up, because adding is the move we reach for and removing is the move we forget. Removals are rare and usually need a special push, like a cleanup day or a rule that automatically expires.

The Add-Don't-Remove Bias

Additive Bias is the pattern that when an agent is asked to improve a system, it reliably reaches first for adding a component and underweights the option of removing one. The asymmetry holds even when subtractive changes are clearly better, and it recurs at organizational, legal, technical, and biological scales, where additions vastly outnumber removals over a system's life. The structural claim is a direction-asymmetric search distribution: when generating candidate changes, agents draw far more from the add region than the remove region of the space of possible changes, even when the better solutions sit in the remove region. This is not a one-off mistake but a bias in the search itself, persisting because its causes recur: additions are visible while removals are not, adding earns clearer credit than removing, and it is harder to reason about an absence than a presence. The result is a steady accretion signature, where components pile up faster than they are removed.

 

Additive Bias is the pattern that when an agent, whether a person, team, organization, evolving lineage, or regulator, is asked to improve a system, the agent reliably reaches first for adding a component (a feature, rule, step, piece, person, or layer) and reliably underweights the alternative of removing an existing one. The asymmetry holds even when subtractive transformations are demonstrably better, and it recurs across organizational, legal, technical, and biological scales, where the ratio of additions to removals over a system's lifespan is strikingly skewed regardless of which would optimize the stated objective. The structural commitment is a direction-asymmetric search distribution: when generating candidate transformations of an existing system, agents draw far more heavily from the add-a-component region than from the remove-a-component region of transformation-space, even when the latter holds better solutions. This is not a one-off error but a bias in the search distribution itself, persisting across substrates because its generative factors recur: the visibility of additions versus the invisibility of removals, the credit-assignment asymmetry between adding and removing, and the cognitive difficulty of reasoning about counterfactual absence versus counterfactual presence. The mechanism produces a characteristic accretion signature in long-lived systems: components accumulate faster than they are removed, the system grows in size and complexity along a trajectory tied more to iteration count than to purpose, and removals, when they happen, are infrequent, deliberate, and require special machinery such as sunset clauses, refactoring sprints, or deliberate purges.

Structural Signature

an agent asked to improve an existing system — a transformation-search distribution over candidate changes — an overweighting of the add-a-component region — an under-visiting of the remove-a-component region — an accretion trajectory tied to iteration count rather than purpose — a baseline-shift invariant: the accreted state is perceived as the system, compounding the bias across generations

The pattern is present when each of the following holds:

  • An improvement task. An agent — person, team, organization, lineage, regulator — is asked to improve a persistent system.
  • A transformation-search distribution. The agent generates candidate changes by drawing from a space of possible transformations of the existing system.
  • An additive overweight. The agent draws far more heavily from the "add a component" region — a feature, rule, step, layer, person — than the alternatives warrant.
  • A subtractive under-visit. The "remove a component" region is systematically underweighted even when it holds better solutions, because removals are less visible and harder to credit, and counterfactual absence is harder to reason about.
  • An accretion trajectory. Components accumulate over the system's history faster than they are removed, so size and complexity grow with iteration count rather than with purpose.
  • A baseline-shift invariant. Each generation perceives the current accreted system as the system; the un-added counterfactual is invisible, weakening the case for removal and compounding the bias across rounds.

The components compose so that the defect is a bias in the search distribution itself, not a one-off error: the structure makes absence-of-removal say nothing about necessity, predicts equilibrium complexity sits above the purpose-optimal level by a margin set by the bias strength, and shows that any process that merely collects proposals inherits the additive skew.

What It Is Not

  • Not bias in general. bias (the near-identical embedding neighbor) is the umbrella of any systematic deviation; additive bias is the specific skew in a transformation-search distribution toward adding rather than removing.
  • Not objective creep. objective_creep is the unauthorized expansion of a goal-set; additive bias is the cognitive preference for adding components when improving a system, whatever the goal.
  • Not scope creep. scope_creep is requirements growing within a project; additive bias is the deeper search-distribution asymmetry that produces such growth across all substrates.
  • Not a violation of parsimony. parsimony_occam_s_razor is a normative preference for simpler explanations; additive bias is a descriptive fact about how agents actually search — they under-sample subtraction even when simplicity is better.
  • Not opportunity asymmetry. opportunity_asymmetry is an uneven distribution of openings; additive bias is an uneven search distribution over add-versus-remove transformations.
  • Common misclassification. Reading accreted complexity as evidence it must be earning its keep. Catch it by asking whether the remove-step was ever actually considered; absence of removal reflects an under-visited search region, not demonstrated necessity.

Broad Use

The pattern appears wherever an agent population can transform a persistent system over time. In cognitive problem-solving it is the original demonstration substrate: across building, grid, schedule, and itinerary tasks, subjects offered both additive and subtractive transformations systematically failed to consider subtractive ones even when those were better and explicitly permitted. In software it is codebase accretion, where deletion is rare and formal refactoring sprints are needed to remove what additions did not earn. In legal and regulatory regimes it is the monotonic growth of statutes and codes, where repeal is far costlier than enactment. In bureaucratic process it is the accumulation of forms, sign-offs, and approval layers added in response to perceived failures and rarely removed when the failures cease. In product design it is "feature creep," where pressure to add user-visible features is sharp and pressure to remove them weak. In curriculum design it is the accretion of content, courses, and requirements that pruning exercises rarely reverse. In biology it appears as an analogous asymmetry — vestigial structures and genomic accretion — where adding features is incrementally easier under selection than coordinated removal. It recurs in language (vocabularies accrete; obsolete terms persist), in urban and architectural accretion, in scientific literature, and in personal possessions and digital files. The substrate varies; the direction-asymmetric search distribution is the same.

Clarity

The label clarifies a confusion between complexity is necessary and complexity is the result of an asymmetric search bias. When a system has accreted to large size, the natural inference is that all this complexity must be earning its keep, since otherwise it would have been removed. Additive bias defeats that inference: the remove-step is under-represented in the search distribution, so absence-of-removal is not evidence that removal was ever considered. The clarifying act is to make absence-of-removal say nothing about necessity.

It also names a recurring intuition that lacked a sharp label. "Bloat," "feature creep," "regulatory accretion," "scope creep," and "complexification" all gesture at the same phenomenon in their respective substrates; additive bias is the substrate-portable structural commitment they each instantiate. By naming the generative asymmetry rather than its surface symptom, the concept lets an analyst recognize that these scattered complaints share one cause — and therefore one family of remedies — rather than treating each as a domain-specific quirk. This reframing changes what a reader sees in a complex system: accreted complexity becomes evidence about the search distribution that produced it rather than evidence of necessary complexity, which changes the diagnosis from "the complexity must be there for a reason" to "we likely need pruning."

Manages Complexity

The pattern compresses a wide range of accretion phenomena — across psychology, software, law, organizations, products, curriculum, biology, language, and architecture — into a common diagnosis: identify the search distribution the agent is drawing transformations from, and recognize that the remove-component region is systematically underweighted unless deliberate machinery is installed to weight it. A long catalogue of domain-specific bloat collapses to one structural question about the distribution over candidate changes.

The compression sorts the interventions, each re-weighting the search distribution toward the underweighted region. Surface the remove option — explicitly prompt agents to consider subtractive transformations, realized as subtractive-only sessions, refactor-only sprints, regulatory-sunset reviews, or curriculum-pruning weeks. Increase the visibility of removal-benefit — make the cost of retained-but-unneeded components observable (storage, cognitive load, maintenance, compliance overhead) so removal earns visible credit. Lower the asymmetric political cost — design machinery (sunset clauses, default-off, opt-in retention) that changes a ledger in which removal triggers vocal opposition and gives only diffuse benefit. Set explicit removal targets — numerical quotas such as one-in-two-out rules that force the distribution toward the underweighted region. Install structural removal machinery — schedule the act of considering removal so it need not be initiated under the bias. Having the structure in hand is what makes these legible as one repertoire.

Abstract Reasoning

Holding additive bias as a unit licenses inferences about the equilibrium complexity of long-lived systems under iteration. A system whose evolution is left to additive-bias-subject agents will accrete components beyond its optimal complexity until either the cost of complexity becomes acute enough to overcome the bias — forcing a removal episode — or exogenous machinery imposes removal. Either way, the equilibrium complexity sits above the purpose-optimal complexity by a margin that depends on the strength of the bias and the cost of the bias-correction machinery. This is a structural prediction about where complexity settles, derived from the asymmetry rather than from any domain's particulars.

The abstraction also licenses a prediction about which transformations to expect: when stakeholders are asked to propose changes to an existing system, the predicted distribution of proposals will be heavily skewed toward additions, so any deliberation process that merely collects proposals inherits the asymmetry. And it predicts self-reinforcement across generations: each generation perceives the current accreted system as the system, the un-added counterfactual is invisible, and the case for removal weakens ("things have been this way for a long time"), so each round of additive bias raises the baseline against which the next is evaluated. Reasoning from the pattern, an analyst can forecast that a system left to iterate will overshoot its optimal complexity, can predict the direction of any proposal-collection process, and can recognize that the baseline-shift makes the bias compound — inferences unavailable to anyone who reads accreted complexity as evidence of necessity.

Knowledge Transfer

The structural roles map across substrates, and with them a constant-shape intervention family travels. The existing system corresponds to a codebase, a legal corpus, an org chart, a curriculum, a product, a genome, a vocabulary, a building stock; the transformation-search distribution to the set of candidate changes an agent draws from; the additive overweight and subtractive under-visit to the reliable skew toward adding; the accretion trajectory to the iteration-count-driven growth; the bias-correction machinery to whatever explicitly re-weights toward removal; the self-reinforcing baseline shift to the rising perceived baseline. Because the roles correspond, a practitioner who has run a refactor-only sprint recognizes a regulatory sunset review or a curriculum-pruning week as the same move.

The interventions inherit that portability. Explicit prompting of the underweighted region generalizes from the original cognitive demonstration to any context with an asymmetric search distribution. Refactoring as the named, separately scheduled act of removal-without-functional-change exports from software to any domain where removal will not happen during ordinary additive work. Sunset clauses and one-in-X-out rules import from regulatory design into product roadmaps and process governance. Zero-based budgeting and scheduling require justifying every existing resource rather than treating it as baseline, transferring the "default-off" logic across organizations. Deliberate feature subtraction, household-style removal rituals, and biology's reductive evolution (the natural cases where selection does favor removal) all illustrate the same structural complement. The constant-shape intervention across every substrate is to install machinery that explicitly weights or prompts the remove-component region, because the default search distribution will not visit it. The transfer is real but bounded: because the bias is rooted in human agents transforming systems, what travels is a repertoire for re-weighting a human (or selection) search distribution, and the substrate is human decision rather than a substrate-neutral structure — yet within that bound the pattern is sharp and portable, recognizable wherever agents add more readily than they remove.

Examples

Formal/abstract

The original controlled demonstration makes the search-distribution asymmetry exact. Present an agent with an existing structure — a Lego bridge that must be stabilized, a grid pattern that must be made symmetric, a travel itinerary that must be improved — and a transformation: each candidate change either adds a component or removes one, and in the designed task a subtractive change is strictly better (it reaches the goal at lower cost). Model the agent's behavior as sampling candidate transformations from a distribution over the transformation space, then selecting the best sampled. The bias is a property of the sampling distribution: the probability mass on the "remove-a-component" region is far below what an unbiased search would assign, so the better subtractive solution is frequently never sampled and therefore never selected — even though it is permitted, available, and superior. The decisive structural prediction follows: the failure persists when the better solution is explicitly subtractive and degrades when removal is made salient (a prompt that names subtraction as an option re-weights the distribution and recovers the subtractive solution). This yields an equilibrium-complexity result for iterated improvement: a system evolved by additive-biased sampling settles above its purpose-optimal complexity by a margin set by the bias strength and the cost of correction machinery, because each round draws disproportionately from the add region and the baseline-shift makes the un-added counterfactual invisible to the next round.

Mapped back: The controlled task instantiates every role — improvement task, a transformation-search distribution, additive overweight, subtractive under-visit, an accretion trajectory under iteration, and the baseline-shift invariant — and locates the defect in the sampling distribution itself, not in any single judgment.

Applied/industry

In software engineering, a codebase under continuous improvement accretes features, abstractions, and dependencies because each improvement task draws first from "add a module / add a flag / add a layer," while deletion is rare — the remove region is under-visited because removed code is invisible in the diff's value story and counterfactual absence is hard to credit. The accretion signature is direct: complexity grows with commit count rather than with delivered purpose. The prime's interventions apply: a separately scheduled refactor-only sprint (surface the remove option as a named act), observability of maintenance cost (make removal-benefit visible), and an explicit deletion target. The identical structure governs regulatory regimes: statutes and codes grow monotonically because enactment is visible and creditable while repeal is politically costly and its benefit diffuse, so the corpus accretes past its purpose-optimal size; sunset clauses and one-in-two-out rules are the bias-correction machinery that forces the search distribution toward removal. And in product design, "feature creep" arises because user-visible additions earn sharp credit while removals risk vocal opposition from the few who used the cut feature; default-off features, deliberate feature subtraction, and zero-based roadmap reviews (justify every existing feature rather than treating it as baseline) re-weight toward the underweighted region.

Mapped back: Across software, regulation, and product design the same roles recur — an improvement task drawing from a direction-asymmetric search distribution, an additive overweight, a subtractive under-visit, and an accretion trajectory tied to iteration — and the same intervention family transports: install machinery that explicitly prompts or weights the remove-component region, because the default distribution will not visit it.

Structural Tensions

T1 — Search Distribution Bias versus One-Off Error (scopal). The prime locates the defect in the search distribution itself, not any single judgment — agents under-sample the remove region. The failure mode is one-off attribution: treating each unnecessary addition as an isolated mistake to correct individually, missing that the distribution will keep generating them. Diagnostic: does the agent reliably under-visit subtraction even when removal is permitted and better? If so, no amount of per-decision correction helps; the sampling machinery must be re-weighted.

T2 — Accreted Complexity versus Necessary Complexity (sign/direction). The frame defeats the inference that accreted complexity must be earning its keep — absence of removal says nothing about necessity. But over-applied, it assumes all complexity is bloat. The failure mode is complexity nihilism: pruning components that were genuinely load-bearing because the frame primed the analyst to read all accretion as bias-driven. Diagnostic: was the remove-step ever actually considered for this component, and does removing it degrade function? Absence-of-removal is not evidence of necessity, but neither is accretion evidence of waste; only testing removal decides.

T3 — Subtraction Prompting versus Over-Pruning (sign/direction). Surfacing the remove option re-weights the distribution toward subtraction, but a removal quota (one-in-two-out) can force removal of still-needed components to hit the number. The failure mode is quota-driven over-removal: cutting valuable components to satisfy a removal target, the mirror of additive bias. Boundary with objective_creep's subtraction coupling. Diagnostic: does the removal target free genuine bloat, or does it force cutting things that earn their keep? A blind subtraction quota over-corrects exactly as a blind additive default over-adds.

T4 — Baseline Shift versus Legitimate Stability (temporal). Each generation perceives the accreted system as the system, weakening the case for removal ("things have been this way") and compounding the bias — but some long-standing components are stable precisely because they work. The failure mode is baseline-shift overcorrection: treating tenure as evidence of bias-driven persistence and pruning components whose longevity reflected genuine value. Diagnostic: is the component un-removed because removal was never considered, or because it has repeatedly survived scrutiny? Baseline shift hides un-examined accretion, but not everything old is unexamined.

T5 — Removal Visibility versus Diffuse Benefit (coupling). The remedy makes removal-benefit observable (storage, maintenance, compliance cost) so removal earns credit, but removal's benefit is often genuinely diffuse while its cost (vocal opposition from the few users of a cut feature) is concentrated. The failure mode is concentrated-opposition capture: removal machinery defeated because the diffuse beneficiaries are silent and the concentrated losers are loud. Boundary with agency_problem. Diagnostic: is the removal's benefit measurable and attributable, or diffuse? Without making the diffuse benefit visible, the concentrated opposition wins regardless of the bias-correction machinery.

T6 — Human Search Bias versus Substrate Optimization (scalar). The bias is rooted in human (or selection) search distributions, so the remedy re-weights a human process — but the frame extends to biology (genomic accretion, reductive evolution), where there is no human agent and the dynamics differ. The failure mode is substrate overreach: applying human-debiasing remedies (prompts, checklists) to a non-human accretion process that has its own selective logic. Diagnostic: is the search distribution a human cognitive process or a selective one? The prompting interventions assume a human agent; selection-driven accretion needs a different lever (changing the fitness landscape, not the prompt).

Structural–Framed Character

Additive bias sits on the framed side of the structural–framed spectrum, matching its aggregate of 0.5. There is a real relational structure — a direction-asymmetric search distribution that over-samples the add-a-component region and under-visits the remove region, producing an accretion trajectory tied to iteration count — but the bias is rooted in human (and, by extension, selection) search distributions, and that grounding is decisive for the grade.

The pinning diagnostic is human-practice-bound, scored at the ceiling. Additive bias is a cognitive bias rooted in human agents transforming systems: it requires an agent asked to improve a system who reaches first for addition, an asymmetry driven by the visibility of additions versus the invisibility of removals and the difficulty of reasoning about counterfactual absence — all features of human cognition. What travels is a repertoire for re-weighting a human search distribution, and the substrate is human decision rather than a substrate-neutral structure. The remaining diagnostics read mid-scale or low. Vocabulary half-travels: "search distribution," "refactoring," "sunset clause," "zero-based budgeting," and "default-off" carry a problem-solving and design lexicon a new domain must partly adopt. Evaluative weight is moderate — "bias" and "bloat" name a distortion to be corrected — without full institutional loading. Institutional origin reads zero, fittingly: the mechanism is individual-cognitive, not lodged in any formal institution, even though it manifests in institutional artifacts. Invoking the prime half-imports a frame (install machinery that prompts the remove region; absence of removal says nothing about necessity) and half-recognizes an accretion already present.

The boundary case is biology — vestigial structures, genomic accretion, reductive evolution — which the prime extends to but which has its own selective logic and no human agent. The entry's own tension flags exactly this: the prompting remedies assume a human agent, while selection-driven accretion needs a different lever (changing the fitness landscape, not the prompt). So the structure reaches a non-cognitive substrate only by changing what "search distribution" means, which is why the grade stays framed rather than mixed-structural. That is the framed signature — a genuine relational asymmetry whose default instances are human cognitive processes and whose remedies re-weight a human (or selective) search, carrying that practice with it.

Substrate Independence

Additive bias is a moderately substrate-independent prime — composite 3 / 5 on the substrate-independence scale. Its domain breadth is wide: the direction-asymmetric search distribution favoring addition over removal is documented in cognitive problem-solving (the original building, grid, and itinerary experiments where subjects systematically overlooked permitted subtractive moves), software (codebase accretion needing dedicated refactoring sprints), law and regulation (the monotonic growth of statutes, repeal far costlier than enactment), bureaucratic process (accreting forms and sign-offs), product design (feature creep), curriculum design, and language and urban accretion. What holds the structural-abstraction component at the middle is that the underlying mechanism is human-cognitive: the pattern is a property of how agents search the space of transformations, and the prime's remedies (prompt explicitly for subtractive options, make removal as cheap and legible as addition) presuppose a human (or human-organizational) agent doing the searching. The biological boundary case — vestigial structures, genomic accretion — is one the prime extends to, but it has its own selective logic with no human agent, and the entry flags that reaching it requires changing what "search distribution" means (re-weighting a fitness landscape, not a prompt), so the structure touches a non-cognitive substrate only by stretching the term. Transfer evidence is moderate: the asymmetric-search diagnostic and the subtraction-prompting remedy carry across cognition, software, law, and design, but the mechanism stays anchored to agents transforming systems. Wide reach across agent-driven substrates, capped by the cognitive anchoring, fixes the composite at 3.

  • Composite substrate independence — 3 / 5
  • Domain breadth — 4 / 5
  • Structural abstraction — 3 / 5
  • Transfer evidence — 3 / 5

Relationships to Other Primes

One-hop neighborhood: parents above, mutual partners to the right, children below.Additive Biassubsumption: BiasBias

Parents (1) — more general patterns this builds on

  • Additive Bias is a kind of Bias

    Additive bias is a specific member of the bias family: a direction-asymmetric search distribution over transformations that over-samples 'add' and under-visits 'remove'. The 0.9991 sim to bias is genus-to-species — bias is the PARENT, NOT a reparent target.

Path to root: Additive Bias → Bias

Neighborhood in Abstraction Space

Additive Bias sits among the more crowded primes in the catalog (34th percentile for distinctiveness): several abstractions describe nearly the same structure, so a description that fits it will tend to fit its neighbors too — transporting it usually means disambiguating within this family rather than landing on it exactly.

Family — Unclustered & Miscellaneous (91 primes)

Nearest neighbors

Computed from structural-signature embeddings · 2026-06-14

Not to Be Confused With

The nearest existing prime by embedding — at a similarity near unity — is bias itself, and the relationship is strictly genus-to-species. Bias names the umbrella structure of any systematic deviation from a norm or true value; it makes no commitment to which deviation. Additive bias is one specific member of that family: the systematic skew, in an agent's distribution over candidate transformations of a system, toward adding a component rather than removing one. The near-perfect embedding similarity reflects that additive bias is literally "a bias," sharing all the umbrella's vocabulary, but the value of naming it separately is entirely in its specific roles — an improvement task, a transformation-search distribution, an additive overweight, a subtractive under-visit, an accretion trajectory, and a baseline-shift invariant — none of which the generic concept carries. To diagnose a problem merely as "bias" is to stop at the genus; additive bias supplies the precise mechanism (under-sampling the remove region) and the precise remedy (install machinery that prompts or weights subtraction). A practitioner who rests at "this is a bias" knows a deviation exists but not that the fix is to re-weight the search distribution toward removal.

A second genuine confusion is with objective_creep (and the closely related scope_creep), because all three describe systems growing past where they should be. But objective creep is about the unauthorized expansion of an actor's goal-set — adjacent objectives accreting beyond what the plan was sized for — and scope creep is its project-level instance, requirements accreting within a deliverable. Additive bias is the deeper, more general cognitive fact that produces such growth: when improving any system, agents reach first for addition and under-visit removal. The relationship is cause-to-symptom across substrates: additive bias is one of the generative engines behind both objective creep and scope creep, but it operates equally on codebases, statutes, curricula, and genomes where no "objective" or "project scope" is in play. The distinction matters because the remedies attach at different levels. Objective creep is managed by authorization parity and subtraction-coupled admission at the level of goals; additive bias is managed by re-weighting the search distribution itself — subtractive-only sessions, removal targets, default-off. A practitioner who frames a bloated codebase as scope creep will reach for requirements governance when the deeper issue is that the team's improvement search never visits the delete region at all.

A third confusion worth drawing is with parsimony_occam_s_razor. The two are intimately related but sit on opposite sides of the is/ought line. Parsimony is a normative principle — a prescription that, other things equal, simpler systems and explanations are to be preferred. Additive bias is a descriptive fact — a claim about how agents actually behave, namely that they systematically fail to reach for the simpler, subtractive option even when it is better and permitted. Parsimony tells you what you should do; additive bias explains why you reliably don't. The relationship is that additive bias is precisely the cognitive obstacle that makes parsimony hard to practice: the razor prescribes cutting, and the bias is why the cut is rarely even considered. The distinction matters because invoking parsimony as a value does little to counteract additive bias — exhorting people to prefer simplicity does not re-weight a search distribution that under-samples subtraction. Only structural machinery (prompts, quotas, scheduled removal acts) closes the gap between the parsimony norm and the additive-biased behavior. A practitioner who treats "value simplicity" as a sufficient remedy mistakes a norm for a debiasing mechanism.

For a practitioner, the distinctions sort by level and modality. If the concern is any systematic deviation in the abstract, it is bias; if a goal-set or project scope is expanding past its authorization, it is objective_creep / scope_creep (govern at the goal level); if the question is the normative preference for simplicity, it is parsimony_occam_s_razor; and if the issue is an agent's transformation-search distribution reliably over-sampling additions and under-sampling removals, it is additive bias — the only one whose remedy is to install machinery that explicitly weights or prompts the remove-component region the default search will not visit.

Solution Archetypes

No catalogued solution archetypes reference this prime yet.