Agonistic behaviour, made inspectable
Before animals clash, they signal, sample, assess and sometimes withdraw. Enter a contest where the strongest-looking animal is not always the strongest, and a sensible decision can still produce an unlucky outcome.
AI-generated natural-history illustration for species context. It is not a photograph, observation or data source.
01 · Contest lab
Commit a prediction, reveal one cue at a time, then advise Contestant A as an observer. Your advice is recorded, not enacted. The seeded animals follow the disclosed teaching rule, allowing decision quality to be separated from a lucky or unlucky outcome.
Resource: breeding access
Two adult stags meet near a harem. Both know their own condition better than they know their rival.
02 · Matched cohort lab
Change one factor, keep the random case stream matched, and compare 120 baseline contests with 120 treatment contests. Claims remain about this teaching model.
Design the comparison
Choose a directional prediction before running the cohort.
All settings are relative teaching-model indices. They are not field-calibrated probabilities or energy units.
MODEL OUTPUT · matched comparison
Commit a prediction and run the paired cohort. A readable table will appear with the chart.
| Measure | Baseline | Treatment | Change |
|---|
03 · Species transfer
Each context changes the resource, contest sequence, cue modality, reliability schedule and evidence boundary. All five currently use one shared teaching engine, so transfer is possible without pretending these are independently fitted species models.
ILLUSTRATION
Original artwork establishes biological context. The simulator’s measured cue readouts, raw records and repeated distributions are the evidence learners use.
Acoustic and visual assessment. Roaring rate and vocal formants can carry different information, followed by parallel walking and possible antler contact.
Current mapping: staged cue reliability and a high-cost final stage inside the shared engine.
A large regenerated claw can resemble an original claw while producing lower force. The context makes imperfect honesty visible without implying conscious deception.
Current mapping: the shared engine plus a low-frequency regenerated-claw cue bias.
Body size, eyespan, ownership and local context can covary with outcomes. The resource-value index represents context broadly, but female presence is not a separate control.
Current mapping: size cue, ownership and subjective value inside the shared engine.
Meral displays may not settle size-matched contests. Ritualised strikes to an armoured telson can reveal persistence without ordinary lethal impact.
Current mapping: N. bredini stages and cue schedules inside the shared engine, not a fitted stomatopod model.
Attackers rap a defender’s shell, and the shell’s value differs between individuals. Oxygen and stamina are important in the source biology but are represented only within composite RHP and cost, not as separate controls.
Current mapping: shell-value cue bias and cumulative stage costs inside the shared engine.
Which visible pattern would distinguish assessment of an opponent from a contestant simply reaching its own cost threshold?
The answer needs experimental comparisons, not a dramatic animation.
04 · Learner guide
The interface separates what exists in the case, what can be observed, what the model infers and what one outcome happens to show.
| Screen element | Type | What it represents | Use it to claim |
|---|---|---|---|
| Cinematic habitat | ILLUSTRATION | Species, resource and contest stage. | Nothing quantitative. |
| Animal anatomy artwork | ILLUSTRATION | Recognisable morphology, fixed across contest stages and not cue-scaled. | Nothing quantitative. |
| Signal overlays and readouts | MODEL STATE | Which signal is active and its seeded measured value. | Which information was available in this model case. |
| Inference bars and ≈ ±1 SD | MODEL OUTPUT | A field observer’s teaching-model estimate and approximate propagated standard deviation. | How uncertainty changed within the shared teaching rule. |
| Cohort distributions | MODEL OUTPUT | Repeated results under matched seeds. | A direction or difference inside this teaching model. |
| Species study note | FIELD EVIDENCE | A bounded result from one named study. | Only the species, setting and variable that study supports. |
State a directional pattern, not “it had an effect”. Example: higher modelled resource value increased escalation across the matched cohort.
Use the sample size and a numerical contrast. Do not use the illustration or a single dramatic outcome as evidence.
Link perceived winning probability, resource value and expected future cost to the decision threshold.
Name stochastic overlap, an omitted variable or another assessment rule that could produce a similar pattern.
No. It includes display, threat, submission, avoidance and withdrawal as well as physical aggression. A contest can be resolved without contact.
Strength shifts the odds. Motivation, ownership, resource value, condition, information error and chance can all affect one outcome.
No. The equation is our inspectable teaching representation of a conditional decision. It does not claim conscious arithmetic or purpose.
05 · Teacher guide
The core move is from an appealing animal encounter to a controlled, repeated comparison and a bounded explanation.
Learners will distinguish agonistic behaviour from aggression, separate RHP from resource value, and use repeated model evidence to explain withdrawal or escalation under uncertainty.
Learners should know variation, competition, costs and benefits, and natural selection. They do not need prior knowledge of any exemplar species or probability equations.
Formative preparation for resource-based AS91603 reasoning. The tool does not grade the standard, establish a real animal mechanism or generate an assessment answer.
| Element | Developing | Secure | Strong |
|---|---|---|---|
| Claim | Names an outcome. | States a direction. | States a bounded, conditional direction. |
| Evidence | Uses one contest. | Uses a cohort measure. | Uses sample size, numerical contrast and matched design. |
| Reasoning | Repeats “costs and benefits”. | Links cue to decision. | Links noisy assessment, value and future costs. |
| Uncertainty | No limit. | Names model simplification. | Names an alternative rule or field test. |
Welfare boundary: learners must not provoke, stage or approach real animal contests. Use published observations, remote footage or ethically collected datasets.
06 · Behind the model
The mathematics is a teaching model, not a claim that animals perform arithmetic. Species evidence anchors cue direction and contest structure. Classroom weights remain disclosed assumptions.
Implementation disclosure: every species context currently uses this same decision engine. Species selection changes the staged behaviour, cue labels, stage reliability, default resource and injury indices, logistic slope, and two bounded cue modifiers: regenerated-claw unreliability for fiddler crabs and shell-value bias for hermit crabs. The uncertainty propagation assumes independent stage-cue errors; clipping at 0 and 100 makes the displayed SD approximate. These settings are teaching assumptions, not fitted population parameters.
| Preset | Field evidence | Teaching assumption | Do not conclude | Possible field test |
|---|---|---|---|---|
| Red deer | Roaring rate and formants can inform rivals; visual escalation follows. | Three cues update one shared estimate. | Pitch alone reveals strength, or all stags use one rule. | Playback of controlled formants and rate with matched rivals. |
| Fiddler crab | Regenerated claws can overstate force while resembling original claws. | A small seeded share of cases carries a bluff penalty. | Crabs consciously lie, or every large claw is reliable. | Compare initiation and escalated outcomes by claw type and residency. |
| Stalk-eyed fly | Body size, eyespan and ownership covary with outcomes; female presence can alter local value. | The shared engine maps alignment to a noisy body-size estimate; female presence is not a separate state variable. | Relative eyespan is an independent strength gauge or all diopsids match. | Factorial manipulation of body size, ownership and lek value. |
| Mantis shrimp | Size-matched contests often escalate to repeated telson strikes; evidence supports mutual assessment in tested N. bredini. | Stage cues update opponent estimates and persistence shifts withdrawal. | Meral spread reliably predicts strike force, or all stomatopods share this rule. | Separate size mismatch, display, strike number, role and residency. |
| Hermit crab | Shell rapping and oxygen availability affect attacker persistence. | The shared engine maps performance cues and shell value to a threshold; oxygen is omitted as a separate variable. | Duration alone proves cumulative or mutual assessment. | Manipulate oxygen, shell value and opponent action independently. |
07 · Product and marketing lens
The strongest proposition is not “watch animals fight”. It is “learn to make a defensible decision from incomplete biological evidence, then test it across cases”.
Decision quality is separated from luck. Learners can make a biologically sensible recommendation and still see an upset, which creates a natural reason to run repeated cases.
Cinematic habitat, fixed stylised anatomy, stage-specific signalling, a calculated model uncertainty measure and a forensic reveal. The interface moves from atmosphere to evidence without confusing the two.
One case captures attention. Matched cohorts support graph and table interpretation. Cross-species transfer prevents memorising a deer-only story.
“Explore how animal contests can be resolved through signals, assessment, withdrawal and escalation, then test explanations against replayable data.”
Species-accurate prediction, validated learning gains, automatic NCEA grading, a universal contest equation or a proven mechanism in every species.
After one 50-minute lesson, can learners distinguish RHP, resource value and outcome luck, then transfer the causal model to an unfamiliar species dataset?
| Graphic | What changes it | What it communicates | Marketing role | Evidence status |
|---|---|---|---|---|
| Habitat artwork | Species selection | Organism, scale, resource and mood | Immediate curiosity and recognisable identity | Illustration only |
| Signal overlay and readout | Contest stage and seeded cue | Information is released over time, while anatomy remains fixed | The signature “read the rival” interaction | Model state |
| Inference bars | Revealed noisy cues | Evidence changes confidence without certainty | Visible intellectual progress | Model output |
| Decision/outcome split | User recommendation and stochastic result | Sensible decisions can have unlucky outcomes | Memorable learning twist | Model output |
| Matched cohort bars | One controlled variable | Distribution-level evidence and trade-offs | Teacher credibility and exam preparation | Model output |
| Study ledger | Active species source pack | Empirical, simplified, omitted and uncertain | Trust and scientific transparency | Field evidence plus disclosure |
The two raster artworks in this prototype were generated for this project and are explicitly labelled as illustration. Production should retain prompts and provenance. Any future field photograph, audio, video or spectrogram needs file-level creator, licence, crop and modification records.
Do not hotlink images or imply that generated art depicts a measured trial. For taonga species, seek suitable authority, provenance and permission before use.
08 · Source ledger
Sources anchor directions and boundaries. The simulator remains an explicit teaching model unless and until parameters are fitted and independently validated for a defined field population.
NZQA Standard 91603 and the 2026 Level 3 Biology assessment specification, which names agonistic behaviour and graphical and tabulated data.
Maynard Smith and Price, 1973; Parker, 1974; Taylor and Elwood, 2003; Arnott and Elwood, 2009.
Backwell et al., 2000; Morrell et al., cumulative-assessment analysis; Lailvaux, Reaney and Backwell, 2009. Sources use the former genus name Uca; this prototype uses Austruca mjoebergi.
Green and Patek, 2015, telson sparring; Green and Patek, mutual-assessment analysis. A 2026 study of related Gonodactylus childi found role and individual variation, so the rule is not generalised across mantis shrimp.
Briffa and Elwood, 2000, oxygen availability and alternative assessment rules in shell fights.
IES Organizing Instruction and Study; Roediger and Karpicke, 2006; Butler, 2010.