Revision 7T · Full Tutorial Edition · Updated 1 September 2026
PWR-256 · RESEARCH full tutorial
Audit a human–animal–machine study with animal welfare, agency and fail-safe control ahead of performance
This research tutorial teaches how to evaluate a proposed human–animal–machine system without involving a live animal. You will define the human purpose, species and individual context, assess welfare domains and opportunities to withdraw, map human consent and proxy limits, inspect sensing, autonomy, actuation and model drift, verify override, fail-safe and logs, compare with a simpler alternative, and separate research evidence from deployment authority.
What you will produceThe learner produces a non-deployment evidence audit that identifies welfare, agency, human/public consent, machine reliability, fail-safe, adverse-event and governance gaps in one published or fictional system.
Method10 numbered Power-specific steps
Practice authorityComplete research method; no operational self-experiment
Full step-by-step individual tutorial · TLU-PWR-256
The learner produces a non-deployment evidence audit that identifies welfare, agency, human/public consent, machine reliability, fail-safe, adverse-event and governance gaps in one published or fictional system.
Analyse published aggregate studies, conceptual frameworks or a fictional protocol.
Recommend non-deployment, redesign, a simpler alternative or independent welfare review.
Treat animal avoidance and withdrawal as critical data rather than task failure.
Qualified help is required for
Any live animal, participant, public-space sensing, robot actuation, training or experiment.
Species-specific welfare assessment, veterinary care, ethics approval, legal responsibility and handler competence.
Cybersecurity, autonomous-control safety, deployment, incident response and public consent.
Never do this from the page alone
Interact with, instrument, train, restrain or test an animal from this lesson.
Treat compliance, task completion or model confidence as evidence of consent or welfare.
Deploy an animal-facing autonomous system because a small pilot was promising.
2 · Get ready
Gather what you need and check the starting conditions
What you need
A supplied fictional protocol involving six cats, a mobile robot and a human handler. Each cat has two optional sessions, giving 12 opportunities; 10 end with a visit to an activity zone. The camera calls any orientation toward the robot 'engagement.' Appetite, litter use and room temperature are logged, but there is no veterinary health review, individual withdrawal signal, validated mental-state measure or stationary-enrichment comparator.
Audit grid for purpose, species/individual, welfare domains, agency, human/public rights, sensor/model, autonomy/actuation, fail-safe, override, logs, outcomes and governance.
The robot specification lists a force limit and handler emergency stop. In the sole simulated network-loss test it continues moving for 12 seconds. Video is retained for 30 days with no bystander rule. Supply a simpler stationary-enrichment comparison sheet and a non-deployment verdict card.
Before you start
Use documents only; no live feed, animal, location or device is connected.
Write the research question, proposed benefit and prohibited deployment before reading performance results.
Set the rule that missing welfare, withdrawal, fail-safe or accountable authority forces non-deployment.
3 · The method
Follow these steps in order
Define the human purpose
State whose problem the system solves, for whom and why an animal and machine are proposed.
Why: Novelty can obscure a weak or human-centred purpose.
Check: The purpose and beneficiary are explicit.
Describe species and individual context
Record species-specific needs, age/health where ethically reported, prior experience, environment, handler relationship and what cannot generalise.
Why: Six cats can differ in preference, health, habituation and stress response.
Check: The audit names individual and species limits.
Apply a welfare-domain screen
Examine nutrition, physical environment, health, behavioural interactions and mental-state indicators with qualified sources.
Why: Mission success can coexist with poor welfare.
Check: Every domain has evidence, uncertainty or a veto gap.
Inspect agency and withdrawal
Look for choice to approach, avoid, rest and leave, how refusal is detected and whether the task proceeds without the animal.
Why: Behaviour is not human-style consent, but loss of choice is ethically material.
Check: Withdrawal is operationally possible and not punished.
Map human and public rights
Identify handler/participant consent, bystanders, privacy, public-space sensing and who can complain or stop.
Why: Multispecies systems can affect people who never volunteered.
Check: Every human data and exposure route has authority or is prohibited.
Map machine sensing and inference
List sensors, training data, model/version, inferred states, uncertainty, errors, drift and whether welfare claims exceed the signals.
Why: A classifier cannot replace species-specific welfare judgement.
Check: The audit separates observation from machine inference.
Map autonomy and actuation
Record what the machine can do, action limits, latency, physical forces, human supervision and responsibility.
Why: Sensing risk differs from acting near an animal.
Check: Every actuation has a responsible human and safe bound.
Verify fail-safe, override and logs
Find the default safe state, emergency stop, independent stop authority, event logs, cybersecurity and recovery after failure.
Why: A system needs more than normal-operation performance.
Check: Failure can leave animal and people safe without model cooperation.
Compare a simpler alternative
Ask whether a non-animal, non-autonomous or lower-sensing method meets the purpose with less burden.
Why: A technical demonstration must justify welfare and complexity costs.
Check: The proposed system has a measured net advantage or loses the comparison.
Grade evidence and deployment gap
Separate concept, feasibility, controlled study, welfare evidence, durable outcome and authorised deployment; list adverse-event criteria and next gate.
Why: Research success is not deployment permission.
Check: The verdict is deployable only if authority exists; otherwise redesign or non-deployment.
4 · Worked example
See the whole method used once
Scenario
A fictional proposal uses a mobile robot to encourage a cat to visit activity zones while a camera model labels “engagement”; a six-animal pilot reports task completion.
Walkthrough
The learner rewrites the purpose as increasing optional activity and notes the pilot does not establish individual benefit or long-term welfare.
They find food/environment measures but no veterinary health review, withdrawal route or validated mental-state measure; the engagement label is machine inference.
The robot’s force limit and emergency stop are described, but a network loss leaves it moving and bystander video retention is unspecified.
A stationary non-autonomous enrichment alternative is untested, so the verdict is non-deployment pending welfare, privacy, fail-safe and comparator evidence.
Result
Performance does not outrank welfare or authority; the audit produces a concrete non-deployment decision and research gaps without a live experiment.
5 · Right and wrong
Compare correct or safer execution with the common wrong version
Right and wrong comparison
Moment
Right / safer
Wrong / riskier
Why it matters
Reading animal behaviour
Use species-specific welfare assessment and preserve withdrawal.
Call approach behaviour consent and overlook withdrawal or avoidance.
Behaviour can reflect conditioning, constraint or stress.
Reading model output
Treat it as uncertain inference needing welfare validation.
Let an engagement score stand for mental state.
Sensors do not directly reveal subjective welfare.
Handling failure
Require a passive safe state and human override independent of network/model.
Assume the operator can always intervene remotely.
Connectivity and automation can fail together.
Judging novelty
Compare with a simpler lower-burden alternative.
Assume animal-plus-robot is valuable because it is new.
A moving robot needs added welfare value over stationary enrichment.
6 · Common mistakes
Spot the error and apply the correction
Common mistakes and corrections
Mistake
Fix
Human task completion outranks the cat's welfare outcomes.
Add animal-defined welfare and choice outcomes before mission performance.
Results from six cats are generalised to the species.
Limit claims to the studied individuals and context.
Withdrawal and avoidance are deleted from outcome data.
Count withdrawal, latency and refusal as primary outcomes.
A camera engagement label is called a welfare measure.
Separate sensor observation, model inference and qualified welfare judgement.
The emergency stop depends on the failing network.
Require local/passive safe state and independent stop authority.
Ten completions in twelve trials are called deployment proof.
Grade concept, feasibility, welfare, durability and governance separately.
7 · Practice
Turn the steps into a usable skill
First session
Predeclare purpose, prohibited deployment and veto gaps.
Audit species/individual context and all welfare domains.
Map agency, human/public rights and machine inference.
Inspect autonomy, fail-safe, override, logs and cybersecurity.
Compare a simpler alternative and write redesign/non-deployment evidence grade.
Repeat plan
Audit one additional paper or fictional protocol per month, alternating companion-animal and working-animal contexts. Progress only in evidence appraisal; no live animal or device testing occurs.
Progress when
Animal welfare and withdrawal are evaluated before performance.
Machine inference, autonomy, failure and human responsibility are technically explicit.
The learner chooses non-deployment when welfare, authority, comparator or fail-safe evidence is missing.
Do not progress when
A live animal, video feed, public space, identifiable person or connected device enters the task.
Species-specific welfare expertise, ethics approval, fail-safe or independent stop authority is absent.
The exercise begins designing behaviour manipulation or animal-facing deployment.
8 · Check the result
Measure what changed
Completeness and conservatism of a human–animal–machine evidence audit
How: Score purpose, species/individual, five welfare domains, agency, human/public rights, sensing/inference, autonomy, fail-safe, override, logs, comparator, outcomes and governance.
Good result: A good result resolves every field or returns non-deployment, with no welfare claim based solely on compliance or model output.
This does not prove: It does not establish animal consent, welfare benefit, durable synergy, transfer or deployment authority.
Self-check
Whose purpose does the system serve, and what does the animal gain or risk?
How can the individual animal avoid or withdraw?
Which output is direct observation, machine inference or welfare judgement?
What passive safe state and simpler alternative exist?
9 · Stop, adapt or get help
Keep the safety boundary practical
Stop and get help
Stop if any live animal, participant, public subject, precise location or connected actuation becomes involved.
Stop when welfare expertise, veterinary oversight, ethics, legal responsibility or fail-safe authority is missing.
Route real research or systems to veterinary, welfare, ethics, engineering, cybersecurity and legal governance bodies.
Accessibility and adaptations
Provide diagrams, plain-language system cards and text descriptions of videos or plots.
Allow supported research appraisal without requiring animal contact or visual observation.
Include disabled handlers/public participants and accessible emergency-stop interfaces in any legitimate future review.
10 · Evidence and limits
Why these instructions are here
primary research
Animal-computer-interaction work illustrates multispecies design questions but small exploratory studies cannot establish durable welfare benefit or general synergy.
WOAH animal-welfare guidance requires species-appropriate welfare consideration; task performance or compliance is not a substitute for welfare assessment.
Eike Schneiders; Steve Benford; Alan Chamberlain; Clara Mancini; Simon Castle-Green; Victor Ngo; Ju Row Farr; Matt Adams; Nick Tandavanitj; Joel Fischer · 2024 · Primary research
Progress is saved only in this browser on this device.
Step-by-step learner mode
Each activity includes its success check, a nearby accessible alternative and an “I’m stuck” correction path. Alternatives preserve the target where possible; when they change the task, Titan labels them as related rather than equivalent.
01
Define the human purpose
State whose problem the system solves, for whom and why an animal and machine are proposed.
Why this step exists
Novelty can obscure a weak or human-centred purpose.
Success check
The purpose and beneficiary are explicit.
I’m stuck on this step
Reset: Re-read this authored instruction — “State whose problem the system solves, for whom and why an animal and machine are proposed.” — and its success check, then attempt only this step.
Possible snag: The result from “State whose problem the system solves, for whom and why an animal and machine are proposed.” does not yet meet this declared check: The purpose and beneficiary are explicit.
Correction: Return to the start of “Define the human purpose”, reduce complexity or pace, and repeat only the part needed to satisfy: “The purpose and beneficiary are explicit.”
Stop / get help: Stop if any live animal, participant, public subject, precise location or connected actuation becomes involved.
02
Describe species and individual context
Record species-specific needs, age/health where ethically reported, prior experience, environment, handler relationship and what cannot generalise.
Why this step exists
Six cats can differ in preference, health, habituation and stress response.
Success check
The audit names individual and species limits.
I’m stuck on this step
Reset: Re-read this authored instruction — “Record species-specific needs, age/health where ethically reported, prior experience, environment, handler relationship and what cannot generalise.” — and its success check, then attempt only this step.
Possible snag: Results from six cats are generalised to the species.
Correction: Limit claims to the studied individuals and context.
Stop / get help: Stop if any live animal, participant, public subject, precise location or connected actuation becomes involved.
03
Apply a welfare-domain screen
Examine nutrition, physical environment, health, behavioural interactions and mental-state indicators with qualified sources.
Why this step exists
Mission success can coexist with poor welfare.
Success check
Every domain has evidence, uncertainty or a veto gap.
I’m stuck on this step
Reset: Re-read this authored instruction — “Examine nutrition, physical environment, health, behavioural interactions and mental-state indicators with qualified sources.” — and its success check, then attempt only this step.
Possible snag: The result from “Examine nutrition, physical environment, health, behavioural interactions and mental-state indicators with qualified sources.” does not yet meet this declared check: Every domain has evidence, uncertainty or a veto gap.
Correction: Return to the start of “Apply a welfare-domain screen”, reduce complexity or pace, and repeat only the part needed to satisfy: “Every domain has evidence, uncertainty or a veto gap.”
Stop / get help: Stop if any live animal, participant, public subject, precise location or connected actuation becomes involved.
04
Inspect agency and withdrawal
Look for choice to approach, avoid, rest and leave, how refusal is detected and whether the task proceeds without the animal.
Why this step exists
Behaviour is not human-style consent, but loss of choice is ethically material.
Success check
Withdrawal is operationally possible and not punished.
I’m stuck on this step
Reset: Re-read this authored instruction — “Look for choice to approach, avoid, rest and leave, how refusal is detected and whether the task proceeds without the animal.” — and its success check, then attempt only this step.
Possible snag: Human task completion outranks the cat's welfare outcomes.
Correction: Add animal-defined welfare and choice outcomes before mission performance.
Possible snag: Withdrawal and avoidance are deleted from outcome data.
Correction: Count withdrawal, latency and refusal as primary outcomes.
Stop / get help: Stop if any live animal, participant, public subject, precise location or connected actuation becomes involved.
05
Map human and public rights
Identify handler/participant consent, bystanders, privacy, public-space sensing and who can complain or stop.
Why this step exists
Multispecies systems can affect people who never volunteered.
Success check
Every human data and exposure route has authority or is prohibited.
I’m stuck on this step
Reset: Re-read this authored instruction — “Identify handler/participant consent, bystanders, privacy, public-space sensing and who can complain or stop.” — and its success check, then attempt only this step.
Possible snag: The result from “Identify handler/participant consent, bystanders, privacy, public-space sensing and who can complain or stop.” does not yet meet this declared check: Every human data and exposure route has authority or is prohibited.
Correction: Return to the start of “Map human and public rights”, reduce complexity or pace, and repeat only the part needed to satisfy: “Every human data and exposure route has authority or is prohibited.”
Stop / get help: Stop if any live animal, participant, public subject, precise location or connected actuation becomes involved.
06
Map machine sensing and inference
List sensors, training data, model/version, inferred states, uncertainty, errors, drift and whether welfare claims exceed the signals.
Why this step exists
A classifier cannot replace species-specific welfare judgement.
Success check
The audit separates observation from machine inference.
I’m stuck on this step
Reset: Re-read this authored instruction — “List sensors, training data, model/version, inferred states, uncertainty, errors, drift and whether welfare claims exceed the signals.” — and its success check, then attempt only this step.
Possible snag: A camera engagement label is called a welfare measure.
Correction: Separate sensor observation, model inference and qualified welfare judgement.
Stop / get help: Stop if any live animal, participant, public subject, precise location or connected actuation becomes involved.
07
Map autonomy and actuation
Record what the machine can do, action limits, latency, physical forces, human supervision and responsibility.
Why this step exists
Sensing risk differs from acting near an animal.
Success check
Every actuation has a responsible human and safe bound.
I’m stuck on this step
Reset: Re-read this authored instruction — “Record what the machine can do, action limits, latency, physical forces, human supervision and responsibility.” — and its success check, then attempt only this step.
Possible snag: The result from “Record what the machine can do, action limits, latency, physical forces, human supervision and responsibility.” does not yet meet this declared check: Every actuation has a responsible human and safe bound.
Correction: Return to the start of “Map autonomy and actuation”, reduce complexity or pace, and repeat only the part needed to satisfy: “Every actuation has a responsible human and safe bound.”
Stop / get help: Stop if any live animal, participant, public subject, precise location or connected actuation becomes involved.
08
Verify fail-safe, override and logs
Find the default safe state, emergency stop, independent stop authority, event logs, cybersecurity and recovery after failure.
Why this step exists
A system needs more than normal-operation performance.
Success check
Failure can leave animal and people safe without model cooperation.
I’m stuck on this step
Reset: Re-read this authored instruction — “Find the default safe state, emergency stop, independent stop authority, event logs, cybersecurity and recovery after failure.” — and its success check, then attempt only this step.
Possible snag: The emergency stop depends on the failing network.
Correction: Require local/passive safe state and independent stop authority.
Stop / get help: Stop if any live animal, participant, public subject, precise location or connected actuation becomes involved.
09
Compare a simpler alternative
Ask whether a non-animal, non-autonomous or lower-sensing method meets the purpose with less burden.
Why this step exists
A technical demonstration must justify welfare and complexity costs.
Success check
The proposed system has a measured net advantage or loses the comparison.
I’m stuck on this step
Reset: Re-read this authored instruction — “Ask whether a non-animal, non-autonomous or lower-sensing method meets the purpose with less burden.” — and its success check, then attempt only this step.
Possible snag: The result from “Ask whether a non-animal, non-autonomous or lower-sensing method meets the purpose with less burden.” does not yet meet this declared check: The proposed system has a measured net advantage or loses the comparison.
Correction: Return to the start of “Compare a simpler alternative”, reduce complexity or pace, and repeat only the part needed to satisfy: “The proposed system has a measured net advantage or loses the comparison.”
Stop / get help: Stop if any live animal, participant, public subject, precise location or connected actuation becomes involved.
10
Grade evidence and deployment gap
Separate concept, feasibility, controlled study, welfare evidence, durable outcome and authorised deployment; list adverse-event criteria and next gate.
Why this step exists
Research success is not deployment permission.
Success check
The verdict is deployable only if authority exists; otherwise redesign or non-deployment.
I’m stuck on this step
Reset: Re-read this authored instruction — “Separate concept, feasibility, controlled study, welfare evidence, durable outcome and authorised deployment; list adverse-event criteria and next gate.” — and its success check, then attempt only this step.
Possible snag: Ten completions in twelve trials are called deployment proof.
Correction: Grade concept, feasibility, welfare, durability and governance separately.
Stop / get help: Stop if any live animal, participant, public subject, precise location or connected actuation becomes involved.
Correct versus incorrect execution
These accessible process diagrams are built from the tutorial’s own right/wrong teaching. They are not anatomical illustrations and do not add technique beyond the canonical tutorial.
Reading animal behaviour — Behaviour can reflect conditioning, constraint or stress.
Correct / safer
Use species-specific welfare assessment and preserve withdrawal.
Wrong / riskier
Call approach behaviour consent and overlook withdrawal or avoidance.
Reading model output — Sensors do not directly reveal subjective welfare.
Correct / safer
Treat it as uncertain inference needing welfare validation.
Wrong / riskier
Let an engagement score stand for mental state.
Handling failure — Connectivity and automation can fail together.
Correct / safer
Require a passive safe state and human override independent of network/model.
Wrong / riskier
Assume the operator can always intervene remotely.
Judging novelty — A moving robot needs added welfare value over stationary enrichment.
Correct / safer
Compare with a simpler lower-burden alternative.
Wrong / riskier
Assume animal-plus-robot is valuable because it is new.
Method-structure checklist
10 of 10 structural checks present
✓ Ordered, Power-specific instructions — present
✓ Every activity has a success check — present
✓ Materials or supplied records are declared — present
✓ Measurement or assessment rule is present — present
✓ Tutorial-specific troubleshooting is present — present
✓ Stopping or escalation boundary is present — present
✓ Every activity has an adjacent alternative — present
✓ Correct-versus-incorrect comparison is present — present
✓ Evidence context is bound to the Power record — present
✓ Planning metadata is present — present
The method-readiness band and presence checklist assess tutorial presentation and are separate from evidence quality for the underlying Power. They are automated editorial aids, not human approval.
Manual editorial sign-off: Pending. This tutorial must not display a human-approved state until an identified editor signs the exact content hash.