Project Epic

1 The project question and purpose

Project Epic asks how far a dog can learn. It does not stop at making a fixed list of behaviors more reliable. It keeps increasing the difficulty, from concrete referents to abstract rules. In the process, the human studies learning, cognition, language, communication, motivation, and emotion—and trains their own behavior. The final subject is not a dog's list of tricks, but how far one dog and one human can grow together.

Raising a dog well is more than feeding, walking, and maintaining health. The human has to understand how the dog receives signals, connects actions with outcomes, reads unfamiliar situations, and makes choices. Looking when called, waiting, and returning on recall are all problems of learning relations among signals, behavior, and consequences. Living with a dog therefore becomes a practical study of learning and cognition, reinforcement, motivation, emotion, and environment.

Knowing the principles is not enough. Late feedback can reinforce the behavior after the intended one. A rule allowed today and blocked tomorrow becomes ambiguous. A signal changes when the person acts from irritation or haste. The human must examine what they actually taught, whether the criterion stayed consistent, and whether feedback arrived at the right moment. Learning to observe before reacting, maintain a rule, and change one's own signal or setup before blaming the dog makes training a form of self-discipline.

Study subject

The learner is Epic, a first-generation Border Collie × Papillon cross often called a Border Papillon. Epic joined the family in Seoul in spring 2024. Project Epic is a longitudinal practice project following one dog–human team over several years. Its result can show that this team reached a level under stated conditions; a species-wide claim requires independent replication with other dogs and handlers.

2 Have we seen the dog's full learning capacity?

Begin with a thought experiment. Imagine aliens keeping humans as companion animals without a shared language. They feed them, house them, and teach only how not to cause trouble and how to follow a routine. A human raised in that environment would be unlikely to build a rocket, make a film, or study quantum mechanics. That outcome would not prove that humans lack those capacities. Their expression depends on language, cumulative knowledge, staged education, tools, peers, and time.

Dog education leaves a similar question. Training systems have become very good at toileting, walking, waiting, not barking, and performing the behaviors needed to live smoothly with people. In a human analogy, however, that may resemble refining kindergarten education indefinitely. Cases in which one concept becomes a prerequisite for the next and difficulty rises for years—from primary school toward graduate work—are comparatively rare. It is difficult to separate the dog's upper limit from the level at which people usually stop teaching.

This analogy does not claim that dogs can build rockets. Dogs and humans are different species with different senses, bodies, brains, and social evolution. It asks a narrower question: Can the current performance of an undereducated population establish the upper limit of its learning? Project Epic tries to answer through long-term teaching and transfer tests. Existing evidence for canine categories, relations, and transfer to unfamiliar stimuli is collected in Concept Training.

3 Working assumption — do not set the limit in advance

It is difficult to prove that “dogs cannot do this.” Failure can arise from a species-level cognitive constraint, missing prerequisites, a task that does not fit canine perception, weak motivation, emotion, pain, or fatigue. Brain imaging can show activity related to a task, but not the maximum performance a dog might reach after years of a different education.

If researchers studied a thousand people who had received only restricted education and concluded that humans could not understand complex mathematics or learn more than a hundred words, their data might reveal the curriculum rather than the species limit. Project Epic avoids the same error. Before treating failure as an endpoint, it checks prerequisite learning, task design, communication, environment, and duration.

Comparisons between people in the past and today illustrate the role of education. “Primitive human” is not one biological category, and real brain evolution occurred across the long history of hominins. Yet a child from an anatomically modern human population, raised today with modern language, education, and cumulative culture, would not be confined to that population's historical way of life. Film and quantum mechanics are achievements of individual brains operating on top of accumulated culture. This does not make dogs equivalent to humans. It shows that education stands between native structure and expressed capacity. The distinction is developed in Learning Principles and Methods.

4 What makes a dog happy — a hypothesis

Project Epic does not treat the indefinite continuation of a state in which every desire is satisfied and nothing is difficult as the completion of happiness. Once basic needs are met, a fuller life may include manageable challenges, occasions when things do not work, renewed attempts, and eventual accomplishment. Difficulty itself is not the value. The value lies in the process of taking on a challenge, recovering, and becoming able to do what was previously impossible.

This is not a scientifically established definition of canine happiness. It is a Project Epic hypothesis used to choose an educational direction. Nor does it justify suffering or deprivation as necessary for growth. Health, rest, safety, a predictable life, and a trustworthy relationship must come first. Without that foundation, difficulty may be a burden rather than a useful challenge.

Maslow's hierarchy is used only as an analogy. The working idea is that after physiological needs and safety are secured, a stable relationship and belonging can support autonomy, competence, exploration, learning, and more difficult goals. This does not apply a human hierarchy directly to dogs. Project Epic uses it as a provisional educational sequence: survival and safety → relationship and stability → autonomy and competence → challenge and growth, revising it as observations accumulate.

The boundary between evidence and hypothesis matters. The Five Domains model connects nutrition, physical environment, health, and behavioral interactions to mental state, and includes opportunities for agency in welfare assessment. Work on choice and control also describes how surmountable challenge may contribute to agency and competence. These sources do not directly prove Project Epic's broader claim that overcoming difficulty makes a dog happy. Mellor et al., 2020 · Webber et al., 2023

Happiness is therefore not inferred from one expression or task success. The project records voluntary re-approach when participation is optional, recovery time after failure, persistence of exploration as difficulty rises, and accumulating avoidance, freezing, or excessive arousal. Greater task mastery with worsening voluntary participation or recovery is not counted as support for the hypothesis.

5 Goals

Every judgment begins with a goal. What to teach, which method to use, and how far the human should intervene all depend on it. The goal is not a dog who has memorized many instructions. It is a dog who chooses and learns, recovers from difficulty, and can solve increasingly complex problems with a human.

1. Enjoy voluntary learning and growth. Do more than repeat behaviors for a visible reward. Explore a new problem and enjoy finding an answer with the human, so learning becomes an ordinary part of life.
2. Live a whole life containing joy, sadness, difficulty, and achievement. The aim is not to remove every frustration. Within safe limits, the dog can encounter difficulty, regain equilibrium, solve the problem, and experience mastery.
3. Make most decisions autonomously. Express wants, choose among available options, and experience outcomes. A very small set of cues directly tied to life and safety have a high reliability target: recall and emergency stop/down. Heel is a structured control task selected by this project, not a universally necessary safety cue for every companion dog.
4. Learn higher-order concepts. Progress from names to time, properties, relations, rules, and imitation, then test new combinations and transfer. Do not define the endpoint in advance, and do not claim beyond what has been demonstrated.

6 Direction

Project Epic is not trying merely to make a dog execute more commands mechanically. Food, toys, social interaction, and the reinforcing value of problem solving are not replacement stages in one developmental ladder. They are coexisting functions whose importance changes with the dog and context. The project aims to broaden the set of outcomes that can sustain learning and cooperation.

Why learn and cooperate?

1. Use external reinforcement to make contingencies clear. Food, toys, play, and access to desired activities provide information and motivation. They are not bait that must disappear once learning becomes advanced.
2. Let problem solving itself acquire value. Measure whether the dog voluntarily returns, persists, and explores when participation is optional. Do not simply equate those behaviors with the human experience of achievement.
3. Make the relationship a resource for cooperation. Social contact, praise, and shared play can reinforce behavior. Measure referencing the human, seeking help, and voluntary cooperation without visible food instead of treating “not wanting to disappoint the leader” as a fact about canine moral emotion.

Observation and interpretation remain separate here as well. In an awake-fMRI study of 15 dogs, 13 showed reward-region responses to a praise-predicting cue that were similar to or greater than their responses to a food-predicting cue, with substantial individual variation. This supports the possibility that human praise and interaction have reinforcing value for some dogs. It does not establish a moral motive such as “not wanting to disappoint.” Cook et al., 2016

How should learning develop?

The learning routes expand in the same direction. Classical and operant conditioning and luring first make relations among cues, actions, and outcomes clear. Shaping then lets the dog offer behavior and use feedback to approach an answer. On top of that foundation come imitation, spoken cues and names, concept and relational learning, and novel combinations of known words and rules.

This sequence does not discard an earlier method when a later one appears. Conditioning continues to operate in shaping and concept learning, while imitation and language also require experience and feedback. The aim is to reduce how often a human must construct every new action from the beginning with food, and increase how often the dog can use a demonstration, a word, or an existing concept to find the answer. Imitation, varied names, choices, relations, and problem solving are therefore introduced early in short and enjoyable forms.

7 Constraints and decision order

Constraints are lines the project does not cross in pursuit of a goal. Better performance accompanied by sustained fear, avoidance, injury, or relational deterioration is not success. When autonomy and control conflict, the following order applies.

• Do not treat one doctrine as the answer, but also do not treat all methods as equally safe or equally supported.
• Begin with the least burdensome method capable of reaching the goal.
• Do not use methods that can cause irreversible physical or psychological harm.
• Include opportunities to opt out, recovery after a session, and voluntary participation in the next session in the outcome.
• Here, pressure or aversive procedures means increasing behavior through removal of discomfort or decreasing it through an unpleasant consequence. Knowing the cue does not by itself make such a procedure safe or justified.

Project Epic does not exclude all pressure in advance. That is a project value judgment that differs from the current AVSAB recommendation for reward-based methods; it is not a claim that aversives are scientifically necessary. Before considering them, the project must examine necessity, cue fluency, prior use of management and less-aversive alternatives, stop and recovery criteria, cumulative welfare, and independent review.

Dogs at schools using more aversive methods showed more stress behavior, higher post-training cortisol, and more pessimistic judgment bias in Vieira de Castro et al., 2020. Casey et al., 2021 also reported associations between aversive-method use and behavior or welfare measures, but its owner-report observational design does not establish causation.

8 How to develop the capacity

Difficulty does not rise at random. The project builds both the material of learning and methods for learning. It moves from concrete objects to properties, relations, rules, and combinations, then asks whether the dog can apply what it knows to a new problem.

1. Teach concepts. Begin with directly referable objects, people, places, and body parts. Build actions and postures, size and direction, same and different, quantity, sequence, time, choice, permission, and prohibition. Teach components separately, combine them, and test unfamiliar examples.
2. Teach ways of learning. Use luring, capturing, shaping, and targeting, but also Do As I Do for a reusable imitation rule, concept training for extracting a rule across examples, and AIC (Augmentative Interspecies Communication) for modeling words and buttons in real events. The goal is to let known learning rules accelerate the next lesson.
3. Increase the precision of two-way communication. Expand not only human cues but the dog's ways to express wants, refusals, discomfort, requests for help, and choices. Study SATS Name & Explain for information about upcoming events alongside button-based communication.
4. Maintain motivation to learn. Make solving problems and learning with the human valuable in themselves. Adjust difficulty so the task is neither boring nor defeating; use choice, success, play, and access to real-life outcomes.
5. Teach emotional regulation. The dog needs to return from arousal to focus, frustration to another attempt, and fear to equilibrium before known skills can survive difficult contexts. Resilience is infrastructure for advanced learning.

What “learning beyond conditioning” means

Classical and operant conditioning are not elementary stages to be discarded. In classical conditioning, the learner acquires a predictive relation between events: when a marker reliably precedes reinforcement, the marker carries information about what comes next. In operant conditioning, the learner acquires a relation between its own action and an outcome: touching a box changes the probability of reinforcement. Both processes remain active in discrimination, concept learning, imitation, and language work. “Beyond” therefore describes a change in the target of learning—from one cue paired with one response toward categories, relations, reusable rules, symbols, and composition—not the absence of conditioning. Rescorla's account of Pavlovian conditioning as learning predictive relations is linked in Rescorla, 1988 and Learning Principles and Methods.

1. Discrimination and categories. Four dogs in Range et al. separated photographs of dogs from landscapes and extended the classification to novel photographs. That favors a category-relevant rule over memorizing individual images, but one visual task with four dogs cannot establish a universal abstract category capacity. Range et al., 2008
2. Relations and abstract concepts. In MTS (matching-to-sample), the learner selects an item matching a sample. Six dogs trained with 48 odors selected novel odor matches above chance. Eight Lagotto Romagnolos transferred a relative-size rule to unfamiliar shapes, with performance varying by shape. MTS success does not uniquely prove an abstract “same” relation; perceptual features, exclusion, and novelty strategies require separate controls. Lazarowski et al., 2021 · Byosiere et al., 2017 · same–different follow-up, 2025 · Concept Training
3. Social learning and imitation. Do As I Do first establishes a reusable rule: reproduce the action just demonstrated by the human. A dog can then acquire a novel action from a demonstration instead of shaping every movement. In a controlled comparison, the methods did not differ clearly for simple actions, while Do As I Do performed better for complex object actions and two-action sequences. Imitation is therefore not universally superior to conditioning; it is a distinct route that can reuse an acquired imitation rule in suitable tasks. Fugazza & Miklósi, 2014 · Do As I Do
4. Symbols, names, and composition. Chaser discriminated 1,022 object names and responded differently to noun–verb combinations, though this remains one intensively educated dog. Soundboard work found contextually appropriate responses to play- and outside-related words, but not the same effect for food. This supports some outcome-linked symbol comprehension, not human grammar or unrestricted abstract language. Pilley & Reid, 2011 · Bastos et al., 2024 · Animal Language
5. Transfer to a new problem. One novel success is not enough. Use several novel probe sets, record the first trial of each, randomize or counterbalance answer positions, compare with chance, and avoid teaching during the analyzed transfer trial. Blind cue-giving and scoring where feasible. Applying a rule in one trial is not the same as acquiring the concept in one trial.

Teaching principles

• Where it improves clarity, explain new material through imitation, names, information, and existing concepts instead of physically guiding every movement.
• Do not discard conditioning. Reinforcement and feedback support all learning. The question is whether they can teach relations and rules, not only one behavior.
• Do not skip foundations. Make attending, selecting, targeting, release, and the response format fluent before presenting an abstract problem.
• Record success after repetition separately from the first response to a novel problem. The latter is stronger evidence of generalization or recombination.
• Do not prefer a method merely because it sounds more advanced. Choose the route that preserves clarity, motivation, and emotional stability.

9 The language and concepts to teach

“Language” is not treated as one all-or-nothing ability. Signal discrimination comes first, followed by a stable label–referent relation, and then symbolic composition in which known labels are recombined in a novel order. Evidence for the first two does not establish human-like grammar.

Project Epic then divides its teaching and records into commands and information. Commands say what to do; information says what exists or will happen. This is a functional curriculum taxonomy, not a universal linguistic classification.

Expand concepts in layers

1. Referents — objects, people, places, and body parts have stable identities and names.
2. Actions and states — sit is a target body state reached from standing, down, walking, or another starting condition.
3. Properties and relations — color, shape, size, quantity, same/different, in/out, over/under, left/right.
4. Categories — group different-looking balls as balls and recognize the same person across clothing and posture.
5. Time and causality — now/later, first/next, and the relation between one's own action and an outcome.
6. Rules and meta-rules — imitate, select the match, wait for permission, complete mandatory cues, try another strategy after failure.
7. Combinations — recombine known nouns, verbs, modifiers, and rules without direct training of the exact sequence.

This is not a claim that dogs understand all human grammar or abstract vocabulary. It is a research curriculum that operationalizes each item and asks whether it transfers to new referents and combinations without body cues or retraining. The history, methods, successes, and controversies of cross-species language work are collected in Animal Language.

10 Teaching “enter the crate” by three routes

Teaching the same final behavior through different learning routes makes the project's central distinction testable.

1. Condition the behavior. Reinforce successive steps from orienting toward the crate to placing all four feet inside. Add a cue such as “crate” or “enter” to the completed behavior. The dog learns which action produces an outcome in this particular cue and environment.
2. Teach it through Do As I Do. First establish the meta-rule “reproduce the human's demonstrated action.” The human demonstrates entering the crate and then gives the imitation cue. Reproduction of a novel action would acquire the behavior through social learning rather than direct shaping.
3. Teach it through language and concepts. Separately discriminate the names of crate, mat, and box. Separately teach “enter” across a room, tunnel, car, and other referents. Then present the untrained combination “crate + enter” and test whether the dog combines the elements.

Is the third route possible?

Possibly, but one success would not establish sentence comprehension. The dog might respond to the word “crate” with a habitual behavior, follow the human's gaze, or rely on a long reinforcement history. The third route must therefore be both training and a controlled test.

1. First verify random discrimination among crate, mat, and box.
2. Verify action and spatial relations such as enter, get on, and touch across several targets.
3. Neutralize gaze, hand movement, and human position; present voice alone.
4. Place an untrained “enter the crate” among familiar combinations such as “get on the mat” and “touch the box.”
5. Prepare several novel combinations and record the first trial of each. Success after repetition may be relearning the new combination.
6. Randomize or counterbalance position and test a new crate, location, and speaker.
7. Where feasible, keep the cue-giver unaware of the correct layout and have condition-blind independent scorers rate the video.

Ken Ramirez on modifiers and transfer, Do As I Do on imitation generalization, and Kay Laurence on learning design provide three different ways to build this experiment.

11 Why dog training is difficult

Results depend on the dog, the human, and the method

Dogs differ in genetics, temperament, education, and history. The same dog can perform differently with different people. What works in practice is a combination of dog, human, and method—not a single procedure guaranteed for every dog and handler. Common principles such as timing and consistency can still exist.

A working case is not a universal answer

Training claims usually rely on two explanations: “I used this method and achieved the result,” or “learning theory predicts this result.” A success case shows that a combination containing the method can work. Theory explains a mechanism by which the result could occur. Neither alone proves universal effectiveness, comparative efficiency, or success for this dog and this handler.

Practice still has to use both cases and theory because controlled repetition with living subjects is difficult. A dog taught method A cannot be reset to an untrained state for method B. Different dogs and handlers add genetic, historical, and technical differences, and dangerous failures cannot be repeated for experimental purity. Project Epic therefore distinguishes “A produced a success” from “A is best,” while making practical decisions from incomplete evidence. See training research and comparisons of reward and correction.

Values enter because the subject is alive

Training decisions contain views about a good life, permissible discomfort, autonomy, and required compliance. Refusing all coercive methods is a value judgment; requiring unconditional obedience is another. Goals and constraints must therefore be explicit before methods are selected.

12 Practice-based research methodology

Project Epic sets goals and constraints first, then tests multiple methods inside them. It does not declare one doctrine universally correct. It searches for a combination that allows this dog and handler to reach their goals. It may not be optimal; the value lies in testing how far the team can go and demonstrating what becomes possible.

Theories share principles and also conflict. Common principles are retained without accepting any school uncritically. Observation is separated from interpretation. Conflicting claims are tested under different conditions, either concurrently where possible or in separate periods where methods cannot be mixed.

The goal determines the means. Questions such as whether physical force can ever be used are not answered by slogans alone. The project asks what must be achieved or prevented, what alternatives exist, what lasting harm is possible, and whether a current constraint would be crossed. Present beliefs can change with new evidence; current constraints remain binding until deliberately revised.

Recording rules that limit overclaiming from one case

Project Epic is an N-of-1 project: repeated longitudinal observation of one dog and one handler, not a controlled group study. It cannot identify the species maximum or the best method for all dogs. It can establish an achieved lower bound—what this team could do under a documented curriculum—and create a record that others can inspect and replicate.

1. Predefine the hypothesis, success criterion, and analysis. Replace “understands” with an observable response, latency, context, number of novel probes, and chance comparison.
2. Separate training material from several test sets. Repeated success with familiar objects is acquisition; repeated first-trial success across novel objects, arrangements, or speakers is candidate transfer.
3. Separate the first trial from subsequent learning. A first-trial success supports reuse of a prior rule. Later success may also reflect rapid learning of the new case.
4. Control alternative explanations. Balance position, gaze, odor, and body orientation. Separate stimulus generalization, exclusion, and novelty bias from the target concept.
5. Use blinding and independent scoring where feasible. Keep the cue-giver unaware of answer placement and the scorer unaware of condition; report agreement between scorers.
6. Record more than accuracy. Include latency, first choice, corrections, disengagement, avoidance, arousal, recovery time, and voluntary re-engagement.
7. Retain every trial, failure, and counterexample. A single success motivates another test; it does not establish conceptual understanding, optimality, or universality.

13 Operational meaning of “must”

No behavior by a living animal can be guaranteed without exception. Pain, fear, sensory failure, environmental interference, and human error can all break performance. In Project Epic, must is not a claim that the dog understands moral duty. It is the highest training criterion applied to a very small set of cues.

1. The cue remains in force until completed or released by the human. A competing desire does not by itself cancel the cue.
2. The criterion is observable. Context, latency, completion condition, and release are written before testing.
3. Safety cues complement management. High recall reliability does not replace a leash or barrier near traffic, and training does not justify engineering high-consequence failures.
4. Failure is classified. Lack of understanding, perception failure, excessive difficulty, reinforcement history, affect, pain, and a competing choice are not collapsed into “disobedience.”

Whether a functional rule of “must” can transfer across cues is an open question. Novel-cue transfer tests are needed to distinguish a shared rule from several separately overtrained behaviors.

14 Methods and levels of evidence

The list below mixes sources of different kinds. Popularity and practitioner success are not the same as independent validation. Project Epic classifies the source before borrowing a testable procedure.

• Empirically studied tasks — ELTE Do As I Do, category and matching-to-sample tasks, odor and relative-size transfer, and soundboard response studies. Interpret only the narrow question and sample tested.
• Practitioner training systems — Susan Garrett, Fenzi, Training Without Conflict, OneMind Dogs, Ken Ramirez, Kay Laurence, SATS, NePoPo, and others. Their procedures and cases are useful without implying independent validation of the whole system.
• Interpretive and welfare perspectives — Suzanne Clothier, Turid Rugaas, L.E.G.S., Absolute Dogs, Control Unleashed, and related work. They provide observational language and tools; causal explanations remain separate claims.

1. Understand systems and research that already produce real outcomes.
   • Agility — Susan Garrett, OneMind Dogs
   • Obedience — Fenzi
   • IGP and play — Training Without Conflict
   • Ring sports — NePoPo
   • Cognition and social learning — ELTE and Do As I Do
   • Abstract discrimination and concept training — Ken Ramirez on matching-to-sample, modifiers, number, and relational concepts
   • Learner-led problem solving — Kay Laurence and Learning About Dogs on shaping, observation, and environmental design
   • Language communication — FluentPet and They Can Talk
   • Prediction and naming — Kayce Cover's SATS and Name & Explain for objects, body parts, sensations, and upcoming events
   • Relationship-centered two-way communication — Suzanne Clothier, Turid Rugaas, Hunger for Words, L.E.G.S., and perspectives that begin with the dog's sensory world
   • Tricks and shaping — My Aussie Gal, Do More With Your Dog
   • Training perspectives — D.I.N.G.O., Absolute Dogs, Control Unleashed
   • Other foundations — nosework and problem solving, cooperative care
2. Test each system against this project's goals. Understand the goal, mechanism, and sequence before using the label.
3. Perform and practice the training. Observation, timing, movement, and criteria are skills that must be rehearsed, not just understood.

15 Operating criteria for pressure

“The dog knows the cue” is only one necessary condition for considering pressure, not a sufficient one. Human anger must never be the contingency. Before pressure is considered, the project must answer all of the following.

1. Is the goal genuinely necessary for life, safety, or unavoidable care?
2. Does the dog discriminate and perform the cue in a low-burden environment?
3. Have management and less-aversive, reward-based alternatives been adequately tried?
4. Are intensity, frequency, termination, and immediate stop signals specified in advance?
5. Are conflict, freezing, avoidance, post-session recovery, and willingness to return recorded?
6. Can the handler stop when judgment becomes emotional and obtain independent expert review?

Meeting these conditions still does not establish that pressure was necessary or optimal. The outcome includes physical health, affect, voluntary participation, relationship, generalization, and long-term maintenance—not cue accuracy alone.

16 Current record

As of August 10, 2026

This is not a final report of Epic's learning capacity. It separates completed foundation work, current curriculum, and unverified future tests.

• Concepts
  • How learning works: correct, fast, on-cue, and sometimes mandatory performance; luring, capturing, shaping, social imitation; agency over outcomes.
  • Choice and rules: request appropriately, accept approval or refusal, follow structured play, distinguish free-choice and required contexts, permission, roles, and responsibility.
  • Signals and concepts: discriminate voice cues without visible action; names for objects; yes/no, right/wrong; now/later; object, place, posture, and waiting cues.
  • Activities: locate odors and solve scent problems; expand tug, retrieve, chase, physical play, imitation, search, and puzzles.
• Emotion
  • Focus quickly and settle quickly; return from distractions.
  • Find value in educational play; recover from difficulty, frustration, and fear.
  • Offer signals joyfully and voluntarily without being destabilized by error.
• Relationship
  • Trust and follow the human, and ask for help when a problem is difficult.
  • Remain comfortable alone while making time together highly valuable.
  • Request, accept approval or refusal, and increase the variety and precision of communication.
• Senses and body
  • Move in varied ways across varied environments.
  • Search for varied odors and experience varied stimuli.

Limits of the public record and the next records

The list above defines Epic's curriculum and observation domains. It is not yet a trial-level dataset containing baselines, accuracy, first-response latency, and raw video. Inclusion in the list therefore does not establish acquisition or transfer. The next public records will preregister success criteria for the following questions and retain failed trials as well as successes.

17 Perspectives currently being studied

• Playing together
  • Training Without Conflict: Chase & Catch and Possession Games make pursuit and possession the game itself; OUT becomes a transition that can restart play.
  • Fenzi Personal Play: discover how this dog enjoys interacting with a person without requiring food, toys, or a predetermined movement.
• Relationship and communication
  • Susan Garrett Recallers: divide recall into games and build reinforcement and distraction systematically.
  • Fenzi engagement: progress from adapting to the environment toward dog-initiated engagement that survives delayed reinforcement.
  • Keep the “Zorba” relationship and communication note in the private study log, but exclude it from evidence comparisons until the exact public source and system are identified.
  • FluentPet and They Can Talk: model outcome-linked words, then record whether uses and combinations remain consistent across contexts.
  • Control Unleashed: predictable patterns such as 1-2-3 and Look At That for returning attention and lowering arousal.
  • Absolute Dogs: train broader tendencies such as arousal regulation, disengagement, optimism, and proximity through short games.
• Training methods
  • D.I.N.G.O. fair training: begin with a real choice between doing and not doing, then design environment and reinforcement so the learner discovers the answer.
  • Do As I Do: build an imitation meta-rule from known actions and transfer it to novel actions. Controlled work found an advantage over shaping for some complex object-manipulation tasks, not every behavior.
  • OneMind Dogs: handle the dog's line rather than a single obstacle, aligning body direction, movement, and voice.
• Drive, rules, and daily life
  • NePoPo: understand the sequence connecting cue, pressure, release at the correct response, and final reinforcement.
  • Cooperative care: make stillness valuable, establish a start behavior, and gradually add touch, tools, and real care while monitoring comfort.

18 Common ground found so far

These are not all settled scientific facts. The first is a practical interpretive stance; the others are recurring operating principles across multiple systems.

1. Practical stance: treat behavior as one of the actions available under genetics, learning history, environment, and present physical or affective state before assigning moral fault.
2. Behavior reflects genetics and life with humans; consistent human behavior can teach a great deal.
3. Timing and consistency are critical.
4. Communication has to be available in a form the dog can understand.
5. When failure is frequent, change the criterion, environment, reinforcement, difficulty, or human execution before blaming the dog.

19 Current judgments and open questions

The statements below are not conclusions directly established by the literature. They are provisional judgments used to design training as of August 13, 2026. Autonomy is the default, a few safety cues receive high reliability criteria, and welfare and physical safety constraints outrank every performance goal.

1. Judge methods by predefined multidimensional outcomes rather than a label or surface appearance. Welfare during training, voluntary participation, relationship, physical health, transfer, and long-term maintenance are part of the result.
2. Once relationship and foundational concepts are learned well, later learning should become easier.
3. Broader exposure creates more possible directions of growth.
4. Do not set limits on canine understanding, learning, and growth in advance.
5. Most learning has more upside when the dog has agency and can change outcomes. Reserve high-control criteria for a few life-and-safety cues such as recall, built first through management and reward-based training.
6. Teach recovery from unavoidable discomfort such as medical care, restraint, and frustration through prediction and graduated exposure. Deliberately increasing discomfort is not the same as teaching recovery.
7. Do not filter out every difficulty. The world contains danger, responsibility, and tasks that must be completed.
8. Give the dog predictable roles in shared life: participate in learning, practice regulation, and cooperate with body care where possible. Do not turn this into a human moral duty to avoid being a burden.

Questions not yet answered

1. How strongly does first-trial success on a novel phrase support noun–verb composition, and how can location, body cues, and learning history be excluded?
2. Does “must” generalize as a functional rule across signals, or is it only a collection of individually overtrained behaviors?
3. When cooperation persists without visible food or toys, what proportion is best explained by the reinforcing value of learning, social reinforcement, intermittent reinforcement history, or relational trust?
4. Which behavioral measures separate productive recovery under manageable pressure from the early development of avoidance or helplessness?
5. Do novel combinations of buttons or spoken symbols show compositional meaning beyond contextual association, and do they transfer to new speakers and first presentations?
6. Which parts of a successful curriculum transfer to other dog–handler teams, and which belong only to this combination?

This is the project's position as of August 13, 2026. Goals stay explicit, constraints stay binding, and methods and judgments change with new evidence.