WAYFARER LABSContact ↗

Independent engineering studio · Switzerland

Software, hardware
and interactive worlds.

Explore the products we build, the questions they answer and the work behind them.

Decades of experience across computer science, data science and product development.

Meet the creator ↗

Explore the products

Choose a product to see it in use
Hardware & joint researchColdFrontCooling, defrost & airflow · In development↗HoldfastWindlass control · In development↗LookoutMasthead vision · Prototype↗EvenKeelWith Myto AG · Prototype↗

About Wayfarer Labs

Meet the creator.

Illustrated portrait of the creator of Wayfarer Labs

Independent engineering · Switzerland

The person behind the work

Computer science.
Applied across disciplines.

Discuss a project ↗

Wayfarer Labs is built on decades of experience turning complex science and data into practical systems. Its creator holds a PhD in computer science and brings extensive director-level experience at a global healthcare company, leading work across strategic intelligence, data science and product development.

That work connects clinical science and operations with portfolio management, finance, manufacturing, regulatory affairs and patient advocacy. It uses digital twins, predictive engines and decision intelligence to understand complex systems, anticipate outcomes and support better decisions.

The same discipline shapes what we build here: make assumptions explicit, test predictions against reality, and give people information they can act on. For our marine systems, that means taking the work aboard boats, where real passages put the models, software and hardware to the test.

Telltale / Sailing intelligence

Local weather.
Routes for your boat.
The systems aboard.

Our coastal wind and wave models combine physics with data science. Boat-specific routing, anchorage assessment and connected instruments put those fields into the context of your passage.

Explore Telltale ↗

Actual application renderer · saved model output, 23–24 October 2026 · not a live forecast.

Passage planning

Route for your boat →

Polars, forecast timing and passage constraints.

Anchorage assessment

Understand the swell →

Directional wave propagation and boat-relative exposure.

Your setup

App or App + Bosun →

Choose how to connect and keep operating aboard.

Get the app ↗Methods and model scale →Discuss your boat ↗
← Wayfarer LabsContact support ↗

Product support

What can we help with?

Find setup instructions, understand a feature or report a problem.

8 suggested topics · 70 topics
Telltale · Getting started

First launch →

Sign in, create an account, or connect a Bosun aboard.

Telltale · Bosun

Working offline →

What keeps working without internet, and what happens after.

Telltale · Instruments and gateways

Supported gateways →

Presets, factory defaults, and why the defaults are often wrong.

Telltale · Keeping it running

Diagnostics →

What to look at, and what to collect.

Curio Studio · Product help

Backups and iCloud →

Export a complete backup or connect your Apple devices.

HoundHunt · Product help

Understand a probability →

Read probabilities alongside calibration and the recorded outcomes.

HoofHunt · Product help

Private beta access →

HoofHunt is available to invited HoundHunt users.

Games · Product help

Join a beta →

Ask about the game and device you want to test.

← All productsHoundHunt support →

Race research / Greyhounds

HoundHunt

Available

A probability for every runner.
A record of every result.

Research Australian greyhound racing with Sergio, our in-house prediction model. Bring form, sectionals and race conditions together, then examine the evidence behind each assessment.

Open HoundHunt ↗
HoundHunt brand artwork
HoundHunt brand artwork

HOW IT WORKS

01

Point-in-time inputs

Form, sectionals, track conditions and market history available before the race.

02

Versioned inference

The same feature path serves model training and inference. Calibration is tracked by version.

03

Recorded outcomes

Compare probabilities with results across the full sample, including losses.

Method & evaluation

An assessment should survive scrutiny.

Preserve the information cutoff, model version, predicted probability and available price. After settlement, inspect calibration, sample size and drift alongside returns. Time-ordered evaluation keeps later races separate from earlier training data.

Research and educational analytics. Probabilities do not guarantee outcomes. 18+.

← All productsHoofHunt support →

Race research / Thoroughbreds

HoofHunt

Private beta

Thoroughbred performance,
modelled in context.

Sergio Thoroughbred combines form, sectional times, race conditions and weather for Australian and New Zealand racing. A dedicated model, within the shared HoundHunt research workspace.

Open HoofHunt · invited users ↗

SYSTEM OVERVIEW

01FORM + SECTIONALS
02CONDITIONS + WEATHER
03SERGIO THOROUGHBRED
04PROBABILITY + OUTCOME

Inputs and assumptions remain attached to the result.

Method overview, not measured model performance

HOW IT WORKS

01

Performance history

Recent form, sectionals, barriers, distance and class changes.

02

Conditions at race time

The going and weather add context to a runner’s history.

03

Evaluate the prediction

Race probabilities, market comparisons and settled results remain connected.

Method & evaluation

The horse model has its own evidence.

Sergio-T is a dedicated thoroughbred model. Its evaluation needs its own races, reference periods and outcomes. Results from greyhound racing are not evidence of performance on thoroughbreds.

Private research beta for invited HoundHunt users. No public subscriptions. 18+.

← All productsColdFront support →

Marine hardware / Cooling, defrost & airflow

ColdFront

In development

Coordinate cooling, defrost
and air circulation.

A retrofit controller designed to manage refrigeration automatically, coordinating compressor effort, intelligent defrosting and air circulation. Use temperature feedback and available power to maintain the required cooling while reducing avoidable energy use.

Discuss your installation ↗
Existing product concept visual · development hardware
Existing product concept visual · development hardware

HOW IT WORKS

01

Measure the thermal system

Five temperature points across the box, evaporator and airflow.

02

Control the cooling work

Coordinate compressor speed, defrost cycles and fan speeds within the configured temperature limits.

03

Connect the energy context

Schedule cooling around available power, with local control and optional Telltale integration for the boat’s energy budget.

Development & evidence

A local control loop, with an energy-aware plan.

Initial integration targets Danfoss BD35 and BD50 compressors. The display brings temperature, power, energy and defrost information together. Savings must be established by comparing equivalent cooling work under measured conditions.

Development proposal. No measured energy-saving percentage is claimed.

SYSTEM OVERVIEW

01TEMPERATURES
02LOCAL CONTROLLER
03COMPRESSOR + FANS
04MEASURED RESPONSE

Thermal feedback closes the loop.
Telltale adds energy context when connected.

← All productsHoldFast support →

Marine hardware / Windlass & thruster

HoldFast

In development

Control the windlass.
Keep the chain count in view.

A wireless handheld and base station for anchoring. Chain length, depth, scope and seabed context sit together, while the original footswitches and wired controls remain available.

Discuss compatibility ↗
Existing handheld concept visual · development hardware
Existing handheld concept visual · development hardware

HOW IT WORKS

01

Command at the bow

Distinct physical controls for windlass and bow thruster.

02

Measure the movement

Chain count follows movement from the handheld or the boat’s original controls.

03

Share with Telltale

Measured chain length can feed the anchor watch and voyage log.

Development & evidence

An additional control path.

The proposed architecture preserves the boat’s original controls. The fitted trial must verify command response, chain count and coexistence across both control paths. Standalone operation does not require an app or server.

In development. Fitted acceptance testing remains part of the release work.

PROPOSED CONTROL PATH

01HANDHELD
02BASE STATION
03WINDLASS
04CHAIN SENSOR

Original wired controls → windlass
Independent command path remains available.

← All productsLookout support →

Marine hardware / Machine vision

Lookout

Prototype

Turn a masthead view
into a position on the chart.

A stabilised thermal and daylight camera concept. Combine viewing angles with mast height and vessel attitude to estimate where a water-level contact lies.

Discuss the prototype ↗
Existing pod concept visual · not a fitted-unit photograph
Existing pod concept visual · not a fitted-unit photograph

HOW IT WORKS

01

Observe

Thermal and daylight imagery from a stabilised masthead pod.

02

Resolve the geometry

Mast height and corrected viewing angle estimate water-level range.

03

Connect the chart

Telltale supplies heading, position and attitude for chart-referenced contacts.

Development & evidence

Make the geometry and its limits visible.

Passive ranging depends on the contact being at water level and on the accuracy of height and viewing angle. A near-horizontal sightline is particularly sensitive to angle error. The proposed laser rangefinder provides another measurement at suitable ranges.

Prototype programme. Published specifications are design targets until verified on a fitted unit.

MEASUREMENT CHAIN

01MAST HEIGHT
02CORRECTED ANGLE
03WATER-LEVEL RANGE
04CHART POSITION

r = h / tan(α)
Idealised water-level geometry, after attitude correction.

← All productsEvenKeel support →

Joint development / Myto AG

EvenKeel

Prototype

Joint research,
before a public specification.

An early-stage product developed with Myto AG. The work is independent of Telltale. Product features and release timing have not been announced.

Contact the lab ↗

SYSTEM OVERVIEW

01RESEARCH QUESTION
02JOINT PROTOTYPE
03OBSERVATION
04SPECIFICATION

Inputs and assumptions remain attached to the result.

Development sequence · no product specification implied

HOW IT WORKS

01

Define the question

Establish the problem and the evidence needed.

02

Build the prototype

Test the proposed approach with the development partner.

03

Publish what is supported

Release a specification when the evidence is ready.

Development & evidence

Keep the public claims proportionate.

EvenKeel has its own development programme. This page provides a clear status and contact route without filling the gaps with speculative features.

Prototype with Myto AG. No public release date.

← All productsCurio Studio support →

Collections, projects & making

Curio Studio

Available

Keep the things you make
connected to what you learn.

A personal studio for hobbies, collections and creative projects. Keep records, photographs, methods and supplies together, from the first idea to the finished piece.

View on the App Store ↗Curious? Explore Curio Studio ↗
Curio Studio cabinet ipad
iPad cabinet preview. Upcoming layout, with fictional example collections.

HOW IT WORKS

Development & evidence

Your studio lives on your device.

Optional iCloud sync connects your own Apple devices. Complete backup exports include records and photographs. The support pages explain backup, recovery and app-lock behaviour before you need them.

iPhone and iPad. See the App Store for current availability and requirements.

The complete portfolio

Explore our products.

Software, research, hardware and games. Each product has its own purpose, evidence and development status.

Available

Telltale →

Wind and wave models, with boat-specific routing.

Available

Curio Studio →

Collections, projects & making

In testing

Wayfarer Labs Games →

Paper worlds and interactive systems.

Available

HoundHunt →

Race research / Greyhounds

Private beta

HoofHunt →

Race research / Thoroughbreds

In development

ColdFront →

Marine hardware / Cooling, defrost & airflow

In development

HoldFast →

Marine hardware / Windlass & thruster

Prototype

Lookout →

Marine hardware / Machine vision

Prototype

EvenKeel →

Joint development / Myto AG

← All productsGames support →
WAYFARER LABSGAMES◆ ◆ ◆

Paper worlds.
Decisions with consequences.

Explore three games in development. See how each one plays and ask about joining the beta.

Paper Tides key artSailing survivor roguelite

Paper Tides →

Steer through the fight while the cannons fire.

Public TestFlight betaCorporate Survivors key artWorkplace survivor roguelite

Corporate Survivors →

Clear a route through the office.

TestFlight testingDead End Department key artLane defence with movable formations

Dead End Department →

Inspect the incoming threats and available budget.

TestFlight testing

Illustrated key art above. Each game page includes a development gameplay capture.

← Wayfarer Labs GamesSupport →

Sailing survivor roguelite / Public TestFlight beta

Paper Tides

Sail, fight and bring
your ship home.

Steer a paper ship through enemy fleets, line up broadsides and turn salvage into a vessel worth bringing home. Wind, tide and depth give the paper world a physical foundation.

Ask about the beta ↗
Paper Tides illustrated key art
01 / PLAY

Position the ship

Steer through the fight while the cannons fire.

02 / PLAY

Build the vessel

Combine weapons, captain choices and refits.

03 / PLAY

Return to harbour

Bank your haul and prepare for the next expedition.

Paper Tides development gameplay capture
Development gameplay capture from the live product page. Artwork, interface and balance may change.

iPhone and iPad beta. Steam for Windows, Steam Deck and Linux is planned.

Beta help →
← Wayfarer Labs GamesSupport →

Workplace survivor roguelite / TestFlight testing

Corporate Survivors

Find a desk.
Finish the assignment.
Survive the interruptions.

Fight through micromanagers and office chaos, then hold a quiet desk long enough to do useful work. Your personal attacks pause while you work, so every productive moment has a cost.

Ask about the beta ↗
Corporate Survivors illustrated key art
01 / PLAY

Fight for space

Clear a route through the office.

02 / PLAY

Work under pressure

Reach the required workstations and complete the assignment.

03 / PLAY

Choose when to leave

Make the exit or risk a little overtime.

Corporate Survivors development gameplay capture
Development gameplay capture from the live product page. Artwork, interface and balance may change.

iPhone beta. PC and Steam Deck are planned.

Beta help →
← Wayfarer Labs GamesSupport →

Lane defence with movable formations / TestFlight testing

Dead End Department

Move the desks.
Change the defence.

Build a team and rearrange its formation as management changes the battlefield. Moving one colleague can rescue a lane, restore a useful bond or expose the rest of the office.

Ask about the beta ↗
Dead End Department illustrated key art
01 / PLAY

Read the lanes

Inspect the incoming threats and available budget.

02 / PLAY

Build relationships

Choose colleagues and neighbours that work together.

03 / PLAY

Rearrange the team

Move desks with a shared cooldown. Every rescue creates a tradeoff.

Dead End Department development gameplay capture
Development gameplay capture from the live product page. Artwork, interface and balance may change.

iPhone beta, with PC and Steam Deck development in progress.

Beta help →

Wayfarer Labs / Methods across the portfolio

Data, physical models
and systems that can be tested.

We work across numerical simulation, machine learning, sensing and embedded control. The common discipline is to represent the system carefully, preserve the evidence behind each output, and test it against the conditions it will encounter.

01 / INFORMATION & TIMING

What could the model know?

  • Use only information available at the decision time.
  • Keep forecast issue time separate from the time being predicted.
  • Exclude later results and revisions from historical inputs.
Race-time features · Forecast run and lead
02 / PHYSICS & LEARNING

What governs the behaviour?

  • Physical models describe mechanisms, from coastal flow to thermal inertia.
  • Learned relationships capture patterns in observations.
  • Check where each approach stops being reliable.
Coastal flow · Runner probabilities · Thermal response
03 / ERROR PROPAGATION

Where does uncertainty grow?

  • Near the horizon, a small camera-angle error can create a large range error.
  • Wind error changes boat speed and arrival time.
  • That timing change alters the weather encountered next.
Contact geometry · Time-dependent routing
04 / INDEPENDENT EVIDENCE

Does it hold up elsewhere?

  • Test later races, different terrain and changing loads.
  • Compare against a simpler baseline.
  • Inspect failure cases and conditions, not just an average score.
Chronological evaluation · Regime-specific error

HoundHunt & HoofHunt / HoundHunt available · HoofHunt private beta

Build the prediction from
what was known at the time.

Machine learning · Temporal data engineering · Probabilistic evaluation

Sergio and Sergio Thoroughbred turn heterogeneous racing histories into runner probabilities. The methodological challenge is preserving the information available before a race, then evaluating every assessment after the result arrives.

01Point-in-time race data
02Versioned feature builder
03Model + calibration
04Settlement + monitoring

HoundHunt + HoofHunt / Data inventory

Individual animals.
Millions of recorded runs.

Runner histories, race conditions and market observations provide the raw material. Versioned feature builders turn selected records into training and evaluation samples.

ANIMAL REGISTRIES70,796

43,464 dogs and 27,332 horses recorded in the product registries.

HISTORICAL RUNNER RECORDS3.46 million

2,441,643 greyhound records and 1,021,552 horse records in the historical market datasets.

RACES WITH STORED PREDICTIONS22,434

21,833 HoundHunt races and 601 HoofHunt races, counted once per product regardless of repeat assessments.

Training samples, variables and the underlying datasets +

What the model builds used

Recorded buildSergioSergio Thoroughbred
Input variables18685
Training rows318,491742,333
Validation rows68,25233,326
Test rows68,245241,224
Total split rows454,9881,016,883

Rows represent runner samples, not distinct animals. The Thoroughbred test split covers 24,915 races. Its form-only variant has 85 inputs; the separate main research variant has 90.

What the data store contains

Greyhound historical market records2,441,643
Horse historical market records1,021,552
Additional horse form records92,814
HoundHunt runner-prediction snapshots3,266,356

Repeated predictions preserve how an assessment changed as new information arrived. They are not additional races. Form and market datasets can overlap the same historical runs, so their counts are not added to the training sample.

Snapshot: 28 October 2026. Registry identities do not count every animal in every historical source. Race coverage means stored probabilistic assessments, not successful predictions. Training figures refer to the identified model builds, not a sum of repeated retraining. Dataset size alone does not establish predictive performance.

Inspect counts and model versions ↗
DATA ENGINEERING

Shape & complexity

Race, runner and venue histories are joined with sectionals, track-and-distance records, box or barrier position, grade or class, conditions, weather and timestamped market prices. Each source can arrive or be corrected at a different time. Comparable-runner cohorts depend on the discipline and race context.

MODELLING

Methods & techniques

Sergio uses CatBoost with a race-grouped objective, turning runner scores into a probability distribution across the field. Sergio Thoroughbred uses a separate CatBoost pipeline for horse racing. Point-in-time feature joins, chronological evaluation and probability calibration address different sources of error; model explanations expose the contributions behind an assessment.

IMPLEMENTATION

Technology & operation

Versioned inference retains its feature schema, model identity and input timestamp. An append-only decision ledger connects probability, available price, policy decision and settlement. Monitoring follows calibration and drift alongside sample size, closing-line comparison, drawdown and returns.

EVALUATION

Validation & uncertainty

Evaluate the complete sequence of decisions, including losses and periods with no action. Keep training and evaluation periods separate. Report model version, sample and uncertainty beside the metric. A profitable interval and a calibrated probability model are different claims.

186

Sergio input variables in the recorded 25 October build

85

Sergio Thoroughbred form-only input variables; the main research variant uses 90

1,060,824

training rows across the two identified builds, excluding validation and test splits

Scope, limitations and evidence status +

Corpus size measures coverage, not predictive skill. Historical sources can describe the same runner or race more than once. Training rows, stored predictions and independent test outcomes are different quantities. Outcomes remain uncertain.

Counts verified from read-only database aggregates and saved model metadata on 28 October 2026. Sergio: build 20261025_160418. Sergio Thoroughbred: form-only build 20261018b. Historical inventory measured 27–28 October. No accuracy or return is inferred from dataset size.

Explore the research products →

Dataset complexity & scale

Count the right thing.
Keep the context.

A forecast cube, a race-history table and a stream of camera frames pose different problems. Volume matters alongside resolution, coverage, information timing, missingness and the independence of observations.

We distinguish source coverage, data actually processed, training samples and independent evaluation samples. Each published figure should carry its period, configuration and purpose.

Actual product capture with demonstration data. Historical interface, not a live forecast.