Development Log

Reviewing every query that decides who gets paid, rather than only the two the report named, turned up the one that mattered: terminating an aircraft lease credited the escrow account and never drained it. Fifteen million dollars accumulated on a balance that must always read zero, which would eventually have made every seized-asset sale in that world impossible.

Also closed: settlement steps that could run twice, an insolvency check that recorded its verdict before doing the work it authorised, capital spending booked to the wrong category, and a codeshare payout that started its arithmetic from zero for a reseller who owned no aircraft.

A recurring class of bug: the interface recomputing something the server had already worked out, and getting a slightly different answer. One screen predicted an 18-hour flight the simulation actually flew in 14.

Almost every fix was a deletion. The server was already sending the right value; the client just was not reading it. Also fixed in the same pass: share prices displayed to the nearest ten million, a 200-row window presented as complete history, and a cabin-rules editor that offered four of the five cabin classes and silently destroyed any rule for the fifth when you pressed Save.

Retiring a route now also warns you what it ends — it could previously kill rotations at entirely different airports with no warning.

Four near-identical copies of the airport-authority logic became one template. Twelve copies of a content-merging loop became one. A geography helper with fifteen users was deleted outright because the standard library already did the job.

The point is the side effect. Six of those consolidations surfaced a genuine bug that the duplication had been hiding: one copy answered a different error code from its siblings, another silently overrode authored data, a spectator could not see a route's schedule, and the test data generator had been creating arrived flights with no arrival time and no cabin breakdown — which blanked the entire airport analytics screen on every test world anyone had ever looked at.

An airframe's state is what an airline actually tracks: hours flown, cycles completed, and when it was last checked. Two tiers of inspection come due on that basis — frequent light checks and rarer, far more expensive heavy ones — and an aircraft that reaches the end of its life is permanently retired.

The cabin ages separately. An interior that has not been refitted in fifteen years is worth less and sells fewer premium seats, which makes refurbishment a real investment decision rather than cosmetic.

Where the work is done matters too: your own hangar at the aircraft's own base does the job quickly, while flying a team in from another continent takes far longer. That is the payoff the maintenance facility tiers exist to sell — and getting it built was the single acceptance criterion an internal review caught as missing.

Replaces: Aircraft "condition" as a stored number.

Its replacement is dates and counters, which is what an airline's own maintenance records contain. Everything the condition number was used for — value, maintenance due, retirement — reads off those instead.

Replaced by: Maintenance — checks, life and cabin age.

There is a difference between "no aircraft is free at Heathrow this morning" — an ordinary busy day — and "this airline does not own a single aircraft of any type this rotation permits", which is structural.

The second suspends the rotation and says so. It also lifts automatically the instant the situation changes: buy the aircraft the message told you to buy, and the rotation resumes on the very next pass rather than after a lost day.

The engine was a genuine integer-programming solver. It worked, in the sense that it produced optimal assignments — but an optimiser that reshuffles your entire operation overnight is not a game mechanic players can plan around, and it required a native mathematics library that complicated every build.

Removing it made the whole system simpler in one stroke: assignment became a lookup against records the player already authored, the entire codebase became free of native dependencies, and the simulation tick got several times faster.

The lesson recorded at the time is the one worth repeating: automation executes, it never designs.

Replaced by: Rotations — you publish the timetable, the game staffs it.

Things drift. An aircraft goes into unscheduled maintenance somewhere unexpected, a rotation is retired, a base closes. Left alone, a fleet slowly ends up in the wrong cities.

A rebalancing pass notices this and dispatches empty repositioning flights to fix it. These are real flights — they appear on the map, they burn fuel, they use a runway slot — because a deadhead is a genuine operational cost and hiding it would misrepresent the cost of a messy network.

It is on this page because it is an honest part of the record. Roughly a fortnight of work was thrown away, and the replacement is better for it: the current model refuses an impossible schedule and says so, rather than quietly producing something the player never asked for.

This is the heart of the game. A rotation is a loop you write: leave your base at 07:15, fly to Munich, turn around, fly to Rome, come back — arriving home so the aircraft can do it again tomorrow. You choose the departure times, which days of the week it operates, and which aircraft types are allowed to fly it.

The game's job is then to staff that timetable from the fleet you own: every day it works out which specific airframe flies each loop. It does not invent routes for you, it does not move your departure times, and it does not decide your network. If your timetable cannot be flown with the aircraft you have, it tells you — before the flights fail — how many more you need and at which base.

That last point is why the design works. The game can predict a shortfall before anything fails to fly, so buying an aircraft is an informed decision rather than a reaction to a mess.

Because a rotation always ends where it began, the game never has to search for where an aircraft is — it already knows. That one guarantee is what let an enormously complicated optimiser be replaced by a simple lookup.

Replaces: The automatic flight-assignment engine.

The single largest performance change in the project. Deciding which aircraft flies which leg is now settled once, in advance, and written down; the simulation just looks it up.

The result is a tick that is fast, predictable, and — importantly for a shared world — does not spend its time solving the same problem repeatedly for airlines whose plans have not changed.

When an airline dies owing money, its assets get sold — but often long after the company itself has ceased to exist. Somebody still has to receive that money and pass it to the people owed it.

The answer is an estate: a frozen snapshot of who held shares at the moment of dissolution, which outlives the company and receives proceeds as its assets are realised. If a repossessed aircraft sells six weeks later, the former shareholders are still paid.

The same work separated the bank's two roles — the institution that lends you money, and the neutral party that holds seized collateral in escrow — which had been the same entity and was quietly mixing the two pots.

Rather than testing against a laptop that gets closed at night, the simulation now runs permanently on a real server with real worlds advancing in real time. Long-horizon bugs — the ones that only appear after a world has been running for weeks — surface on their own instead of being imagined.

It also means what you would see if you visited is the actual current state of the build, not a demo.

World Port installs like a native app and carries everything that is not live game data on your device: the whole interface, the entire world map and its geography, and the complete rulebook of airports and aircraft.

So it starts up on a plane, on a train, or on a bad hotel connection, and asks the network only for the things that genuinely change — your money, your flights, and what other players have done.

A large audit across airport capacity, facilities, the AI companies and the company model produced hundreds of findings, which deduplicated to 260 root causes. Every critical and high-severity one was fixed and independently verified.

This is the least visible work on this page and among the most valuable: it is the difference between a simulation that looks right for an evening and one that is still correct after a world has been running for months.

When an aircraft is due for a check, the game works down one list: your own hangar at that airport, then a partner you have a contract with, then whoever is there at walk-up rates, then the manufacturer's own network, then a last-resort fallback at a punishing price.

The ladder is the same everywhere, so the answer to "why did this cost so much" is always readable. Before this, several parts of the game each had their own opinion about the ordering.

A new world is not empty. It contains aircraft lessors with fleets to lease out, manufacturers taking orders, ground handlers at major airports and a bank that lends money — and all of them are ordinary companies in the database, subject to the same rules as a player's.

That matters more than it sounds. Because the bank is a real company with a real balance sheet, money it holds has to come from somewhere and go somewhere; because the lessors own real aircraft, the used market has genuine supply. A world with two human players still feels populated.

Failure is a process, not a deletion. A failing airline stops selling seats, finishes the flights already in the air, has its leases terminated and its collateral seized, and its remaining assets sold off. What is left is distributed to whoever held shares in it.

The wind-down deliberately cannot complete while anything is still unresolved — an aircraft mid-flight, an unsold facility, an outstanding debt — because a company that vanishes with loose ends attached breaks the world around it.

Periodically the whole build is audited by a reviewer who did not write it, hunting specifically for the kinds of bug this codebase is prone to: game-time confused with real-world time, money paths that fail open instead of closed, half-finished database transactions, and race conditions between the simulation and the API.

The findings are then deduplicated down to root causes and fixed at the source rather than patched where they were spotted. This audit found a critical one the automated checks could not: the API did not start at all under a particular configuration.

Previously every airport carried a hand-written importance tier that drove its lease pricing and authority fees. That meant a designer's guess decided how expensive a hub was, forever, regardless of what players did there.

Now it emerges: an airport that airlines genuinely build up becomes prominent, and prominence feeds back into what it costs to operate there. A quiet regional field that a player turns into a hub becomes an expensive place to be — and that is a consequence the players created.

The tier drove gate and slot lease pricing and airport-authority fees. It was removed entirely rather than kept alongside the emergent measure, because two sources for the same fact is how they end up disagreeing.

A lounge serves everybody who passes through, so modelling it per-flight was both wrong and expensive. It became a facility that accumulates usage and settles weekly — simpler, faster, and closer to how a lounge actually works.

There had been parallel lists of countries in different parts of the codebase, drifting apart from each other. They were collapsed into one, with a region derived from the country rather than stored separately — so it cannot disagree with itself.

Real aviation is full of nationality restrictions, and they shape who can buy whom. World Port models this as a default-closed rule: taking control of a licensed airline across a border is prohibited unless that country's rules allow it.

It is a genuine strategic constraint on the share market — you can hold a stake in a foreign carrier, but there is a line past which the acquisition is simply not permitted.

When you just need an aircraft somewhere else, you ferry it. It leaves, it arrives some hours later, it costs money, and it does not appear on the map or in any statistic as a revenue flight.

Deliberately, a ferry ignores aircraft range limits. That is how a short-range turboprop bought in Europe can end up flying in Hawaii, which is exactly what happens in reality — via a slow multi-stop delivery nobody wants to micromanage. Charging for it and moving on is the right level of detail.

This is a different mechanic from a repositioning flight, which is the game's own tool for correcting a fleet that has drifted out of place, and which does fly for real.

A merger moves everything the acquired airline owned onto the acquirer and dissolves the original. That includes the awkward parts: contracts where the two airlines were each other's customer, overlapping gate allotments at the same airport, and duplicate routes.

The first phase handles the mechanical transfer. What it deliberately does not yet do is tell the player what a merger will destroy before they confirm it — that is still ahead.

Unglamorous but load-bearing. A world holds millions of interlinked rows, and deleting one has to leave no dangling references anywhere. This is what makes it safe to spin up experimental or short-lived worlds and throw them away afterwards.

Money can move between the group and the companies it owns, recorded as equity rather than as revenue. That lets a player run a profitable handling business and use it to bankroll a loss-making airline — a completely legitimate strategy, and one the books now describe accurately.

An A320 can come from Hamburg, Mobile, Tianjin or Toulouse; a 737 comes from Renton. The game models that: you choose the assembly line, and the delivery flight starts from there. It is a small piece of authenticity that has real consequences for delivery cost and time.

Ordering a jet in a dense all-economy layout costs meaningfully less than the same jet with a full flat-bed business cabin. The configuration decision therefore has a capital cost as well as a revenue consequence, instead of being a free choice made after purchase.

A320, B77W, E190, AT76 — the same codes used on real flight plans and in real airline systems. Anyone who follows aviation will recognise them immediately, and it removes a whole class of invented-identifier confusion.

Freighter variants of aircraft carry cargo instead of passengers, against their own demand model and their own revenue. Cargo behaves differently from passengers — it is less schedule-sensitive and more price-sensitive — so an all-freight operation plays quite differently from a passenger airline.

Consolidated the various one-off ways things changed hands into a single marketplace with listings, and reorganised the fleet screens around the idea that an airline is a portfolio of assets — aircraft, facilities and contracts — rather than just a list of aeroplanes.

Aircraft are fitted out with a mix of First, Business, Premium, Economy and a basic-economy cabin. What you fit decides how many of each seat you have to sell, and denser layouts trade comfort for volume.

On the other side, demand is split into business, general and budget travellers who value price, schedule and comfort differently. A route full of business traffic and a route full of holidaymakers reward completely different aircraft configurations — which is the point.

The idea was that you could pay extra to be served before other airlines when a facility was busy. In practice the only thing that shipped was a doubled fee on maintenance events, with no queue or precedence logic behind it at all.

Facilities stopped being a property of your airline and became assets with their own identity. That means they can be listed on a marketplace, sold to another player, pledged as loan collateral, repossessed by the bank, or moved between sister companies in the same group.

It opens up a genuine second career path: build handling operations at busy airports and make money from other people's airlines rather than running one yourself.

Codesharing lets you sell a seat on a flight you do not operate. You request it on the route page, the operating airline approves it, and from then on your brand pulls passengers onto their aircraft with the revenue split between you.

It is genuinely useful rather than cosmetic: it lets a small airline offer destinations it could never serve profitably itself, and it gives a large one a way to fill seats it was going to fly empty.

Internally a codeshare booking is bookkeeping, not a second aircraft — an important distinction that keeps every per-flight statistic honest.

Holding capacity at an airport is not free. Each gate and slot allotment carries a weekly fee, locked in at the rate that applied when you signed, and set per-airport by whoever holds the authority role in that world.

Fall behind on those fees and your allotments are revoked automatically — which is a slower, more recoverable failure than bankruptcy, and a much more common one.

A shared persistent world needs governance. This shipped a three-tier permission system — game administrator, world administrator, and per-world staff split into moderators and airport authority — plus the surfaces they act through: grants, suspensions, an audit trail, world settings, join codes and world deletion.

The airport-authority role is the interesting one: it is a player-facing job, setting gate and slot pricing, capping frequencies and banning country pairs, rather than a hidden admin switch.

An airport can only handle so many movements an hour, park so many aircraft, and clear so many departures per runway. When demand for those exceeds supply, aircraft queue — on the ground before departure and in a hold before landing.

This is enforced at the moment a flight tries to depart or arrive, not estimated afterwards. Congestion at a hub is therefore a real operational cost that shows up as delay, and it is caused by all the airlines at that airport collectively, not just yours.

Whether you are served by a contracted partner, paying walk-up rates, or falling back on the default operator, the money moves through the same daily settlement path. Before this, one of those three routes silently charged nothing at all.

An early version moved cash around at the moment each flight landed, which made the books almost impossible to explain to a player: money appeared and vanished with no readable trail.

This inverted the model. Costs and revenue accumulate against the thing that caused them and settle on a schedule, each producing a labelled ledger row. It is why the finance screens can now tell you why you lost money last week rather than just that you did.

At an airport where you have no facilities of your own, someone has to handle your aircraft. That someone is another player's ground-services company, an independent operator, or — if nobody wants the business — a fallback provider at an unattractive price.

Contracts are open-ended and priced at signing, so a rival who builds a handling business at a busy airport before you do has genuine leverage over your costs there. This is the point at which World Port stops being a solitaire tycoon game.

It also shipped the aircraft-on-ground flow: when a jet breaks somewhere you have no maintenance cover, it sits there, earning nothing, until someone is paid to fix it.

You do not simply receive an aircraft into your fleet. It is assembled at the manufacturer's factory airfield and then has to travel to one of your bases, which takes time and costs money.

For a short-range turboprop bought for a network on the other side of the world, that is a real consideration rather than a formality.

Not every airline can afford new metal. Aircraft can be listed for sale by their owners and bought by anyone, at whatever price the market bears, with age and condition reflected in the value.

Leasing is the alternative: far less cash up front, a rent bill every week forever, and the aircraft goes back at the end. Choosing between the two is one of the sharper financial decisions in the game — and getting it wrong is one of the more common ways to go broke.

An in-world bank will lend against your fleet and facilities. How much, and at what interest rate, depends on a credit assessment of your airline: what you own, what you already owe, and whether you have been running a deficit.

Miss enough payments and the bank seizes what it lent against. That seizure path is fully modelled rather than being a game-over screen — the aircraft genuinely change hands and come back onto the market.

A player is not an airline. A player owns a group (a holding company), and that group owns companies — airlines, and later ground-handling and maintenance businesses. Each company is then activated inside a specific world.

The layering is what makes the interesting things possible later: selling a company without selling the player, moving cash between sister companies, letting one player hold a stake in another player's airline, and winding a failing airline down without deleting the person behind it.

It also shipped the universal transfer primitive — one mechanism that moves money, aircraft or facilities between any two owners — and the first version of the wind-down state machine that handles an airline going out of business.

Ground services (check-in, baggage, catering, pushback) and maintenance hangars exist at tiers. A higher tier turns aircraft around faster and does heavier maintenance work, but costs more to build and staff.

Crucially, construction is not instant — you commit money now for a capability that arrives weeks later, which is the first real forward-planning decision the game asks of you.

This also set the airline-wide catering policy: what you serve in each cabin, priced per passenger, and penalised if you promise a standard you cannot deliver.

Four payroll categories — flight crew, cabin crew, ground staff and maintenance staff — sized from what you actually operate, plus fixed weekly upkeep on every facility you own.

This is the floor under the whole economy: it is the reason an idle fleet bleeds cash, the reason overbuilding hurts, and the reason a route that merely covers its fuel is still losing you money.

Before an airline can exist, the game needs to know what an airport is. This laid down the catalogue: real-world airports with their runways, gates and the aerodrome reference code that decides which aircraft are physically allowed to operate there — a 747 cannot use an airfield built for turboprops, and the game enforces that rather than trusting the player to know.

It is deliberately shipped as authored data files rather than database tables, so the whole rulebook travels with the app and can be read on the player's own device without asking a server.