Why I rebuilt
HeavyMach

I chose development as my third occupation. I worked as a magazine journalist in 2015 and worked in the Philippines in 2016. In July 2017, I changed careers to become an iOS developer. The world I experienced in jobs far removed from software felt nondeterministic and irrational to me. There were several reasons I left that life behind and chose another occupation.

Events at the magazine had an even greater influence than HeavyMach. Editorial designers with skills that were hard to replace collectively defied management, and their working conditions improved as a result. They were guaranteed to leave on time except for the two days spent finishing the monthly issue. I also wrote about this in issue 9, “The Price of Defiance”.

A month later, the editor-in-chief of the editorial department erupted over drinks. He said it was impossible for six people to put three monthly magazines and one daily newspaper to bed every month and do the reporting as well, and that everyone was working late every day. The result went the other way. The whole editorial department was ordered not to leave before ten at night for a month. It was a retaliatory work order for having annoyed someone, rather than a formal disciplinary penalty.

There were also things I saw and heard while interviewing executives at factory-automation companies such as Rockwell Automation, B&R and Yokogawa. They spoke about the benefits of automation through engineering. Those reporting experiences also influenced my choice to become a developer.

When I share glasses of soju with people I know, they sometimes tell me that if I had continued doing the work I did in 2015 and 2016, I would have ended my life by jumping to my death. I still do not know whether they are right.

The two people who made HeavyMach extended my life by nine years and three months.

Lately, I feel that I do not have much time left. It is not because of a health problem or extreme lethargy. If anything, I feel my mind is clearer than it was a month, two months or half a year ago. So I think about what I want to try now and what I want to leave behind. Restoring HeavyMach is one of the things I chose. A game once showed me a direction; now, having become a developer, I make it work again. To me, that connection brings the ending back to the beginning.

I rebuilt in 2026 a game released in 2009 that had shown me a path. It is no longer on the App Store, and few people around me remember it. I wanted to run that game on the iPhone I use today. I had Codex write the source code anew, then kept pointing out and correcting the parts that behaved differently from the original.

Someone called mesabloo had worked on a similar project. He obtained assets from HeavyMach Defense and rewrote the source code from the ground up. He seemed to have stopped while implementing the algorithms for combat between units. I had a better starting point than he did, since he had to write the source code from scratch.

I had a dream while sleeping one weekend. In it, I was seeing myself first encountering the iPod touch and Cydia from a third-person perspective. When I woke up, I typed heavy Mach into Google. Gemini in Google Search told me about a specialist web-archiving site where I could get the original HeavyMach. That was the beginning of this restoration project. I started the restoration with a HeavyMach 1.13 build downloaded from the HeavyMach series collection on Internet Archive (archive.org).

That is why I think I did almost nothing. It is also why I think I have even less business trying to make money from it. In my own words, all I did was grab Codex by the collar and slap it across the face when it strayed from the original source code and its creators' aesthetic achievement and did something arbitrarily wrong.

If an original creator or another stakeholder asks for it to be taken down, I will delete this repository without hesitation.

I do not know what Byun Hae-jun, the developer who made the series, or Park Jae-cheol, its illustrator, would think of this restoration project. I left an inquiry on a messenger profile that appeared to belong to Park Jae-cheol himself. When I wrote about this in the developer's statement, I had not yet received a reply.

When I began the restoration, the creators I knew about were Byun Hae-jun and Park Jae-cheol. That was why I recorded their names first. I knew about Mesabloo's restoration project, but only learned that WeMade had acquired the HeavyMach IP while organizing the credits this time.

In 2011, WeMade Creative, a subsidiary of WeMade Entertainment, acquired Indiapps assets including the IP for HeavyMach and HeavyMach 2. The credits distinguish the original development and illustration from the later acquisition.

In the initial implementation, the game saved a file every time the tank fired or took a hit. Notifications that ammunition or experience had changed led straight to file writes. I asked, “You save a file every time I fire?” Codex separated memory updates from saving and changed it to save when choosing a slot, moving to the next stage, reaching game over or returning to the menu. After the fix, automated tests recorded zero saves during 180 frames of firing and one on returning to the menu.

The working version of the original game that I had tested myself had no frame drops, but the early restored app did. I asked whether using UIKit to draw game graphics was even appropriate and told it to compare that with the existing path. Codex replaced the CPU image creation performed every frame with OpenGL and prepared textures in advance for reuse.

Firing still caused delays after the rendering method changed, so Codex continued investigating the cause. Codex measured the calls that restarted sound effects and identified the cost of starting playback through AVAudioPlayer. It changed them to reuse buffers and sources prepared in advance in OpenAL. It fixed the missing sound when changing weapons by adding the omitted sound-effect playback calls.

I also asked why it had added accumulated-time calculations. When a delay occurred, that implementation updated the game state multiple times between display calls. It changed the original feel of movement and firing. Codex changed it to update the game state once each time a frame was drawn. When the device is locked or the app goes into the background, the game, sound and held-input states stop; after returning, the player resumes manually.

I asked it to show a wider world instead of enlarging the artwork. The wider screen also needed to let me fire at the point I touched in the newly visible areas on either side. Menu backgrounds and mission-result artwork also had to fill the wider screen. I looked at them directly and pointed out discontinuous image boundaries and black lines. I also had it fix the menu transition, where the original and new areas opened and closed separately. Codex changed the four panels that close like doors to cover the whole screen.

In mission 3-5, the aircraft boss approached the left side too quickly and did not return well to the right. Codex initially mistook the enemy I meant for a helicopter. Only after I clarified that it was an aircraft did it compare the correct boss. The wider screen was satisfying both the condition for displaying the enemy and the condition for starting combat too early. Codex separated the two conditions to fix it.

There was also an issue where air-support attacks landed behind the tank instead of at the touched point. Fake coordinates passed on touch release were entering the attack-position calculation. Codex separated input release from aiming coordinates and retained the last attack position. I also required a procedure for checking all weapons in the cheat app. Even when core calculations were correct, actually pressing and releasing controls produced a different result.

The illustrations after completing a mission were something I remembered. When I pointed out that they were missing, Codex fixed JPG/PNG suffix matching in resource names. I then checked the illustrations on a real phone. When the wider screen exposed the battlefield outside the illustration, Codex corrected the background and visible area so the results screen covered the whole display.

I had the game code and resources shared between iPhone, iPad and Mac. If movement, ballistics, damage and enemy placements are copied onto each device, a correction on one can be missed on another. Codex put a shared C++ core, Apple runtime and resource bundle into one workspace. The game code is linked through static frameworks, and artwork, audio and maps use the same resource bundle. Mac was built as an AppKit app.

For iPhone, we added landscape rotation and touch, tilt and keyboard input. iPad can use touch, tilt and a keyboard together. Mac supports windowed and full-screen modes. Movement, jumping, barrel adjustment, firing, weapon selection and menus were mapped to the keyboard, with Shift allowing finer angle adjustment.

The regular app and cheat-enabled validation app install separately. The cheat app lets you specify missions, weapons, parts and upgrades, so you do not have to replay the game from the start just to see the same enemy. Movement and firing still use the same processing paths. App IDs and save storage are separate too.

On Mac, Codex once described a testing-tool failure as though something was wrong with the whole game. I told it to finish testing the actual game first. When signing failed through the CLI, I pointed out that it had already run through Xcode. Codex compared the team, signing and device states of the successful and failed paths before proceeding with installation.

I specified the Yak, Harrier and F35 and told it to follow the existing units' colors and texture. I did not ask it to reinterpret their appearance. But Codex itself added conditions to the first generation prompt such as aircraft inspired by real models, wide and low silhouettes, and exaggerated armor panels. It referenced the existing helicopter artwork, but did not include photographs or drawings of the actual aircraft.

I asked why the shapes differed from the real models. Codex added rules for preserving the real shape. It separated photographs and drawings used to check form from the existing game artwork used to match color and texture. It compared the positions and proportions of noses, wings, tails and intakes. After checking the new drafts, I told it to use that process for future additions of real-world units too.

In terms of aesthetics, visual elements and tone, I think Codex leans too heavily on the style of OpenAI's image-generation model even when it has perfectly working source code, assets and visual clues such as sprites. Lux Lunae was a relatively lucky case. For the Yak, Harrier and F35, I assumed that a human would look at the original designs and depict something similar. The clues were already all there. Instead, it brought back science-fiction depictions in the opposite direction, drafts that even ignored my direction, and I had to wrestle with it.

After leaving my last company in 2025, I rode a motorcycle around the country. Unlike Seoul or Gyeonggi Province, riding regional national roads and empty paths through fields while depending on moonlight feels different from a simple trip or a change of mood. Riding through darkness so deep that I cannot see my own hand, judging the road only by moonlight and headlights, can create the illusion that only the motorcycle and I exist in the world.

When you ride a narrow, unfamiliar road at 120 kilometers an hour, your mind automatically empties itself to stay alive. You could say it becomes close to an extreme meditative state that the body instinctively enters when facing death. In a place you have never been, all you can see ahead is moonlight; riding under it in a state of complete inward immersion changes a person. The word moonlight became a peculiar symbol in my life.

My ride around the country under moonlight and Namco naming the bosses' squadron Lux Lunae in 8 must be coincidences: independent events with zero correlation. It is also a coincidence that one of the major enemies and visually striking symbols in a game I was waiting for carries that name.

From a gameplay perspective, 8 was a well-made work that did not break my underlying premise of “I'll die a little later.” Lux Lunae combined well-designed looks, brutally formidable skills to match them, and actors' lines that were enjoyable to listen to. The game as a whole showed a comprehensive level of polish, and Lux Lunae, one of its symbols, contributed to that.

I wanted to recreate the threat and appeal I felt when facing them myself. Looking at contemporary popular culture, I think twisted, powerful yet compelling villains stay in public conversation longer than appealing protagonists or good guys alone. I think Josh Brolin in Sicario and the F-22 in Transformers serve that intention too.

The F-22, Su-57 and MiG-31 were the aircraft that felt most villainous among Lux Lunae's squadron in 8. Still, I sometimes think I should have included one with a different design, such as an F-16 or Typhoon, for variety. I think the three have broadly similar designs.

The original boss battles combine one fairly difficult boss unit with regular units that appear in other stages. I thought that structure was a little ordinary. Before joining the magazine, my first workplace, I once passed the application-document screening for a game-planning role at Netmarble. This time too, I thought from the perspective of player experience and game planning.

I wanted two effects while intervening as little as possible in the original assets, unit-behavior code and game experience. One was to raise boss-battle difficulty. The other was to create tension both during and before boss battles by introducing enemies that share a design language with existing enemies but feel different in appearance and performance and are difficult to kill.

The three VTOL types were inserted between placements without replacing original enemies. They reused the existing helicopters' movement and weapons; Codex calculated the additions at 113 aircraft across 14 sub-missions. The 1,182 existing enemy-placement entries remained. Fighters in Lux Lunae liveries were added as reinforcements in boss battles and the final stretches of the preceding missions.

The fighters reuse the existing helicopter missiles with burst firing added. Durability was set to 1.7 times the maximum for regular helicopters, and missile movement speed to a 25% increase. I told it to reduce the initial 25% increase in firing rate to 10%. After seeing aircraft suddenly appear inside the screen in the regular app, I told it to fix that; now the whole aircraft enters from beyond the right edge.

From the perspective of keeping the original feel while changing or tuning its source code as little as possible, I think Lux Lunae's movement and weapon-use sequence stay within the original. I did not direct the fast flight in from off screen. But it reminded me somewhat of Lux Lunae's high-speed hit-and-run attacks and vicious turning fights in Ace Combat 8, so I gave no correction on that and kept it. The formation that bunched up on entry was revised in v0.16 by spreading out spawn positions and entry destinations. After combat begins, the aircraft can still converge around the player.

Beyond making the player pay more attention to movement and evasion, I wanted the tension of thinking, “I really need a special attack for this, or I need to take a deep breath and concentrate,” when facing bosses that could originally be beaten with control, standard weapons and enough time. Even I, who hardly used special attacks except when I could not be bothered, started using them actively to clear the game against the new enemies.

I am considering whether to devise unconventional behavior algorithms, such as fleeing off the map and reappearing. The current fighters have been corrected to enter from off screen; leaving and returning is an idea to consider next.

How accurately can it produce a design when I direct it through text alone? Can it reuse existing source code to create working functionality? Can it reuse existing assets to create working visual effects? What percentage of elements, from the design to unit behavior, ignore my instructions? Those were the technical things I wanted to examine.

Unlike my other projects, I also wanted to see how far model capability and the existing source code alone could get, without setting up a separate harness or agents. Enemies reusing the existing helicopters' weapons and movement did actually make it into the game. At the same time, I had to make it redo drafts that broke my instructions to follow the aircraft shapes and the original colors and texture.

Even while working on something with great personal meaning, I kept checking how accurately Codex followed my instructions. I judged the results myself: whether it had read and reused the existing code and artwork properly, and whether it had implemented the appearance and behavior I wanted. I had it remake the incorrect drafts and kept the fighters’ fast entry.

Entry spacing has been corrected, but bunching during combat remains, and I am still considering behavior that leaves the screen and attacks again. Both games also influenced the combat in HeavyMach. I continue to play the game myself to decide what to change next.

The public release covered here is v0.16 as of October 29, 2026. It includes the ground units, HUD editing, new icon and revised formation-entry spacing. Codex wirelessly updated the regular and cheat apps on iPhone and iPad while preserving existing saves and HUD layouts. I did not notice problems when playing on both devices myself.

During the earlier restoration work, Codex passed 33 Mac checks covering combat, input, display, audio and full screen, and found no leaks in apps signed with a development certificate. On October 27, the automatic iPhone and iPad checks each passed 84 of 85 cases, with one display-time comparison failing. Later measurements on October 29 still showed a p95 frame-callback interval of 21.39ms during formation entry on iPad, above the 20ms threshold. Comparison with the pre-change version under the same conditions remains to be done.

A later measurement after HUD editing was added used local ad-hoc signing. Three of six conditions exceeded the 16.67ms processing-time threshold; five had exceeded it before the change. Remeasuring under the earlier development-certificate conditions and investigating window-display delays remain to be done.

Tests with actual Bluetooth or USB keyboards, long-duration heat and memory changes, and difficulty across the entire campaign with added enemies remain. I also reported a new issue in which game sound disappears after using the alarm app and returning, and its cause needs investigation.

Auto-aim status feedback and tutorials 0-0 through 0-4 have not been implemented. An Android icon is ready, and the plan is to reuse the shared C++ core; there is no Android app or APK. The formation now has clearance during entry, but no AI has been added to maintain formation throughout combat.

We focused on restoring the original and making it playable, so we have not comprehensively investigated and fixed build warnings. Some warnings appear to involve APIs whose use is no longer recommended. Their exact cause still needs investigation. I also want a script combining download, resource preparation, build and launch. It would let users choose Mac execution or iPhone/iPad installation and guide them through tools, development teams and signing settings.

I asked Codex whether someone else could obtain the repository and build it. Even with source code, they cannot run it if artwork, audio and maps exist only on my computer. Codex put the required resources into Git and LFS and recorded file lists and hashes. After freshly cloning public main, it checked building and simulator execution with access to the existing working folder and original package blocked.

On an Apple Silicon Mac, use Xcode, CMake 3.20 or later, Python 3 and Git LFS. Clone the repository or download the source-and-resource archive from v0.16 Releases. Fetch the LFS resources, run the resource validator, then open the workspace and choose your own development team and run destination. There are separate run configurations for the regular and cheat apps. The commands and four configurations are in the Apps installation section.

The version history from v0.9 through v0.16 is also collected in GitHub Releases. After v0.16, Codex added a tool to main that packages a tag's code and assets and prepares release notes and hashes. Package and failure-recovery checks were completed; actually publishing a new release with the tool remains for the next version. A tool that takes another person all the way from downloading to building and running the game has not been made yet.

I kept the game code, images, audio and maps together and recorded what we restored and what problems we found and corrected. Task branches merge into develop, develop merges into main, and release tags are placed on main's merge commits. That makes it possible to rebuild that edition from its tag.

There are also eight images, two GIFs and three MP4s with sound extracted from my gameplay recordings. The GIFs play at their original speed. They preserve the original flow of jumping, choosing weapons and upgrading, as well as the new fighters and auto-aim, in actual game footage.

The ground-unit and HUD-editor images were captured separately afterward in cheat apps on physical devices.

I do not intend to make money from this restoration. In my developer’s statement, I wrote: “I want to release the code and assets together so that others can build and play it themselves. I also want to leave a record of how this game was restored and what we corrected.”

I applied MIT to newly written code. The scope for original-derived materials and external code and documentation is separated. Original artwork, audio and game data, and additional artwork, follow the asset-use terms; OpenAL Soft, stb and other dependencies follow the third-party notices. The credits also distinguish the original creators' development and artwork from the work I planned as additions.

If I later have reason to create an App Store account for sass.studio, and the original creators and stakeholders permit distribution, I will still offer it for free.

I do not have much time left, and money is something you only need in this world.