Methodology

I'm Dean, and this page is how everything in the Codex gets made, from the moment a game arrives in the post to the moment it appears on this site.

All of it happens in my office, home of the one true master of the Mega Drive Codex. (I'm also its only member of staff.)

To show what that looks like in practice, I'll follow one game through the whole process: EA Hockey. I chose this because it's one of the first games archived on the site. It's also a good example because it doesn't have a sound test, which makes recording its music harder than most, and its box held the first insert I ever archived.

Every box, manual and cartridge on this site passed through my hands. Every soundtrack was recorded from an original cartridge on an original console. Nothing here comes from anyone else's scans or recordings.

What one game takes

In September 2026 I timed a whole game, Ghouls'n Ghosts, from start to finish. It took about 15 hours. It has a 52-page manual and a sound test, so a game without one takes longer.

There are 568 games in the PAL catalogue. If every one took as long (and most do), that would be more than 8,500 hours of work.

When a copy arrives

Buying it

I buy every game second-hand, from Vinted, eBay, Facebook Marketplace, CeX and anywhere else one turns up.

Opening the parcel

The first thing I do is photograph the game from every angle, exactly as it arrived. Those photos come back later, when I sell it on.

Then I go through it. Is the manual ripped or written in? Are any pages missing? Is the spine faded? Are there inserts, and are any missing? Is it a pressing I haven't seen before? And I open the cartridge to check the circuit board inside is genuine, because reproduction games are common and some are hard to tell from the real thing.

If a copy is too faded, torn or damaged beyond repair, it goes back or gets discarded, and I find another.

EA Hockey arrived in excellent condition. All 110 manual pages were there, none had been written in, the board was genuine and the registration card was still in the box.

Choosing between copies

Sometimes every copy I can find has a fault. When I have to choose, I take a colour shift over blurred print every time. A colour shift can be measured and corrected. Print that left the factory out of register, with its colours slightly offset from each other, can't be fixed, because the detail was never printed in the first place.

The copy I scan stays the master until a better one turns up. When it does, I scan that one instead.

Pressings

Some games were made in more than one factory, with real differences in the packaging. I scan and document every pressing I have separately, and none is treated as the real one.

Selling it on

Once a game is scanned, restored and archived, I sell it on through Vinted or eBay for exactly what I paid, postage included. I don't make a profit on anything.

Scanning

Everything is scanned at 600 pixels per inch (ppi), in 16-bit colour, on a calibrated scanner, and nothing is cut or damaged along the way.

The standard I follow

I follow the framework of the Video Game Scanning Collective (VGSC scans on archive.org), with two differences. The VGSC unbinds books by cutting them at the spine and scanning the loose pages. This destroys the books. I never cut anything, and nothing that came into the Codex's hands has been damaged. The VGSC also corrects colour by eye, adjusting Levels in Photoshop until the result "looks" right. I correct colour from measurements, which I'll come to under Restoring.

Setting up

Before anything touches the glass, I clean the item with 70% isopropyl alcohol on a microfibre cloth. The alcohol evaporates almost straight away. Then I clean the scanner glass.

The scanner is run from VueScan by Hamrick Software on its professional settings. Every automatic correction and geometric adjustment is switched off, so what's saved is what the scanner saw.

It's calibrated with an IT8 target: a printed card with several hundred colour patches, each one measured precisely by the company that makes it. Scanning that card shows how my scanner's colours differ from the real ones, and VueScan builds a profile that corrects the difference. Without it, every scan would carry the scanner's own colour cast. This gets the colours close to pinpoint accuracy, leaving less work to do later.

Why 600 ppi

Mega Drive boxes were printed by offset lithography, which builds an image out of tiny dots of cyan, magenta, yellow and black. Scan below about 600 ppi and the scanner's grid of pixels clashes with the grid of printed dots, giving moiré, the rippling pattern you see in poor magazine scans. At 600 ppi the scanner picks out the dots themselves and it's a recognised industry standard for archiving.

It also gives me room to repair. At this size I can mend a scratch at the level of individual dots, and the file is detailed enough to reprint a cover at full size on a commercial press.

Keeping it straight

I keep a 90-degree builder's set square on the glass and butt every item up against it, so each scan goes on as square as possible.

How each scan goes

For every scan, VueScan takes a quick preview first. I draw the scan area around the item, leaving a little room to crop, then scan and save. Each one takes about 20 seconds.

The file is a 16-bit TIFF. That means each colour holds 65,536 shades rather than the usual 256, which leaves room for colour correction later without smooth areas breaking up into visible steps.

The order things are scanned in

1. The cover. I slide the cover sheet out of the case and scan it whole: back, spine and front in one go.
2. The cartridge. I open it with a game-bit, the security screwdriver bit Sega's screws need, and scan the front, back and top of the shell flat on the glass. While it's open I clean the circuit board: a silicone scrub block on the contacts to lift off dirt, then contact cleaner and isopropyl alcohol.
3. The manual's covers, front and back.
4. The circuit board. I photograph rather than scan it, on a Canon 6D with its stock lens, in an LED light box like the ones used for product photography. One shot of each side, in RAW format, which keeps all of the camera's sensor data for processing later in Photoshop.
5. The manual's pages, one by one (below).
6. Everything else. Inserts, posters and anything else in the box come last. Anything bigger than the scanner goes on in overlapping sections, which are joined together afterwards in Photoshop.

The manual pages

If a manual is stapled, I take the staples out carefully by hand so every sheet lies flat. Once scanning's done, a long-arm stapler puts new staples back, and an overhead LED light helps me line them up with the original holes.

I scan all the left-hand pages first, then all the right-hand ones. A sheet of black sugar paper goes on top of each page to stop reflections, and heavy books go on top of that to hold it flat. Nothing heavy ever touches the paper itself. Manuals without staples are flattened under the books in the same way.

EA Hockey's manual is 110 pages, and like most EA manuals of the time it's printed both ways round: halfway through, the pages turn upside down and run from the other cover. I had to decide whether to show it the right way up or as printed. The point of the Codex is accuracy, so it's shown exactly as printed.

After the pages came the registration card, the first insert I archived, scanned front and back.

The raw scans

Every scan is kept as it came off the scanner, and it's never edited. All the work that follows happens on copies, so any decision I make can be checked against the original, or done again.

Restoring

The aim is each item as it was on the day it was sold: the artwork as close as possible to what the publisher sent to the printers, on clean paper.

Factory fresh, on purpose

Paper goes back to white and black ink goes back to near-black. That's a deliberate choice. A 30-year-old box has yellowed, and the Codex shows it as it was sold, not as it aged. The aged version is never lost, because it's in the raw scan.

Actual size

Every item is measured, and every digital file is made to exactly that size. If a cover measures 27 cm by 17.2 cm, its file prints at 27 cm by 17.2 cm.

Restoring a cover, step by step

1. Two copies. I make two copies of the raw scan. The raw scan itself is never touched. One copy is where most of the work happens; the other is kept for masking.
2. Straightening and cropping, if the scan needs either.
3. Descreening. I run Sattva Descreen on the first copy. It removes the pattern of printed dots and gives back smooth, continuous colour, the way the artwork looked before it was printed.
4. Colour, by measurement. I place colour samplers on parts of the print that should be white, black and neutral grey, then use Curves to bring each one to its correct value. With the real box in my hands to compare against, every correction rests on a number taken from the print, not on how it looks on a screen. The same correction is then copied to the second copy, since the two are identical.
5. The grid. Older Mega Drive boxes, before the blue covers, have a grey grid printed across the background. Sonic the Hedgehog's box is one. Using the second copy, I mask the grid out, so I can repair the cover without damaging the grid pattern.
6. Cleaning and repairs. Everything that isn't text gets cleaned up and repaired: dust, dirt, scratches and creases, using the Spot Healing Brush, the Remove tool, the Patch tool, brushes and more masking. At 600 ppi I'm working at the level of individual print dots.
7. Bringing the grid back. I run Sattva on the second copy and merge the two together, which puts the original grid back over the repaired cover.
8. The text. Finally I mask out the text, correct the colour around it, and bring the text back in. That keeps every letter sharp, where descreening would soften it.
9. The master. The finished cover is saved as a layered Photoshop master, and every web image of it is made from that.

The cartridge, the inserts and everything else go through the same process.

Faded print

Sun damage is the most common fault on a 30-year-old box, and it's where restoration comes closest to guesswork. So I only ever correct it by measurement.

I sample the faded area and an unfaded area of the same print at several points, each sample an average of 5 by 5 pixels, and compare them channel by channel. If every reading is within about four points of its reference, I correct nothing, because a difference that small is the scanner, not fade.

When the fade is real, the readings show what kind it is. If all three colour channels have drifted together, Curves can correct it. But the usual Mega Drive problem is different: yellow ink fades first, so the blue readings rise wherever yellow was printed. Mortal Kombat II's spine is an example. That needs a correction aimed at the yellow ink itself, which Curves can't do.

The correction stays on its own layer, so what was changed can always be seen and undone. The raw scan keeps the fade for good, because the fade is part of that copy's history.

Text

Broken or faded letters are repaired using the same letter from elsewhere on the same page, copied over the damage, or by filling in what's faded. I never type anything in, and I never use a similar-looking font. Every letter on the finished page is the printer's own.

What I never repair

If something is torn, a piece is missing or there's a sticker or someone's writing on it, I find another copy. I won't invent detail that isn't there.

Checking the work

I check every repair zoomed in to individual pixels, against the real item in my hand and side by side with the untouched raw scan.

The manual pages

Manual pages are prepared together as a batch first. On light pages, with dark ink on paper, the scanner's uneven lighting is evened out and the paper set to white, without touching the ink. Dark pages, with light text on a dark background, only get honest black and white points, because the same treatment would work backwards on them.

Then I go through every page by hand: straightening or cropping where needed, removing dirt and repairing faded text in the same way as on the cover.

Box renders

The box renders, the 3D pictures of each game in its case, come from a template I built in Photoshop from a real Mega Drive case and its clear plastic sleeve, both scanned at actual size. Each game's own restored cover is dropped into the template, so every render uses a real case and real artwork.

Logos

Where a manufacturer's original vector artwork exists, I use it. Otherwise the logo is taken from the scans.

Making the web images

The images on this site are made from the masters by a script, so every one is sized and named the same way. Each master is resized at full 16-bit depth, converted to standard web colour (sRGB) while still 16-bit, and only reduced to 8-bit at the very last step, because doing that earlier causes banding, visible steps across smooth areas of colour.

Then every set of images is checked automatically before it goes on the site: the names, the true pixel sizes, the colour, the compression quality and that nothing is missing.

Manuals as searchable PDFs

Each manual's pages are put together into a PDF at the full 600 ppi. Optical character recognition (OCR), software that reads printed words, then adds a hidden layer of text in whichever languages the manual is printed in, so you can search every manual and copy from it. EA Hockey's comes out as 110 pages, upside-down half included.

Screenshots

Capturing them

Screenshots are the one place I use a flash cartridge, a Mega EverDrive Pro, because it gets me around a game quickly. The console's picture goes over S-Video (a video connection that carries brightness and colour separately) into an I-O Data GV-USB2 capture device. Every soundtrack, by contrast, is recorded from an original cartridge.

What I capture

I capture whatever is unique to that game: its levels and stages, menus, character select screens, maps, option screens or the playing field in a sports game. A bigger game gets more screenshots.

For EA Hockey that meant the menus, the matches themselves and the playoff tree.

The CRT filter

The screenshots you see on this site have my own CRT filter applied in Photoshop, to give the look of the televisions these games were played on. It curves the picture slightly, adds scanlines and the glow of a television's phosphor coating, and softens the edges the way a CRT softened the Mega Drive's dithered graphics (patterns of dots the games used to blend colours). It's there for the look of it, not for accuracy, and I keep every capture without it as well.

Recording the soundtrack

Every soundtrack is recorded from an original cartridge, on an original PAL console, through a signal chain I've measured.

Why record from hardware

The usual alternative is a VGM file (Video Game Music), a log of the commands a game sends to its sound chips. A VGM tells a chip which notes to play. It doesn't capture what the console's analogue circuitry then did with them: the digital-to-analogue converter, the capacitors, the amplifier, all the small things that give a real machine its sound. An emulator playing a VGM imitates those in software, and every emulator imitates them differently. Recording the hardware captures the real thing.

The console

I record on a PAL Model 1 Mega Drive with the VA6 board, and I chose it for three reasons. The VA6 has the YM3438, a version of the YM2612 sound chip with measurably less distortion than later revisions. The Model 1 still has the headphone amplifier Sega removed from the Model 2 to cut costs. And later revisions added a high-frequency noise the Model 1 doesn't have.

It's been fully recapped, meaning every electrolytic capacitor has been replaced. A capacitor more than 30 years old no longer works to specification, so an original console records the sound of an ageing circuit board rather than the sound it was designed to make.

The modifications

It carries a Triple Bypass v2.1, which is three modifications in one:

• Video. A modern THS7374 amplifier and bypass circuitry clean up the RGB picture (red, green and blue sent separately, the console's sharpest video signal), removing the faint vertical stripes called jailbars, sharpening pixel edges and stopping colours smearing.
• Audio. Its Mega Amp 2.0 circuit lowers the noise floor (the background hiss beneath the sound), corrects the stock console's poor mixing levels and gives clean, balanced stereo, tuned using measurements from the MDFourier project.
• The connector. It reroutes the output connector for cleaner signals.

There's also a DFO multi-region modification, which I keep at 50 Hz, the PAL standard, for every recording. That way every game plays at the speed and pitch it was sold at in Europe.

Power

The console runs from a Triad Magnetics WSX090-2500 regulated power supply. The stock supply put electrical noise into the audio. I could remove that noise afterwards in software, but I'd rather it was never recorded in the first place. I want as raw a capture as I can get.

The signal chain

The sound comes out of the console's headphone socket, through a 3.5 mm to twin 1/4-inch TS cable (TS being the standard jack plug on studio and instrument cables), into a Motu M2 audio interface, and onto a MacBook Pro running Adobe Audition. It's recorded at 96 kHz and 24-bit (96,000 measurements of the sound every second, each finely graded), far more detail than the music needs, which leaves room for every later step to work without losing anything.

Measured with MDFourier, a tool that compares a console's audio against reference recordings, the chain's noise floor is -84.11 dBFS (decibels below full scale, the loudest a digital recording can go): 14.60 dB quieter than the community's reference hardware.

A recording session, step by step

1. Warming up. I switch the console on and leave it for 15 minutes, so its components settle at their working temperature.
2. The room. Nothing else electrical is plugged in in the room. The MacBook runs on battery, because mains power brings electrical interference with it.
3. Connecting up. The Motu M2 goes to the MacBook, the cables go to the console and the console's volume slider is set. My headphones go into the M2, so I can hear exactly what's being recorded.
4. Finding the music. With the television on, I find the sound test, or wherever the music plays, and work out which buttons control it. Then the television goes off and, unless a game can't be recorded without it, the video cable comes out.
5. Setting the level. I make a rough recording of the game's music to find its loudest track, and set the level so that track peaks at around -12 dBFS. That's a widely used working level in recording studios: high enough to keep the music well above the noise, with plenty of room left for editing. Once it's set, the volume isn't touched again for the rest of the session.
6. The noise print. I record 10 seconds of the game's silence. That's the noise print: a sample of the background interference in this particular session, which the editing later removes.
7. The tracks. Then each track is recorded in turn.

A session lasts as long as the music does, anywhere from 30 minutes to 2 hours, depending on how many tracks a game has and whether they can all be recorded in one go.

Recording every loop

Most game music loops. Tracks under 30 seconds are heard three times on the finished soundtrack and longer tracks twice, and I record every one of those loops from the game itself. I don't record one loop and copy it, as some recordings do. Each track runs until 15 seconds after its last loop has finished, to leave room for the fade.

I know a track has looped because it starts over. If I'm not sure, I let it run longer and check it in the edit. I'd rather get it right at the console than fix it afterwards.

In Road Rash II, one track seems to start over at about the four-minute mark. Only when I checked it did I hear that the drums and bass were slightly different: it hadn't looped yet, so the whole track had to be recorded again.

Games without a sound test

Without a sound test, the music has to be recorded while the game is running. I play through, use a pause screen where the music keeps playing when paused or use Game Genie codes.

EA Hockey is one of these. Its four tracks play at different points of the game, so each was recorded where it naturally plays. The game drops into its demo mode if it's left alone, so while each track recorded I had to keep pressing the d-pad at regular intervals to stop it.

Road Rash II's race music only plays while you're on the track, and if you sit still for long, the police arrive, the sirens start and the game ends. I found a Game Genie code that stops the police appearing, so each track could play out in full.

Flashback does have a sound test, but it leaves out a lot of the game's short danger and level cues, which only play in the game itself. Those cues carry on playing if the game is paused just as they start. So with careful timing, I'd start recording just before a cue began, then pause the game to let it play out in full.

As far as I know, mine is the only recording of Flashback with every cue intact.

Naming the tracks

Track titles come from the most authoritative source there is: an official soundtrack release, the game's own sound test, its manual or titles shown in the game. Some I've confirmed by asking the composers themselves. When none of those gives a title, I name the track after the level or stage it plays on. The title music of a game is named after the game, not called "Title Screen".

EA Hockey's titles came from its composer, Rob Hubbard: "EA Hockey", "Playoffs", "End of Period/Game" and "EA Hockey (Reprise)".

Editing the soundtrack

Editing takes away what the recording chain added, and nothing the console produced.

Raw and finished recordings

Every recording exists twice. The raw recording is the untouched 96 kHz, 24-bit capture. I never alter it and never publish it, and I keep it so every later decision can be checked against what the console produced. The finished recording is what you hear, made by the steps below.

Adobe Audition handles recording and editing. iZotope RX 12 handles everything else.

Editing a soundtrack, step by step

1. Checking the recordings. First I make sure every loop is there in full, and there's no stray sound just before or after each track.
2. Azimuth correction. I open every track in RX and run its azimuth correction. The recording is analogue, and the Motu M2 has separate level controls for the left and right channels, so the two never match exactly. Azimuth correction measures both channels and lines up any difference in their level and timing.
3. Removing the noise. RX's Spectral De-noise learns the 10-second noise print from that session, then removes that interference from every track: mains hum, electrical noise, the hiss of the cables. None of it is in the game, and none of it came out of a television in 1992.
4. The start and the fade. Each track gets half a second of silence at the start. Then I fade out the final loop, starting about four seconds before the loop point and running for 11 to 12 seconds. Tracks that end naturally aren't faded.
5. The high-pass filter. A filter removes everything below 30 Hz. Our hearing reaches down to about 20 Hz, but in these recordings what sits below 30 Hz is mostly rumble and DC offset (a constant shift in the signal) rather than music. It adds nothing you can hear, but it moves speaker cones and uses up headroom (the space between the loudest peak and the point where the sound distorts).
6. Resampling. Each track is converted from 96 kHz to 48 kHz for listening.
7. Setting the level. RX's Waveform Statistics shows the loudest peak anywhere in the whole soundtrack. Every track is then raised by the same amount, so that peak lands at -1.0 dBTP (true peak). If the loudest peak is at -11, every track goes up by 10 dB. This always happens after resampling, because resampling can push peaks slightly higher, and setting the level first could let them clip. Sound-effect collections are levelled separately, so a loud explosion can't decide the level of the music.
8. Dither. Each track is dithered to 24-bit, a tiny amount of noise that keeps the quietest passages clean when the processing is reduced to the file's final bit depth, and saved as an uncompressed 48 kHz, 24-bit WAV.
9. Tagging. In Mp3tag, every file gets as many tags as can be filled: title, artist, album, year, track number, genre, comment, album artist, composer and the cover art. Titles and capitalisation follow the Apple Music Style Guide, a recognised industry standard.
10. Encoding. XLD turns the WAVs into FLAC files for the archive. FLAC is lossless: it reproduces the audio exactly, just in a smaller file. (I also make ALAC copies, Apple's lossless format, for my own music library.)

Why fade like that

Our ears judge loudness by proportion, not by fixed amounts, so a sound has to drop by the same proportion every second to seem to fade evenly. A logarithmic fade does exactly that. A straight-line fade seems to hang near full volume, then fall away suddenly at the end.

If the fade started right at the loop point, you'd hear the track begin again at full volume and then get pulled away. Starting it about four seconds earlier means the music is already easing off as the loop comes round.

Why level the whole album at once

If every track were pushed up to the same peak, a quiet ending theme would come out as loud as the title music. Raising every track by the same amount keeps the differences between them exactly as the composer wrote them.

What I don't do

No EQ, no compression, no stereo widening, no reverb and no attempt to make anything sound "better" than the console did.

The cover art

Each soundtrack's cover is made by the Codex from the game's own cover or key art.

Listening

Every finished soundtrack is checked on ADAM A7V studio monitors and Audio-Technica ATH-M50x headphones. Then it gets played the way people listen to music: in the car while I'm driving, on a phone speaker, through cheap earbuds. If it doesn't hold up on all of them, it gets recorded again.

Uploading to YouTube

YouTube is where the soundtracks are easiest to listen to, so each one goes up as an album: a video for every track, gathered into a playlist.

One video per track

People use YouTube for music much as they'd use any streaming service, picking the track they want. A video for each track lets them do that.

Making the videos

Every video is rendered in Adobe After Effects from one template, so every soundtrack in the Codex looks the same. It shows the album art, the track's details and a waveform that moves with the music. The waveform's colour is the average colour of the album art, measured by a script. Each soundtrack also gets its own thumbnail.

The titles, descriptions and tags are all generated from the game's own record, so the details match on the Codex, YouTube and archive.org. Every description links back to the game's page here and to the lossless files on archive.org.

Before and after uploading

Before anything goes up, I check the videos have rendered properly and play without problems, and that every script has done its job. Then the videos are uploaded into a playlist in track order, and the playlist gets a pinned comment linking to the full soundtrack.

After uploading, I check everything arrived, then watch for copyright claims, which often need appealing. False ownership claims, where someone with no right to the music claims it anyway, happen a lot.

Uploading to archive.org

archive.org is the permanent home for every soundtrack. Each one goes there first, before it goes to YouTube.

Why archive.org

YouTube is the convenient place to listen. archive.org is where the lossless files are kept for good, free for anyone to download. If it ever went down, the YouTube uploads and my own copies would still hold the recordings.

What goes into each item

Each soundtrack's archive.org item holds every track as a FLAC file, the cover art, a README explaining how it was recorded, a metadata file and a checksum file. archive.org then makes its own MP3 copies for listening in the browser. The item's description lists the tracks and the hardware they were recorded on, and links back to the game's page on the Codex.

The checksums

The checksum file holds a SHA-256 fingerprint of every FLAC file: a long code worked out from the file's exact contents. Change one bit of the file and the fingerprint no longer matches, so anyone can check their download is identical to mine.

Licence

The music belongs to its copyright holders and is archived for non-commercial study. The README and metadata files I write are licensed Creative Commons Attribution 4.0, so anyone can reuse them with credit.

The VHS tapes

Alongside the games, I preserve VHS tapes from the Mega Drive era: Sega's promotional tapes, magazine cover tapes and anything else on tape about the Mega Drive. Most run between 15 and 60 minutes.

Cleaning the tape

A tape that's sat in a loft for 30 years can carry mould, and playing a mouldy tape can ruin both the tape and the machine. So every tape is cleaned first, on a VHS is LIFE mould cleaning system, which wipes the tape itself with isopropyl alcohol as it winds from one spool to the other.

I run it slowly, so the tape isn't strained, and clean both sides with two full passes. Depending on the tape's length, that takes anywhere from 10 minutes to an hour. One tape had mould so bad it needed several passes, and the plastic shell itself needed a deep clean too.

Playing it

Tapes play on a JVC HR-S7722EK S-VHS deck, and I clean its heads before every recording. Its time base corrector (TBC) and noise reduction are both switched on. The TBC steadies the timing of the picture signal, so the lines of the image stop wobbling and jittering.

Capturing it

The deck feeds the same I-O Data GV-USB2 capture device used for screenshots, and the tape is recorded in SwiftCapture at 720 by 576 pixels and 25 frames a second, the PAL standard, as Apple ProRes 422 LT video (a high-quality format made for editing) with 24-bit, 48 kHz audio.

Restoring and publishing

The only thing I change is the colour, corrected in DaVinci Resolve. Then the finished tape goes up on the Codex's YouTube channel.

Putting it all on the site

Everything on a game's page comes from a primary source wherever one exists. When a fact can't be confirmed, its field is left blank.

Magazine scores

I went through every issue of every magazine I could find, page by page, in archived scans of the originals, and typed each review score into a spreadsheet by hand. That took about six months, and it carries on as more magazines are scanned and put online.

Every score is recorded with the magazine it came from and the issue it was in, so any score on this site can be traced back to the page it was printed on. One-off specials and books had no issue numbers, so those are left blank.

Scores stay on the scale the magazine used. A score out of five stays a score out of five, because turning it into a percentage would claim a precision the review never had. The averages on the site are worked out only from magazines that scored out of 100, for the same reason.

Release details and credits

Catalogue numbers and credits come from the games themselves wherever possible: the box spines, the credits pages in the manuals and the staff rolls at the end of the games, which means playing through to the ending, or using cheat codes to get there. Only when a game doesn't say do I use secondary sources such as Sega Retro, the LaunchBox Games Database or Wikipedia, and only as a last resort.

What each page shows

Every game is in one of three states, worked out from what exists rather than set by hand. A Wanted game has no copy in the collection yet. A game In preservation has been bought and scanned. A Published game is complete and on the site. A blank field means the answer isn't known yet, not that there isn't one.

Where the limits are

Most of what I do rests on measurement. These are the parts that rest on judgement.

Faded colour is reconstructed

Ink that has faded is gone, and nothing brings it back. My correction is measured against unfaded print on the same item, so it's based on evidence, but it's still a reconstruction.

Loop points are judgement calls

I set loops by ear, from the best recording I can make. Another careful listener might place some of them differently.

"No sound test" means I haven't found one

Every sound test code on this site has been tested on real hardware, and works. But a game listed as having no sound test only means I haven't found one. If you know of one, tell me.

Screenshots are filtered

The screenshots carry a CRT filter for the look of it. The unfiltered captures are kept.

The catalogue isn't finished

Games I don't have yet are marked Wanted, not presented as finished.

It's one person's work

The Codex is one person's work, which means one person's mistakes. If you spot one, tell me, and you'll be credited for the correction.

One game, start to finish

So when you open EA Hockey's page, here's what you're looking at. A copy checked page by page and board by board before I'd keep it. A cover, cartridge and registration card scanned square against a builder's set square, then restored to measured colour with the real box in my hand. A 110-page manual unstapled, scanned page by page under black sugar paper and a stack of books, re-stapled and kept upside down halfway through because that's how EA printed it. Four tracks recorded from the original cartridge, warmed up for 15 minutes first, with my thumb tapping the d-pad to keep the demo mode away and named by the man who wrote them.

Then multiply that by 568, and you've got the Mega Drive Codex. Happy trails!