Castlequest exhumed!

“Lost game” Castlequest (1980) has been found!


In the 1990s, my parents had a subscription to the online service “GEnie”. I fondly remember the “GEnie Games” area. There you could play either 350-point Adventure or David Platt’s 550-point Adventure (and as a kid I never knew there was a difference, which made it extra special and spooky to stumble into one of Platt’s new rooms that I just knew wasn’t there yesterday) — and also several games that are now lost, such as Black Dragon and Castlequest.

Castlequest, written by Mike Holtzman and Mark Kershenblatt (but ported and credited-on-GEnie-to Bob Maples), was a parser adventure similar to, and inspired by, Adventure (but coded differently; Holtzman had never seen Adventure’s actual source code). Castlequest opens like this:

You are in a large, tarnished brass bed in an old, musty bedroom.
cobwebs hang from the ceiling.  A few rays of light filter through
the shutters.  There is a nightstand nearby with a single wooden
drawer.  The door west creaks in the breeze.  A macabre portrait
hangs to the left of an empty fireplace.
The shutters are closed.
There is a silver bullet here.

Similar to Adventure, there are wandering nuisance monsters — but this time they’re werewolves. One of the most memorable bits of Castlequest is what happens when you kill one:

You killed a werewolf. An old gypsy woman
appears and drags away the body.

Anyway, after the demise of GEnie (and The Source and Compuserve), it seemed that the game had been lost forever. But in fact it had not!

Give me the source code

Here’s the code. This code is still under copyright to Mike Holtzman and Mark Kershenblatt, and provided by Mark’s extensive and generous assistance.

In fact, had this code not been copyrighted, it would still be lost today! Funny story…

See, one of the few references to Castlequest on the Internet prior to this decade was on the website of the U.S. Copyright Office. Back in 1981, Mike Holtzman had submitted the program for archival — a service the USCO still provides, even though all creative works are automatically protected under U.S. copyright law whether they’re submitted for archival or not. So it was overkill on Holtzman’s part, legally speaking… but I’m sure glad he did it! The documentation submitted by Holtzman was assigned the number TXu000091366, and remained on file somewhere in the depths of Washington, D.C., for forty years. However, the USCO entry doesn’t say what kind of documentation was submitted. (I assumed it would be a game transcript or walkthru.)

I made contact with Mike Holtzman for just a couple of months in 2016, through an email address which is now defunct. He answered some questions about the history of the game, but I failed to interest him in retrieving the deposit. (He, also, assumed it was just a transcript: it was so long ago that he’d forgotten.)

Then, in 2020, I got an email from the other author, Mark Kershenblatt, whom I did persuade to attempt retrieval! The retrieval process required USCO employees to search the physical archive, which meant that it was very slow: the coronavirus pandemic had reduced the USCO’s headcount and hours. But just a week ago, Mark emailed me to say that the records had been found, copied, and snail-mailed to him!

And guess what? Mike Holtzman’s 1981 copyright deposit wasn’t just a transcript — it was the full Fortran source code and data files for the entire game, all 78 pages, printed out in remarkably high quality! (Naturally, Mike did it on his office printer. Times haven’t changed a bit.)

So the USCO sent Mark copies of all 78 pages; Mark went to Staples, scanned them in, and uploaded them to Dropbox; I downloaded them and…

…well, it turns out that even 16 years after the launch of Google Books, the state of the art of OCR is still ridiculously bad. After a miserably failed attempt to get Google Drive to OCR the PDF, I just went the manual route. Three days of eyestrain later, I had a plain-text version of the code!


  • 1979: The initial “release” of Castlequest at RPI, according to Mike Holtzman’s best recollection. Mark Kershenblatt guesses “between September 1979 and May 1980.” (Holtzman was in the graduating class of 1980.)

  • 1980-02-??: The date on the first page of the source code.

  • 1981-10-22 (“81.295”): Holtzman, now employed at Grumman Data Systems, prints out the source code for the deposit.

  • 1981-11-06: The USCO stamps Holtzman’s deposit.

  • 1983-07-??: Compute! magazine publishes a BASIC game by Timothy G. Baldwin, coincidentally also named “Castle Quest”. Holtzman (and Kershenblatt?) prank the magazine with “a cease and desist letter on a fake legal letterhead”; Compute!’s October issue carries an apology for having “inadvertently used the name Castle Quest,” crediting the wrong authors, which they then have to correct again in their January 1984 issue. Good times.

  • 2016-01-06: I post my “In Search of LONG0751” webpage, with a mention of Castlequest.

  • 2016-08-19: Mike Holtzman first emails me (from an address). He sends four emails through September, filling in the history of the game.

  • 2016-09-06: Holtzman’s last email to me.

  • 2018-07-11: Holtzman’s address bounces.

  • 2020-07-28: Mark Kershenblatt first emails me (from a address).

  • 2020-10-06 (less “a couple of days”): Mark Kershenblatt begins the USCO retrieval process.

  • 2021-03-02: Mark Kershenblatt receives the packet of source code from the USCO.

  • 2021-03-09: I finish transcribing Mark’s scanned PDF and upload the code to GitHub.

Bug bounty program

I certainly introduced some mistakes during that manual transcription process. Therefore I’m offering a “bug bounty” of $5 per error to anyone who reports mistakes in the transcription. Note that my goal is fidelity to the original, so “mistakes” in the original (e.g. the word STAUS on page 1) do not count. Vice versa, if you point out where my fingers accidentally corrected the spelling of a word that was misspelled in the original, you get a bounty! Submit reports to me via email, or as GitHub pull requests.

So, can it be compiled?

Sadly, the code can’t be compiled right out of the box. If you’re an expert in antiquarian Fortran and would like to assist, send me a pull request and/or an email!

Known pain points, as of this writing, are:

(1) The use of character strings such as 'BULL' in DATA statements where an array of two integers is expected. My impression is that this was common in old Fortran code. Adventure famously packed five 7-bit characters into a 36-bit word on the PDP-10, which is why it looks at only the first five characters of each vocabulary word. But gfortran 4.9 doesn’t approve:

Error: Incompatible types in DATA statement at (1);
attempted conversion of CHARACTER(1) to INTEGER(4)

(2) How to read the data files. The original code calls out to a (VS Fortran?) builtin to hook up the HINT file to input unit number 8, and then reads it a record at a time. To modernize this code is probably not difficult, but isn’t something I know how to do off the top of my head.

      CALL CMS('FI      ','8       ','DISK    ',
     2         'HINT    ','CQDATA  ','(       ',
     3         'LRECL  ','80      ','RECFM   ',
     4         'F       ')
      DO 110 I=1,50

I’m also puzzled by the way that many (but not all) strings, both in the data files and in the code, begin with the digit character 0.

(3) The use of builtins for getting the time and date (TOD), initializing the random number generator (RDMIN), generating random real numbers (RDM), and maybe a couple other builtins I failed to notice.

UPDATE: The game can be compiled! See:

Posted 2021-03-09