The SAILDART archive cited in the question does indeed contain files from the days of TeX78. A TeX78 job, like a current one, can be "modular", that is, it can be composed of multiple input files instead of just a single file; that is likely the reason that some of the files you looked at are not self-contained.
Both the "vocabulary" and syntax of TeX78 differ from that of the current program. The original manual for TeX78 was published as a Stanford Computer Science report; a scanned pdf file can be found here (thanks to David Purton for finding it). Some early changes that moved in the direction of TeX82 are included in the scan (file "ERRATA.TEX; see physical page 117 of the scanned file); these changes were incorporated in the program and the original "public" manual published by the American Mathematical Society.
Here are some of the significant differences.
The original program was written in SAIL; the keyboard for the SAIL computer had different non-alphameric characters than were on the standard U.S. "qwerty" keyboard. In particular, there were arrows pointing up and down (assigned to do the same things now done by the
_ keys). There was no
&; instead, there was a circled times (called familiarly "splat") that was used as a table column divider. (There may be more.)
As for command names, the first character was required to be a letter, but after that, any other character was permitted (TeX82 allows only letters). The initial letter could be either upper- or lowercase, but after that, the case was not distinguished; thus,
\TEX, although consistency was recommended.
Some basic control sequence names were different:
\chcode instead of
\hjust instead of
\hbox (although that was changed before the AMS manual was released), and the syntax was different from that of TeX82 --
\hbox to <dimen>.
Font access was defined by
\:x, where "x" represents any letter; since the length was uniform, no space was required between the font directive and the text, so
\:bbold would be treated the same as
No list of primitives is given in the manual, but primitives can be inferred from the definitions shown in appendix B ("
basic.tex", which corresponds to "
plain.tex" of TeX82). These are also not referred to as "primitives", so it's not productive to search for "primitive" in the file.
It might be possible to define a parsing mechanism to recognize and process the old syntax, but it wouldn't be easy; it would likely be easier to edit in the necessary changes if a document really needs to be reconstructed.