Earliest known eval interpolation

[BB+]

Since men (in general) desire to be in vogue with their chatterings, the cognoscenti now refer to a Phalanx-like mixing of middlegame/endgame scores, rather than Fruit-like.

Here is prior art:

ICCA Journal 14/2 (June 1991), pg 89,
Report on the 11th World Microcomputer Chess Championship
THE PARTICIPANTS (programs, that is)

The King: [...] Separate evaluation terms are used for opening/middle/endgame with smooth transitions between them by using 3 weight which sum to 100%, giving a fuzzy definition of the phases.

It is also not quite correct to refer to "Phalanx-like" in the Fruit context, as the former mixes the amounts only in (approximately) the middle third of the game.

[hyatt]

There were others. In Cray Blitz (and blitz) which dates back to the 70's, we "interpolated" some of the eval terms. I did not think that things like centralization and such should change, but clearly king safety needs to "phase out" as material (and danger) decreases. We used almost the same exact formula I use in the current "full interpolation model" with cray blitz to phase king safety out. The formula was something like

score += king_safety_score * opponents_material / 31 (max material for one side, Q=9, N=3, not counting pawns).

For endgame stuff, such as passed pawns, distant passed pawns, candidate passed pawns, isolated pawns/pawn islands and such, the inverse was done:

score += passed_pawn_scoring * (31 - opponent_material) / 31

As time went on, I ended up doing that in multiple places, and decided to go "fruity" in 05 or 06, and just compute two separate scores, one assuming all pieces are on board, one assuming none are on the board, and then interpolating between them based on material to smoothly transition from hard opening to simple endgame. This approach was used by most everyone in the 70's in some shape or form.