What's the quickest possible draw?

I'm not sure if I understand you correctly. In any event, when I say 'forced stalemate' I mean a situation like your first position in which the players can still move but a stalemate is unavoidable. At first glance it had seemed to me your 2nd one was also leading to stalemate.

Here both can move but stalemate is inevitable:

A better example, but whether both can move depends on the rule set assumed. If it's FIDE rules or occurs as a position in a retro-problem under WCFF rules, then the game has finished under the dead position rule, so neither player can move. (Obviously they could physically, but it wouldn't be part of the game and there's no reason in that case to apply any rules; they can play tiddlywinks on the board if they like.)

By the way, my 2nd. position is also leading to stalemate in the sense that, absent the dead position and agreed draw rules and the other rules you exclude, all complete games continuing from the position would necessarily end in stalemate. (I'm counting infinite games as incomplete even though no further moves can be added; no position or result is arrived at, not to mention nobody would have the stamina.)

From the original post:

"The 4 move 3-move-repetition draw, and the 10 move stalemate are both well known. I'd like to know what's the quickest possible forced draw without using the 3 move repetition or stalemate rules. You must reach a position in which it's impossible for either player to deliver a checkmate."

(My highlight.)

NO, there is, knowingly and willingly - according to Andrew Buchanan - no definition of what counts under that heading and what not. The idea is that it is "self-evident" (untrue of course) and today's definition will be overtaken by tomorrow's innovation (true).

My opinion is that the IDENTITY classification needs to be replaced by a SOLVING classification:

1. (+R) = solve by retro conventions
2. (-) = solve by self-evident conventions but only if really beyond all reasonable doubt, for instance by stipulation type. like #2=non-retro, SPG#6=retro if you want that.
3. (-R) = solve without retro-conventions

Just examples to drive home the idea!

Why am I not surprised?

Even more interesting is inevitable "win" positions, like this:

Should white automatically win here if he doesn't move?

Do you know any long ones?

By "irrelevant" I simply mean that because the Dead Position rule doesn't apply in such cases, it has no bearing on the solutions of these proof game compositions. IF the rule applies, then clearly that would affect the solutions, but that's besides the point since the rule does not apply. As implied in the WFCC Article, many retro-problems do specifically employ the DP rule, so obviously in such cases the rule is relevant.

The basic reason behind the WFCC Article for Dead Positions is that there are a lot of interesting compositions that wouldn't work with the DP rule in place, while there are also many interesting retro-problems that specifically use the rule. So the aim of the Article is to keep both types valid while setting some default assumptions for convenience. The last thing the WFCC wants is to invalidate some interesting composition types based on some rigid interpretation of its Codex.

The OP is posing an interesting synthetic game problem (a relative of proof games): how to attain a Dead Position in the quickest way from the opening array. No problemist will misinterpret the Codex to suggest that such a problem task is invalid, regardless of what "retro-problems" means precisely.

Hi cool question,
I think the answer is 15.0.

Francois Labelle made a computer analysis of massacre games, see: http://wismuth.com/chess/problems-massacre.html. At ply 31 (that's 15.5) he identified 161 positions with 28 captures. Clicking through, to see the list, there are 11 I think where only 2 bishops with the same parity remain, i.e. the game is dead. The number to the left is the number of proof game which result in this position, with or without the DP rule. The DP rule only reduces the number of PGs in the 11th example:

9 wKf2, wBf7, bKe7, bBc8
391 wKf1, wBb2, bKe7, bBf8
391 wKf1, wBb2, bKf7, bBe7
675 wKd2, wBf1, bKg7, bBc8
1139 wKf2, wBg8, bKe7, bBc8
1458 wKc1, wBf1, bKf7, bBc8
1625 wKd2, wBf1, bKf8, bBc8
2547 wKd2, wBf1, bKe7, bBc8
5163 wKf1, wBd2, bKe7, bBf8
5163 wKf1, wBd2, bKf7, bBe7
11103 (15477) wKd2, wBf1, bKf7, bBc8

Determining an exact game to result in these positions is a trickier business for us amateurs. Most of the main engines (Stelvio, Natch, Euclide & Jacobi) are not optimized for massacres. Popeye is probably the best, but is still likely to take a lot of cycles.

So the most interesting example is the 11th. I just spotted that out of all the 161 4-piece positions reachable in 15.5 moves, it is the only one where the number of solutions is greater with DP rule switched off. What this means of course is there is at least one DP in 15.0 moves, in which the last move is forced capture of bQ. This is most likely to be wKe1, wBf1, bKf7, bQd2 bBc8. It's possible instead or as well, that it's wKd1. This is really cool, and suggests to explore the 11th example above, and identify the 15477-11103=4374 PGs that are dead at 15.0.

Francois was not aware of this position, so we've agreed to co-own it. It's not a unique proof game though, so interest in it is finite.

I don't know what your attitude is to a position which is dead because inevitable stalemate is otherwise looming. If you're ok with that, then in 10.0 (A.Buchanan after S.Loyd) have 5bnr/4p1pb/4Qpkr/7p/7P/4P3/PPPq1PP1/RNB1KBNR. White can only capture the checking Q, after which it would be stalemate. This is just half a move longer than the famous actual stalemate which Loyd found at 9.5. There are 28 ways to reach this. A unique proof game takes 12.0 moves.

It's certain that there is no definition of "retro".
I've have never thought of it as "self-evident". To me, the right approach is some kind of "case law", where a ruling is interpreted in various contexts, and those serve to guide in the future. Particularly, need to have a flexible framework based on interpretation as a foundation for increasing appearance of fairy retros.

Some things which are to me definitely retro:
(1) Composed proof games (both unique & non-unique, & also synthetic games)
(2) Problems where game history is explicitly or implicitly asked for
(3) Retractors
(4) Problems relying on Codex Article 15 (if retropat, it must be the other side that starts)
(5) Proof that castling rights have been lost, or e.p. right is maintained.
(6) RS, PRA, AP, etc
(7) Assumption of DP rule allows for retro behaviour to be enabled, but without that assumption, can't access that. (E.g. DP means last move would be illegal, so Art 15 would trigger.)

Things which are not retro:
(1) "Simple" check that a position is legal
(2) "Simple" check that castling is ok, or e.p. is not ok (i.e. castling & en passant conventions are not retro - they serve to protect non-retro positions from irrelevant retro concerns).

(3) Adding / removing units

Honestly that suffices for 99% of all cases. However there are some grey areas, which serve to indicate that this is not cut and dried:
(1) Somewhat "Non-trivial" retro required to prove legality, castling rights ok or e.p. not ok, but still not intended.
(2) Retractions illustrate some cool effect, which is not required for soundness of forward play.

These, and (7) from the list above are really saying that it's all about *intention*. Codex Article 17A has been magnificent at allowing different communities to co-exist. "Good fences make good neighbours", as they say. And the idea that DP rule relies on as rich a concept as "retro" has provided large and fertile design space for this tiny rule.

All regulators in all walks of life must make exercise judgement to determine down to what level to specify. And I think GMs Kjell Widlert & Michel Caillaud made exactly the right call in 2015.

The current record is 30 pieces (A.Buchanan):

@anselan Hi Andrew, Long time no see in these parts! Your modification of the Loyd is a funny way to achieve a DP quickly, and it's a nice record at 10.0 moves. Quite a coincidence that you published that problem just a few months ago and now someone decided to open such a discussion. The OP seems newish to constructing proof games but clearly is talented.

What are you saying, Rocky?

For almost all PGs, the DP rule is indeed irrelevant. There are a few massacre PGs only which don't work because of DP, but these are balanced by a few massacre PGs which become sound only because of DP. Look at the details in http://wismuth.com/chess/problems-massacre.html if you want. Most PG composers are not interested in massacres, so this is somewhat of a corner case.

But if DP matters, the convention is there. There is actually a lot of design space for PGs which combine DP rule with 3Rep & stalemate. For example here's one recently published in The Problemist by A.Buchanan:

PG in 12.5 with the additional constraint: "Game over!"

Or here's another from Idee & Form by the same person:

PG in 13.5
Bottom line: DP has pretty much just upside for PGs

When I said that the DP rule is irrelevant "in these cases", I was referring to the massacre PGs like the Wilts & Geissler I quoted. I'm not familiar with the exceptional cases where the DP rule is required for a sound massacre PG. Sounds interesting! I suppose this means there's a "try" sequence that fails because it would end the game with a DP.

Yes exactly. I have three examples of this, all by Francois Labelle:

PG in 15.5
Unique solution (forgive the German piece notation):
1. d4 Sh6 2. Lxh6 d6 3. Lxg7 Le6 4. Lxf8 Lxa2 5. Lxe7 Lxb1 6. Txa7 Lxc2 7. Dxc2 Tg8 8. Dxc7 Txg2 9. Dxd6 Txh2 10. Dxb8 Txf2 11. Dxb7 Dxd4 12. Txh7 Dxb2 13. Txf7 Txf1+ 14. Txf1 Dxe2+ 15. Kxe2 Txa7 16. Dxa7= patt

Forbidden solutions (one of the 32 DP tries): 
1. d4 Sh6 2. Lxh6 d6 3. Lxg7 Le6 4. Lxf8 Lxa2 5. Lxe7 Lxb1 6. Txa7 Lxc2 7. Dxc2 Txa7 8. Dxc7 Tg8 9. Dxd6 Txg2 10. Dxb8 Txh2 11. Dxb7 Txf2 12. Dxa7 Dxd4 13. Txh7 Dxb2 14. Txf7 Txf1+ 15. Txf1 Dxe2+ DP 16. Kxe2=? would be patt, as would 16. Sxe2=

PG in 17.0

Unique solution: 
1. d4 e5 2. Lf4 exd4 3. Lxc7 Dxc7 4. Dxd4 Dxh2 5. Dxg7 Dxg1 6. Dxg8 Txg8 7. 
Txh7 Txg2 8. Txf7 Txf2 9. Txd7 Txe2+ 10. Kxe2 Dxf1+ 11. Ke3 Dxb1 12. Txb7 
Dxa2 13. Txb8 Dxb2 14. Txa7 Dxc2 15. Txc8+ Dxc8 16. Txa8 Dd8 17. Txd8+ 
Kxd8 DP

Forbidden solutions (one of 427 DP tries): 
1. d4 e5 2. Lf4 exd4 3. Lxc7 Dxc7 4. Dxd4 Dxh2 5. Dxa7 Dxg1 6. Dxb8 Dxf2+ 
7. Kxf2 Txa2 8. Dxb7 Txa1 9. Txh7 Txb1 10. Txh8 Txb2 11. Txg8 Txc2 12. Txg7 
Lxb7 13. Txf7 Lxg2 14. Txd7 Lxf1 15. Kxf1 Txe2 16. Kxe2 Kxd7 DP game over 17. Ke3?? Kd8??

PG in 17.0
I don't know the solution or DP tries for this 3rd one, but I am told there are 6216 candidates of which exactly 1 avoids premature demise. If someone has some computing power, would appreciate to see.

@Rocky, Thanks - nice to encounter you. Just by chance I wander into chess.com today after long absence: either a coincidence or chess.com guys talk about DP every day I agree about OP: it is far from trivial to find a massacre down to insufficient-material DP in just 16 moves

You read it here first: Dmitri Turevski has just kindly run Popeye and confirmed in 1.932 seconds (blindingly fast) that as expected, the following position can be reached in 4374 ways in 15.0 moves. (The other candidate with wKd1 cannot be reached.)

PG in 15.0
Record holder for shortest way to draw that does not depend upon pat or 3Rep.

An example solution is: 1.e2-e4 d7-d5 2.e4*d5 Qd8*d5 3.Qd1-h5 Qd5*a2 4.Qh5*h7 Qa2*b1 5.Ra1*a7 Qb1*c2 6.Ra7*a8 Qc2*b2 7.Qh7*g8 Rh8*h2 8.Ra8*b8 Rh2*h1 9.Rb8*b7 Rh1*g1 10.Rb7*c7 Rg1*g2 11.Qg8*g7 Rg2*f2 12.Qg7*f8 + Ke8*f8 13.Rc7*e7 Rf2*d2 14.Re7*f7 + Kf8*f7 15.Bc1*d2 Qb2*d2 +

Right! And as long as the "case law" has not been produced, widely published and agreed (in the absence of an authorized lawyer) the proposed law will need to be appended to the individual case stipulation - by what I named "solving identity" - though there are certainly better terms. Practically (at this stage) the terms "case law"and "solving identity" refer to different stages of the same thing. The "-" (blank) "solving identity" would assume that some agreement was reached on the applicable case law which therefore has become "self-evident". The other "solving identities" refer to either overrides of the default case law or cases for which no laws were yet established - but possibly will in the future.

off-topic: Did you know that "@jetoba" (FIDE arbiter) will finally bring the item to the FIDE table which caused the WFCC to eliminate the dead rule for non-retro compositions? Caused by unavoidable 5R and 75M automatics on the horizon which would not be dead without them!

That's gotta be the most complicated dead position there is lol