JavaRush /Courses /Claude Code SELF /“Done” That Isn’t Done

“Done” That Isn’t Done

Claude Code SELF
Level 9, Lesson 1
Available

1. A middling spec — and a “done” report

Tommy goes to Alex — no need to drag Sam in right away; first they need to figure out what they are even ordering.

“So, chess for two. Where do we start?”

“With the task,” Alex replies out of habit. “Sam won’t build anything out of thin air; he needs a brief. Let’s sketch a short spec — not a novel, just enough to make the game clear.”

He scribbles down literally three lines:

Chess for the browser, a two-player game on one screen.
Board, pieces, alternate turns — White, Black.
Button “New Game”.

“That’s it?” Tommy glances sideways. “Nothing about how the pieces move? I mean, the knight in an L, the bishop diagonally…”

“And why would we need more?” Alex shrugs. “Everyone knows chess. Sam knows it too. If we write ‘chess,’ he’ll figure out how the pieces move himself. It’s not his first day.”

That sounds reasonable. Tommy gives the spec to Sam. Sam gets to work, rustles around — and soon reports: done, the chess game is built.

Tommy opens it — and gets hooked. The squares are even, the pieces are in place, the board is black and white, it shimmers, and the “New Game” button gleams.

“Beautiful,” he breathes out. “That’s Sam for you, that was fast. It’s done, right?”

2. The check shows: it doesn’t work

There is no one to brag to; Oliver has left — and Tommy decides to just play against himself and test it out. He moves a white pawn. Then he switches to the black side to respond — and, by habit, reaches for a black knight.

He reaches — and his hand itself moves the knight not in an L, but in a straight line, three squares to the side. Just to check. The knight obediently lands on the new square.

“Whoa.” Tommy freezes. “That’s not how a knight moves.”

He tries more: he moves that same black knight a second time in a row, even though black just moved and it has been White’s turn for a long time already. It goes through. He moves a black bishop in a straight line, like a rook — sure. He places his own king directly under attack from an enemy rook — and the game does not react at all, no “check, you can’t do that.” He hurls pieces around the board any way he wants, and the board obediently swallows everything — no move rules, no turn order, no check.

Tommy leans back, gloomy:

“This isn’t chess. It’s just a board where pieces slide around wherever you want. And Sam said it was ‘done.’”

He goes to Sam — in a polite way, to explain what is wrong:

Listen, there are bugs here. The knight moves randomly, turn order is not respected,
it doesn't detect check. Fix it.

Sam rustles around again, fixes something — and once more, brightly: done, fixed it.

Tommy opens it with hope, sits down to play — and after half a minute, the same picture. The turn order became a little stricter, but the knight still jumps incorrectly, castling is missing, checkmate is not detected. Again “done” — and again not done.

“What the heck is going on?” Tommy snaps. “Second time ‘done,’ and second time half of it doesn’t work!”

Alex joins the conversation — calmly, as if he had been waiting for this moment:

“That’s where you ran into something important. Notice? Sam said ‘done’ twice. And both times he wasn’t lying — he honestly thought he was finished. It’s just that ‘done’ from an AI and an actually finished game are two different things. The AI says ‘done’ when it believes it is done. Whether it really finished or messed up — that still needs to be checked and checked again.”

3. First law: “looks done” ≠ “works”

Tommy digests that. And the most annoying part is that he believed it himself. Twice. He looked at the even squares, heard “done,” and decided the job was finished.

“That stinks,” he says. “It really looks like real chess. And Sam was so confident — ‘done.’”

“That’s where all beginners get burned.” Alex grows serious. “They confuse ‘looks done’ with ‘works.’ A pretty board is not working chess yet. And ‘the AI said done’ — that is still not done.”

He lets that sink in and adds:

“Burn this into your memory, because this thing will keep popping up all over the place in real work: a ‘done’ report by itself is not proof. Not from Sam, not from anybody. Proof is when you sit down yourself, go through it, and see that it works. Just like you did when you played.”

“Looks done” ≠ “works.” The agent says “done” — that means “I wrote it,” not “I checked it.” Checking by eye, one frame: the squares are even, the pieces are in place, the “New Game” button gleams — it seems done. But at the start, a broken board and a working one look the same. Playing like a player, move by move: White moves a pawn — fine; a black knight moves in a straight line — it goes through; the same knight moves a second time in a row — it goes through; a bishop moves like a rook — it goes through; the king steps into attack — the game stays silent. Verdict: it doesn’t work. A “done” report is not proof, not from the agent, not from anybody. Proof is when you sit down, go through it and see for yourself that it works “Looks done” ≠ “works” Agent: “done” means “I wrote it,” not “I checked it” CHECKING BY EYE one frame PLAYING LIKE A PLAYER a game, move by move squares are even pieces in place the “New Game” button gleams at the start, broken and working boards look alike Seems: “done” 1 2 3 4 5 White: pawn forward black knight — straight line, three squares same knight — a second time in a row bishop moves like a rook king under attack — game stays silent In fact: it doesn’t work A “done” report is not proof. Not from the agent, not from anybody. Proof is when you sit down yourself, go through it and see that it works.

“Well, at least I checked.” Tommy snorts unhappily. “I sat down to play — and it all fell apart. But I was just clicking randomly. How do you check it properly, so you know for sure?”

4. A rule list — the ruler we measure with

“Then let’s break down what you were just doing.” Alex catches the moment. “You weren’t checking ‘in general.’ Look: you checked the knight move — bad. You checked turn order — bad. You checked check — and it didn’t see it. You checked some things, and you didn’t touch others at all — castling, for example, or en passant. So you still don’t know whether they work or not.”

“True.” Tommy realizes. “I was clicking whatever came to mind. I could have missed half of it.”

“That’s why we’ll write down everything real chess is supposed to be able to do. A list. And then we’ll check it point by point, without forgetting anything.”

Alex writes down next to the project:

WHAT REAL CHESS MUST BE ABLE TO DO (checklist):
1) 8×8 board with correct coloring
2) pieces set up correctly
3) alternate turns, White goes first
4) each piece has its own movement rule (knight in an L, bishop diagonally...)
5) castling
6) check and checkmate
7) details: en passant, a pawn's first move of two squares

“Now you don’t have ‘well, it seems to work,’ but a ruler,” Alex says. “You take any version of the game and measure it against the points: this one is there, that one isn’t. And you check it not with your eyes from the picture, but by going through the game like a live player — from the first move to the end of the match. The path a real user takes is called a scenario: start a game → move white → pass the turn → move black → and so on until checkmate. Until you personally trudge through the whole scenario to the end, you won’t know where it trips. Want to run your ruler over Sam’s build?”

They run through the list, honestly, step by step, through the game. Board, coloring — there. Pieces in place — there. Alternate turns — well, not always; Sam did not quite finish that one either. And the movement rules, castling, check and mate, details — zeroes.

“Only two out of seven points work,” Tommy sums it up. “‘Done,’ huh. Done to a quarter.”

“But now you know exactly how done it is.” Alex is not gloating. “Not ‘kind of bad,’ but specifically: it passes two points and fails five. With that, we can already work. Now we just need to figure out how to get those five missing things into the game.”

The ruler: what the agent’s build passes. The agent said “done.” We apply the checklist. Visible in a screenshot: 1 — 8×8 board with correct coloring — yes; 2 — pieces set up correctly — yes. Visible only in play: 3 — alternate turns, White goes first — not always, so no; 4 — each piece has its own movement rule — no; 5 — castling — no; 6 — check and checkmate — no; 7 — details: en passant, a pawn’s first move of two squares — no. By the ruler: passes two, fails five. “Not always” doesn’t count: working means working every time. One ruler for any version of the game The ruler: what the agent’s build passes Agent: “done” applying the checklist 1234567 8×8 board with correct coloring pieces set up correctly alternate turns, White goes first each piece has its own movement rule castling check and checkmate details: en passant, pawn moves two squares yes yes not always no no no no visible in a screenshot visible only in play By the ruler: passes two, fails five “Not always” doesn’t count: working means working every time One ruler for any version of the game: apply it and measure point by point
1
Task
Claude Code SELF, level 9, lesson 1
Locked
Chess checklist 📋
Chess checklist 📋
1
Task
Claude Code SELF, level 9, lesson 1
Locked
Run the game through the checklist 📏
Run the game through the checklist 📏
Comments
TO VIEW ALL COMMENTS OR TO MAKE A COMMENT,
GO TO FULL VERSION