Cornerstone · II

How Do You Even Begin? The Two-Phase Architecture of Chess Analysis

You have been told to “calculate forcing moves first.”

You have read that “candidate moves are the foundation of chess thinking.”

You have heard masters say “look at every check, every capture, every threat.”

Every one of these instructions is correct. Every one of them is also useless.

They share a fatal property: they assume you can already do the thing they’re telling you to do. They assume you can already see which moves are forcing, which positions are tactical, which threats are real. They tell you what to do after you’ve already noticed what to look at — and the entire problem in chess thinking is that nobody has ever been told what to look at first.

This is the problem the Dimitrov Method solves that no other chess theory solves.

The Method has a two-phase architecture. First you scan the position to identify candidate threats by their geometric signatures, in a specific order. Only after that do you calculate which of them is actually most important. Most chess instruction collapses these two phases together — and that collapse is exactly why most chess instruction fails to actually improve players who follow it.

This article walks through the architecture, explains why the two phases are separate, and shows what changes when you start using them correctly.

The Hidden Problem in Chess Thinking

Picture yourself sitting at a board. Your opponent has just made a move. It is your turn.

What do you do?

If you’ve read enough chess books, you know the answer in principle. You should consider candidate moves. You should look at forcing moves first. You should calculate concrete variations. You should ask what your opponent is threatening.

In practice, what actually happens is this: you stare at the position for a few seconds. Some moves jump out at you because they’re “obvious” — captures, checks, the move that maintains your plan from last turn. You consider those. You play one of them, or maybe one that came to you on second look. Sometimes the move you played turns out to be terrible. When you analyze afterwards, the engine shows you a move you didn’t even consider.

You weren’t bad at calculation. You weren’t even bad at evaluation. You were bad at the step before calculation — you didn’t see the move at all. It was never in the candidate set you were comparing.

This is the most common failure mode in adult chess and the one chess instruction is least equipped to fix. Books and courses teach you what to do with candidates. Almost none of them teach you how to generate candidates. The implicit assumption is that pattern recognition will eventually surface the right moves. Your job is to recognize patterns from playing more games. The teacher cannot help you with this step.

Except it turns out the teacher can. The Method’s first contribution is showing how.

The Two Questions That Aren’t The Same Question

A chess threat exists at two levels of analysis. You can ask:

  1. What kind of threat is it, geometrically? — “Is my piece in a position to deliver check, threaten capture, or set up a check threat?”
  2. How dangerous is it, given the full position? — “Is this threat indefensible? Does my opponent have CABs available? What’s at stake?”

These are different questions, and they have different difficulty levels.

Question 1 is fast. It’s a perceptual operation. You look at a piece, you trace its attack lines, you see whether those lines reach an opposing king or piece. The skill is geometric: do you recognize the laser pattern? Once you’ve internalized that pieces are lasers, the answer comes in well under a second.

Question 2 is slow. It requires counting defenders, tracing future move sequences, checking for interference patterns, evaluating piece values. It can take several seconds to several minutes depending on position complexity. The skill is calculation: can you walk forward in the variation tree without making errors?

Most chess players, instructed to “look for forcing moves first,” try to do both questions at once. They look at a piece, ask “is this dangerous?”, and try to answer both the geometric and the analytical questions simultaneously. This collapses the cheap operation (geometric pattern matching) into the expensive operation (calculation), and it turns out that’s the wrong sequence.

The Method separates the two. Phase 1 answers Question 1, fast, across all candidate moves. Phase 2 answers Question 2, slow, only on the candidates that survived Phase 1. This is the architecture.

Phase 1 — Perception

Phase 1 is the perceptual scan. You go through the board looking for laser patterns that match each L-level, in lower-L-first order:

Notice what these scans share. They’re all asking “does this piece’s geometric attack pattern reach a target?” The pattern recognition is the same; only the target type and chain length vary. With practice, all five scans run in seconds.

The scans are ordered by two properties at once: simplicity of the geometric pattern, and forcing-ness of the resulting threat. L1 patterns are simplest (direct laser to king) and most forcing (chess rules require response). L5 patterns are most complex (chained piece-capture setup) and least forcing (long time horizon, opponent has many responses).

The order is not arbitrary. It’s the order in which threats are both easier to see and more likely to demand response. It’s the perceptual scaffold that makes Phase 2 even possible.

What comes out of Phase 1 is a list of identified threats — yours and your opponent’s — without yet knowing which is most important. The list is structured (each threat is tagged with its L-number) but unprioritized.

This is what classical chess literature has never given you: a startable position scan with a defined output.

Phase 2 — Classification and Prioritization

Phase 2 is calculation. Now that you have a candidate set, you can refine and prioritize.

For each identified threat, you compute its suffix: – An L1 either delivers mate (L1#), wins material in a forced sequence (L1), or simply gives check that the opponent will CAB. Counting CABs and checking for forced sequences turns the bare L1 into one of these. – An L2 either is L2 (one CAB or more available next turn), L2# (zero CABs — mate threat in 1), or L2## (only L1(d#) defends). – An L3 either is L3 (target has a defender) or L3 (target is loose). – An L4 either is L4, L4#, or L4##.

The suffix is what carries the actual urgency. It’s also what costs calculation. You cannot compute it during the perceptual scan because counting CABs requires looking at the position’s defensive structure, not just the laser geometry.

Once each threat has its suffix, you can sort them into priority tiers — the ontological order:

And finally — for each candidate move you might play — you run the cross-track priority calculation: does my move’s net king-capture progress exceed the cost of allowing the opponent’s highest-tier threat to execute?

That comparison is the actual decision. The output is the move you play.

Why The Phase Split Matters

The reason chess instruction has historically failed at this is that the two phases require different cognitive operations and skill people develop separately:

Phase 1 skill is visual and perceptual. Trained by drilling laser-pattern recognition. Improvable through pattern-recognition exercises (vision puzzles, board sight drills, geometry-only checkmate puzzles).

Phase 2 skill is analytical and calculatory. Trained by working through forcing sequences, counting defenders, evaluating tactical motifs. Improvable through tactics puzzles, calculation drills, and forced-sequence problems.

Most chess training programs train one of these skills heavily and assume the other will catch up. Tactics trainers like Lichess Puzzles and ChessTempo train Phase 2 — given the candidate moves are visible, find the right one. Visualization apps train Phase 1 — given a position, see the geometry clearly. Neither program gives you the connection between the two phases — the structured handoff where Phase 1’s output becomes Phase 2’s input.

The Method’s L-system is that handoff. The L-numbers are the typed output of Phase 1. The suffixes are the required input for Phase 2. Without the typed handoff, the two skills don’t compose into a thinking system. With it, they do.

This is why the L-notation looks like a priority scale on first reading and breaks down when you compare across cross-overs. The L-numbers are not a priority scale. They are a typed handoff between perceptual scan and calculation. A type, not a value.

What Changes When You Use It Correctly

Most adult-improver blunders, on close inspection, turn out to be Phase 1 failures, not Phase 2 failures.

The player did not miscalculate. They did not misjudge piece values. They did not fail to see the consequences of their move. They simply did not consider the move that would have been correct. It never made it into their candidate set. They computed accurately on the wrong inputs.

When you run a Process Audit on these mistakes — and the Process Audit format depends entirely on this distinction — you find a recurring pattern. The player ran a partial Phase 1 scan. They looked for L1s, maybe L2s. They didn’t scan for L4 setups, or didn’t notice the discovered-check pattern in the position, or didn’t see that one of their pieces could move to an L3 square to threaten material in a way that would have addressed an opposing threat at the same time.

The fix isn’t more calculation practice. It’s more structured perception. Drill the L-level scans. For one week, on every move you make in online games, audibly count through the L-scans: “L1 for me — anything? L1 against me — anything? L2 for me — anything? L2 against me — anything? L3 for me — anything?” You will lose games on time. You will identify candidates you would have missed. After a week, the scans become subvocal. After a month, they’re reflexive. After three months, you cannot stop running them — and you have stopped making the kinds of blunders that defined your previous chess.

This is what’s actually happening when chess players “improve.” It’s not that their evaluation got better. It’s that their candidate generation got more reliable. They see more moves now. The Method names the skill that improvement consists of.

The Generalization

Once you see the two-phase architecture in chess, you start seeing it everywhere.

Every domain that requires fast decision-making under partial information has the same problem: you can’t act until you’ve identified candidates, and identifying candidates is a different skill from choosing among them. Doctors triaging patients. Lawyers spotting issues in a contract. Fighter pilots scanning for threats. Negotiators reading the room.

The successful systems in each of those domains all separate perception from decision. The doctor’s differential diagnosis is a typed output of perception (a set of suspected conditions tagged with prior probability) handed to a separate decision engine (test ordering, treatment selection). The lawyer’s issue-spotting checklist is a perceptual scaffold; the legal analysis follows. The pilot’s threat-scan-pattern produces a typed list of contacts handed to engagement protocols.

Chess is not a special case. Chess is just a domain where the gap between perception and decision is unusually visible because the game is small enough to see the whole problem at once. The Method articulates the structure that already exists implicitly in expert play in every domain — and uses it to make the structure teachable.

What’s Next

Two things follow from internalizing the two-phase architecture, and you can do them today.

First, download the Pre-Move Checklist take the diagnostic and get yours. It is the explicit Phase 1 scan in printable form. One page. Fits next to your screen during online play. The checklist makes Phase 1 audible until it becomes reflexive.

Second, submit a position to the Process Audit take the diagnostic and get yours. Pick a recent game where you blundered. The audit will identify which phase you skipped — and within Phase 1, which specific L-scan you ran incompletely or missed. That diagnosis is information you cannot get from any engine, any coach, or any other chess training resource. It is the unique product of the two-phase architecture being made explicit.

You’re not going to stop blundering. Nobody does. But once you have the architecture, every blunder becomes diagnosable. You’ll know exactly which step you skipped. You’ll know exactly which scan to drill more. The improvement curve, instead of being a function of accumulated games, becomes a function of accumulated audits.

That’s the difference between a thinking system and a thinking habit.

That’s the Dimitrov Method.