How MyChessPlan Analyses Your Games
An honest, technical walkthrough of the methodology — what we measure, how we classify you, and what you get back.
MyChessPlan.com
TL;DR
- We pull your last 100 public games from the Chess.com API — no login, no password, no engine running on your machine.
- We compute ~17 deterministic metrics (win rate, time-class share, opening diversity, tilt indicator, and more). No AI is involved at this stage.
- A rule-based classifier places you into 1 of 8 weakness archetypes × 3 ELO bands = 24 profiles. The shape of each rule is described below.
- You see a free web report immediately. You can opt in to a free PDF (email-gated) and to a paid $14.99 personalised 30-day workbook.
- The free report is fully deterministic. The paid workbook adds Stockfish move evaluation and a Claude-AI written narrative on top of the same metrics.
A note on transparency: we describe what each rule looks at and the kind of pattern that triggers it, but we don’t publish the exact numeric cutoffs. The classifier is meant to read your real play, not a tuned input — leaving the thresholds out keeps the diagnostic honest.
1. Where the data comes from
We call the official Chess.com Published Data API. Specifically:
/pub/player/{username}— public profile (used to confirm the account exists and pull the country/title)./pub/player/{username}/stats— current ratings for rapid, blitz, bullet and daily./pub/player/{username}/games/{YYYY}/{MM}— game archives. We walk the most recent months until we have up to 100 finished games.
That’s it. We never ask for your password and we don’t read anything that isn’t already public. If your Chess.com profile is set to private, the API returns nothing and we tell you so on the form.
Why 100 games? Below a few dozen games the metrics get noisy — one bad weekend dominates the signal. Above 100 the API throttles us and the older games stop reflecting your current play. 100 is the sweet spot.
2. What we measure
Every game is projected from your point of view (your colour, your result, your rating, your opponent’s rating). From those projected games we compute a fixed battery of metrics — all of it in plain Python, all of it deterministic, no model in the loop. The categories:
Outcome metrics
- Overall win rate.
- Loss-type breakdown — what share of your losses ended on the clock, in checkmate, or by an early resignation.
Game-length metrics
- Average length of the games you lose vs the games you win — short losses point to tactical leaks, long losses point to endgame technique.
- Loss rate in long games (the conversion problem) and in short games (the vigilance problem), tracked separately.
Opening metrics
- Opening diversity — how often you switch openings. A high value means you’re rediscovering the same problems in many lines; a low value means you have a stable repertoire.
- Worst recurring opening — for any opening you play repeatedly, the loss rate in that line.
- ECO distribution — the share of your games in each ECO bucket A, B, C, D, E. (Roughly: A = flank openings, B = Sicilian/Caro-Kann, C = open games / French, D = Queen’s pawn, E = Indian defences.)
Time-control metrics
- Time-class mix — the share of your games that are bullet, blitz, rapid and daily.
Behavioural metrics
- Longest losing streak in chronological order.
- Tilt indicator — your win rate in the games you played immediately after two consecutive losses. A low value means losses snowball; a higher value means you reset between games.
There are also ~20 secondary “advanced” numbers (FIDE-style performance rating, current streak, win rate vs higher-rated opponents, colour split, recent form, etc.) — they don’t drive classification but they’re shown in the “Numbers in depth” section of the results page so you can verify everything yourself.
3. How we pick your archetype
The classifier is a hand-written decision tree. Rules are evaluated top to bottom; the first one that matches wins; if none matches you fall through to The Balanced. This is intentional — we’d rather have a rule we can explain in plain English than a black-box model that’s right 2% more often but you can’t audit.
The shape of each rule:
| # | Archetype | Looks for… |
|---|---|---|
| 1 | The Tilter | A meaningful losing streak combined with a clearly reduced win rate immediately after consecutive losses — losses snowball instead of resetting. |
| 2 | The Blunderer | A low overall win rate driven primarily by losses inside the opening / early middlegame — vigilance problems, not strategy. |
| 3 | The Bullet Addict | An overwhelmingly bullet-heavy mix with little to no rapid play — pattern recognition without depth of calculation. |
| 4 | The Lost Opener | Many different openings tried alongside short losses — repertoire is unstable, the same problems recur in different lines. (The “short” threshold is relaxed for players who play mostly daily, since daily games naturally run longer.) |
| 5 | The Failed Converter | Long-running losses where checkmate is rare — equal or winning middlegames drift into lost endgames. Endgame technique is the leak. |
| 6 | The Impatient Attacker | An opening repertoire dominated by sharp 1.e4 lines (ECO B + C), combined with either short losses or a high mate-rate — attacking instinct outruns calculation. |
| 7 | The Passive Solid | A quieter opening repertoire (ECO A + D), a high share of losses on the clock, and long games — careful play that runs out of plans before it runs out of position. |
| 8 | The Balanced | None of the patterns above stand out. No single dominating leak — work the smallest weakness first. |
Why we don’t publish the exact cutoffs: the classifier is meant to read your real play. If we listed the numbers, the diagnostic would only tell you how well you can engineer your inputs. The cutoffs do exist, are well-defined, and live in the source code — they just don’t belong on this page.
Want the prose version of each archetype with examples and fixes? See the free archetypes guide.
4. ELO band
On top of the archetype we tag you with a band so the advice fits your level: Beginner, Intermediate or Advanced. The cutoffs roughly correspond to the standard improvement plateaus on Chess.com.
The band uses your rating in your primary time class (the one you play the most). If you don’t have a clear primary, we fall back in this order: rapid → blitz → bullet → daily. That’s why a player who only plays daily gets their daily rating used as the reference, not a missing rapid number.
Eight archetypes × three bands = 24 profile slots. The same archetype reads slightly differently for a beginner, an intermediate and an advanced player — the same leak, different fixes.
5. What you actually get back
Free web report (no email needed)
- Your archetype + ELO band, with a plain-English description.
- The full table of metrics — the same numbers the classifier saw.
- A walk-through of one of your real games that exemplifies the archetype.
- 3-5 prescriptive next steps (e.g. for The Tilter: a cooldown rule after consecutive losses).
Free PDF (email-gated)
- The same web report formatted as a printable workbook.
- Email goes onto our list (MailerLite). One welcome email, monthly tips, no spam.
Paid 30-day workbook — $14.99
- ~47 pages personalised to your username and your specific metric values.
- A 30-day training plan — daily schedule of tactics / opening study / endgame drill / one annotated game.
- Stockfish move evaluations on three of your real games, picking out the critical moments and the moves you missed.
- A narrative written by Claude AI using your numbers — tone is direct and coach-like, not generic.
- Curated Chess.com lessons + YouTube videos that match your archetype.
The paid PDF doesn’t change which archetype you got — same rules apply. It deepens the analysis (engine evaluation) and makes the plan personal and actionable. If you’d rather pass and just work the free advice, that’s fine — the free report is a complete product on its own.
6. What we deliberately don’t do
- We don’t run a full Stockfish evaluation on every game in the free pass. Engine analysis on 100 games would take 5-15 minutes per user and cost real money — we’d rather keep the free product fast and free, and reserve engine work for the paid workbook.
- We don’t claim a magic prediction. The classifier is a heuristic over 100 games, not a model trained on millions. It’s good at flagging the dominant leak; it’s not going to tell you “your next move should be Nf3.”
- We don’t store your games or share your data. The analysis runs on demand; we keep only the username, the email if you gave us one, and the resulting report (so you can come back to it). No PGNs, no raw API responses.
- We don’t replace a coach. A human coach watching your games over Zoom will catch things our metrics never can — body language, time pressure, opponent-specific holes. What we offer is the diagnostic before you spend on coaching, so the coaching session starts from a known weakness, not from “let me watch a few games.”
7. Caveats and edge cases
- Very few games means the metrics are noisy. We still classify but flag the result as low-confidence.
- Mixed time classes can pull metrics in opposite directions (e.g. you blunder in bullet but not in rapid). The “primary time class” logic mitigates this, but if you want a clean read, run two analyses — one when you’ve been playing mostly rapid, one when mostly bullet.
- Rules drift over time. As we see more users we re-tune the cutoffs. The shape of each rule rarely changes; the numbers do. If a future analysis reads slightly differently for the same input, that’s expected.
- The paid workbook depends on Stockfish + Claude AI being available. If either service is down at purchase time we queue the report and email it within a few hours.
Ready to see your own numbers?
Drop your Chess.com username and we’ll run all of the above on your last 100 games — free, in under 60 seconds.