popsicle-bot-lab

PGN → personality fingerprint → a bot that actually plays like you.

Live Demo →

Now you might ask "why popsicle bot lab? why not chess human exterminator?"

And that's a valid question. I decided to name this repo "popsicle-bot-lab" cuz that's me (or my account) on chess.com

An open prototype for Chess.com's 3rd 2026 AI focus area — mapping a player's historical PGN data to a personality vector, then encoding that vector into a Stockfish-driven bot spec.

Built by Ankita Pal aka @popsicle_guy

---

How it all works (not that you care but maybe you do hehe)

chess.com /pub/player/{user}/games    PGN file
                    └──────────────┘
                           ▼
               [ python-chess parser ]
                           ▼
              per-game heuristic metrics
                           ▼
         normalise → 8-dim trait vector ∈ [0,1]
              ┌────────────────────────────┐
              ▼                            ▼
  nearest-centroid cluster       param-mapping function
              ▼                            ▼
          ARCHETYPE              STOCKFISH BOT SPEC
                              skill / depth / contempt
                              move-bias vector
                              blunder injection prob

Pipeline

Step	What happens
01 PGN ingestion	Pull last N games via Chess.com's public archives API, or accept a raw PGN dump. Parses mainline into board states + move metadata.
02 Heuristic feature extraction	Per-game: captures, checks, sacrifice ratio, early-queen moves, castling speed, central pawn pushes, endgame reach + result, opening (ECO) diversity, time-control base.
03 Trait normalisation	Raw stats compose into 8 traits in [0,1]: aggression, tactical_vision, positional_play, time_mgmt, opening_diversity, endgame_strength, risk_tolerance, blunder_resistance.
04 Nearest-centroid clustering	8 hand-seeded archetype centroids (Tactical Tornado, Patient Strategist, Endgame Surgeon…). Euclidean distance → confidence score.
05 Bot spec generation	Trait vector maps to Stockfish runtime params: skill_level, search_depth, contempt, move_time, blunder_injection_prob, opening-book width, and a 6-dim move-selection bias vector.

Archetypes

Archetype	Personality
Tactical Tornado	Relentlessly aggressive, sacrifices for initiative
Patient Strategist	Pawn structure devotee, positional grinder
Endgame Surgeon	Trades into endings with frightening precision
Opening Theorist	25 moves of prep, slightly lost after
Blitz Berserker	Pre-moves on move 3, clock as weapon
Solid Defender	Never loses a won position. Actually never loses.
Chaotic Gambiteer	King's Gambit on move 2. Sacrifices are felt, not calculated.
Balanced Pragmatist	Does everything competently. Terrifying in aggregate.

---

What's next

• Plug Stockfish + the bias vector into a move-sampler (top-k softmax over engine PVs weighted by bias)
• Add engine-eval-based blunder labelling for true tactical_vision
• Train an embedding head on millions of Chess.com users to replace hand-seeded centroids
• Replace heuristic ECO diversity with proper opening tree analysis

---

Stack

• Vanilla JS + CSS (no framework — keeps it inspectable)
• Chess.com Public API (/pub/player/{user}/games/archives)
• Stockfish 16 (spec generated; live move-sampler = next weekend)
• GitHub Pages (zero infra)

---

Run locally

git clone https://github.com/0xafraidoftime/popsicle-bot-lab
cd popsicle-bot-lab
# Any static server works:
python3 -m http.server 8080
# → open http://localhost:8080

Note: The Chess.com API is CORS-friendly for public endpoints, so this runs fully client-side.

---

Links

• GitHub: 0xafraidoftime
• LinkedIn: Ankita Pal
• Twitter: @afraidoftime_
• Chess.com: @popsicle_guy

---

v0.1 · open-source · built in a weekend