Glossary (Plain Language)

Core building blocks

• Column — a small, independent “witness” that walks a grid and writes notes.
• Grid map — a 2D notebook. Each cell stores a tiny summary of recent tokens.
• Cell — one square in the grid. It keeps the top‑K token IDs that visit it.
• Trajectory — the path a column takes over time across the grid.

Tokens and steps

• Token — a stable ID for a piece of information (e.g., “Alice”, “room_A”).
• Step token — a combined token representing the whole event at a step.
• Window — the last N step tokens, used for fast localization.

Movement and indexing

• Displacement — a small, deterministic move computed from recent tokens.
• Location index — a token→location lookup table for fast search.
• Localization — finding candidate locations from a token window.

Memory and replay

• Replay — rebuild state by re‑running events from a checkpoint.
• Checkpoint — a saved snapshot that lets replay start closer to “now”.
• Slow map — coarse summaries for very old history (approximate memory).

Frames and semantics

• Frame — a semantic lens, like “emotion” or “conflict type”.
• CNL — Constrained Natural Language; a small list of allowed frames.
• Frame trend — a simple time series of how a frame changes.

Reasoning and answers

• Consensus — voting across columns to reduce noise and ambiguity.
• Verifier — rules that check for contradictions (e.g., dead actors acting).
• Verdict — the answer label: supported, conflicting, or unsupported.

Two tiny examples

Example 1 (step token): “Alice enters room_A” becomes tokens for Alice, enters, room_A, then a single step token that represents the whole event.

Example 2 (consensus): If 4 out of 5 columns say Alice is in room_A and 1 column disagrees, the system reports room_A but notes the disagreement.