The Thinking Map
Node types, relationships, the four lenses, and the four moves a node opens.
The Thinking Map is a typed graph of your reasoning, extracted fresh from the conversation on every turn. Everything on it is something you said or clearly hold — labeled in your own language, two to eight words a node.
Node types#
Thirty types, grouped by the kind of thinking they belong to. A conversation uses the handful that fit it:
| Group | Types |
|---|---|
| Deciding | goal · decision · value · belief · idea · assumption · evidence · question · tension · consequence |
| Arguing & researching | claim · counterpoint · source |
| Learning | concept · misconception |
| Making | theme · character |
| Planning | constraint · milestone |
| Mathematics | given · unknown · equation · definition · transformation · theorem · step · inference · verification · result · error |
A few distinctions carry weight: a belief is something you hold; an assumption is taken as true but unexamined. A source is the container, not the fact — the study you cited, separate from the evidence inside it. And an error in math is kept and repaired in place, never deleted — mistakes are part of the record of the work.
Nodes carry status (open → supported / resolved / revised), optional LaTeX, an optional note, and a “+N folded in” badge listing wordings that were merged into them. Maps hold up to 16 nodes and 22 edges — 26 and 34 for math, because a solution chain is long by nature. On the free tier the map stops taking on new nodes at four; Socria One lets it keep growing, and nothing already drawn is ever taken back.
Relationships#
Eleven edge types, each read in plain language on the map itself:
supports- evidence or reasons holding a node up
conflicts- two things pulling against each other
depends- one thing needing another to hold
relates- a connection that matters but fits no sharper type
leads_to- consequence; one thing producing another
revises- a changed mind — the new thought pointing at the old one
precedes- chronology; comes before
part_of- structure; sits inside
transforms_to- math: one expression becoming the next, the operation written on the edge
implies- logic: A logically implies B, the spine of proofs
justifies- a theorem or definition licensing a step
Edges have strength — weak, normal, strong. A relationship that stops carrying weight is weakened, not deleted; strong is reserved for connections the whole argument rests on.
How the map stays honest#
Before adding anything, each extraction is required to reorganize: merge duplicate ideas (keeping the absorbed wording visible), mark a supported claim supported, mark an answered question resolved — kept on the map, never silently deleted — and when you change your mind, create the new thought and point it at the old one with revises, so the history of the reasoning stays legible.
The lenses#
One map, up to seven readings of it. Lenses appear only when the map has something for them to show:
| Lens | Reads | Appears when |
|---|---|---|
| Graph | everything at once — a live force-directed layout that settles as you watch | the map has any node |
| Structure | what rests on what, as a layered tree | more than one node |
| Tensions | each conflict as an opposing pair, facing each other | a conflict, tension or counterpoint exists |
| Evidence | claims in columns with their actual support beneath | evidence or sources exist |
| Solution | the worked chain, step by step, operations written beside the arrows | math, with a chain to follow |
| Plot | the functions on the map, graphed | math, with something plottable |
| Board | the work laid out by hand — see Mathematics | math, with any node |
Math work leads with the Solution lens automatically — until you pick a lens by hand, after which the choice is yours and stays yours. On the free tier only the map’s lead lens is open — Solution for math work, Graph otherwise; the other readings are Socria One’s.
what rests on what
What am I trying to accomplish?
The four moves#
Click any node and a menu opens: four ways to act on it, all of which stop short of resolving it for you. Each returns the same anatomy — a concept, a framing (a frame, never a conclusion), a connection back to your own words, and one question only you can answer — plus its own specialty:
- Explore
- What this idea is, and what it isn’t. Searches the web for conceptual context; research informs the frame, never the decision.
- Challenge
- Where this would break: up to four pressure points, each a condition to test rather than a criticism. It steelmans the other side and is forbidden from being contrarian for its own sake.
- Research
- What the evidence actually says, from a live web search. Sources it actually cited are marked; disagreement is reported as disagreement, and thin evidence is called thin.
- Trace
- Where this came from: verbatim quotes from the transcript. The model only chooses which moments — the quote text is sliced from the real transcript by the server, so a fabricated quote is impossible.
The menu also offers Add context (ground the node in your own material) and, while a draft is open, Hold in view. Each node keeps one conversation thread across all four moves — switching from Explore to Challenge is a change of lens, not a new conversation — and anything you type there is folded back into the session so the map can use it.
A raise measures what a market will pay for the work you can already do. Progress measures what you will be able to do next. They move together often enough to feel like one thing, and come apart exactly when a role pays well for skills you have finished learning.
You put the money and the stagnation in the same sentence, as though the first answered the second.
If the offer paid the same as your current job, would you still want it?