Socria Logos 2

The conversation, a live map, and models the engine builds from what you say — what you can build, what to ask for, and what you can do with one.

Logos 2 is the same surface as Logos — the conversation and the Thinking Map, side by side — with one thing added: a model workspace. Structured, computed objects that you own, edit and undo, rather than pictures that are generated again each time you ask for a change.

At a glance#

• Say what you are working through. Logos 2 maps the reasoning as you talk, and when what you describe is a system — a shape, a mechanism, a set of equations, a relationship in data — it builds a model of it beside you.
• Move it. Every control recomputes the model, and what depends on it moves with it.
• Read it. Each model opens a row of views — the surface, its level sets, a cross-section, the slope, the residuals — and an inspector that says what each part is and where its number came from.
• Change it by saying so. “Make the damping zero”, “delete the second spring”, “undo that” — the model changes, not a picture of it.
• No depth modes. Unlike Core 3.1 there is no depth to set. Logos 2 answers at one depth and puts the effort into the model.

What you can build#

Every figure below is the same renderer drawing a model the engine built and solved, live on this page — there is no saddle component and no attractor component. Each caption ends with a sentence that builds one like it.

Three things have a simulation of their own, with a camera and physics in three dimensions: a black hole, the expanding universe and an orbiting system. Ask for one — “generate a black hole” — and it opens: a Kerr hole with its horizon, photon orbit and innermost stable orbit, light traced as real geodesics, and mass, spin, disc and rays as controls.

Things to ask for#

These land on shapes the engine knows how to build, which makes them the reliable ones. Name the thing and, if you have them, the numbers.

Surfaces and shapes

“Model a saddle, z = a·x² − b·y²” · “Plot a bivariate normal and let me change the correlation” · “Draw a torus I can reshape”

Things that move

“Model a mass on a spring with damping” · “Three masses joined by springs” · “Why is a double pendulum unpredictable?” · “Simulate an orbit”

Systems over time

“How does lowering the transmission rate flatten the curve?” (an SIR epidemic) · “Model a driven RLC circuit” · “Show me the Lorenz system”

Relationships in data

“wage = β₀ + β₁·educ + β₂·exper + u — set β₁ to 0.08” · “Add experience squared” · “Fit it with unit effects”

Objects in space

“Generate a black hole” · “Simulate the big bang” · “Simulate an orbit”

Then, about it

“Why does the critical point stop being a saddle when b goes negative?” · “What does the damping do to the energy?” · or select a part and press Ask about this.

What you can do with a model#

Move every control

Sliders for each parameter and each free input, with their units — “stiffness = 20 N/m”, “educ = 10 years”. The model is solved again as you drag.

Read it from every side

The Understand row lists the views a model supports: surface, level sets, cross-section, values, curve, path, phase portrait, against time, slope, residuals, coefficients, diagnostics, dependencies, sensitivity. Each is the same model, not a second drawing of it.

Axes with numbers and units

Every axis carries the variable’s name and, where the model declares one, its unit. 3D views carry round-numbered ticks on the cage, so a height can be read rather than guessed.

Ask about any part

Select an object and press Ask about this: the answer comes from the model — what it is, what it depends on, its value now, and whether that value was computed or chosen.

Edit by saying so

Change a value, remove or replace a part, add a body, undo, redo, duplicate into a branch — all from the conversation. See Editing one below.

Edit the map too

Right-click any card on the Thinking Map to mark it resolved or remove it, or say “delete the funding node” in the chat. What you remove stays removed.

Keep it in a project

Lines of thinking sit in the same rail as your chats and can be filed into projects.

Save as image

Save as image, in the panel head, exports the Thinking Map as a picture to share.

Models you own#

Everything else Socria draws is a picture of an idea. A model is different: it is structure the engine has validated and computed — objects, controls, equations, a solver — and it has a name that survives being changed. Move a control and it is solved again. Take a part out and the governing equations are assembled without it. Undo, and you get the previous state of the thing rather than the previous drawing.

That figure is the real engine, running in this page. The block hangs on a spring with a damper; the parts were assembled into equations of motion; a Runge–Kutta integrator solved them; the panels underneath are the same run as the motion above, not second drawings of it. Drag the stiffness and the integration happens again.

A document, not a picture

A model has an id, a title and a stack of revisions. Edit it and it is the same model one revision later — which is what makes undo, history and branching possible at all.

A revision, not a regeneration

Nothing is re-authored when you change something. The structure you had is the structure you keep, with the one thing you changed changed.

A branch, not a replacement

“Duplicate this and make the copy twice as stiff” leaves the original exactly as it was, and records what the copy came from.

Where a model comes from#

You ask for one in ordinary words — a mass on a spring, three tanks draining into each other, an epidemic, a regression on data you have given it. What happens next is the part worth knowing, because it is what separates a model from a convincing drawing:

1. Socria proposes the structure: what the objects are, what the controls are, which equations or parts govern it.
2. The engine — not the language model — sanitises it, checks that every expression resolves, works out what can actually be computed, and picks the solver.
3. If it computes, you get a model. If something is missing, you get that, named.

Everything in a proposal is marked as proposed — “Socria suggested this and nothing has confirmed it” — until you change it. A value you chose and a value it chose never read alike.

Editing one#

You edit a model by saying what you want, in the conversation, and the change happens to the model rather than to the picture:

“Change this mass to 4 kg.”

The parameter moves, everything that depends on it is recomputed, and the revision is remembered.

“Delete the second spring.”

The spring leaves the model. The equations are assembled again with one fewer term, so the motion is genuinely different — not the same motion with something hidden.

“Replace this damper with a spring.”

A dissipative term becomes a restoring one. The system is rebuilt and re-solved.

“Undo that.”

Back one revision. Redo goes forward again; it is your history, not a re-derivation.

“Duplicate this and make the copy twice as stiff.”

A branch under its own name, with the original untouched.

“Delete this model.”

It goes.

Clicking a part works the same way: select the damper and ask what it is, and the answer comes from the model — what it is, what it depends on, what its number is at this instant, and whether that number was computed or chosen.

When it will not build#

A refusal is a real answer here, and usually a more useful one than a picture. There are three kinds, and each says what would fix it.

Something is missing

A mass with no value, a state with no starting point, a system with no rule for how one of its parts changes. The structure is kept and the model waits.

Nothing here computes

The engine can hold what you described and no solver it has runs it. It says so, rather than drawing something that implies otherwise.

The method is yours

Ask for a model of data without saying how to estimate it and Socria will not pick. You get the candidates, what each one assumes and what each one gives — and the choice stays yours, because choosing the specification is the work.

What it does not do yet#

Said plainly, because the gap between what a product can do and what it appears to do is where trust is lost.

• Not everything becomes a model. A curve, a limit, a market or a distribution is still drawn as a figure — those are already exact and a model of them would be a worse version of the picture.
• Mechanisms are one-dimensional. Bodies on a line, with springs, dampers and forces. Pendulums, linkages and anything rotating are written as systems of equations instead.
• Adding reaches parts, not everything. You can add a body, a spring or a damper. Adding a variable to a fitted model, or a column to data, is not there.
• Estimation stops at least squares. Ordinary, within-unit and lagged. No instruments, no clustered errors, no tests with critical values — and it says so rather than implying otherwise.
• Undo is per model. Undoing a change to the map itself is not a thing yet.
• Comparison is computed, not yet shown. Two revisions can be compared; nothing puts them side by side on screen.

Access and plans#

Logos 2 needs an account. Pick it from the model menu beside the chat box; leaving it returns you to whichever Core model you were on. The Try Logos 3.5 invitation beside the chat opens Logos 3.5, which is everything here with a second seat, thinking depth and the everyday displays.

The conversation, the lenses and the moves on each card are the ones the original Logos had — see that page and the Thinking Map. What is stored, and where, is on accounts and data.