Threads¶

A thread is the durable home for one research direction. It keeps the question, context, decisions, plans, experiments, reports, and interpretation together, so the work remains understandable even after a long break.

Think of it as the place where the investigation learns. An individual experiment may succeed or fail, but the thread preserves what that outcome means for the next question. Folders help you organize threads; the research context itself lives inside each one.

Why the thread boundary matters¶

Research rarely follows a straight line. A baseline reveals a confound, the confound suggests a control, and the control changes the original interpretation. Keeping that sequence in one thread makes the later conclusion legible.

The boundary should still be narrow enough that the work has one through-line. A thread called “Understand confidence failures in structure model X” can hold several related experiments. A thread called “Model research” will eventually contain too many unrelated assumptions to summarize well.

What a thread contains¶

A thread can contain:

the original question and follow-up messages
files, repositories, papers, datasets, models, and earlier work attached as context
questions from Silico and the answers you provided
plan drafts and revisions
experiments and their current state
reports, including diagrams and interactive views
a codebase that accumulates reusable code
interpretation, competing explanations, and follow-up ideas

Context stays with the message that introduced it. This makes it easier to see not only what Silico used, but why it mattered at that point in the research. A future reader can tell that a paper motivated a follow-up, or that a new dataset entered only after the first experiment.

Context stays explicit across threads¶

Experiments in the same thread can build on that thread's earlier work. Other threads remain separate unless you attach one of them, or one of their reports, to the message that needs it.

Attached cross-thread context is read-only. Silico can inspect it to answer the new question, but work in the new thread does not rewrite the source. This boundary helps prevent a convenient prior result from quietly becoming an input to unrelated research.

Threads, plans, and experiments¶

	Purpose
Thread	Preserves the research question, reasoning, decisions, and continuity between pieces of work.
Plan	Describes proposed work before it runs, including the goal, inputs, method, controls, and expected output.
Experiment	Preserves the execution record and evidence after a plan launches.
Reports	Communicates durable outputs from that work.

Launching an approved plan creates an experiment. Its reports return to the thread, where you can interpret the result and decide what question should come next.

For example, one thread might contain an initial representation analysis, a control experiment after a data-overlap concern, and a final steering experiment. Keeping them as separate experiments preserves their conditions. Keeping them in one thread preserves the argument that connects them.

Summary and Timeline¶

The thread has two views:

Summary shows the current synthesis, active work, and recent reports.
Timeline shows plans, experiments, decisions, and reports in order.

Start with Summary when you are returning to the work. It should answer “What do we currently believe, what is still running, and what matters next?” Use Timeline when you want to reconstruct a decision or understand how a conclusion changed.

The thread codebase¶

A thread codebase gives related experiments a shared foundation. Each experiment works on an isolated branch, and useful code can be merged back when later work should build on it.

Use the code icon to browse files and history or download an archive. Ask Silico to push reusable code to GitHub as a draft pull request.

Continue, fork, or start a new thread¶

Continue when the next question depends on prior reasoning or evidence. This lets the planner interpret new work in light of what the thread has already learned.

Use Fork to branch from a specific completed response. The fork is independent and excludes later source turns.

Start a new thread for an independent question. Attach earlier work when its evidence matters, and state what role it should play rather than importing an entire history by habit.

Copy the thread link to share the full, read-only research record.

Continuing creates your own thread with the source attached as context. It does not grant private credentials or compute access.

See Working with a thread for the practical workflow and Sharing for access boundaries.