Other subjects

Dum's conversation isn't limited to code. The skill tree and evidence system are built around programming, but the conversation itself can be about anything you're genuinely working on — a proof, a composition, a research paper.

We're exploring subject-specific skill trees and evidence. These examples describe that direction, not a shipped assessment system for every subject.

What can carry over

Conversation can discuss a proof, a score, or a historical source. The current prompts and verified-source catalog are programming-oriented; a plausible answer isn't a subject-specific assessment.

Project memory (:remember) persists across sessions regardless of subject. You can use it to record decisions, open questions, and context for any kind of project.

Sharing work: :inspect path/to/file shares a bounded excerpt from project text files, such as LaTeX, LilyPond, or an essay draft. :changes shares saved changes. The screen-aware wizard can see a diagram or application view while screen advice is enabled.

What's coding-oriented right now

The skill tree (:tree, :skills), the three evidence levels (recognize / build / apply), and :submit are built for programming. They track concepts and tools in specific languages. Subject-specific assessment outside code is not implemented.

The generic evidence machinery can store custom skills, but there aren't curated math, music, or history tracks or validated assessment standards for them. A custom note doesn't establish that a proof, performance, or historical claim is sound.

For each subject, the question is what independent work would demonstrate a skill, how prerequisites fit together, and which sources make a review checkable. Memory should shape substantial projects, not turn every topic into a short quiz.

Mathematics

Working through a proof, a problem set, or a research document? Dum can discuss approaches, follow reasoning steps, and remember context across sessions. The wizard can point out a gap in an argument — its judgment can be wrong, so check its reasoning.

What it looks like

• Share a LaTeX excerpt with :inspect; ask about a specific step or ask dum to follow the argument.
• Describe the structure of a proof; get wizard feedback on whether the cases look complete.
• Record which approaches you've tried in project memory so dum doesn't repeat suggestions you've already ruled out.
• Share a plot or diagram with screen sharing and ask what dum sees.

Current limits

There's no curated math track or proof checker. A useful direction is a proof notebook that spans several techniques: completed proofs and counterexamples would be evidence, not a confident explanation alone.

Music

Composing, arranging, transcribing, or analyzing? Dum can discuss theory questions, arrangement choices, and structural decisions. Share notation in a text format and ask specific questions about it.

What it looks like

• Share LilyPond, ABC notation, or a plain-text lead sheet with :inspect; discuss a specific bar or harmonic move.
• Describe what you're trying to achieve; ask about common approaches in that style or period.
• Record compositional decisions in project memory — why you chose a particular form, what you've ruled out and why.
• Use screen sharing to show a score or DAW view and ask for observations.

Current limits

There's no dedicated score or performance assessment. A candidate project is an arrangement with an annotated score and recorded performance, with separate evidence for harmony, voice leading, and interpretation. Current text and screen advice cannot verify a performance or replace a musician's review.

History and research

Writing a research paper, analyzing primary sources, or building a historical argument? Dum can discuss interpretations, source context, and essay structure. Project memory makes it useful across multiple sessions on the same topic.

What it looks like

• Share a draft paragraph with :inspect; ask about clarity or argument structure.
• Describe a source; discuss its likely provenance, limitations, and how to read it critically.
• Record research threads, working hypotheses, and sources you've already checked in project memory.
• Use screen sharing to show an archival image or a dense passage and ask dum to describe what it sees.

Current limits

There's no scholarly-source verification system or curated history track. A substantial direction is a sourced essay or small critical edition: primary-source comparisons, traceable citations, and argued interpretations would be separate evidence. Verify dates and provenance yourself; a model's agreement isn't historical proof.

Input and sending

Wizard screen advice is the default source while advice is enabled. It samples the screen for unprompted advice, subject to OS permission. You can pause it or switch to saved project files. This is separate from manually previewing and sending a screenshot with a message.

Voice dictation is deliberate input. You initiate recording, review the transcript, and decide when to send the message. It isn't an auto-send channel. If screen advice is enabled, its independent screenshots may include visible drafts.

The two paths are independent. Screen sampling doesn't record audio. Voice dictation doesn't capture the screen.