Sigil: Research Tower

• the user wants durable understanding rather than a short summary;
• source claims, related papers, notation, and definitions need separation;
• the result should teach Arcanum operators in the user's design language;
• a future sigil, spell, ontology, runtime, or implementation decision depends on the research;
• subagents may be spawned and must be reconciled before closure.

• primary source URL, PDF path, paper title, corpus path, or research target;
• intended output folder;
• desired depth: compact, standard, full, or deep;
• known related papers or concepts;
• existing local research artifacts;
• target audience or operator lens;
• promotion boundaries and non-goals;
• subagent strategy or delegation preference.

Route to research-evidence-harness when the target needs:

• append-only run schemas;
• JSONL fixture validation;
• objective-vector, scoring, or reviewer-rubric checks;
• dry-run result summaries;
• live-run data integrity;
• claim-adjudication readiness from measured results.

The handoff should name the source/tower artifacts that support the experiment, the evidence mechanics still missing, and the claim or paper section blocked by missing run data.

• primary-source: stated by the main source;
• related-source: stated by a related paper or external source;
• local-inference: inferred from source-backed material;
• analogy: useful Arcanum mapping, not source meaning;
• operator-reading: practical design interpretation for the user;
• open-residue: unresolved question or future task.

• subagent/lane name;
• assigned question;
• evidence returned;
• integration decision;
• rejected or blocked material;
• final status;
• owner of follow-up residue.

A research tower cannot be marked closed while any subagent lane is implicit, unreviewed, or missing an integration decision.

• verify existing artifacts before creating new ones;
• cite or path-link the source evidence used;
• distinguish source claims from Arcanum readings;
• make important notation learnable before using it heavily;
• produce glossary, definitions, distilled knowledge, and final learning pack;
• close named residue or move it into explicit future work;
• reconcile every subagent lane;
• preserve local-only promotion boundaries;
• hand off to an evidence harness when data mechanics, metrics, or claim adjudication become the blocker;
• validate required artifacts and links;
• leave another agent able to reproduce or audit the learning path.

• summarizing the paper without a reusable artifact spine;
• turning analogies into source claims;
• promoting terminology because it sounds useful;
• skipping notation until the final pack;
• hiding uncertainty inside polished prose;
• using related papers as unbounded background;
• spawning subagents without a closeout ledger;
• calling the tower closed while residue is unnamed;
• producing final recommendations without borrow/block decisions.

## Research Tower Summary

- Target: <source-or-topic>
- Output root: <path>
- Depth: compact | standard | full | deep
- Source record: pass | partial | block
- Existing artifacts checked: yes | no
- Required artifacts: pass | partial | block
- Notation bridge: pass | not-needed | block
- Glossary/definitions: pass | partial | block
- Related work: pass | not-needed | partial | block
- Subagent closeout: pass | not-used | block
- Promotion boundary: local-only | promotion-candidates-listed | block
- Validation: pass | partial | fail | not-run
- Final learning pack: <path or none>
- Open residue: <count>

### Produced Or Updated

- <path> - <role>

### Next Governed Route

1. <next action>