RulesLawyer
Beta readers catch three kinds of problems. They catch pacing issues. They catch character motivation issues. And they catch places where your prose breaks your own rules — “wait, didn’t the book say this spell costs ten mana? He just cast it for five” — which is the kind of thing that haunts revision passes. The RulesLawyer is the agent that exists to catch that third category before your beta readers do. It reads your active ruleset, reads a document, and flags places where the prose contradicts what the ruleset says is possible. It’s not a replacement for a real beta reader. It’s a way to spend your beta reader’s time on things they can catch that the agent can’t, instead of on things the agent should have caught.
What it checks
Section titled “What it checks”RulesLawyer runs three kinds of check on every document you scan:
Strict mode. Numerical and structural consistency. Is a character described using stats that aren’t in their stat block? Is a spell cost mentioned that doesn’t match the ruleset? Is a character using an ability they don’t have? Is a formula being applied wrong? Strict mode catches the mechanical mismatches — the kind of thing a careful author would catch on a re-read but usually doesn’t.
Narrative mode. Prose-to-mechanics mismatches. Is the character described as “weak” but has Strength 20? Is a spell described as “easy to cast” when the ruleset says it requires a 10-minute ritual? Is a character shrugging off damage that, per the ruleset, should have killed them? Narrative mode catches the soft contradictions — prose that isn’t technically against the rules but feels wrong.
Both mode. Runs both. This is the default for a full pre-submission audit.
How to run it
Section titled “How to run it”From the Magic System panel, open the RulesLawyer tab. Pick a document (or a section of a document) and a check mode. Click Run. The agent streams back results as it works — a structured list of issues, each with a severity level, a description, a pointer to the source location in the prose, and a suggested fix.
Checks are not instant. Depending on document length and the current LLM model, a full-chapter check might take anywhere from 30 seconds to several minutes. You can keep working in the editor while a check runs — the panel updates when results come in.
What an issue looks like
Section titled “What an issue looks like”Each issue RulesLawyer surfaces has a consistent shape:
- Severity. Critical, major, moderate, or minor. Critical issues are hard contradictions — a stat that’s wrong, a spell effect that’s impossible. Minor issues are suggestions — a prose description that’s mildly inconsistent with the stat block but could be interpreted either way.
- Issue type. Which category the problem falls into — stat mismatch, ability misuse, formula error, narrative drift, missing ability, prerequisite violation.
- Description. What the agent thinks is wrong.
- Location. The paragraph and character offset in the document where the problem was detected.
- Affected entities. Which characters, items, or abilities the issue touches.
- Suggestion. A proposed fix, usually in the form of “the prose should say X” or “the ruleset should say Y.”
- Confidence. How sure the agent is. Useful for sorting — high-confidence issues are worth looking at first.
Every issue can be marked open, acknowledged, resolved, or false positive. Open means the issue is new and unreviewed. Acknowledged means you’ve seen it but haven’t fixed it yet. Resolved means you fixed the source. False positive means the agent was wrong — the prose is actually consistent, the agent misread something.
False positive isn’t a dismiss button. It’s a record that you reviewed the issue and decided the agent was wrong, and it persists across future checks so the same false positive doesn’t re-surface every time you run the analysis.
Check history
Section titled “Check history”Every check runs against a specific document and ruleset version, and the results persist. You can see previous checks in the history panel — “Full project check, 2025-11-04, 23 issues found, 18 resolved.” Click a historical check to see its full issue list. Click an issue to see its full context.
This matters more than it looks like it does. When you’re doing a second pass on a manuscript months later, you want to know which issues were flagged last time and which ones you resolved — you don’t want to re-check from scratch and have to re-mark every false positive. The history keeps that state.
Strict mode vs. narrative mode
Section titled “Strict mode vs. narrative mode”The two check modes have different strengths and it’s worth thinking about which one you actually want to run.
Strict mode is cheap, fast, and reliable. It catches hard contradictions that are objectively wrong. It won’t catch soft narrative issues that don’t violate the rules but still feel off. Run it frequently — every major draft, every time you finalize a chapter.
Narrative mode is slower, more expensive (more LLM tokens), and less reliable. It catches subjective stuff the agent has to make judgment calls on — “the character is described as weak but has Strength 20.” Narrative mode will generate false positives. It will miss things. It’s still useful because it catches issues strict mode can’t, but treat its output as suggestions rather than verdicts.
Both mode runs both and tags each issue with which check found it. Use it for pre-submission audits where you want maximum coverage and don’t mind spending the tokens.
What RulesLawyer can’t do
Section titled “What RulesLawyer can’t do”A few things the agent is bad at, in order of how often they matter:
- It doesn’t read tone. A character described as “too weak to lift the sword” when they have high Strength isn’t necessarily wrong — maybe they’re exhausted, or the sword is cursed, or the narrator is unreliable. RulesLawyer flags it, but you have to decide if it’s really an issue.
- It doesn’t understand temporary effects well. If a scene earlier in the chapter established that a character is debuffed, and a later scene uses the debuffed stat values, RulesLawyer may flag the later scene as inconsistent with the ruleset even though it’s correctly applying the debuff.
- It struggles with retcons. If you’ve updated the ruleset since the last time you edited a document, and the document uses the old rules on purpose (it’s a flashback, it’s a different character’s perspective), the agent will flag the old-rules prose as wrong.
- It’s model-dependent. A cheaper model will miss subtleties a more expensive model catches. If you’re running RulesLawyer on a critical chapter before submission, run it with your best model configured in Etherforce.