# Contradiction pages and a read-only wiki check (free sample) A free sample from Compile Your Research: a cited, checked second brain your agents keep, by PaperChaseWebb. Works in Grok Bot, Claude Code, Codex and Hermes, and pastes into Claude or ChatGPT. Version 1.0, 2026-10-27. Most AI second brains fail in the same quiet way: two notes disagree, and the agent picks one without telling you. This sample gives you the two pieces that stop that: a contradiction page that holds both statements until a person decides, and a health check prompt that lists problems without fixing them. Use them on any folder of markdown notes you already keep. ## The rules to paste first Paste this first, or put it in the file your harness reads first. ```text WIKI RULES (SAMPLE) 1. Never edit, rename or delete a source file. 2. Every factual line ends with (src: ). What you infer goes under a "Synthesis" heading, labeled as yours. 3. Never smooth over a conflict. Two sources that disagree get a contradiction page. Do not say which is right. 4. You propose; a person ratifies. Write changes to a proposed/ folder. Never mark anything ratified. 5. Never send, post, publish or copy notes outside this folder. ``` ## The contradiction page One page per conflict. Copy this shape: ```markdown --- type: contradiction title: status: draft sources: , last_compiled: ratified_by: --- ## Statements - A: "" (, ) - B: "" (, ) ## Why both cannot stand Resolution: ## Links [[]] ``` The Resolution line stays blank until a person writes it, with their name and the date. Every page that prints either statement links here, so nobody reads one version without seeing the other. ## Prompt: open a contradiction page ```text These two statements disagree: and . Write a contradiction page in proposed/: both quotes, exact and short, with file and date; one sentence on why both cannot stand; a line "Resolution:" left blank. List every note that prints either statement and propose a link from each. Do not say which is right. ``` ## Prompt: read-only health check ```text Read every note in this folder. Without editing anything, list: 1. Factual lines with no (src: ...) at the end. 2. Notes that cite a file that does not exist. 3. Notes nothing links to. 4. Contradiction pages with a blank Resolution line. 5. Pairs of notes that seem to disagree in meaning, with both quotes. For each item, propose one fix as a diff in proposed/. This is a list, not a fix. ``` ## Try it in two minutes Make two files. `notes/pricing.md`: "Starter package: $1,200, two rounds of changes." `notes/call.md`: "Told the prospect the starter package is $1,500 with three rounds." Ask your agent "What does the starter package cost?" with the rules loaded. A good answer says the notes disagree, quotes both and offers to open a contradiction page. An answer that just picks a price is the failure this sample exists to catch. ## Where it runs Claude Code: put the rules in CLAUDE.md and keep file edits on ask. Codex: put them in AGENTS.md and run with approvals on. Hermes: add them as a skill. Grok Bot: add them to the bot's instructions, with no send or publish connectors. Claude or ChatGPT: paste the rules, then the notes. ## In the full pack The folder layout and scope list, ten compile rules with the approval steps, seven page types (concept, entity, decision, contradiction, question, reading, answer), ten prompts, a Python health check script that needs no AI, a weekly routine, harness notes for four tools, a seeded test vault with nine expected findings, a failure table and an honest note on where it stops scaling. $24 at paperchasewebb.ai. Credit: the second brain pattern this grew from is Andrej Karpathy's LLM wiki idea, as explained in a short video by @seb.ai and built out in public GitHub vaults; rewritten in our own words. Made by PaperChaseWebb at paperchasewebb.ai.