How to Use Arc Codex
Your guide to reading intelligence, understanding analysis, and contributing to the feed.
Reading the Feed
The main feed shows articles collected from 2,350 sources worldwide, analyzed by the A.R.C. framework as they arrive. Each card represents one article with its full analysis inline.
The Chimera Difficulty Score (circle gauge)
The circular gauge on each card shows reading difficulty on a 0–100 scale, synthesizing four grade-level readability metrics into a single number. This is a purely informational signal — not a quality or manipulation score.
- Flesch–Kincaid — architectural complexity.
- Coleman–Liau — character density.
- SMOG — polysyllabic gravity.
- Dale–Chall — lexical rarity.
- 0–10 — Kindergarten
- 11–20 — Elementary
- 21–30 — Middle School
- 31–40 — High School
- 41–50 — College
- 51–60 — Graduate
- 61–70 — Academic
- 71–80 — Expert
- 81–90 — Specialist
- 91–100 — Quantum Electrodynamics
Higher scores indicate denser, more technical prose — not better or worse content. The 41–55 band tends to be the sweet spot for quality long-form journalism.
Directive tags (footer)
The tag at the bottom of each card shows which editorial directive matched the article — e.g. Active Threat Campaigns or Economic Policy. Click it to navigate to the wiki page for that directive.
Infinite scroll
The feed loads in Tribonacci batches — 2, 3, 5, 10, 18… articles at a time as you scroll. This keeps the initial load fast while giving you a deep feed on long sessions.
Click any article title to go to the full article page, where all analysis sections are expanded by default.
Understanding the A.R.C. Analysis
Every article is run through three independent AI analytical passes. Expand any section on a card to read it. We do not tell you what to think — we show you how the thinking was constructed.
Facts Only — Red Team
Extracts verifiable facts only — who, what, when, where. No interpretation, no opinion. If it can't be independently verified, it's not in here.
Executive Summary — Blue Team
A balanced, journalist-style summary written for an educated general reader. Covers the key points without editorializing. Good starting point if you're short on time.
Full Take — Purple Team
The deep analysis. Covers: steelmanning the narrative, scanning for 48 A.R.C. cognitive anti-patterns, root cause analysis, implications for human dignity, bridge-building questions, and a hypothetical influence campaign analysis. This is where Arc Codex earns its keep.
Counter-Analyst comment
Every article gets an adversarial AI comment, labeled with a robot emoji and cyan styling. It's deliberately provocative — designed to give readers something to push back against. The “empty dance floor” problem: nobody wants to be the first to comment. The Counter-Analyst solves that.
Sentinel — AI Content Detection
Sentinel runs an automated pass on the wording of the original source article and gives a rough estimate of whether it shows signs of machine generation. It's tuned conservative — false positives (calling human writing synthetic) are treated as worse than false negatives. The estimate is uncalibrated and experimental: a small local AI model makes it, it has not been checked against known human and machine texts, and it can be wrong in both directions. It cannot show who wrote an article, and it looks only at the source article, not at this site's own AI-written sections.
How to read Sentinel's estimate
- Provisional no strong signs of machine writing were found (a raw score below about 0.6 and no synthetic flag from the model). This is not a finding that a person wrote it.
- Inconclusive mixed signals: a middling score, or the score and the model's own reading disagree.
- Provisional flag the source article shows possible signs of machine generation (a raw score of about 0.8 or more with a synthetic flag). An estimate, not proof.
- Unavailable the check did not finish, so no result is shown.
Sentinel never labels an article “Human”. The number shown is a raw score from 0 to 1, not a probability and not a percentage. The list of signals shows what the model noticed in the source article; treat it as a starting point, not a conclusion.
Sentinel is not a spam filter — a flag means the writing pattern resembles AI output, not that the content is wrong or malicious.
Translation
Arc Codex translates articles and their analysis on demand, into a wide range of languages, through its inference pipeline. Each translation is cached per article and language for 24 hours, so a repeat request is instant.
How to translate an article
- Click the Translate button below the article title.
- If you have a preferred language set in your account, it fires immediately.
- Otherwise a language picker appears — select your language.
- Once translated, a language pill appears. Click it again to switch languages or reset to the original.
Translating foreign-language articles into English
English is first in the dropdown specifically for this use case. Articles from German, Swedish, Spanish, and other sources can be read in English instantly. Arc Codex sources include Aftenposten, NRC, Der Spiegel, and others.
Setting a preferred language
Sign in with Google or GitHub, then open the account menu (top right). Set your preferred language and Arc Codex will auto-translate articles when the source language differs from your preference.
Publishing to Arc Codex
Signed-in users can submit content directly to the Arc Codex pipeline via the Publish page. All submitted content goes through the full A.R.C. analysis pipeline.
Share a URL
Paste any article URL — Arc Codex fetches the content, runs A.R.C. analysis, and publishes it to the feed. YouTube URLs are also supported (metadata and description are analyzed).
Write Text
Paste or write article text directly. Good for content behind paywalls or content you've written yourself. The full A.R.C. pipeline runs on your text.
Upload a File
Upload .txt, .md, .pdf, .docx, or .odt files. Arc Codex extracts the text and runs the full analysis.
Write a Prompt
Describe what you want Arc Codex to write — it generates a full article using AI, then runs it through the A.R.C. pipeline. The title is derived automatically from the generated content.
Make Public vs Keep Private
When you click Publish, a confirmation modal asks how to publish:
- Make Public — article appears in the public feed for all readers.
- Keep Private 🔒 — article is visible only to you when signed in. Nobody else can see it.
Private articles appear in your feed with a lock icon, visible only to you when signed in.
Submissions are processed asynchronously — your article will appear in the feed within 1–2 scribe cycles (up to ~10 minutes). Priority queue items are processed first at the top of each cycle.
Your Account
Arc Codex uses soft authentication — the site is fully public and requires no login. Signing in with Google or GitHub unlocks preferences, publishing, and private articles.
What signing in gives you
- Set a preferred translation language (auto-fires on foreign articles).
- Publish articles to the feed (public or private).
- See your private articles in the feed.
Privacy
No tracking, no ads, no data sales. Arc Codex stores: your email, name, profile picture (from OAuth provider), preferred language, and account timestamps. GDPR self-service deletion is available in the account menu.
Questions not answered here?
ross@arc-codex.com