Atlas Research Notebook: Six Distinct Views Per Symbol And An Honest Traffic-Truth Audit

Two build passes on 2026-06-02 locked a permanent guard that keeps the six customer views of every Atlas symbol structurally distinct, then added an audit that reports 10 verified engaged humans against 1,470 raw nginx hits over the same 24 hours.

The first number is the audience. The second is load. The point of the work logged below is that the Atlas research notebook now refuses to confuse the two, and refuses to serve the same text under six different labels.

Table Of Contents

• What "Six Distinct Views" Actually Means
• The Traffic-Truth Audit: 10 Humans, 1,470 Hits
• Coverage: 171 Symbols, 75 Indexable, 82 Sitemap URLs
• The Free-Source Spine Behind The Pages
• Caveats: What Is Not Proven Or Not Live

What "Six Distinct Views" Actually Means

Every symbol page in the Atlas research notebook can be served in six shapes: Snapshot, Source Pack, Report, Authored Report, Fast Brief, and Thread Cards. A free reader sees the Snapshot. A Pro reader can pull the authored report, the source pack, and the export trail. The risk that creeps into any tiered system is that these labels drift toward the same body of text, so a paying view ends up being the free view with a new heading.

The hardening pass on 2026-06-02 turned that into a test that has to pass before the backend restarts. The file /root/Atlas_Backend/atlas_no_leak_smoke.py now asserts that all six text views stay structurally distinct, that all Pro PDF views stay distinct, and that all Pro CSV row views stay distinct. The checks it runs are named free_snapshot_public_renderer, all_text_views_distinct, all_pro_pdf_views_distinct, all_pro_csv_row_views_distinct, and cold_deep_request_refuses. The last one matters most for honesty: a cold snapshot-only deep request fails closed to a visible limited-state message rather than fabricating depth it does not have. The rich fixture used for the run was XMR.

This is the spine of an honest research surface. A symbol page that promises a report should contain a report, and a snapshot should look and read like a snapshot.

The Traffic-Truth Audit: 10 Humans, 1,470 Hits

The second change addresses a question every public research surface eventually has to answer out loud: how many real people read this. The new script /root/Atlas_Backend/atlas_traffic_truth_audit.py compares Atlas beacon humans with raw nginx hits drawn from /root/Traffic_Monitor/data/traffic.db, and writes both a JSON and a Markdown report under /root/Atlas_Backend/cache/traffic_truth/.

The latest run reported these figures for the trailing 24 hours:

• 24h verified engaged humans: 10
• 24h raw Atlas nginx hits: 1,470
• 24h raw bot hit percent: 22.86
• 7d verified engaged humans: 10

The operating rule recorded with the audit is plain: use beacon verified engaged humans as the audience number, and treat raw nginx hits as load, crawl, and API activity rather than visitors. A read-only endpoint at GET /api/atlas/traffic/summary exposes the same shape to the workstation, filtered to atlas.freedomcore.io, returning verified-human counts, multi-page counts, and referrer buckets. The endpoint deduplicates referrers by a daily hash inside each source, so one visitor browsing several URLs is not counted several times. For the same window it returned 10 verified humans today, 6 of them multi-page, across 67 events, with referrers split freedomcore=8 and direct=2. The top pages were the Atlas home, the symbol coverage index at /stocks/, and the Apple page at /stocks/AAPL/.

Ten readers is a small number stated without decoration. That is the intended behaviour of cross asset market intelligence built for trust rather than for a vanity dashboard.

Coverage: 171 Symbols, 75 Indexable, 82 Sitemap URLs

The Atlas symbol coverage build on 2026-06-02 produced 171 symbol pages. Of those, 75 stayed indexable and 96 top-100 crypto expansion pages stayed noindex, follow, which keeps crawl budget pointed at the pages that carry authored research. The build emitted 82 sitemap URLs and the Atlas Pulse JSON rebuilt with 171 of 171 validated coverage.

Health and quality checks backed the build. The endpoint curl -fsS http://127.0.0.1:8081/api/atlas/health returned ok=true with atlas_substrate_v2.counts.authored_ready=171. The full writer QA report at /root/Atlas_Backend/cache/qa/atlas_full_writer_qa_20260601_230652.json recorded symbol_count=171, depth_counts={"authored_ready": 171}, all six views returning SCOUT_RESEARCH_READY, and empty issue_counts and warning_counts.

A static spot-check confirmed the indexing policy on disk: /var/www/atlas/stocks/AAPL/index.html carries <meta name="robots" content="index, follow">, while /var/www/atlas/stocks/XMR/index.html carries <meta name="robots" content="noindex, follow">. The split is deliberate, not accidental.

The Free-Source Spine Behind The Pages

The research notebook is assembled from named public sources rather than a private feed. The recent-files trail shows the spine refreshing through the morning of 2026-06-02: SEC company facts and submissions files such as sec_companyfacts_0000320193.json for Apple, paired with per-symbol substrate files like aapl.v2.json and daily-context files like AAPL.json, plus CoinGecko coin profiles under cache/market_universe/coingecko_coin_profiles/. The same window touched substrates for MSFT, GOOGL, AMZN, NVDA, META, TSLA, AMD, INTC, and AVGO.

This is the OpenBB DefiLlama RSS market spine pattern: regulator filings, on-chain and market aggregators, and dated RSS context, blended per symbol and stamped with provenance. The methodology behind those source choices is documented openly, and the upstream tooling is worth reading directly. The OpenBB platform docs at docs.openbb.co and the DefiLlama API at defillama.com describe the public interfaces this spine draws from. The page structure itself follows the Schema.org BlogPosting and article conventions so search engines can read the surface cleanly.

What Pro Adds On Top Of The Free Snapshot

The public symbol pages now state the free and Pro split clearly. Pro adds the authored report, the source pack, a risk register, a catalyst map, a PDF and CSV export trail, a receipt email, a watchlist, and archive controls. A PDF export probe through the Pro gate on 2026-06-02 returned 200 OK for GET /api/atlas/export/AAPL.pdf?template=operator_sheet&style=authored with content length 23,297 bytes, and pdftotext confirmed a signed cover stamped SIGNED BY and FreedomCore Atlas Research Desk. The snapshot PDF export, by contrast, correctly returned 403 FORBIDDEN.

You can browse the public side now through the Atlas symbol coverage index and the Atlas notes archive. The wider system, including the autonomous engine work at Maverick and the franchise hub at FreedomCore, sits alongside Atlas but is governed by separate surfaces and separate rules.

Caveats: What Is Not Proven Or Not Live

Several things are deliberately unfinished, and stating them is part of the methodology.

• Gmail and external customer email delivery remains deferred and untouched as of 2026-06-02. The Postmark approval guard in atlas_receipt_worker.py still holds external recipients, and the system will not falsely mark a Gmail message as sent while delivery is blocked.
• The traffic figures are a single trailing-window snapshot, not a trend. Ten verified humans over 24 hours and 7 days is a starting baseline, and the honest read is that crawl-and-index proof from search engines is still the open metric.
• The 96 expansion pages are intentionally held at noindex, follow. They exist for internal linking and future promotion, and none of them should be read as live indexable coverage yet.
• None of this is investment advice. The Atlas research notebook publishes methodology, source provenance, and dated build notes. It does not publish trade calls.

The thread running through both passes is the same. A research surface earns trust by making its own claims checkable: distinct views that are actually distinct, a visitor count that is a visitor count, and a coverage number that matches what is on disk.