Atlas · MACRO NOTE
Published 2026-06-19

Atlas Research Notebook Build Log: 225 Symbol Pages, GSC Crawl Truth, and a Macro Market Structure Spine

The Atlas research notebook now publishes against a measured pipeline state: 225 symbol pages rebuilt, 177 of them indexable, and a stocks sitemap carrying 184 URLs, all confirmed by the production service run logged at 07:46 BST on 2026-06-18.

Table of Contents

Symbol Page Pipeline State on 2026-06-18

The Atlas writer pipeline was reworked so that a healthy queue backlog no longer reads as a hard failure. The atlas-dynamic-news.service runs in queue mode, but atlas_full_writer_qa.py previously treated queued material writer work as missing output, which made the daily service fail after the source and substrate stages had already succeeded.

The patched quality assurance now detects pending and running writer-queue symbols under cache/daily_writer_queue/{pending,running}. Missing snapshot, report, or authored views on queued symbols are now classed as warnings named writer_queue_backlog:* rather than hard failures. Completed and non-queued symbols still fail normally on missing reports, wrong renderers, duplicate views, policy markers, or banned dash punctuation.

After the patch, the first full quality assurance pass found 19 missing-view symbols that were not already queued: MSFT, MU, CRM, BRK.B, V, COIN, HOOD, COST, HD, UNH, JNJ, PFE, XOM, CVX, BA, CAT, GE, RDDT, and UBER. Standard pending writer jobs were added for those under cache/daily_writer_queue/pending with the queue tag qa_missing_view_backfill. The follow-up full pass then reported ok=true across 171 symbols with issue_counts={} and writer_queue_backlog_symbol_count=79.

The production service run exited status=0/SUCCESS, rebuilt 225 symbol pages, 177 indexable, and wrote a stocks sitemap of 184 URLs. The batch report landed at /root/Atlas_Backend/cache/daily_batch_reports/atlas_daily_batch_20260618_063915.json. Live routes verified at 200 after the change include https://atlas.freedomcore.io/, /workstation/, /datasets/, /free-account/, and /stocks/SPY/.

GSC Crawl Truth for the Atlas Research Surface

Search Console exports were parsed from /root/Atlas_Backend/https___atlas.freedomcore.io_-Performance-on-Search-2026-06-18.zip and /root/Atlas_Backend/https___atlas.freedomcore.io_-Coverage-2026-06-18.zip. The performance export is hourly and shows 48 impressions, 0 clicks, and a weighted average position of about 63.5.

The coverage export still reports only through 2026-06-12, with 81 indexed, 137 not indexed, 111 discovered-currently-not-indexed, and 3 crawled-currently-not-indexed. The seven-day exports immediately before this were rising: 323 impressions in the 2026-06-16 seven-day export and 417 in the 2026-06-17 seven-day export. The honest read is that the drop from the 111-impression hourly spike to about 48 is short-window volatility, not evidence that the dataset authority pages hurt the surface.

The nginx access logs do give concrete crawl proof. Googlebot has hit /datasets/atlas-symbol-coverage/, /datasets/atlas-market-pulse/, and /datasets/atlas-security-posture/, and on the audit day Googlebot also hit /economic-calendar/. Bingbot hit /datasets/sitemap.xml, and Amazonbot, Semrush, Ahrefs, MJ12, and Yandex also touched dataset and feed routes.

Public Dataset Authority Layer and External Mirrors

The Atlas public proof layer is wired for discovery. The robots.txt file advertises /sitemap.xml, /stocks/sitemap.xml, /notes/sitemap.xml, /sitemap-apex.xml, and /datasets/sitemap.xml. The /datasets/authority/ page returns 200 with index,follow, a self-canonical tag, and ordinary HTML anchors to dataset routes and GitHub. The /datasets/sitemap.xml carries 5 canonical URLs with lastmod 2026-06-18.

The dataset manifest at /datasets/atlas-authority-manifest.json reports version 20260618-f93d0d2ea67f, generated 2026-06-18T06:13:24Z, with 175 public rows and 226 total rows. The atlas-dataset-push.service ran successfully at 07:15 BST and pushed HuggingFace, Kaggle, and GitHub (github.com/freedomcoreai-lang/freedomcore-datasets) for all three datasets. Zenodo was intentionally skipped because it re-versions weekly, and data.world was skipped because DW_TOKEN and DW_OWNER are not configured.

The honest framing on these mirrors is that GitHub, HuggingFace, and Kaggle pages should be treated as discovery, provenance, and AI-citation surfaces, not as guaranteed ranking authority. That distinction is the difference between a real cross asset market intelligence record and a link-juice claim that the logs cannot support.

Economic Calendar Coverage and the US Indicator Page Family

The macro market structure notes lean on the economic calendar stack, which is the public product name behind Atlas Radar. The /economic-calendar/ route returns 200, and /api/atlas/economic-calendar?region=US&days=14&impact=all returns 200 with official-source macro event data. There are 15 generated indicator pages under /economic-calendar/us-*, all serving 200, and the atlas-indicator-pages.service ran at 07:20 BST with count=15 and skipped=[].

These 15 Radar pages are already listed in /sitemap-apex.xml, so discovery is not broken. The economic calendar research surface is a long-tail target across queries like US CPI latest, nonfarm payrolls date, jobless claims today, PCE inflation latest, and FOMC calendar. The calendar's surprise column now renders an animated hawkish, dovish, or neutral badge with the sigma value kept as small print, and the direction flips per indicator in economic_calendar._policy_lean, where hot inflation or jobs reads hawkish and rising unemployment, claims, or weak housing reads dovish.

Data Source Methodology and Freshness Gates

Source quality is enforced with dated freshness gates rather than blind passthrough. Euro-area inflation is wired from the ECB Data Portal HICP feed inside uk_eu_values behind a 75-day freshness gate. That feed is currently stale, with its latest reading at 2025-12, so the surface shows an honest n/a and is set to auto-activate when the feed catches up. The gate has been proven to render 1.9 percent year on year with sigma when relaxed, which is exactly the behaviour a methodology-led surface should show before publishing a number.

The OpenBB DefiLlama RSS market spine and the wider Atlas source stack are documented in public methodology pages and dated build notes rather than asserted. For the underlying source tooling, see the OpenBB docs and the DefiLlama site. Public notes like this one are structured as a Schema.org BlogPosting so search engines can read the Atlas research surface as a dated, attributable record.

Theme and Navigation Fixes Across Public Atlas Pages

Two public-page corrections shipped on 2026-06-18. Published note pages dated before about 2026-06-11 did not load /js/theme-init.js, so they could not read the shared fc-theme cookie and follow the application day and night toggle. The fix backfilled <script src="/js/theme-init.js"></script> into the 13 older note pages under /var/www/atlas/notes/, so every public note now resolves theme from the cookie.

Separately, the Atlas public data catalog and proof-layer pages, including /datasets/, /datasets/authority/, atlas-symbol-coverage, atlas-market-pulse, and atlas-security-posture, were hardcoded to light mode. The builder atlas_authority_assets.py, run by atlas-authority-refresh.service, was patched to add theme-init.js, a data-theme="dark" default, and a full dark override block, and all five pages were regenerated. The Atlas mobile burger menu was also unified into one identical 11-item menu across pages/atlas.html, workstation/index.html, and economic-calendar/index.html, dropping duplicate research-terminal anchors. Coverage and notes can be browsed directly at the Atlas symbol coverage and Atlas notes routes, and the wider stack sits under FreedomCore alongside Maverick.

Caveats and What Is Not Yet Proven

The coverage export lag means index counts are dated to 2026-06-12 and the newer authority links cannot yet be confirmed in Search Console. The euro-area HICP number stays n/a until the ECB feed clears the 75-day gate. The data.world mirror is not live because DW_TOKEN and DW_OWNER are unset. GitHub, HuggingFace, and Kaggle mirrors are provenance and discovery surfaces, not proven ranking authority, and the post-spike GSC figures should be judged over a 7 to 14 day window rather than from a single hourly export. The supervised reading is that the surface is crawled, dated, and honest, and the next Search Console window is the test.

The daily note /notes/atlas-daily-note-2026-06-18-0516/ is live at 200, the writer-queue worker completed at 06:20 BST, and the authority refresh published at 07:13 BST. Browse the full Atlas research notebook at Atlas notes.

Browse the Atlas research notebook

FreedomCore Atlas Research →