Operator's guide to ATLAS

Research templates

ATLAS does not accept open-ended free-text prompts from the public surface. Every request is a whitelisted template plus a ticker (or list of tickers) plus an output style. This is deliberate: it keeps queries auditable, prevents prompt injection, and lets the data rail stage the right pulls before the model writes the brief.

Output style is one of: operator, brief, tweet, council, or table. Style governs format, not content. The same template + ticker returns the same numbers regardless of style.

Tiers & caps

ATLAS uses the standard FreedomCore subscription tiers. Free templates run anonymously up to a small daily cap; gated templates require a logged-in subscriber on the matching tier. There is one bill, one account, one rail across the whole suite.

Caps reset at midnight UTC. The workstation status pills show your current tier, today's used / cap, and how many briefs you have remaining. When a cap is hit, ATLAS returns a rate_limited response and the workstation renders a clear "DAILY CAP REACHED" message — no fake content is generated to fill the gap.

Provider state

ATLAS's request/response loop, validation, tier gating, daily caps, and audit ledger are all live. The no-paid public-source rail is also live for SEC company lookup, recent filings, and XBRL company facts on US-listed tickers. Paid market data, redistribution-sensitive feeds, and model-composed briefs remain deliberately gated until terms and keys are cleared.

Until those keys are wired, every POST /api/atlas/query returns a structured 503 PROVIDER_NOT_CONFIGURED response and the workstation renders a "BACKEND ONLINE / PROVIDERS PENDING" panel. The market center can still show SEC public filings and company facts where available. This is intentional: ATLAS will not fabricate prices, event dates, or statement values to hide a missing provider.

What unblocks live briefs:

1. Paid market-data redistribution terms cleared and the REST key installed server-side.
2. Model-provider commercial-use terms cleared and the corresponding key installed server-side.

Both are commercial decisions, not engineering. When they land, the same exact request/response loop you see today produces a real brief — no public-side change needed.

v1 scout-seat note. ATLAS's scout research currently runs on a single working NVIDIA seat — llama4_maverick — selected because it returned cleanly under the direct-route uptime conditions during this build. Lane diversity comes from prompt and task separation, not model diversity: each lane (source pack, filings and fundamentals, news catalysts) gets its own prompt and produces its own cited evidence row in the audit ledger. Additional scout seats listed in models.yaml will rejoin the rotation as their NIM model availability and uptime stabilise and their commercial terms are clean — at which point lane diversity becomes model diversity too, with no public-side change needed.

Data & output lifecycle

Every brief is composed from numbers fetched on demand. ATLAS does not maintain a long-lived cache of asset values that you might be reading "live" while they are actually stale. The data flow is:

1. Browser submits {template, ticker, style} over HTTPS to the ATLAS backend on the same origin.
2. Backend validates the template, ticker, and style against a strict whitelist; unsupported fields are rejected with 400 unsupported_fields.
3. Backend checks subscription tier and daily cap; tier or cap failures return structured errors before any provider call is made.
4. Free public-source endpoints can pull SEC ticker identity, recent filings, and company facts without paid credentials. Paid quote/news/provider calls remain off until commercial terms are cleared.
5. When gated providers are enabled, the backend hands source data and template instructions to the model rail and returns the composed brief in data.brief.
6. The response is rendered into the workstation response panel verbatim. Numbers in the brief come from the data rail; narrative comes from the model.

ATLAS does not generate placeholder numbers, recycle yesterday's cached values, or simulate a brief when providers are disabled. Anything you see in the workstation outside that flow is clearly labelled (idle demo / pending / cap reached / tier gate / network error).

Query privacy

ATLAS keeps an audit ledger so we can tune caps, debug failures, and refuse abusive traffic. The ledger records, per query:

• Account email and tier only when authenticated; anonymous queries record no email.
• Template ID, ticker(s), and output style.
• Provider counters (cache hit / miss / refused) and HTTP status.
• A payload hash for replay detection.
• A salted IP hash for daily cap accounting on anonymous traffic. The salt comes from a server-side secret; raw IPs are not retained.

The ledger does not retain:

• Raw request bodies, free-text prompts, or anything outside the whitelisted template fields.
• Generated briefs (the brief is sent to your browser; it is not archived per-user).
• Raw IP addresses, browser fingerprints, or third-party tracking identifiers.

Ledger rows are kept for operational tuning only. They are not sold, syndicated, or used to profile users beyond cap accounting and abuse prevention.

Compliance & limits

ATLAS is a research surface. Several things are off the table by design:

• No financial advice. Briefs are research output. They do not constitute personalised advice, a recommendation, or an inducement to transact. Decisions remain yours.
• No brokerage. ATLAS does not place trades, route orders, custody assets, or execute on your behalf. There is no "buy" button anywhere on this surface.
• No guaranteed returns. Past performance, valuation snapshots, and forward-looking statements are descriptive, not predictive.
• No tax, legal, or jurisdictional advice. Local rules vary; consult a licensed professional before acting on anything ATLAS surfaces.
• Data freshness limits. Public SEC data, macro sources, and future paid providers each have their own refresh cadence. ATLAS does not invent values to fill gaps; missing data is marked.

API surface

ATLAS's public API is REST, same-origin, and reachable from atlas.freedomcore.io, maverick.freedomcore.io, and the FreedomCore apex. There is no WebSocket today; all calls are synchronous request / response.

Error envelope is consistent: an HTTP status plus a JSON body with error, message, and (where useful) request_id and fields. The workstation maps every error to a human-readable response panel state.

FreedomCore family docs

ATLAS is one of several FreedomCore product surfaces. Each product publishes its own documentation, scoped to its own product:

• SHADOW docs → — copy-trading evidence, wallet tiers, attribution, risk controls.
• ARENA docs → — public leaderboard methodology, wallet pages, scoring, limits.
• You are here: ATLAS docs — market intelligence terminal.