How the research desk works

Ticker Scan is the starting point of any Desk workflow. It replaces the ten tabs you would normally open before taking a position: quote, profile, valuation metrics, social and news sentiment, analyst consensus and the latest headlines are fused into one scrollable report with a transparent composite score.

• Stock or ETF ticker, 1–5 characters (AAPL, NVDA, TSLA, MSFT, AMZN, HOOD, SPY).
• Legacy contract-address lookups are still accepted for on-chain assets and route to the original analysis flow.

• Factor scores are computed deterministically from the fetched metrics — no model is involved in the numbers.
• Every factor is shown individually so a composite can always be traced back to its inputs.
• The deep research layer makes exactly one language-model call per scan. Every source is sent as a labelled block and a single JSON object is returned; the response is parsed tolerantly and markdown fences are stripped.
• The model call is only persisted and charged after a successful parse. An errored or unparseable response returns a clean error and deducts nothing.
• Every source fetch runs in parallel with individual failure tolerance. A missing source degrades its own panel to a short factual note and never fails the scan.
• Charts use a single consistent accent colour in both directions. There is no red/green up-down colouring, because this is a facts tool rather than a signals tool.

Reports, including the full deep research layer, are cached for 60 minutes per symbol. Re-running the same ticker inside that window returns the cached report instantly and costs nothing.

• Tickers are validated before any work is done: uppercased and matched against ^[A-Z0-9.-]{1,10}$, then confirmed against live market data. Unknown symbols return a clear not-found message instead of an error.
• The empty search box carries no example symbol — it simply invites a ticker.
• Deep research panels are US-listed equity and ETF oriented. Non-US listings may return partial panels.
• Coverage follows the upstream market-data provider: illiquid OTC names may return partial metrics.
• AI analysis runs under a daily spend cap; when the cap is hit the rest of the report still renders.
• Quotes may be delayed depending on exchange and session.

Most research tools confirm what you already believe. Risk Desk is built to do the opposite: it takes a single free-text thesis and returns a structured bull-versus-bear grid so the weakest part of your reasoning is visible before you size a position.

• One free-text thesis input, e.g. "NVDA stays above 140 into earnings on datacenter demand".

On demand. Each submission is a fresh analysis; nothing is cached between runs.

• Runs under a daily credit cap; the module reports remaining budget when it is close to exhausted.
• Output is an argumentation aid, not a price forecast and not investment advice.

Fundamentals is the comparison view. Instead of scanning one symbol at a time, you see the whole universe ranked on the factor you care about, then jump straight into Ticker Scan for the name that stands out.

• No input required — the universe loads automatically. Sort by clicking any column header.

• Scoring is deterministic: each metric is normalised against the universe, then weighted into the composite.
• Identical inputs always produce an identical score — there is no model randomness in this module.

The universe snapshot is cached for 15 minutes and shared across all viewers.

• The universe is intentionally curated for coverage quality rather than breadth.
• Tickers with missing upstream metrics score on the factors that are available and are flagged as partial.

Market Overview answers the first question of the session — what is moving and why. It pairs a live watchlist with an aggregated news stream so you see the crowd turning before it shows up in your positions.

• No input. The module tracks a fixed 10-ticker watchlist of high-volume names.

Refreshes on load and on manual refresh; upstream data is polled on a short server-side interval.

• Breadth is limited to the watchlist rather than the full market.
• News tone is a heuristic aggregate, not a fundamental judgement.

Warning Signals is the exit-side counterpart to Ticker Scan. It surfaces the patterns that historically precede a drawdown — distribution by insiders, a cluster of downgrades, a sentiment roll-over — and ranks them by severity so the loudest signal is at the top.

• No input. The module continuously screens its coverage list.

Screens refresh every 15 minutes.

• A signal is a prompt to investigate, not a sell instruction.
• Insider filings are reported with a regulatory lag inherent to the disclosure process.

Smart Flow is the calendar-and-positioning view. It tells you which names have a scheduled catalyst in the near term and where professional opinion has moved recently, so you are never surprised by a print you could have seen coming.

• No input. Coverage is a fixed earnings watchlist of large, widely traded names.

Refreshed on load with a short server-side cache to stay inside upstream rate limits.

• Confirmed dates can move; treat pre-announcement dates as provisional.
• Estimate coverage varies by name and is omitted where the provider has none.

Market Feed is the ambient layer. It runs alongside the analytical modules so you can keep one eye on the wire while you work, filtered down to the categories that matter to your book.

• No input. Filter the stream by category using the toggles at the top of the module.

Polled continuously while the module is open.

• The feed reflects publisher coverage — absence of a story is not evidence of absence of an event.
• Headlines are not verified or editorialised by Desk.