Analyst Task Guide

An Analyst task is a prompt that runs on a schedule. Writing a good task is different from writing a one-off research prompt — it needs to work reliably across many runs, adapt to changing market conditions, and produce output that is useful without further editing.

Anatomy of an effective task

A strong Analyst task has four elements:

1. What to research — the specific subject or question
2. What sources to use — which data sources or document types to draw from
3. What to look for — thresholds, events, or signals that matter
4. What to produce — the format and length of the output

Example: Daily macro briefing

Each morning, summarise the key overnight developments across FX, rates, and equities relevant to a G10 macro portfolio. Focus on central bank communication, significant data releases, and large moves in key pairs (USD, EUR, JPY, GBP). Output a structured briefing: one paragraph per asset class, then a “What to watch today” section with three bullet points. Limit to 400 words.

Example: Earnings monitor

Each week, identify any companies in the S&P 500 that have reported earnings in the past seven days where actual EPS or revenue deviated from consensus by more than 10% in either direction. For each, provide: the actual vs consensus figures, the stock’s reaction on the day, and a one-sentence summary of management’s explanation. Format as a table.

Example: Credit watch

Every Monday, check for rating agency actions (upgrades, downgrades, outlook changes) on European investment-grade corporate issuers from the past week. For each action, note the issuer, prior and new rating, agency, and the stated rationale. Flag any issuers we hold in our portfolio. Output as a bullet list, most significant first.

Writing the persona

The Analyst’s persona sets the context for every task it runs. Think of it as the standing brief that every task inherits.

A useful persona:

• Names the domain and focus area
• Describes the intended reader
• Sets expectations for format, tone, and depth

Example persona:

You are a macro research analyst focused on G10 FX and rates markets. Your audience is portfolio managers who need concise, actionable briefings. Lead every output with a one-paragraph summary. Use plain financial English — no jargon that would require explanation to a senior analyst.

Scheduling principles

• Daily tasks work well for market monitoring, news digests, and price alerts.
• Weekly tasks suit deeper analysis like earnings roundups, credit watch reviews, and sector updates.
• Ad-hoc manual runs are available for tasks you want on demand without scheduling. Set the cadence to Manual and run the task from the dashboard when needed.

Avoid scheduling high-Quality tasks at very high frequency — long-running tasks that consume significant model context are better suited to daily or weekly cadences.

Using Date Range in tasks

Most scheduled tasks should leave Date Range unset and instead phrase the time reference in the prompt itself (e.g., “in the past seven days”, “since the last run”). This gives TwoFiftyTwo flexibility to find the most recent relevant material.

Set a fixed Date Range only when the task is intended to analyse a specific historical period that should not change between runs.

Testing before going live

Save the task as a Draft and run it manually once before activating it. Check:

• Is the output the right length and format?
• Are the citations pointing to the right sources?
• Does the structure hold up if there is nothing notable to report?

The last point is important. A good Analyst task degrades gracefully — if there are no material developments in a given week, the output should say so clearly rather than hallucinating content to fill the template.

Common pitfalls

Too broad a scope. A task that covers everything produces output that covers nothing well. Narrow the scope to one asset class, geography, or question type per task.

No output format specified. Without a specified format, output varies between runs. Always include length guidance and structure requirements.

Ambiguous time references. Avoid phrases like “recently” or “lately”. Use “in the past 24 hours”, “since last Monday”, or “in the past seven calendar days” so the window is unambiguous.

Over-relying on a single source. If your primary data source is unavailable, the task will produce thin output. Where possible, specify primary and fallback sources.