AI Trading Library Apply for a private system →

The trading team · Guide 03

News agent, Analyst and Researcher: facts, context and checkable rules

13 min read · The seats

In one lineThe News agent turns the calendar you paste into an event sheet, the Analyst describes today's market in the same six sections every day, and the Researcher turns your setup from a feeling into yes/no rules, then checks today's chart against them.

What you'll get

  • The News agent prompt: today's events and blackout windows, built only from what you paste.
  • The full Analyst prompt and what a good brief looks like.
  • The two Researcher prompts: build your setup sheet once, then run the daily setup watch.
  • A worked example on NQ, from a messy description to a clean rule sheet.
  • Red flags for both roles, so you catch bad output before it reaches a trade.

1The News agent: events from your calendar, never from memory

A chat model does not know what is on today's calendar. Ask it anyway and it will often produce a plausible list from the shape of a typical month. That is how a trader ends up avoiding a release that isn't happening, or sitting through one that is.

So the team has one rule for news: the only events that exist are the ones you paste. Each morning, copy today's rows from the calendar you use, as text. The News agent turns them into an event sheet: times in your time zone, inside or outside your session, and the blackout windows your own news rule creates. The same sheet then goes to the Analyst and to the Risk Manager, so both work from identical events.

News agent prompt (chat A, during the five platform minutes)

Prompt32 lines
Role: News agent.

Build today's EVENT SHEET from the calendar text below and nothing else.

Inputs:
- Today: {date, weekday}
- Calendar I use: {name of the calendar}
- Time zone shown on the calendar: {e.g. New York / London / UTC}
- My time zone for the team: {e.g. New York}
- My session window: {e.g. 09:45–11:30 New York}
- Currencies or markets I care about: {e.g. USD for MNQ; EUR and USD for EURUSD}
- My news rule: {e.g. no entries from 10 minutes before to 10 minutes after any high-impact event; no new trades on rate-decision days}

Calendar text (pasted by me, unedited):
{paste the rows for today}

You may not:
- Add, rename or "correct" any event from memory. If an event is not in the pasted text, it does not exist for this sheet.
- Guess an impact level or a time. If either is missing in the text, write UNKNOWN.
- Forecast the number, the reaction or the direction. No comment on what the event means for price.

Reply in exactly this format:
EVENT SHEET — {date} — times in {my time zone}
1. Events, in time order: time · currency/market · event name as pasted · impact as pasted · INSIDE or OUTSIDE my session window.
2. Blackout windows from my news rule: start–end for each event the rule covers, and the event it comes from. Show the arithmetic (event time minus and plus my minutes).
3. Whole-day rules triggered: any "no trading today" line in my news rule that an event sets off, or "none".
4. Time-zone conversions made: from {calendar zone} to {my zone}, hours added or subtracted. If the two zones are the same, write "none".
5. Could not read: rows that were cut off, duplicated or unclear, quoted as pasted.
6. Count check: "{n} rows pasted, {n} events listed."

If the pasted text is empty, reply only: "No calendar pasted. I can't build an event sheet from memory."
Before you send: confirm every event in section 1 appears word for word in the pasted text, and that the count in section 6 matches.

2The Analyst: same shape, every day

The Analyst's job is boring on purpose. It describes structure, levels, events and a regime label. It doesn't suggest trades. Its value comes from consistency: when every brief has the same six sections, you start to notice what's different today.

Before you run it: have 2–3 screenshots ready (higher timeframe first) and the event sheet.

Analyst prompt (chat A)

Prompt26 lines
Role: Analyst.

Inputs:
- Instrument: {NQ / MNQ / EURUSD / ...}
- Session I trade: {e.g. New York, 09:45–11:30 ET}
- Today: {date, weekday}
- Attached screenshots: {e.g. "image 1 = 4H, image 2 = 1H, image 3 = 15m, latest bar on the right"}
- Exact levels from my data (optional): {e.g. prior day high / low / close, overnight high / low, or "none"}
- Today's event sheet (from the News agent, pasted by me): {paste, or "none pasted"}
- Headlines (optional, pasted by me): {paste, or "none"}

You may not:
- Suggest a trade, an entry, a direction to favour or a bias stronger than the evidence.
- Write any price that is not on my screenshots or in my data. Tag every price (chart) or (data).
- Mention any event or headline I did not paste.

Write the MARKET BRIEF in exactly this format:
1. Structure, higher to lower timeframe: one line each (up / down / sideways, last swing high and last swing low as prices).
2. Key levels: up to 5. Price, source tag, and what it is (prior day high, overnight low, range edge...).
3. Scheduled events: copy the INSIDE-window events and blackout windows from my event sheet. If none pasted, write "none pasted".
4. Regime label, one of: TRENDING, RANGING, QUIET, EVENT DAY. One sentence why, citing sections 1–3.
5. What would change this read: one observable event.
6. Unknowns: everything you could not read from the images, and any timeframe that disagrees with another.

If an image is too small or blurred to read a price, write CAN'T READ for that line. Do not estimate.
Before you send: check that every price has a tag, that section 3 contains nothing I did not paste, and that no line tells me what to do.

The regime label matters more than it looks. Four words only: TRENDING, RANGING, QUIET, EVENT DAY. Most setups are built for one or two of them. Once your journal has enough rows, you can count how your cards were scored and followed in each.

3The Researcher, job one: build the setup sheet

Most setups live in the trader's head as a feeling: "a clean retest with a strong candle". You can't check a feeling against a chart, and neither can an AI. The Researcher's first job is to rewrite your setup as rules that a stranger could check by looking at a chart or a clock.

You do this once per setup, then paste the result into your rules file.

From a feeling to rules you can checkTap to enlarge

Researcher · setup sheet prompt (chat A, or any chat in the workspace)

Prompt27 lines
Role: Researcher. Job: SETUP SHEET.

Here is my setup in my own words:
"{describe the setup the way you would explain it to a friend}"

You may not:
- Add a condition, filter, indicator or number I did not say. If the sheet needs one, ask.
- Say whether the setup is good, or whether it works. Your job is wording, not judgement.
- Fill a gap with an assumption. A gap becomes a question.

Reply in exactly this format:
SETUP SHEET (draft {n})
- Name: a short, fixed name for my journal.
- Context rules: what must be true before the setup can exist (timeframe, session, structure).
- Trigger rule: the exact event that makes it a trade.
- Entry: where, and with which order type.
- Invalidation (stop): where the idea is wrong, as a rule I can point to on the chart.
- Target: where I take profit, as a rule.
- Time exit: when I close it if neither stop nor target is hit, or "none stated".
- No-trade conditions: when I skip even if everything else passes.

VAGUE WORDS: every judgement word in my description ("strong", "clean", "looks like", "big"), each with two measurable replacements for me to choose between. Do not choose for me.
QUESTIONS: up to 3 things you need from me before the sheet is complete.
TRACE: next to each rule, quote the words of mine it came from. A rule with no quote is yours, not mine. Delete it.

Every rule must be a yes/no question I can answer by looking at a chart or a clock.
Before you send: reread each rule. If two traders could disagree about the answer on the same chart, it is still vague. List it under VAGUE WORDS.

Answer its questions. Run it again with your answers. Two or three rounds is normal. Stop when every rule is a yes/no question.

The numbers are yours. For each vague word the Researcher offers two measurable replacements and you choose. It must not choose for you, and neither it nor this library knows which value is better on your market. If you want to find out, that is a research question, and Guide 11 has the method.

4The Researcher, job two: the daily setup watch

Every morning, the Researcher reads the Analyst's brief and today's chart, checks each context rule, and marks where your trigger could happen today.

Researcher · setup watch prompt (chat A, right after the brief)

Prompt23 lines
Role: Researcher. Job: SETUP WATCH.

Inputs:
- My SETUP SHEET (in the team setup, or pasted here): {paste if needed}
- The Analyst brief above in this chat.
- Attached: {current chart screenshot(s), timeframe labelled}
- Exact levels from my data (optional): {paste, or "none"}

You may not:
- Upgrade a CAN'T TELL to PASS, or soften a FAIL.
- Propose a watch zone at a price that is not on my chart or in my data.
- Comment on whether today "looks good".

Reply in exactly this format:
SETUP WATCH — {setup name} — {date}
0. First line: "SETUP POSSIBLE" or "NO SETUP TODAY: {the no-trade condition met}".
1. Rule check: each context rule with PASS, FAIL or CAN'T TELL, and the evidence (a price with its tag, or a time).
2. Watch zones: up to 2 places the trigger could happen today, each with the price area (tagged) and the trigger rule it needs.
3. No-trade conditions: each one, MET or NOT MET, with evidence.
4. What I must see live: the trigger as one sentence I can watch for.

If more than one context rule is CAN'T TELL, say the chart is not good enough to check and ask for a specific better screenshot or data.
Before you send: check that every PASS has evidence next to it and every price has a tag.

CAN'T TELL is a real answer. It means the chart you gave it doesn't show enough. Give it a better screenshot or exported data. Never let it, or yourself, turn a CAN'T TELL into a PASS because you want the trade.

5Worked example: from a feeling to a sheet

Illustrative. MNQ, New York session, prices are examples.

The trader's words: "I buy when NQ breaks the opening range and comes back to retest it with a strong candle, as long as the overall trend is up. I don't trade big news mornings."

Vague words the Researcher found, and the replacements agreed:

VagueMeasurable replacement
opening rangeHigh and low of 09:30–09:45 New York time
breaksA 5-minute candle closes above the opening-range high
comes back to retestPrice trades back to within 3 points of the opening-range high after the break
strong candleA 5-minute candle closes back above the opening-range high, with its close in the top third of its range
overall trend is upOn the 1H chart, the last swing low is higher than the one before it, and price is above the prior day's close
big news morningsAny high-impact event between 09:30 and 11:00 on my calendar

The setup sheet:

  • Name: Opening-range retest (long)
  • Context rules:
    1. 1H last swing low is higher than the previous swing low.
    2. Price is above the prior day's close.
    3. Time is between 09:45 and 11:00 New York.
  • Trigger rule: after a 5m close above the opening-range high, price returns to within 3 points of it, and a 5m candle closes back above it with its close in the top third of its range.
  • Entry: limit at the opening-range high, or market on the trigger close if price is within 2 points.
  • Invalidation (stop): below the low of the trigger candle.
  • Target: prior day high, or 2R, whichever is closer.
  • Time exit: flat at 11:30 New York if neither stop nor target is hit. (From the trader's answer to the Researcher's question.)
  • No-trade conditions: a high-impact event between 09:30 and 11:00; the opening range is wider than 60 points; two losses already today.

The event sheet, that morning: four rows pasted, four listed. None inside 09:30–11:00. One at 14:00, outside the window, blackout 13:50–14:10.

The Analyst brief (abbreviated):

  1. 4H up, last swing high 21,520 (chart), swing low 21,310 (chart). 1H up, higher low at 21,402 (chart). 15m sideways inside the opening range.
  2. Prior day high 21,492.25 (data); prior day close 21,418.00 (data); overnight low 21,396.50 (data); opening range 21,431.25–21,452.00 (data).
  3. Events inside window: none on the event sheet. One at 14:00, outside window.
  4. Regime: TRENDING. Higher highs and higher lows on 4H and 1H.
  5. Would change: a 1H close below 21,402.
  6. Unknowns: volume isn't visible on the screenshot.

The setup watch:

  1. SETUP POSSIBLE.
  2. Context 1: PASS. 1H higher low at 21,402 (chart) above 21,310 (chart). Context 2: PASS. Price 21,447 (chart) above prior close 21,418.00 (data). Context 3: PASS at time of check.
  3. Watch zone 1: 21,449–21,455 (data: opening-range high 21,452.00). Needs the full trigger rule.
  4. No-trade conditions: event in 09:30–11:00 NOT MET (event sheet). Opening range wider than 60 points NOT MET, range is 20.75 points (data). Two losses today NOT MET (trader's count).
  5. What I must see live: a 5m close above 21,452, a return to within 3 points of it, then a 5m close back above it in the top third of its range.

Notice what's not there: no opinion, no "this looks great". Just rules, evidence and a sentence the trader can watch for.

6Common mistakes

  • Asking the AI what's on the calendar today. It will answer. Paste the rows.
  • Skipping the setup sheet. Without it, the daily watch is the AI guessing what you mean.
  • Letting the Researcher design your strategy. It rewrites your rules. It doesn't invent new ones. If it adds a rule you like, adopt it deliberately and write down the date.
  • Feeding it charts at random timeframes. Same timeframes, same order, every day.
  • Running the Researcher before the Analyst. The watch needs the brief for levels and the regime.

7Checklist

Your reading list is saved in this browser only.

Not financial advice. The founder is not a licensed financial advisor. Software and education only. Trading involves substantial risk of loss. Backtested results are hypothetical.
The AI Trading Library is published by Co trader LLC (Wyoming, USA) · Legal & disclaimers