deep-market-research

name	deep-market-research
description	perform deep research for a polymarket market, event, or thesis. use when the user wants fair-value work, evidence gathering, catalyst mapping, linked-market comparisons, or a detailed explanation of whether the market looks mispriced. for wide candidate sets or triage questions, use opportunity-classifier first.

deep market research

Use this skill to answer one question:

after doing the real evidence work, what should this market be worth and why?

This skill is the heavy analytical layer:

• downstream of opportunity-classifier and market-memo
• upstream of strategy-draft and order-ticket

It should produce a research synthesis with a defended fair-value view, not a casual narrative and not an execution plan.

default mode

• Default to research mode.
• In research mode, optimize for evidence quality, resolution discipline, catalyst mapping, and fair-value synthesis.
• Do not drift into broad triage, shallow memoing, or live order construction.

what this skill is for

Use this skill when the user is effectively asking:

• is this market mispriced?
• what evidence really matters here?
• what is a disciplined fair-value range?
• what catalysts could move this market?
• how do related markets change the view?

Use this skill for:

• one market or one tightly linked thesis cluster
• evidence-heavy work that goes beyond a memo
• fair-value synthesis from market data plus external evidence
• resolution-sensitive markets where mechanics and facts both matter

Do not use this skill to:

• rank a broad candidate set
• improvise a numerical forecast from vibes
• place, preview, submit, or cancel orders
• skip sourcing because the market already has a price

minimum viable input checklist

Do not finalize research until you have, or explicitly mark as missing:

• one resolved market identifier
• live market state
• resolution text or equivalent primary settlement source
• at least one primary or authoritative external evidence source
• at least one clear argument against the working thesis
• a fair-value view or a clear reason why it cannot yet be formed

For stronger research passes, also pull:

• related markets when cross-market consistency matters
• historical price or trade context when timing and drift matter
• persisted state when prior developments or past syntheses exist

outside-in frame

Before writing the synthesis, inspect the market from these lenses:

• resolution lens: what exactly settles this market, and where can interpretation break?
• evidence lens: which sources are primary, which are secondary, and which are noise?
• model lens: can this be priced with real numerical discipline, or is it still too underdetermined?
• catalyst lens: what future developments are most likely to move the probability?
• workflow lens: is deep research the right next step, or should this still be triage, memo, modeling, or strategy?

If the workflow lens says the main missing piece is numerical forecasting, use actuarial-forecasting rather than improvising percentages.

dependency and handoff rules

If the main missing piece is:

• broad triage or ranking across many markets -> opportunity-classifier
• a compact live-state brief before deeper work -> market-memo
• explicit probability modeling, scenario weights, threshold pricing, or visible numerical anchors -> actuarial-forecasting
• entry timing, invalidation, or exit framing -> strategy-draft
• exact preview-safe execution parameters -> order-ticket

For expensive or ambiguous names, consider narrow read-only subagents such as:

• rules_auditor
• catalyst_researcher
• microstructure_analyst
• related_market_mapper

Spawn them only when their output could materially change the fair-value view or the next handoff.

If actuarial-forecasting is not installed locally, use the fallback at ossianravn/actuarial-forecasting.

main workflow

1. Normalize the assignment

Write a 1-2 line intent statement for yourself covering:

• what market or thesis cluster is being researched
• what pricing or decision question the research must answer
• whether the goal is a first-pass synthesis or a higher-conviction fair-value call

2. Refresh live market context

Before external research, confirm the live setup with the minimum needed reads:

• always start with get_market_snapshot
• add get_orderbook when spread or tradability affects interpretation
• add get_recent_trades when recent flow may reveal stale pricing or catalyst-driven activity
• add get_price_history when timing, drift, or volatility matters
• add get_market_state when existing stored research or developments could save repeated work

3. Lock down resolution mechanics first

Identify:

• what counts as a win
• who or what likely determines resolution
• what edge-case wording matters
• what timing matters for both settlement and trading

Do not do probability work on a market whose settlement mechanics you have not understood.

4. Build the evidence map

Separate evidence into:

• supports yes
• supports no
• unresolved

Prefer:

• official documents
• primary data
• company filings
• regulator or election authority materials
• direct timeline evidence

Use secondary reporting only to point toward stronger sources or to capture genuinely current developments.

5. Form or import the numerical view

If the question requires real forecasting:

• use actuarial-forecasting for explicit probability modeling, scenario weights, and visible numerical anchors
• then bring that output back into the research synthesis

If a full model is not possible, say why and bound the uncertainty rather than manufacturing precision.

6. Compare fair value with the live market

Make the pricing comparison explicit:

• live implied probability
• low / base / high fair value
• main reason the market may still be right even if your base case differs

Never use the market price itself as evidence for the thesis.

7. State what would change the view

Name:

• the next catalyst
• the strongest disconfirming evidence
• the single biggest assumption
• the main reason not to trade yet, if applicable

8. Persist reusable work when available

When state tools are available:

• call record_research_synthesis with the final thesis, fair-value range, and evidence map
• call record_development for standalone catalysts or developments that should be queryable later

Persist only the pieces likely to remain useful beyond the current thread.

hard rules

• Do not place, preview, submit, or cancel orders in this skill.
• Do not skip resolution analysis because the market looks obvious.
• Do not treat headlines, comments, or market chatter as sufficient evidence on their own.
• Do not improvise exact percentages when the problem needs real modeling.
• Do not hide uncertainty inside a polished narrative.
• Do not let related-market comparisons replace primary evidence.
• Do not forget the strongest counter-case.

output contract

Return exactly these sections in this order.

[market title]

research intent

State in 1-2 sentences:

• what pricing question the research is trying to answer
• what the current scope is

current market pricing

List:

• live implied probability
• recent move
• spread or liquidity note
• any flow feature that matters for interpretation

resolution and mechanics

List:

• what resolves the market
• what counts as a win
• key timing detail
• main ambiguity or wording risk

evidence map

Use flat bullets under:

• supports yes
• supports no
• unresolved

linked markets

List:

• related markets that should move with this one
• any meaningful inconsistency
• whether the cross-market read strengthens or weakens the thesis

fair-value view

List:

• low
• base
• high
• whether this came from direct modeling, bounded inference, or actuarial-forecasting

market versus fair value

State:

• whether the live market looks under, over, or roughly fairly priced
• the strongest reason the market could still be right

what would change the view

List:

• disconfirming evidence
• next catalyst
• biggest assumption
• reason not to trade yet, if any

next handoff

Choose one:

• remain in deep-market-research
• hand off to actuarial-forecasting
• hand off to strategy-draft
• hand off to order-ticket

Explain why that is the cheapest useful next step.

completion criteria

The research pass is complete only when:

• mechanics and resolution risk are explicit
• external evidence is separated from market observations
• the strongest counter-case is present
• the fair-value view is defended or clearly bounded
• the comparison with live pricing is explicit
• the next handoff is singular and justified

common failure modes

• writing a long memo with no real evidence map
• skipping primary sources because current news feels enough
• smuggling market price into the evidence base
• outputting fake precision on an under-modeled question
• ignoring the strongest reason the market may still be right
• drifting into strategy or execution before the research is done

activation tests

Should trigger

• "do a deep research pass on this market."
• "is this market mispriced?"
• "build the evidence map and fair-value range."
• "compare this market to related names and tell me what it should be worth."
• "research this thesis before we build a trade plan."

Should not trigger

• "which of these markets are worth looking at?"
• "give me a short memo on this market."
• "turn this thesis into an entry plan."
• "place a 25c bid."
• "just summarize the book."

resource map

• opportunity-classifier — use when the main question is whether the name deserves research at all
• market-memo — use when a compact live-state brief is the missing prerequisite
• strategy-draft — use when the fair-value view exists and the next problem is execution framing
• order-ticket — use when the strategy is ready to become a guarded trade plan