weekly-scan

name	weekly-scan
description	Run a weekly thesis-relevant news scan. Reads all active thesis/theme docs, runs targeted Tavily searches per thesis, checks for hypothesis triggers and disconfirming evidence, and writes a dated scan artifact to scans/. Use when the user says "run the weekly scan", "scan for news", "check thesis triggers", "what happened this week", or when the research methodology weekly cadence is due.

/weekly-scan

Scan thesis-relevant news, check for hypothesis triggers, and explicitly look for disconfirming evidence across all active positions.

Step 1 — Read active theses

Read all active thesis and theme docs from strategy/ to understand current positions and open hypotheses.

For each file matching strategy/thesis-*.md or strategy/theme-*.md:

1. Read the file
2. Extract the thesis name (from the # heading)
3. Extract the ## Active Positions table — list of tickers and roles
4. Extract the ## Hypotheses table — focus on rows with Status "open" and their Timeframe, If Confirmed/Disconfirmed columns
5. Extract any [!unverified] or [!gap] callouts that might be resolved by recent news
6. Note the thesis tags from frontmatter for search query construction

Organize a summary: for each thesis, you should have its name, key tickers, 2-3 top open hypotheses, any claims awaiting verification, and any near-term catalysts that should be added to docs/calendar.md.

Step 2 — Build search queries

For each thesis, construct 1-2 recurring targeted search queries based on:

• The thesis subject area (e.g., "AI infrastructure spending", "space economy launch")
• Active positions and their near-term catalysts (earnings, product launches, regulatory events)
• Open hypotheses nearing their timeframe deadlines
• Unverified claims or gaps that recent news might resolve
• Bear-case checks: failure to execute, missed milestones, adverse regulation, margin/financing pressure, valuation excess, or better competing expressions of the same thesis

Query construction rules:

• Use specific entity names where possible (company names, tickers, program names)
• Keep queries concise (5-8 words is ideal for search relevance)
• Do NOT search for price movements or trading signals
• Prefer stable recurring queries week to week, so scans are comparable over time. Only add ad hoc queries when a specific catalyst or hypothesis needs it.

Recurring query plan:

Thesis / Theme	Recurring queries
AI	hyperscaler capex AI infrastructure Amazon Microsoft Google Meta; AMKR NVTS AI data center advanced packaging power semiconductor
Crypto	Bitcoin ETF flows weekly Farside BlackRock IBIT; G20 crypto regulation exchange ban stablecoin
Energy Transition	Vistra data center PPA hyperscaler; spodumene lithium price AMPX commercial battery contract IRA storage tax credit
Space	AST SpaceMobile commercial service BlueBird launch; Rocket Lab Neutron Intuitive Machines IM-3 MDA government contract
Tech Conviction	NVIDIA AMD AI GPU market share TSMC A16 N2; Google Meta antitrust remedy regulatory action
Nuclear	Oklo Aurora INL criticality NRC ADVANCE Act; uranium spot price Sprott hyperscaler nuclear PPA
EV	NIO Li Auto deliveries gross operating margin; China EV price war BYD NIO Li Auto
Quantum	IonQ logical qubits SkyWater acquisition; Infleqtion revenue guidance quantum advantage
Robotics	Symbotic Walmart deployment customer diversification; Intuitive Surgical procedure growth Ion systems earnings

Step 3 — Run targeted searches

Use mcp__tavily-mcp__tavily_search for each query.

Search parameters:

• search_depth: "basic" (this is a scan, not deep research)
• max_results: 5 per query
• time_range: "week" (last 7 days — this is a weekly scan, not a monthly catch-up)
• topic: "general"

Rules:

• Run 1-2 searches per thesis (no more)
• Process theses in order: AI, Crypto, Energy Transition, Space, Tech Conviction, then themes (Nuclear, EV, Quantum, Robotics)
• If a search returns no relevant results, note it and move on — silence is fine
• Check article dates. Tavily's time_range filter is approximate. Before surfacing a finding, verify the article publication date falls within the scan window (last 7 days). Discard or clearly label anything older as "prior context, not new this week."

Do NOT use:

• mcp__snaptrade__list_positions or any SnapTrade tools
• mcp__tavily-mcp__tavily_research (that's for deep research, not scanning)

Step 4 — Evaluate findings

For each search result that is relevant to a thesis, filter by source quality FIRST, then evaluate.

Source quality filter (apply before evaluating):

• filing: SEC/SEDAR filings, regulatory filings, official government releases. Highest trust.
• primary/company: company investor releases, earnings materials, official product or mission updates.
• trade/wire: Reuters, Bloomberg, industry trade publications (DigiTimes, SpaceNews, NucNet, S&P Global, WNN).
• data provider: Farside, CoinGecko, Sprott, EIA, FRED, etc. Use with date and methodology caveat where relevant.
• commentary: Motley Fool, SimplyWallSt, analyst blogs, crypto-native publications. Only surface if the underlying data is verifiable.
• discard: social media — LinkedIn posts, X/Twitter, Reddit, Instagram — unless they link directly to a primary source.

Evaluation:

1. Check against the Hypotheses table in the relevant thesis doc:

   • Does the news confirm, disconfirm, or relate to any open hypothesis?
   • A hypothesis is "triggered" if the news provides conclusive evidence that would change its Status from "open" to "confirmed" or "disconfirmed"
   • Partial evidence (directional but not conclusive) is a directional signal, not a trigger

2. Check against open backlog items in docs/research-backlog.md:

   • Does any result answer an open research question?
   • Flag items that might be resolvable, but do NOT resolve them — that is what /research is for

3. Check against unverified claims ([!unverified] and [!gap] callouts):

   • Does any result provide source material for an unverified claim?
   • Note it for the user but do NOT update the callout — that requires /research

4. Check against existing records:

   • Search decisions/log.md, recent reviews/, and existing scans/ for the same signal.
   • Label each finding as New this week or Carried from prior log/review.
   • Do not duplicate a carried signal in the summary unless it remains thesis-relevant and no scan artifact has captured it yet.

5. Check for calendar items:

   • If a search surfaces an upcoming earnings date, mission window, regulatory deadline, or other catalyst tied to a hypothesis, add it to docs/calendar.md.
   • Calendar additions should link the relevant thesis/theme and hypothesis when possible.

6. Check for disconfirming evidence:

   • Identify evidence that weakens, falsifies, or complicates the thesis even if it is not a formal hypothesis trigger.
   • Separate adverse evidence from positive directional signals so reviews can see whether a thesis is strengthening or merely getting attention.
   • If no material adverse evidence appears, record that explicitly and name the thesis areas checked.

Step 5 — Present findings to user

Present a structured summary of findings. Categorize each finding by type:

Hypothesis triggers (conclusive evidence):

• Present the finding with source citation
• Ask the user for confirmation before updating the thesis doc
• If confirmed: update the thesis doc (Hypotheses table Status + Evaluated, ## Updates dated entry). The thesis doc is the permanent record — no separate log entry needed.

Directional signals (partial evidence):

• Summarize in 1-2 sentences with source citation
• These will be written to the scan artifact and accumulate for monthly review
• Label New this week vs. Carried from prior log/review
• Include the source tier

Potential backlog resolutions:

• Note which backlog item and what the finding is
• Recommend running /research {item_id} to properly resolve it

Disconfirming evidence:

• Summarize any adverse or thesis-weakening evidence with source citation
• Say whether it is a formal hypothesis trigger, a directional warning, or a kill-criteria review item
• Include it in the scan artifact even if the rest of the thesis section is positive

If nothing significant was found for a thesis:

• Say so explicitly: "No significant developments for {thesis name}."

Step 6 — Write scan artifact

Write the scan results to scans/YYYY-MM-DD.md using the template at templates/scan.md. Fill in every thesis section — even if the content is "Nothing significant."

Structure per thesis:

• Hypothesis Triggers — any conclusive evidence with source citation (publication + article date)
• Directional Signals — partial evidence with source citation
• Disconfirming Evidence — adverse evidence or "No material disconfirming evidence found" with areas checked
• Potential Backlog Resolutions — which backlog items have new source material

Every finding must include:

• Newness label: New this week or Carried from prior log/review
• Source tier: filing, primary/company, trade/wire, data provider, or commentary
• Source publication name
• Article publication date (verified against the scan window)
• 1-2 sentence explanation of what it means directionally

Frontmatter: fill in theses_scanned (count of active thesis/theme docs scanned), signals_found (count of hypothesis triggers + directional signals).

Counterweight: fill in the top-level ## Disconfirming Evidence section with a cross-thesis summary. This section must be non-empty.

Summary denominator: count active strategy/thesis-*.md and strategy/theme-*.md files. Do not hard-code 10.

Step 7 — Report scan summary

Report the final scan results to the user:

Weekly Scan Summary — YYYY-MM-DD
Scan written to: scans/YYYY-MM-DD.md
Theses/themes scanned: X/Y
Hypothesis triggers: N (list them)
Directional signals: M (list them)
Disconfirming signals: D (list them)
Nothing significant: K theses

Next steps: [any recommended actions — research items, hypothesis evaluations to watch]

Constraints

• Don't pull SnapTrade data. This is a news scan, not a portfolio valuation.
• Don't change conviction levels. Surface findings; the human decides if conviction changes.
• Don't deep-dive into any single item. If something needs deep investigation, recommend /research for it.
• Don't update callout types. Changing [!unverified] to [!source] requires the full /research workflow with source archiving.
• Don't log to decisions/log.md. Signals live in the scan artifact. Hypothesis triggers update the thesis doc directly. Only portfolio actions go in the log.
• Do update docs/calendar.md for upcoming catalysts found during the scan when they are tied to a thesis/hypothesis.
• Keep total search time under 30 minutes. 1-2 searches per thesis, max 5 results each. Prioritize theses with nearer-term hypothesis deadlines.
• Silence is fine. If nothing significant happened for a thesis, say so — no need to manufacture findings.
• Cite sources explicitly. Every finding must include source publication, article date, link, source tier, and newness label. Filter by quality before presenting: primary sources over commentary, discard social media. If a finding only appears in low-quality sources, flag the provenance gap rather than presenting it as fact.