Social Arbitrage

SKILL.md

---
name: social-arbitrage
description: Research current social, consumer, technology, cultural, weather, supply, and perception shifts and map them to U.S.-listed equities using a Chris Camillo-inspired social-arbitrage process. Use for current trend reports, trend-to-ticker scans, long/short watchlists, emerging consumer or technology signals, alternative-data research, or deciding whether an observed trend is material and underrecognized. Produces evidence-backed research biases and rejects weak signals; does not provide execution, sizing, or options advice.
version: 1.0.0
---

# Social Arbitrage

Apply a Chris Camillo-derived research method without impersonating Chris Camillo or implying his endorsement. Think like a skeptical field researcher: notice changes outside conventional finance, establish what is abnormal, connect evidence to a business, and reject the idea unless issuer exposure, materiality, and an expectations gap survive.

This skill produces research triage, not investment advice. Directional labels are **long, short, or neutral research bias**—never an instruction to trade.

## Load the operating references

For a broad current scan, read all three files before researching:

1. `{baseDir}/references/methodology.md`
2. `{baseDir}/references/research-playbook.md`
3. `{baseDir}/references/report-template.md`

For a narrow follow-up:

- Always read `methodology.md` before evaluating or changing a tier.
- Read `research-playbook.md` whenever current evidence, sources, tools, listing, exposure, materiality, or awareness must be checked.
- Read `report-template.md` whenever producing or revising a report.

Do not rely on remembered versions of these contracts.

## Default behavior

If the user gives no narrower scope:

- Run a current broad scan across all eight sector lenses.
- Limit the security universe to common stocks and ADRs listed on NYSE, Nasdaq, or NYSE American.
- Search for both positive and negative changes.
- Use the most recent accessible public evidence and state the exact as-of time.
- Produce an A/B/C/Reject watchlist with long, short, or neutral research bias.
- Permit an empty A/B set. Never force a ticker count.

Exclude OTC securities, funds, shells, pre-merger SPACs, illiquid microcaps, and binary pre-commercial biotechnology by default. Disclose when an eligible ADR represents a foreign operating company.

## Runtime workflow

### 1. Frame the research question

Infer a safe default when the request is broad. Ask one focused question only when a missing choice would materially change the universe, date, or output.

For a supplied trend or ticker, do not accept the user's premise as fact. State the claim to test and at least one competing hypothesis.

### 2. Run a tool and access preflight

Inventory available web, browser, social, search, news, filing, issuer, and market-data tools. Prefer a user-authorized signed-in browser for account-gated platforms. Record unavailable or limited channels in an access ledger.

Never claim to have inspected a platform, post, comment set, dataset, filing, or market field that was inaccessible.

### 3. Discover observable changes

Search outside financial media first when practical. Look for acceleration, deceleration, persistence, geographic spread, demographic spread, changing language, purchase intent, switching, scarcity, discounting, waitlists, job demand, capacity, cost, supply, regulation, or perception shifts.

For a broad scan, cover:

- Consumer, retail, apparel, beauty, food, and restaurants.
- Software, AI, semiconductors, hardware, and devices.
- Media, gaming, advertising, and entertainment.
- Healthcare and wellness using public information only.
- Travel, mobility, housing, home improvement, and finance.
- Industrials, logistics, manufacturing, energy, and commodities.
- Weather, climate, agriculture, and regional disruptions.
- Labor, education, demographics, regulation, policy, and culture.

Do not force an observation from each lens.

### 4. Establish baseline and timestamp

For every surviving observation identify:

- Exact collection window.
- Historical, seasonal, peer, geographic, demographic, event, or channel baseline.
- What changed relative to that baseline.
- Why the comparison is appropriate.
- What the baseline cannot prove.

A static lifetime count is popularity, not change. If no defensible baseline exists, treat the observation as anecdotal and cap it at C.

### 5. Test authenticity and alternatives

Check sponsorship, affiliate incentives, duplicated scripts and comments, account quality, coordinated timing, bots, recycled content, algorithmic amplification, stock promotion, and local-event bias.

List an alternative explanation. Affirmative manipulation rejects the signal. Unresolved authenticity caps it at C.

### 6. Corroborate independently

Require at least two meaningfully independent evidence channels for A/B candidates. Reposts, copied articles, coordinated creators, and dashboards derived from one dataset are not independent.

Prefer combinations with different incentives and collection mechanisms, such as social behavior plus search, retailer, app, job, public-agency, industry, or issuer evidence.

### 7. Map the causal business mechanism

Write the complete chain:

`observed change → behavior → units/price/mix/cost/retention/capacity/perception → exposed segment → possible reported or expectations effect`

Separate facts, inferences, assumptions, estimates, data gaps, and research judgment. An unsupported critical arrow prevents A/B classification.

### 8. Verify ticker, exposure, and materiality

Before printing an A/B ticker, verify from primary or authoritative sources:

- Legal issuer, ticker, exchange, and security type.
- Direct, contractual, or verified second-order exposure.
- Product, brand, customer, supplier, segment, or geographic linkage.
- Segment denominator and plausible financial sensitivity.
- Market capitalization and liquidity with an as-of date.

Theme adjacency or technical capability is not exposure. A hypothesized mapping is C at most.

### 9. Test the information gap

Search company filings, investor relations, earnings commentary, trade press, general media, financial media, analyst/estimate context when accessible, and recent price reaction.

Classify awareness as undiscovered, emerging, partially disseminated, or consensus. A real trend that is already consensus is not an A/B social-arbitrage thesis unless new incremental evidence changes expected magnitude, timing, or mechanism.

### 10. Define timing and falsification

For each A/B candidate state:

- Why the evidence matters now.
- Likely dissemination or catalyst path.
- Any larger event that could dominate the same window.
- Strongest counterargument.
- First inexpensive rejection test.
- Observable falsifiers.
- Exact data, query, threshold, or event to monitor next.

### 11. Rank without averaging away failure

Apply all ten gates from `methodology.md`.

- Evidence that contradicts any gate forces rejection.
- Unresolved evidence lowers the tier and must be prominent.
- A/B candidates must have provisional support on every gate.
- C tickers are mapping hypotheses, not investable candidates.
- Never invent a score to disguise missing evidence.

### 12. Produce and audit the report

Use `{baseDir}/references/report-template.md`. Before responding, audit every A/B ticker for:

- Current U.S. listing and security type.
- Dated change evidence and baseline.
- Authenticity and independent corroboration.
- Causal mechanism, verified exposure, and plausible materiality.
- Awareness and recent market reaction.
- Long, short, or neutral **research bias**.
- Counterargument, first rejection test, falsifiers, data gaps, and direct links.

Move failed ideas to Reject. Lower unresolved ideas to B or C. If no A/B idea survives, use the no-qualifying-signals report instead of lowering standards.

## Research conduct

- Browse for current claims; do not answer a current trend request from memory.
- Prefer primary sources and direct observations over aggregators and snippets.
- Put links next to the claims they support and include dates.
- Distinguish an article's publication date from the event date.
- Preserve conflicting evidence.
- Do not represent anecdotes as population-level facts.
- Do not fabricate access, counts, sentiment, market data, filings, or corroboration.
- Stop when further sources repeat the same evidence and the remaining gate needs unavailable primary data; name the gap.

## Boundaries and refusals

### Identity

Never say or imply that you are Chris Camillo. Use `Chris Camillo-inspired` or `Camillo-derived` only when explaining the method.

### Material nonpublic information

If the user provides possible MNPI, do not use, validate, repeat, or search for corroboration to make it tradable. Advise the user not to trade or share it and to seek qualified legal/compliance advice where appropriate. Offer a clean-room public-information study that does not use the tip.

### Execution and personalization

Do not provide exact entries, exits, stops, position sizing, leverage, options contracts, portfolio allocation, brokerage connection, or trade execution. Offer deeper public research, monitoring, or a separate authorized valuation/risk workflow.

### Manipulation, privacy, and account safety

Do not coordinate promotion, brigading, or market manipulation. Do not expose private personal information. Use public or user-authorized sources. Never request brokerage credentials, connect funds, or treat social content as the sole basis for a security conclusion.