Editorial Policy

1. How Content Is Created

Stock Expert AI uses a multi-stage AI pipeline to generate all analytical content on the platform. The process works as follows:

• Data Ingestion: Raw market data is collected from Financial Modeling Prep (FMP), with Yahoo Finance and Alpaca Markets as fallbacks (SEC EDGAR for filing links only) (see Data Sources below).
• AI Analysis: Language models (listed in section 5) write the qualitative parts — dossiers, theses, commentary. Scores such as MoonshotScore are computed by published rules from provider data, not by a language model.
• Quality Checks: Automated validation compares AI-generated metrics against source data to flag discrepancies.
• Publication: Content is published to the platform and indexed for search engines.

2. Data Sources

Stock Expert AI aggregates data from the following sources:

• Financial Modeling Prep (FMP): Primary source for fundamentals, financial statements, analyst ratings, insider transactions, and price quotes.
• Yahoo Finance: Fallback only — a price or fundamental when FMP has no figure for a ticker.
• SEC EDGAR: Filing links and company identity details (legal name, address) only — never a source of figures.
• Alpaca Markets: Fallback only — a price when neither FMP nor Yahoo Finance has one.
• News Aggregation: 200+ news sources in the archive feed Market Journal articles and sentiment analysis.

Data accuracy depends on these upstream providers. While we apply validation checks, we cannot guarantee the accuracy of third-party data at all times.

3. Update Frequency

• Stock Prices: Each price is the last quote we recorded, shown with the trading session it belongs to. Pages are served from a cache, so the copy you are reading can lag that quote. Each quote is a provider snapshot, not an exchange feed.
• MoonshotScore: The V2 five-pillar model is recalculated from the most recent completed US trading session, so scores move on US trading days. The score's computation date shows its last successful calculation; a newer price does not mean the score has refreshed. A failed update leaves the previous dated score in place. Legacy V1 nine-signal calculations are retained as historical records only. They are not used as a fallback when a current V2 score is unavailable.
• Dossiers and authored research: Dossiers and authored stock research are dated analyses, not continuously rewritten reports. Read each analysis as of its displayed date. Price and score updates do not refresh the written research; an updated analysis or correction carries its own date.
• Market Journal: The Market Intelligence Journal is paused: its last edition was published on August 3, 2026. The archive stays online — 5,000+ dated stories drawing on 200+ news outlets (counts as of August 2026).
• Financial Statements: Refreshed after quarterly and annual filings are processed.

4. AI Editorial Personas

Market Journal articles are attributed to AI editorial personas (e.g., Alex Sterling, Sam Rivera, Taylor Brooks). These are not real people. They represent distinct analytical styles and editorial perspectives within our AI system:

• Each persona specializes in a specific area (market strategy, equity research, earnings analysis).
• Personas provide consistent editorial voice and analytical framework.
• All persona-generated content is clearly labeled as AI-generated.

5. AI Models in Use

This shared list describes configured product models and models recorded in saved authored research. A dated document retains its original model attribution; a newer model configuration does not rewrite that history. Read the model and analysis date disclosed on the individual page.

• GPT-5.6 Sol (OpenAI, gpt-5.6-sol) — Dated stock research prepared with Codex; model recorded per document.
• Claude Haiku 4.5 (Anthropic, claude-haiku-4-5) — Dated stock research retained with its original model attribution.
• GPT-5.3 Codex Spark (OpenAI, gpt-5.3-codex-spark) — Dated stock research retained with its original model attribution.
• Gemini 2.5 Flash (Google, gemini-2.5-flash) — Configured company-dossier generation; each dossier page discloses the model and date that wrote it. OpenAI GPT-4o mini is the fallback when Gemini is unavailable.
• Gemini 2.0 Flash (Google, gemini-2.0-flash) — Company dossiers generated before the current configuration. Recorded on earlier documents; this entry describes that historical use.
• GPT-4o mini (OpenAI, gpt-4o-mini) — Company dossiers generated by the fallback model before the current configuration. Recorded on earlier documents; this entry describes that historical use.
• Gemini 2.5 Flash (Google, gemini-2.5-flash) — Machine translation of English pages into the other site languages.
• Gemini 2.5 Flash (Google, gemini-2.5-flash) — Moon AI conversational assistant. Accessed through the Gemini OpenAI-compatible endpoint; OpenAI is the configured fallback provider.
• Claude 3 Haiku (Anthropic, claude-3-haiku-20240307) — Legends Council and sector-insight narratives.
• GPT-4o (OpenAI, gpt-4o) — Portfolio screenshot reading (OCR) — first of two readers run in parallel.
• Gemini 2.5 Pro (Google, gemini-2.5-pro) — Portfolio screenshot reading (OCR) — second reader; the two results are reconciled.
• GPT-4o mini (OpenAI, gpt-4o-mini) — Quick news analysis, company summaries and prediction assistance.

6. Human Oversight

What human review actually covers today: every error reported by a reader, every page the automated checks flag (a figure that disagrees with its source, a competitor that is not a peer, advice-like wording), and every methodology change before it ships. Outcomes are logged in the public corrections log. In addition, an automated sample review runs weekly: 50 randomly drawn published dossiers are checked for advice-like wording, untranslated interface keys, source-emptiness prose, competitor-list errors, and market-cap magnitude errors. Each run's counts are stored and the latest run is shown here; a monthly summary in the corrections log is not yet automated. Latest automated run: 2026-09-28 (weekly-random), 50 dossiers checked, 0 check flags recorded (per-dossier audit evidence unavailable or unsupported; legacy-compatible count). The first cycle ran on September 1, 2026 over 200 dossiers and found one defect — a source-emptiness sentence in one fund description, a class the render-time guard already removes on display; the stored text was flagged for regeneration.

7. Machine Translation

English is the authoritative version of every page. The other site languages are machine-translated from the English page by Gemini 2.5 Flash and are not reviewed by a human translator. Numbers, tickers and links are carried over unchanged and checked by a six-layer validation (numeric carry-over, ticker consistency, length ratio, paragraph preservation, markup validity, title sanity); a translation that fails is not published. Translated pages state their translation status and the date of the English analysis they were made from; where the English narrative is newer, the translated page says so.

8. Conflicts of Interest

• No paid coverage: no company, fund or broker pays for a page, a score, a placement or a mention.
• Scores cannot be steered: every score is computed by published rules from provider data with no manual override, so nobody's holdings — including the founder's — can move a number.
• Founder positions: the founder may hold positions in securities the site covers; those positions are not used as inputs anywhere and are disclosed on request.
• No affiliate links inside analysis: analysis pages carry no affiliate or referral links. If that ever changes, the link will be labelled where it appears and this section updated first.

9. Legends Council and Munger's Mindset

AI simulations built from the named investors' published principles. Not affiliated with, endorsed by, or the opinion of these individuals. Each lens scores a fixed set of provider metrics and offers a Favorable, Neutral or Unfavorable read — never a buy, sell or hold instruction. The public sources every lens was distilled from, what it checks and where it stops are listed on the lens methodology page.

10. Error Correction Policy

We take data accuracy seriously. If you find an error in any content on Stock Expert AI:

• Report: Email sedat@stockexpertai.com with the page URL and a description of the error.
• Review: Reported errors are reviewed within 24 hours on business days.
• Correction: Confirmed errors are corrected promptly. If the error is in upstream data, we note the discrepancy and update when corrected data becomes available.
• Systemic Issues: If an error reveals a systemic problem in the AI pipeline, we address the root cause to prevent recurrence.

11. Content Disclaimers

All content on Stock Expert AI is for informational and educational purposes only. It does not constitute investment advice, financial planning, or personalized recommendations. See our full disclaimer for complete legal disclosures. Errors we have published, and errors still open, are listed in the corrections log (JSON).

12. Contact

For questions about this editorial policy or to report content issues:

Email: sedat@stockexpertai.com

Founder: Sedat ANAK

13. Changelog

• 2026-09-28 — Automated sampling now records the selected content versions, completion state and distinct flagged dossiers. Missing, failed and stale runs are disclosed. Automated checks are described as screening selected issues, not independent verification of every statement.
• 2026-09-27 — MoonshotScore refresh is described as moving on US trading days rather than at a clock time. The Quality Gate answer now names the seven page dimensions the gate scores and its two thresholds (below 10 not generated; below 30 generated but not submitted for indexing). The Market Journal line is the one dated status sentence used across the site.
• 2026-09-23 — The model list now separates configured models from models recorded on earlier documents (Gemini 2.0 Flash, GPT-4o mini). MoonshotScore refresh is described per engine, as the methodology states it. The sample-review paragraph now says that a monthly summary in the corrections log is not yet automated; each run's counts are stored and the latest run is shown on this page.
• 2026-09-22 — Distinguished dated stock research from live price/score updates, and added the models recorded in authored research documents to the disclosure registry.
• 2026-09-01 — The promised scheduled sample review went live: 50 random published dossiers weekly, checked automatically for advice-like wording, untranslated interface keys, source-emptiness prose, competitor-list errors and market-cap magnitude errors; monthly counts publish to the corrections log. First cycle (200 dossiers) found one source-emptiness sentence.
• 2026-09-01 — AI models now listed from the code registry (Gemini 2.5 Flash for dossiers, translation and Moon AI; Claude 3 Haiku for Legends narratives; GPT-4o + Gemini 2.5 Pro for screenshot reading). Added machine-translation, conflicts-of-interest and Legends Council sections; human oversight described as it actually operates.
• 2026-08-25 — Retired the per-page "Analysis by" founder byline; pages now carry a machine-generation disclosure with model and date, and an editorial-accountability line.
• 2026-03-09 — First published.