Methodology
// data_sources · ai_model · squeeze_score · gex · fact_check · refresh · corrections · disqualification
Tapeboard sources market data from named primary vendors with disclosed fallback chains, summarizes filings with Anthropic’s Claude family without producing investment advice, computes the Short Squeeze Score from a fixed weighted formula calibrated 2024-11-15, and reverifies every comparison page within 30 days or pulls it from the index.
Why this methodology page exists
This page is the trust anchor for everything Tapeboard publishes. Every comparison page, ticker page, and data widget on the site links back here so you can audit the math before you trust the output.
If a number on Tapeboard does not have a path back to a primary source (an exchange feed, an SEC filing, a FINRA report, a Federal Reserve series) we treat that as a documentation bug and fix the page rather than the number.
Where the data comes from
Every numeric surface in Tapeboard resolves through a named primary source with at least one disclosed fallback. The full vendor table (including refresh cadence and license posture) lives on a dedicated page so this methodology document does not get blown out by a per-vendor matrix.
See Data sources for the full table and per-vendor disclosure. The summary: equities quotes route Schwab primary and Finnhub fallback; SEC filings come direct from EDGAR; Form 4 insider transactions and Form 13F holdings come direct from EDGAR; macroeconomic series come from FRED with the published series ID cited inline; options chains and GEX inputs are vendor-licensed (vendor name disclosed on /data-sources).
Public-tier real-time posture. Public (logged-out) views of US equity quotes are 15-minute delayed by Schwab license. Authenticated paid-tier users get real-time quotes consistent with the Schwab licensing terms.
AI summarization model: what we use, what we will not do
Tapeboard uses Anthropic's Claude family of large language models for summarization, classification, and structured extraction tasks. The current production models are Claude Sonnet 4.5 for high-throughput pipelines (filing summaries, news classification) and Claude Opus 4.7 for longer-context analysis (filing-set diffs, multi-document research briefings). Anthropic's terms guarantee that prompts and completions submitted via the API are not used to train future models.
What the AI is allowed to do
Summarize. Compress a 10-K, an 8-K, or a news cluster into a structured summary that preserves the named entities, dollar figures, and dates the source contains.
Classify. Tag a filing or headline with a small, fixed taxonomy (earnings, insider buy, secondary offering, capital raise, merger, regulatory, litigation, other).
Extract. Pull named entities into a structured object that downstream code can validate against the underlying filing text.
What the AI is not allowed to do
No extrapolation. The summarizer is prompted to refuse rather than guess when a value is not in the source. If a 10-K does not state a number, the summary will not invent one.
No advice. The model is system-prompted to never produce a buy / sell / hold recommendation, a price target, an entry / exit, or a position-sizing suggestion. Tapeboard is not a registered investment adviser; see Risk Disclaimer.
No silent rewriting. Quoted passages from a filing are passed through verbatim, never paraphrased into the model's voice.
How the AI is validated
Every AI-generated summary is cross-checked against the source filing for entity-level consistency before it ships to a public surface. Discrepancies route to a human review queue rather than to the page.
Short Squeeze Score: the formula
The Tapeboard Short Squeeze Score is a 0-100 composite computed daily across roughly 500 actively-traded US equities. Each input is z-score normalized against the universe before being weighted, so borrow-fee outliers do not dominate the sum.
| INPUT | WEIGHT | SOURCE |
|---|---|---|
| Short interest as a percentage of float | 35% | FINRA bi-monthly settlement, retrieved via Yahoo Finance defaultKeyStatistics |
| Borrow fee (annualized) | 25% | Interactive Brokers stock-loan inventory via iborrowdesk.com |
| Float utilization | 20% | Interactive Brokers stock-loan inventory |
| Days to cover | 15% | FINRA short interest divided by 30-day average daily volume |
| Trailing 5-day price momentum | 5% | Schwab daily bars |
Last calibrated: 2024-11-15. Weights are fixed, disclosed, and chosen from the squeeze literature plus cross-sectional analysis of the 2021-2023 named events. Weights do not adapt per ticker.
The full derivation, normalization details, FAQ, and known limitations live on the dedicated page: /methodology/short-squeeze-score.
Gamma Exposure (GEX) calculation
Gamma Exposure is the dollar-gamma of all open option contracts on a given underlying, signed by dealer position. Tapeboard computes GEX from CBOE-published end-of-day open interest and a Black-Scholes second-derivative for gamma, summed across the active option chain.
Source posture. Open interest, strike, and expiry come from the licensed options-data vendor disclosed on /data-sources. Implied volatility per contract uses the vendor-supplied IV when available. We do not silently fall back to historical-vol when IV is missing. Those contracts drop out of the sum and the page discloses the partial coverage.
Dealer-position assumption. Tapeboard assumes dealers are short calls and long puts at index-equity scale (the standard retail-flow posture). The page surfaces the assumption rather than presenting GEX as observed dealer positioning, because dealer positioning is not publicly observable.
Refresh cadence. GEX rebuilds end-of-day from settled open interest. Intraday GEX figures are estimates that interpolate against the morning's snapshot; they are labeled as estimates in the UI.
Fact-check process for comparison pages
Every Tapeboard comparison page (the /alternatives/* slugs that compare Tapeboard to Bloomberg, Koyfin, TradingView, Finviz, Unusual Whales, Benzinga Pro, and similar) carries pricing, feature, and competitive claims that go stale when the competitor changes their pricing page. The fact-check process keeps these honest.
Pre-publish requirements
Every numeric or factual claim on a comparison page has a row in the citation registry (citation-registry.csv) with the source URL and the retrieved date. Primary sources only: vendor's own pricing or documentation page, SEC EDGAR, FINRA, FRED, CBOE, exchange feeds. Listicle aggregators (G2, Capterra, "best of" pages, Wikipedia, Reddit) are disallowed.
Refresh policy: every comparison page reverified within 30 days OR noindex
If a comparison page has not been reverified within 30 days, the page automatically flips to <meta name="robots" content="noindex"> until the next reverify pass clears it. Stale pricing claims are worse than no claim. We would rather lose the index entry than mislead a reader on a competitor's current price.
Reverify cron
A scheduled cron at scripts/recheck.py runs on the 15th of January, April, July, and October. The cron diffs each comparison page's last-verified pricing against a fresh scrape of the competitor's pricing page and opens a pull request for any page that has drifted, blocking merge until a human reviews the diff.
Refresh cadence
Different surfaces refresh at different rates depending on the upstream publisher schedule. The table below is the contract. If a surface is not refreshing at the cadence shown, that is a desk failure and a correction-page entry.
| SURFACE | CADENCE |
|---|---|
| Equities quotes | Real-time during regular hours (9:30 AM-4 PM ET) for authenticated paid users; 15-min delayed for public views per Schwab license. |
| SEC filings | EDGAR RSS poll, sub-90 seconds post-publish. |
| Form 4 insider transactions | EDGAR Form 4 feed, sub-90 seconds. |
| Form 13F holdings | EDGAR Form 13F feed, quarterly per filer. |
| Macroeconomic series | FRED daily refresh for high-frequency series; weekly refresh for low-frequency series (CPI, PCE, NFP, GDP) on the publisher release schedule. |
| Squeeze score components | Daily after 9 PM ET (borrow + utilization + price). FINRA short-interest input lags ~10 business days. |
| Comparison pages (pricing) | Quarterly cron on Jan 15 / Apr 15 / Jul 15 / Oct 15. Off-cycle reverify on any vendor pricing change within 14 days. |
Editorial corrections policy
Tapeboard maintains a public correction log. Every correction to a published page (pricing, feature claim, methodology calculation, or AI-generated summary) gets an entry on the corrections page with the date, the original claim, the corrected claim, and the source the correction was verified against.
Public correction log: /corrections (page stub goes live on first correction; until then, no log to publish).
Reporting an error. Email [email protected] with the URL, the claim that looks wrong, and (if possible) the source you would expect us to use. We acknowledge during US trading hours; off-hours mail queues until the desk opens.
What we do with the report. Re-fetch the underlying source, compare against the published value, either correct the page or publish a disclosure explaining the discrepancy. Silent edits are not permitted. Corrections that change a claim get a dated entry on the corrections log.
Conflicts of interest
Zac Nicholson is the founder and operator of Tapeboard. He trades penny stocks for his own account on a five-cent scalp strategy, separate from the platform.
Not a financial adviser. Zac is not a registered investment adviser, broker-dealer, or commodity trading advisor. Nothing on Tapeboard is investment advice; nothing on Tapeboard is a recommendation to buy, sell, or hold any security. See the full Risk Disclaimer.
Position disclosure. Zac may from time to time hold positions in securities that appear on Tapeboard's screeners, scanners, or comparison pages. No content published on the site should be assumed to be free of position-related conflicts of interest. Editorial bylines on opinion content (the daily Tape Open commentary, the weekly filings newsletter) carry an inline position-disclosure footer when relevant.
No paid placement. Tapeboard does not accept payment from competitors named on the site, and does not accept payment to feature or favor any ticker, asset class, or third-party tool.
Where Tapeboard does not compete
Every product has a frontier where it stops being the right tool. The table below is the canonical list. These claims render verbatim on the corresponding /alternatives/* page in a "Where [competitor] wins" section. We win where we win; we say so where we do not.
| COMPETITOR | TAPEBOARD DOES NOT COMPETE ON |
|---|---|
| Bloomberg Terminal | Chat / IB messaging, fixed income coverage, FX dealer desks, the buy-side workflow integrations Bloomberg has owned for 30 years. |
| Koyfin | International equities (European, APAC, EM single-name coverage) and global macro dashboards. |
| TradingView | Pine Script custom indicators, charting-first workflows, broker integration for direct order routing. |
| Finviz | Long-history fundamental screening (10-year-plus financials backtesting at scale). |
| Unusual Whales | Options flow data depth, dark-pool prints granularity, congressional trade tracking. |
| Benzinga Pro | Audio squawk, fastest-to-tape breaking news, headline-driven trader workflow. |
Last reviewed and review cycle
This methodology document is reviewed quarterly and immediately whenever a vendor changes, a fallback path is added or removed, a model version is upgraded, the squeeze-score weights are recalibrated, or the disqualification matrix changes. Material changes are stamped at the top of the affected section and reflected in the page's dateModified JSON-LD field.
Methodology last reviewed: 2026-05-03. Editor: Marcus Reilly. Operator: Tapeboard.