术语表

为指数、信号、公司页与方法论中用到的术语提供的通俗解释。站内其他地方带点状下划线的术语会直接链接到这里。

Conviction: How strongly the swarm holds its directional lean, on a 0–100 scale. It multiplies four factors: how much the agents agree, how many of them had data to vote (coverage), how far the aggregate score is from neutral (strength), and a discount for imminent event risk. It is deliberately NOT a probability that the call is correct — empirical hit-rates are shown separately, each next to its base rate and sample size.
Centrality: A network measure of how connected a company is within its industry supply-chain graph. A company that many others buy from, sell to, or depend on scores higher. It is one of the four inputs to the Industry Index score.
Chokepoint factor: How concentrated and hard-to-route-around a company’s position is — a supplier that the rest of the network has no substitute for (a bottleneck) scores high. It rewards irreplaceability, not just raw connection count. One of the four Industry Index inputs.
Market weight: The size component of the Industry Index — a company’s market capitalization relative to its industry peers, min–max normalized across the day’s active universe. One of the four Industry Index inputs.
Momentum: The recent-trend component of the Industry Index — how a company’s score has moved lately. When a company has no price history the momentum is derived from news activity only (shown as “news-only”). One of the four Industry Index inputs.
Coverage: The share of the signal-agents that actually had data to vote on a given company and horizon. Low coverage means most agents abstained (no fundamentals, no 13F, no price history, …), which lowers conviction — an honest “we don’t know enough yet” rather than a fabricated vote.
Agreement: How aligned the participating agents are on direction. When the agents that voted mostly point the same way, agreement is high and conviction rises; when they conflict, agreement (and conviction) falls. It is one of the multiplicative factors in conviction.
Lean: The swarm’s stored directional output for a company and horizon — Bullish, Bearish, Neutral, or No signal. A directional lean requires the aggregate |score| to clear the lean gate plus minimum participation, coverage, and conviction; otherwise it is shown as “Low-confidence · Neutral.”
13F: Form 13F is the quarterly SEC filing in which institutional investment managers (over $100M in US equities) disclose their long positions. It is filed up to 45 days after quarter-end, so the data is delayed; the swarm’s smart-money agent reads net accumulation from it.
13F coverage: Form 13F discloses only long positions in US-listed 13(f) securities (equities and listed options). It excludes short positions, cash, bonds, foreign-listed shares, currencies, commodities, and private holdings — so a 13F is a partial, long-only snapshot of a manager, never its whole book.
Confidential treatment: Managers may apply for confidential treatment to delay disclosing certain positions (e.g. while accumulating). When a filing flags this, some holdings are omitted from the public 13F, so the reported total understates the real book. TradingPilot surfaces this flag when present.
Corporate 13F book: Some operating companies (NVIDIA, Alphabet, Amazon) manage over $100M in US-listed equities and must file Form 13F like any institution. That filing discloses only long US-listed public-equity stakes: private venture investments, foreign-listed stakes, subsidiaries and cash never appear — so a corporate 13F book is a narrow window into strategic stakes, never the company’s full investment portfolio.
STOCK Act disclosure: Members of Congress report securities transactions under the STOCK Act as Periodic Transaction Reports: amounts appear only as statutory ranges (e.g. $1,001–$15,000, up to open-ended “Over” categories), within 30–45 days of the trade, covering the member, spouse, joint and dependent-child accounts; sales may be partial, and filings are sometimes amended. Federal law (5 U.S.C. § 13107) restricts commercial use of these reports — TradingPilot displays them for general public dissemination only, and a disclosed trade is never investment advice or an endorsement.
$100M 13F threshold: Institutional investment managers with discretion over more than $100M in Section 13(f) securities must file Form 13F quarterly, up to 45 days after quarter-end. Smaller managers never appear.
% of reported 13F portfolio: TradingPilot-computed: a holding’s reported value as a share of the filer’s cover-page 13F total (tableValueTotal). Because a 13F omits shorts, cash, bonds, foreign and private holdings, "% of 13F" is NOT "% of AUM" — it can materially overstate a manager’s real allocation.
Sold out (no longer reported): A holding dropping out of a 13F means only that it is no longer reported: it may have been sold, moved below the reporting threshold, converted, or omitted under confidential treatment. TradingPilot never asserts an outright sale from a 13F alone.
Passive filer: Vanguard, BlackRock, State Street and Fidelity file 13Fs reflecting broad index and ETF holdings — their quarter-over-quarter changes are index rebalancing flows, not active conviction. TradingPilot tags them "passive," excludes them from the "guru" framing, and excludes them from the smart-money signal so their flow is not mistaken for a bullish/bearish read.
Substitutability: An editorial 0–1 estimate of how easily a company’s role could be filled by an alternative. Low substitutability (near 0) means few or no substitutes exist, which raises the company’s chokepoint weight. It is a curated judgment, not a disclosed figure.
Quartile bands (P25 / median / P75): The peer distribution for a financial ratio, split into bands. P25 and P75 are the 25th and 75th percentiles of the peer set, the median is the middle, and min/max are the extremes. They show whether a company’s value is high or low relative to comparable companies, computed at request time from filed XBRL.
Out-of-sample (OOS): A result is out-of-sample when it is measured on data that was not used to fit the model’s weights. The backtest fits weights on an earlier window and applies them only afterward; the live prediction ledger is out-of-sample by construction. It is the honest way to estimate real-world accuracy (though the backtest still carries survivorship bias).
Sharpe ratio: A risk-adjusted return measure: the strategy’s excess return divided by the volatility of those returns. Higher is better return per unit of risk. On this site Sharpe is withheld until the curve spans ~6 months, because annualizing a short sample is meaningless.
IV skew: The difference between the implied volatility of puts and of calls (put IV − call IV). A positive skew means puts are relatively more expensive — often a defensive / hedging tilt. It is read from a delayed, aggregated options-chain snapshot, not a real-time feed.
Volume: The total number of option contracts that changed hands in this company today, summed across every strike and expiry on the chain. It is the raw size of the day’s options flow — how much traded, before any comparison to open interest or to the name’s own history. It is computed from delayed, end-of-day, aggregated chain data, so it never distinguishes an opening trade from a closing one, nor a buy from a sell.
Vol/OI: Volume divided by open interest at the COMPANY level: the sum of every contract that traded in this name today over the sum of all positions already outstanding across the chain at the prior close. A ratio above 1× means more contracts changed hands today than existed beforehand — the day’s flow is mostly NEW positioning rather than churn in old bets, which is the core "unusual activity" tell and needs no price history to read. Summing the whole chain (rather than reading one strike) is deliberate: a single near-zero-open-interest weekly can print a huge per-contract ratio that says nothing about the name, so the board ranks on this name-wide figure. A ratio is only as good as its denominator, so it is WITHHELD entirely — the cell shows "—" — whenever the whole chain holds fewer than 1,000 open contracts. A handful of open contracts cannot carry a ratio: dividing a full day of volume by one or two of them yields a five-figure "×" that is really just the volume with a multiplication sign after it, and this column is read as a measurement. Every real chain we measured clears that floor, though the thinnest names do so narrowly — the four thinnest we saw held 1,152, 1,373, 1,551 and 1,972 open contracts — so the em dash is the deliberate trade: a genuinely thin name loses a small true ratio rather than have a fabricated one printed for it. It is computed from delayed, end-of-day, aggregated chain data, so it never distinguishes an opening trade from a closing one, nor a buy from a sell.
Premium: The dollar value of the options that traded — the sum of (volume × contract price × 100) across the flagged contracts. It answers "how much money is actually behind this flow," which a raw contract count hides. On the leaderboard it is shown per contract in the by-contract view, or rolled up to the company’s daily total in the by-company view. It is derived from delayed, aggregated end-of-day data and is notional traded value — not a cash outlay and not a directional bet.
Relative volume: The company’s total options volume today compared to the average of its prior days’ daily totals — e.g. 3× means the name is trading three times its normal options flow. Unlike Vol/OI it is a NAME-WIDE, history-relative read: it measures the whole chain against that same company’s own recent baseline, not a single contract. It needs history to be meaningful, so until about ten sessions of options data exist the cell reads “building” rather than a not-yet-trustworthy number. Computed from delayed, end-of-day, aggregated chain data.
Implied volatility (IV): Implied volatility is the annualized volatility implied by an option’s market price under a standard pricing model — the size of the move the market is pricing into that contract. Higher IV means richer premium and a wider expected move. On the by-contract view it is that contract’s IV; on the by-company view it is the standout contract’s IV (the roll-up carries its most-unusual contract’s value), not a name-level average. It is read from a delayed, end-of-day, aggregated options-chain snapshot, so it is a coarse read, not a real-time surface, and it is not itself a directional signal.
IV rank: IV rank places a contract’s current implied volatility within its own recent range (computed as today’s IV over its trailing baseline). A high rank means the option is unusually expensive relative to how it normally trades, which is part of what makes activity “unusual.” It needs prior sessions to be non-null, is derived from delayed, aggregated chain data, and is context, not a directional call.
Score: The Score column ranks how UNUSUAL a company’s options activity is, 0–100. It multiplies a magnitude term — the flow’s size relative to open interest, its total premium, and its volume versus the name’s own baseline — by a character term that rewards short-dated, out-of-the-money contracts (the shapes that carry leverage and urgency). The company figure is the premium-weighted blend of its flagged contracts. It measures UNUSUALNESS, not direction: a high score says a lot of conspicuous option activity happened, not that it was bullish or bearish — read Signal, Call/put premium and Side for that. It is computed from delayed, end-of-day, aggregated chain data and is not a forecast.
Standout: Each board row rolls up every unusual contract a company flagged that session; Standout names the ONE with the highest unusualness score — its strike, whether it is a call or a put, and its days to expiry. It is the single most conspicuous bet behind the row, shown so a name is not just a number: the neighbouring # column says how many contracts flagged in total, and Standout says which one stood out most. A call standout is not itself a bullish verdict — it can be bought or sold, and it is one contract among # — so read it alongside Signal and Side. Drawn from delayed, end-of-day, aggregated chain data.
Persistence: How many recent sessions the unusual activity has persisted — distinguishing a single-day blip from a multi-day accumulation in the same name. It counts the distinct days this company drew a stock-level unusual-activity flag within the trailing twelve sessions, so more days suggests a sustained position rather than one print. Computed on-read from delayed, end-of-day, aggregated data; it is descriptive, not a prediction.
Flat tape: Options volume was large relative to open interest on this contract while the shares closed within 1.5%. This delayed end-of-day aggregate feed cannot tell whether those trades opened or closed positions, or whether they are one leg of a spread. It is descriptive — options got busy without the stock reacting — and it is not a prediction, not a claim of informed trading, and not a directional call. It is shown ONLY on a company’s options panel, beside that contract’s expiry, volume, open-interest change and premium — the grain at which it was measured and at which its evidence is visible. It is deliberately NOT rolled up to a company badge on the leaderboard: the company row scores a premium-weighted average across its flagged contracts, a different quantity from the per-contract score the flat-tape threshold was calibrated on, and a company-wide “the tape was flat” badge would rest on contracts a reader cannot see from that page.
Call/put premium: How the day’s option premium in this company split between calls and puts: the dollar premium that traded in calls, and in puts, as two shares of the total. It is arithmetic on what traded — Σ(volume × price × 100) on each side of the chain — and nothing more. THE POPULATION IS THE CHAIN WE CAPTURE, WHICH IS NOT THE COMPANY’S FULL OPTION CHAIN. Each session the capture job asks the provider for at most the 6 nearest expiries falling within 45 days, and keeps the contracts that actually traded. LEAPS, quarterlies and every monthly beyond that window are in no denominator on this site. The cap is not theoretical: on 2026-07-24 it bound for 128 of the 155 companies on the board, which used all 6 expiries. Inside that window the split does span every traded contract, which is far more than the contracts listed on the company’s Options tab: that table shows at most 20 flagged contracts, and across all 155 companies on the board on 2026-07-24 those visible rows carried a median of just 13.7% of the company’s own captured premium (a quarter of companies under 4.5%, a quarter over 29.7%, and 72 of the 155 hold more than 20 flagged contracts). Summing the rows on screen will therefore NOT reproduce this percentage, and that is expected — they are a sample of a much larger captured chain. WHAT IT IS NOT. It is not a directional verdict, and TradingPilot no longer renders it as one: it was previously banded into “bullish / bearish / mixed” and coloured green or red, which invited the reader to take a premium measurement for the site’s call on the company. Crucially it cannot tell whether those options were BOUGHT or SOLD — call premium can be sold as easily as bought — so a call-heavy split is not a bullish position, and a put-heavy one may be a hedge against stock rather than a bearish bet. It says nothing about whether positions were opened or closed, and nothing about which side of the spread the trades hit (that is the separate, and separately caveated, inferred side). It is computed from delayed, end-of-day, aggregated chain data.
Inferred side (buy / sell): Whether the day’s options flow looks buyer-initiated or seller-initiated — inferred, never observed. For each contract TradingPilot compares that contract’s single last traded price against the bid/ask spread captured in the SAME delayed snapshot as that print. A print in the UPPER QUARTER of the spread reads as buyer-initiated, a print in the lower quarter as seller-initiated, and the middle half — the most common case — as no side at all. A print that lands modestly outside its quote is still read (the two are captured together but the feed is delayed and aggregated, so a gap of about one spread-width is expected); a print further out than that belongs to a different quote regime and is discarded. The company-level lean then combines those per-contract reads. Two things about it matter: each contract’s side comes from its SINGLE LAST PRINT, and that one inferred side is then weighted by the contract’s WHOLE-DAY notional — so one closing trade’s inferred side is applied to that contract’s entire day of volume. (The weight uses the midpoint of the quote, not the print itself, so a print near the ask does not also earn a bigger vote.) The company lean is published only when contracts with a usable quote account for at least a quarter of the day’s premium; below that it is withheld rather than extrapolated from a sliver of the chain. This is an INFERENCE and NOT confirmed buy/sell flow. It rests on one delayed quote and one last print — not on the trade tape — so it cannot see whether a print was a sweep, one leg of a spread, a hedge against stock, a market-maker’s two-sided quote, or a trade negotiated between the quotes. Real side classification requires trade-level (OPRA) data with each print time-stamped against the prevailing national best bid and offer, which this site does not have. When the bid, ask, or last is missing, when there is no bid at all (a one-sided book), when the quote is crossed or locked, or when the print is far outside the quote, no side is inferred and the cell is left blank — withheld, never guessed. It also says nothing about whether a position was opened or closed: that would need the change in open interest, which is a whole-contract net figure across every participant for the previous settled session, and is shown as its own separate column for the reader to weigh. Finally, side is NOT direction: buying puts is buy-side and bearish, so the inferred side is shown in its own colours and must not be read as a bullish or bearish call.
Signal (swarm short-horizon lean): This column is the swarm signal engine’s stored directional lean for the company on the SHORT (~10-trading-day) horizon — the very same verdict the /signals board and /signals/explain render, not a second calculation. See “Lean” for how a directional lean is gated. It is shown as a coloured chip ONLY for Bullish or Bearish; neutral, no-signal, and a company with no computed signal all render a muted “—”, because the board makes a directional claim or none — and the short horizon is the noisiest and most often neutral, so “—” is the common, honest cell. The chip links to /signals/explain so you can see exactly which agents voted. CRUCIALLY it is a SEPARATE thing from the two other directional-looking columns on this same board and must not be read as one verdict with them: “Call/put premium” is a premium MEASUREMENT (which side of the chain carried more dollars, not a bought/sold or bull/bear call), and “Side (inferred)” is the inferred buy/sell aggressor of the options flow (buying puts is buy-side AND bearish). Only THIS column is the site’s actual directional verdict on the stock. Finally, the signal and the unusual-options flags are produced by different jobs on different cadences, so the signal can be dated earlier than the board’s flagged session; when it lags by more than a few days it is shown muted with its own “as of” date rather than as a current read, never a confident colour on a stale call.
Event risk: The risk from an upcoming, scheduled catalyst — such as earnings — within the signal’s horizon. It is not a directional vote: a high/medium-impact catalyst nearby only raises event risk, which discounts conviction, reflecting the extra uncertainty around a known upcoming event.
Effective agents (n_eff): The effective number of independent agents standing behind a signal. Because some agents read overlapping information, their votes are partly correlated and are down-weighted, so n_eff is usually below the raw count of agents that voted. A low n_eff means the signal rests on fewer truly independent views.
Moat: A company’s durable competitive advantage — what keeps rivals from eroding its economics (brand, network effects, switching costs, scale, patents). On this site the moat profile is a curated judgment, not a disclosed figure, grounded in cited filings and industry sources and marked “editorial estimate.” The bands feed a bounded long-horizon vote at weight 0.20 — never the short or medium horizons; the narrative and moat type stay descriptive.
Moat rating: A qualitative band for how durable a company’s competitive advantage is: wide (strong and lasting), narrow (real but limited), none, or eroding (weakening). It is a curated judgment, not a disclosed figure, and carries a citation and an as-of date.
Moat type: The classic source of a company’s moat — network effects, switching costs, cost/scale advantage, intangibles/IP, efficient scale, or none. It names WHY the advantage persists. It is a curated judgment, not a disclosed figure.
Leadership: A qualitative band for how far ahead of its competitors a company sits — clear leader, co-leader, fast follower, at parity, or behind — instead of a false-precision “leads by N months.” It is a curated judgment, not a disclosed figure.
Moat prior: The moat prior turns the editorial moat bands (moat rating, leadership, pricing power) into a bounded directional vote on the long (~6-month) horizon. It is a curated judgment, NOT a disclosed figure and NOT backtested: it is capped (±0.6), low-confidence, carries a fixed 0.20 weight, and never votes on the short or medium horizons. It is shown, decomposed, and discountable in the signal explanation so you can see exactly how it nudged the call.
Pricing power: A company’s ability to raise prices without losing customers — strong, moderate, or weak — often visible in a durable gross-margin trend. It is a curated judgment, not a disclosed figure, and carries a citation.
Market-share band: An editorial band for a product’s market position — leader, top 3, challenger, niche, or unknown — used deliberately instead of an invented market-share percentage. The band is a curated judgment tied to a cited basis in the product’s share note, not a disclosed figure. Products are presentation-only and never feed a signal.
Technical barrier: An editorial band for how hard a product is for a competitor to replicate — deep, moderate, or low. It reflects the process, IP, capital, or ecosystem hurdles described in the product’s cited barrier note. It is a curated judgment, not a disclosed figure, and never feeds a signal.
Guru: A hand-curated famous investor (e.g. Berkshire, Scion, Pershing Square) whose complete reported 13F portfolio is surfaced on /gurus. Gurus are distinct from passive index managers, which are tracked but excluded from the guru framing. Guru pages are presentation-only and never feed a signal.
Breakout profile: Whether one company owns all five of the things that compound in an AI application business: the customer relationship, the proprietary data, the evals, the workflow, and a domain-specific model. Each is banded owns / partial / none / unknown and carries its own citation and as-of date. A separate paired axis, AI adoption depth, records how deep the deployment actually runs; it is read beside the five and never counted among them. It is a curated judgment, not a disclosed figure, it is weightless — shown on the page and the graph but never an input to any signal — and it is not a forecast of returns.
AI adoption depth: How deeply AI is actually deployed in the business, banded deep / moderate / shallow / unknown with its own citation. It is a PAIRED AXIS, not a sixth ownership dimension, and it is never counted in the “N owns of 5” pattern: the five ask whether the value would accrue to this company if the thesis works, and this asks whether it is working at all and how far it reaches. The two are never summed or averaged — a company can band none on proprietary data and none on a domain model and still run one of the deepest deployments anywhere, which is precisely the reading a single blended number destroys. It is a curated judgment, not a disclosed figure; unknown means the evidence was not found, not that the deployment is shallow, and a profile with no reading at all is shown as “not yet assessed” rather than being defaulted into either.
Customer relationship: Whether the company holds the commercial relationship with the end customer rather than reaching them through a platform, reseller or incumbent. It is a curated judgment, not a disclosed figure; unknown means the evidence was not found, not that the answer is no.
Proprietary data: Whether the data the product depends on is genuinely proprietary — accumulated through use or exclusively licensed — rather than purchasable by any competitor. It is a curated judgment, not a disclosed figure; unknown means the evidence was not found, not that the answer is no.
Evals: Whether the company owns the evaluation methodology that defines correct output in its domain — the benchmark customers and regulators come to trust. It is a curated judgment, not a disclosed figure; unknown means the evidence was not found, not that the answer is no. This dimension is the least documented of the five and is banded unknown most often.
Workflow: Whether the practitioner’s actual work happens inside the product rather than being exported to another tool to finish. It is a curated judgment, not a disclosed figure; unknown means the evidence was not found, not that the answer is no.
Domain-specific model: Whether the company owns a model adapted to its domain rather than calling a general frontier model unchanged. It is a curated judgment, not a disclosed figure; unknown means the evidence was not found, not that the answer is no.
Traction evidence: Observable, dated, cited business facts — funding rounds, named customer deployments, hiring, disclosed revenue or usage — recorded so the structural bands can be checked against something falsifiable. The selection of which facts to record is a curated judgment, not a disclosed figure; each fact links to the source it was read from.
ADR / depositary ratio: An American Depositary Receipt is a certificate traded on a US exchange that represents shares held on deposit in the company’s home market. The DEPOSITARY RATIO says how many ordinary shares one receipt stands for, and it is very often not one: SK hynix’s ADR represents one-tenth of a common share, so ten receipts make one share. A receipt price and a home-market price therefore differ by that ratio before anything real has happened, and usually by a currency as well. TradingPilot shows one listing at a time, states the ratio with its filing, and never converts, differences or compares the two — a gap between them is not an arbitrage you can take, and presenting it as one would be the lie this page exists to prevent.