Methodology

Data source and tracked universe

The active mention lane uses 15-minute GKG 2.1 source files published by the GDELT Project. The current mapping covers 57 deliberately curated targets: 47 equities and 10 crypto assets. It uses exact normalized company or asset names and, for targets with a reviewed mapping, exact GKG theme codes—not ticker substrings or free-form keyword searches. Short or ambiguous names are intentionally omitted when they would create too many collisions.

Coverage includes general company and entertainment news as well as financial reporting. A record about a Netflix show, for example, can count for Netflix when it matches the configured company name. A mention does not have to discuss the company's shares.

The curated universe is a product boundary, not a claim to cover every security, crypto asset, publisher, article, conversation, or public mention on the internet. A source file can also omit or misclassify a real-world reference before Mentioned receives it.

What counts as one mention

For each valid GDELT GKG source record, Mentioned checks normalized entity names and the exact GKG theme codes configured for each curated target. If either exact match succeeds, that record contributes one mention to the target. Repeating a match—or matching both a configured entity name and theme for the same target—does not make that record count twice for that target.

One record can contribute a mention to more than one tracked target when it contains an exact configured match for each of them. This is a count of matched source records, not a count of words, readers, unique events, opinions, or people. Syndicated copies and similar coverage remain separate records; Mentioned does not deduplicate stories across publishers.

The 48-hour window

A current feed requires 192 consecutive 15-minute source intervals, covering 48 hours. The chart combines every six intervals into one 90-minute bucket, producing 32 bars from oldest to newest. The total shown for a ticker equals the sum of those 32 bars. This is a rolling window: older intervals leave as new ones arrive, so a count can fall without any correction to the source. A bar shows a 90-minute period, not the exact time of each reference.

“Complete” refers only to successful receipt and processing of all 192 expected GDELT source files for this curated mapping and window. It does not mean complete coverage of the web, the market, or all public discussion. Publication and collection delays mean the view is not a real-time feed; its displayed snapshot time is the freshness authority.

Missing intervals and integrity checks

Mentioned verifies the expected interval, archive identity, checksums, archive structure, mapping revision, and aggregate payload before an interval can enter the measured window. Reprocessing the same exact interval is idempotent, so it does not add duplicate counts.

If an expected interval is missing, malformed, conflicting, or too old, Mentioned does not silently replace it with zero and label the result complete. The current feed remains unavailable until a valid consecutive window exists; any retained snapshot is labeled by its own timestamp.

Aggregate-only privacy boundary

The collector sends per-target record counts, service-allowlisted GKG topic-label counts, and interval-level integrity metadata. Mentioned does not retain or display raw article text, article URLs, article titles, or publisher identities in this lane. Product output is limited to aggregate ticker counts, timing buckets, source attribution, freshness, and eligible recurring aggregate topic labels.

For a matched ticker, each allowlisted label counts at most once per source record. The ticker's own identity and match theme are excluded. Across a complete window, labels must appear in at least two matched records; Premium can receive the top four, while Free responses strip them. These labels are not article excerpts, raw keywords, sentiment, causation, exhaustive coverage, or an explanation of why an entity appeared. A count alone cannot establish that separate source records describe separate events.

How ranking works

The current public feed ranks tracked entities by their aggregate matched-record count inside the complete 48-hour window. Ties and low counts should be read as descriptive measurements, not as an investment ranking or a claim that one entity matters more than another.

How mention alerts work

Customer notifications are currently paused during prelaunch. The following describes stock alert evaluation; a current feed or a saved notification preference does not mean delivery is enabled. Crypto mention alerts are not available in this release.

Stock alerts use a separate, provider-neutral authority that runs only while the complete 192-interval source receipt chain is current and an operator has explicitly activated it. Each evaluation compares the latest six hours of aggregate mentions with the prior 42 hours, normalized to the same six-hour duration. A saved 2×, 3×, 5×, or 10× setting is applied to that baseline, and an account-wide minimum-surge preference may make the trigger stricter. Missing or stale source authority stops evaluation instead of treating missing data as zero.

When delivery is active, at most one alert event is queued per ticker every six hours, with at most six across the account in any rolling 24 hours. These limits include events that fail or are cancelled; retrying the same event does not use another slot. A surge held by a limit is checked again using fresh data when capacity returns. After an event is queued, activity must fall below a reset band before another surge can qualify. Events expire after one hour if they cannot be delivered. Aggregate counts describe measured attention; an alert is not sentiment, investment advice, or evidence of a price move.

What we do not do

• We do not issue buy or sell recommendations
• We do not generate price targets
• We do not infer sentiment or price direction from a count
• We do not claim that mentions caused a market move
• We do not trade ahead of subscribers
• We do not sell user data
• We do not track private watchlists in order to build trading positions

Limitations

Exact matching reduces ambiguous ticker collisions but can miss spelling variants, aliases, translated names, and references that the upstream source did not classify as an entity. It can also count syndicated or substantially similar coverage as separate source records. The feed therefore describes matched records within the stated source, mapping, and time window—not unique stories, unique speakers, demand, conviction, or future returns.

How this page changes

Material changes to the source, curated universe, counting rule, timing window, or alert state are documented here and versioned with the implementation.