> ## Documentation Index > Fetch the complete documentation index at: https://docs.focusalpha.ai/llms.txt > Use this file to discover all available pages before exploring further. # Events Interface (merged stream) > One message per event revision — the conclusion. A stable event_id, monotonic revisions, the full label state, what changed, accumulated facts, and our per-ticker impact reading. The Events interface (`stream: "merged"`) is FocusAlpha's conclusion layer. Everything that turns out to be about the same thing — the wire write-ups, the 8-K, our reading of the guidance, the analyst reactions — hangs onto one stable `event_id`, and you receive one message per **revision** of that event, not one per article. Measured on a real replay, an event averages **about two pushes**: the first arrival, then one more when a materially new label lands. Nine arrivals about one earnings print become three messages, not nine. ## Payload Real output — the second push of an earnings event, where the guidance label arrived: ```json theme={null} { "event_id": "CSCO:results:20260812", "revision": 2, "reason": "new_label", "closed": false, "symbols": ["CSCO"], "company_id": "US_17275R102", "subject": null, "origin": { "at": "2026-08-12T20:05:00+00:00", "basis": "published", "precision": "minute", "what": "CISCO REPORTS FOURTH QUARTER AND FISCAL YEAR 2026 EARNINGS", "ref": null }, "delta": { "added": [["guidance", "guidance_raise"]], "dissent": {}, "since_revision": 1, "since": "2026-08-12T20:05:00+00:00" }, "state": { "labels": [ ["earnings", "earnings_result"], ["guidance", "guidance_raise"] ], "facts": { "revenue": "$17.3 billion", "eps_gaap": "$0.97" }, "evidence": ["news:90101", "filing_8k:0000858877-26-000106", "news:90102"] }, "trigger": { "id": "news:90102", "kind": "news", "title": "Cisco forecasts upbeat annual revenue", "why": "Cisco guided FY27 revenue above consensus on AI infrastructure demand.", "link": "https://www.reuters.com/technology/cisco-forecast", "impacts": [{ "ticker": "CSCO", "direction": "bullish", "score": 80 }] }, "at": "2026-08-12T20:12:50+00:00", "channel": "wire", "publisher": "reuters.com", "title": "Cisco forecasts upbeat annual revenue" } ``` ## Fields | field | meaning | | ---------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `event_id` | Stable across every revision. Format `TICKER:family:YYYYMMDD` (date of the event's first item). | | `revision` | 1, 2, 3… per event. Never reused, never renumbered. | | `reason` | `first` or `new_label`. | | `closed` | A terminal label arrived (a deal closed or terminated). No further revisions will follow. | | `symbols` / `company_id` | Who this is about — your distribution list. A list, because one event can involve several names (both sides of a deal). A ticker only enters after both the entity layer found it in the text and the impact model scored it as relevant, so name-collision junk is kept out. | | `subject` | The entity when there is no ticker — an unlisted target, a government. | | `origin` | The event's own start, repeated identically in every revision — receiving revision 3 out of the blue, you can still tell which event it updates. `origin.basis` names the clock: `published`, `filing_accepted`, or `stated`. Never compare timestamps across two different bases. | | `delta.added` | What is new **in this revision** — the "what changed" line to highlight. | | `delta.since_revision` | The previous revision number. Holding r1 and receiving r3, you see `since_revision: 2` and know to backfill exactly one. | | `delta.dissent` | A second value seen on an axis where one is already set. Almost never a reversal — surfaced so it is queryable, never pushed as its own event. | | `state.labels` | The full label set as of now, as `[family, subtype]` pairs — see [Event labels](/websocket/event-labels). `subtype: null` is a positive statement: family known, direction not yet readable. | | `state.facts` | Numbers accumulated from whichever arrivals carried a body (revenue, EPS…). | | `state.evidence` | Every raw item seen on this event so far, by `id` — pull any of them in full from the [Raw news interface](/websocket/news-stream). You never have to guess which items belong to an event by time window. | | `trigger` | The arrival that caused **this** revision, with its content — title, one-line `why`, link, and per-ticker `impacts`. | | `at` / `title` / `channel` / `publisher` | The triggering arrival's timestamp, headline and provenance. | `trigger.why` and `trigger.impacts` are **this system's reading** of the material, not the outlet's words and not a price forecast. Attribute them to FocusAlpha, never to the publisher. The payload deliberately carries no article text — title, labels, `why` and the link are the content; the article is one click away. ## When a revision fires * **Revision 1 fires on the first arrival and carries facts only, no direction.** Most first arrivals have no readable direction yet, and shipping the first outlet's verdict would mean shipping a verdict the source never gave. The exception is anticipatory labels (`earnings_preview`, `merger_acquisition_rumour`): "this has not happened yet" *is* the fact, and it is kept. * **A new label fires a revision.** A repeat of a known label does not, however many outlets carry it. * **A second value on the same axis does not fire** — it lands in `delta.dissent`. * After roughly a day an event stops moving; each family has its own quiet window, after which an arrival opens a *new* event rather than revising this one.