Reading a briefing
A briefing is built to be read critically. Every source that answers is timestamped, every gap is named, and what's missing is as visible as what's present. Here is how to read what comes back.
Live and cached data
Legate caches the values it fetches and, within a short per-source window, reuses a value instead of re-fetching it. You don't read this per figure inside the briefing text — the signal lives in the Data Freshness table below, where each source is marked live if it was freshly fetched or hit if it was served from cache.
A cached value is one fetched earlier and reused while still inside its window, so it can be as old as that window allows — being cached is not a promise that the value is fresh. The window is tuned per source: a price caches for about a minute, while verified contract source code, which never changes, caches for up to 30 days.
The Data Freshness table
When at least one source returns data, a briefing ends with a Data Freshness table. Legate appends it by machine after the briefing text is written, so it reflects the fetches that actually happened — not the model's account of them. It has exactly four columns:
- Source — the data provider (CoinGecko, DefiLlama, Etherscan, and so on).
- Endpoint — the URL that was called. It's shown in full unless it runs past 60 characters, in which case the first 57 are kept and an ellipsis is appended.
- Sampled — an ISO timestamp of when Legate sampled the data during this run, not when the provider recorded it. For a cached source it's regenerated on read, so it reports this run's read time.
- Cache —
hitif the value came from cache,liveif it was freshly fetched.
On a hit row, the Sampled time is when Legate re-served the value for this run, not when the source last changed it. Read the Cache column first — it tells you whether that timestamp marks a fresh fetch or a value reused from cache.
If no source returned anything, the table is omitted entirely rather than printed empty — the briefing ends with the agents' own text.
When a source doesn't answer
Sources fail quietly. A timeout — 4 seconds by default — a rate limit, a bad status code, or a network error all make a source return nothing, with no crash and no error surfaced in your briefing.
Nothing is hidden, though. The missing source is named in a Sources unavailable line on the agent's data bundle, its section is dropped, and the briefing's Data Gaps section spells out what came back empty. When fewer sources respond, confidence drops; when a whole agent fails, the briefing is flagged partial.
A named gap is a feature, not a failure — it tells you exactly what the read is missing, so you can go check that source yourself. Legate never patches the hole with a guess: an agent omits a metric rather than invent a number.
No confidence score, by design
A briefing carries no numeric confidence score and no "Confidence Assessment" section. A single number like that is confidence theater — it reads as precision the data doesn't support. Judge a briefing instead from its Data Gaps, its Data Freshness table, and the citations inline in the text.
A briefing will never show you:
- A fabricated number. When a metric has no live data, the line is omitted, never filled with a guess.
- A confidence percentage. There is no single score to trust or distrust — the honest signals are the gaps, the freshness table, and the citations.
- A sentiment figure stated as fact. Lexicon sentiment is always flagged indicative only, never quoted as a verdict.
Read this way, a briefing is a fast, honest starting point — the fuller list of what Legate is and is not lives in What Legate can & can't do.
Next: Research a token by address →