When a benchmark, signal, or channel is rejected
If a call fails because a benchmark name,signal_type, or channel is unknown:
1
Call vertical_legend
Re-read the live manifest — the benchmarks, signal catalog, and channels it advertises.
2
Use only advertised values
Pick the correct value from what the legend returns. Many tools also return an
available / error list on rejection — use it.3
Never guess a replacement
Do not invent a substitute string. If nothing advertised fits, say the requested value is not available rather than approximating it.
Legend advertises a channel the client schema doesn’t expose
The legend reflects the live server; the client’s tool list can lag it. When they disagree, the legend is the truth and the client view is the thing to refresh.Sources and evidence links
include_details=trueonly when the user needs evidence. It attaches source refs and enlarges responses — use it when the user wants source links, not by default.- Return the public
source_refsthat were returned. Cite what the tool actually gave you. - Never fabricate a URL and never claim a source that was not returned. If no source ref came back, say the evidence link is not available.
Related
- AI Operating Rules — rule 1 (never invent) and the labeling rule.
- Tool Reference — where
include_detailsandvertical_legendare described. - Periods, Windows & Confidence · Vertical tool list
