A complete research workflow

Read the state.
Check the research.

An agent can retrieve market history, find a relevant publication, and cite what each source actually supports. This example uses IONQ.

Give your connected agent this task.

Use Chart Library to read IONQ's completed-session market state for 2026-05-12. Find the published IONQ noon Casebook and read its full article. Explain what each source supports, with dates, sample sizes, unknowns and citations. Keep the noon intraday case separate from the completed-session state. Does either establish a profitable trading rule? If a source is unavailable, report that explicitly.

Connect Chart Library through MCP, then use the prompt above. The calls below are examples, not a recorded model response.

1. Read the completed session.

market_state(symbol="IONQ", date="2026-05-12")

Preserve the returned status, actual session, sample sizes, packet hash, empirical outcome ranges and warnings. If the packet is empty, keep that result in the answer.

2. Find the related publication.

search_research(query="IONQ noon", kind="casebook")
read_research(
  research_id="casebook:ionq-noon",
  section="article",
  version="FULL_VERSION_FROM_SEARCH"
)

Choose casebook:ionq-noon from the search results and copy its documents.article.read_arguments into the read tool. The example's FULL_VERSION_FROM_SEARCH stands for that complete 64-character version. For further chunks, use content.next_read until it is null. If the publication changed, restart from its current overview instead of mixing revisions.

Search summaries help find evidence; they do not replace reading it. For studies, read any available guide alongside the result and inspect related_research for limiting findings. Cite the source URL and full publication version, and preserve the document hash when saving evidence.

3. Answer with both clocks intact.

The published noon case inspects 50 intraday comparisons. It reports that 41 have complete ten-minute windows, and that preceding paths and elapsed time differ. It did not test subsequent returns. Its corpus also uses full-session eligibility and normalization fitted after the seed.

The completed-session packet is a separate observation with its own method and sample. Its numbers must come from the actual returned packet. It cannot describe what was known at noon. Do not pool its analogs with the Casebook's 50 members or call its historical ranges calibrated forecasts.

A useful answer explains the relevant history and the limits of the comparison. These sources do not, by themselves, establish a profitable trading rule.

Use the same evidence through REST.

curl "https://chartlibrary.io/api/v1/state-packet?symbol=IONQ&date=2026-05-12"
curl "https://chartlibrary.io/api/v1/research/search?query=IONQ%20noon&kind=casebook"
curl "https://chartlibrary.io/api/v1/research/read?research_id=casebook%3Aionq-noon&section=article&version=FULL_VERSION_FROM_SEARCH"

Replace FULL_VERSION_FROM_SEARCH in the REST example with the version returned by your search. Omitting it reads the current revision without checking it against the earlier response.

See the research we plan to test with agents →