Known limitations
These are the confirmed limits of the product and of the measures it produces.
Product scope
NL Analytics currently includes English earnings-call transcripts. You cannot upload your own text, and other document types are not currently available. See corpus availability.
There is no public end-user API. The product is used through the UI and CSV downloads; these docs do not cover API authentication, endpoints, schemas, SDKs, rate limits, or automation workflows because no such public interface exists.
Availability
The corpus consists of the English earnings-call transcripts available in NL Analytics. Availability varies by country, sector, firm, and time. Check sample availability before interpreting variation.
The refresh cadence is roughly biweekly but not guaranteed.
Query and filtering
Query design affects every measure. You may need to refine keywords, include plurals manually, validate suggested terms, and balance precision against recall. See query quality.
Specific companies cannot be pre-selected before a search. Sector, country, section, speaker, and adjacent-sentence settings are part of the query. To focus on particular firms, filter the dataset export or the Snippet Tool view after the search.
Very broad queries are rejected when their estimated matches exceed a share of all sentences in the corpus (the default threshold is 3 percent). Narrow the query or contact support if a broad search is essential.
Generated dataset results are retained for 90 days. The durable dataset definition remains available, but an older dataset must be refreshed before its matched sentences can be inspected again in the Snippet Tool. Refresh updates the same dataset and counts as one search.
Weighted datasets cannot be combined or compared with other datasets, and their values are weighted sums rather than sentence counts. The dictionaries are fixed; their terms cannot be edited.
Measures
The core measures are raw integer sentence counts built from explicit queries and curated word lists. See measure definitions. They are not black-box LLM scores, causal estimates, investment advice, forecasts, compliance determinations, or complete summaries of business risk.
nr_of_sentences always measures the full transcript. It does not adjust when a query restricts matching by section or speaker or includes adjacent sentences. Use nr_of_sentences_filtered as the denominator in those cases, as described in normalization guidance.
Researchers own construct validation, sample design, normalization choices, crosswalk checks, and empirical interpretation.
Exports
Dataset CSV exports include all earnings calls in the selected date range, including rows where no sentences satisfy the query. See Keep zero rows.
Exports can contain provenance columns and columns for measures under development that these docs do not define.
The GVKey crosswalk is a convenience output for downstream joins. Verify match quality before relying on joined results.
Compliance
NL Analytics does not currently hold formal compliance certifications, and these docs make no compliance, certification, or regulatory-assurance claims.