Panel

The researcher endpoint — long (id, date, metric, value) observations mixing B3 tickers and CVM fund CNPJs in one call.

Parameters, response columns and types are generated from the SQL catalog and render under Endpoints → Panel in the sidebar. This page carries what a schema cannot: the conventions, the nulls that are deliberate, and the traps.

Overview

panel returns one row per observation, (id, date, metric, value), so B3 prices and fund fundamentals for the same window come in one request. Pivot locally; there is no server-side corr, rank or regression.

api.panel(p_ids         TEXT[],
          p_metrics     TEXT[]  DEFAULT ARRAY['close','nav'],
          p_from        DATE    DEFAULT CURRENT_DATE - 365,
          p_to          DATE    DEFAULT NULL,  -- null = clamp fund rows to complete periods
          p_freq        TEXT    DEFAULT 'month',
          p_entity_type TEXT    DEFAULT NULL,  -- keep one fund family
          p_min_nav     NUMERIC DEFAULT NULL,  -- universe mode: latest NAV floor (BRL)
          p_min_months  INT     DEFAULT NULL,  -- universe mode: min observations of the first metric
          p_after       TEXT    DEFAULT NULL)  -- cursor: '' = first page, then the last row's key

The grain is (id, asset_class, date, metric). 385 CNPJs file under two families in the same month, and the panel returns one row per family. Pivoting on (id, date, metric) alone raises on the duplicate or silently averages two vehicles. Pass p_entity_type, or keep asset_class in your pivot key.

Over 1,000 rows the panel refuses; it never trims. A window over one page raises 22023 (HTTP 400): page with p_after, or narrow the request. See Paging.

Parameters

Parameter	What to know
p_ids	Tickers, derivative codnegs and CNPJs, mixed. After stripping non-digits, exactly 14 digits means CNPJ; anything else is a B3 code, upper-cased (cash ticker, option series or termo contract is decided by the market segment it prints in). 3 ids anonymously, 50 signed in; over it raises 22023.
p_metrics	Lower-cased and trimmed. A name not in the Metrics table is silently ignored, so a typo shows as missing rows, not a 400.
p_from	Snapped down to the first of the month for fund metrics; as-is for quote metrics.
p_to	Omitted: fund metrics end at each family’s latest complete period; quote metrics run to today. Explicit: served verbatim, partial months included. coverage reports the bound as complete_through.
p_freq	day, d, daily mean daily; every other value, including a typo like week, falls back to month.
p_entity_type	fi, fidc, fii, fip or fiagro; anything else raises 22023. With p_ids empty it selects universe mode.
p_min_nav	Universe mode only. Latest non-null net assets in the window, in BRL; a filter, not a rank.
p_min_months	Universe mode only. Minimum non-null observations of the first requested metric.
p_after	null is whole-result mode and refuses above one page. '' is the first page. '<date>|<id>|<metric>|<asset_class>' from the last row is the next. A malformed cursor raises 22023.

Universe mode

Leave p_ids empty and name a family with p_entity_type: the panel selects that family’s funds itself, optionally floored by p_min_nav and p_min_months. Nothing is ranked server-side. Signed-in callers only (each page re-scans the family’s whole window); catalog().limits.tiers.<tier>.panel_universe says which you are.

{"p_ids": [], "p_metrics": ["delinquency", "nav"], "p_entity_type": "fidc",
 "p_from": "2025-09-01", "p_min_nav": 10000000, "p_min_months": 12, "p_after": ""}

Metrics

Metric	Id type	Asset class	Grains	Source	Meaning
close	ticker, option, termo	equity, unit, bdr, fund_quota, cash_security, derivative	day, month	b3_cotahist	Unadjusted close. Cash tickers: the instrument’s published BDI board. Monthly = last session of the month.
volume	ticker, option, termo	equity, unit, bdr, fund_quota, cash_security, derivative	day, month	b3_cotahist	Session financial volume in BRL. Monthly = the last session’s volume, not a sum.
close_unit	ticker	equity, unit, bdr, fund_quota, cash_security	day, month	b3_cotahist	close / quotation_factor: close per single quoted unit. Compares price levels across papers (one quoted per lot reads 1000× its unit price). Not corporate-action adjusted.
close_return	ticker	equity, unit, bdr, fund_quota, cash_security	day, month	b3_cotahist	p_t / p_(t-1) - 1, with p_(t-1) divided by the share ratio of any split, grouping or bonus between the two prints (a 1:4 split from 100.00 to 26.00 is +4%, not −74%). Price only, not a total return. Cash tickers only.
settlement_rate	future	derivative	day, month	b3_price_report	Futures settlement as a rate, as B3 publishes it: for DI1, % a.a. on 252 business days. Ids are B3 contract codes (DI1F27); future_curve lists a session’s. Final (F) settlements only. Nothing is rolled: a contract stops at maturity. The default for a future.
settlement_price	future	derivative	day, month	b3_price_report	Settlement price as published: for DI1 the PU. Final settlements only.
open_interest	future	derivative	day, month	b3_price_report	Open contracts at the session’s close. Monthly = the last session’s, not a sum.
nav	cnpj	fi, fidc, fii, fip, fiagro	month	cvm	Fund net assets (vl_patrim_liq).
quota	cnpj	fi	month	cvm	Unit quota value. Comparable within a subclass only.
delinquency	cnpj	fidc, fiagro	month	cvm	Delinquent portfolio value, not a rate; divide by nav yourself.
yield	cnpj	fii	month	cvm	Monthly yield percentage as published.
inflows	cnpj	fi	month	cvm	Gross monthly subscriptions.
redemptions	cnpj	fi	month	cvm	Gross monthly redemptions.
quotaholders	cnpj	fi, fii	month	cvm	Number of unit-holders. Not served for fidc, fiagro or fip.
receivables	cnpj	fidc	month	cvm	Receivables portfolio total (tab II TOTAL), the denominator for a concentration ratio. See FIDC concentration.
sacado_top1	cnpj	fidc	month	cvm	Exposure to the single largest debtor (tab VIII rank 1, as filed). Anonymized in the source.
sacado_top25	cnpj	fidc	month	cvm	Sum of the ranks the fund filed (1..n, n ≤ 25); nothing imputed. Can exceed receivables for ~2% of funds, served as filed, never capped.

A metric asked outside its asset classes (catalog().applicability) gives no rows for those ids, not nulls or zeros. FIDC delinquency starts in 2013-01 and is on every row from 2020-11; before that a fund with no tab VI row is null, not zero (catalog().regime_breaks).

freq=day is quotes only. No CVM fundamental is daily, so nav with p_freq: "day" returns the quote rows and nothing else. Mix asset classes at freq=month, where the B3 value is the last real session in the month. Aligning a monthly NAV to a daily calendar would require filling, and this API does not fill.

Response fields

Field	What to know
asset_class	Cash tickers: the conservative B3 type (fund_quota deliberately does not guess ETF versus FII). derivative for option and termo codnegs and futures contract codes. Funds: the vehicle’s entity_type.
date	At freq=month always the first day of the month, for every id. This is normalisation: CVM publishes FI/FII/FIAGRO at first-of-month, FIDC at month-end, FIP at year-end. fund_nav serves the period as filed.
value	Nullable for close and volume; for every other metric a null is no row at all.
source	b3_cotahist, b3_price_report (futures) or cvm.

Rows are ordered by date, id, metric, asset_class.

Example

curl

curl -s -X POST "$SILO_URL/rpc/panel" \
  -H "apikey: $SILO_ANON_KEY" \
  -H "Content-Type: application/json" \
  -d '{
        "p_ids": ["PETR4", "12345678000190"],
        "p_metrics": ["close_return", "delinquency", "nav"],
        "p_from": "2021-01-01",
        "p_to": "2026-06-30",
        "p_freq": "month"
      }'

Paging

Send p_after: "" for the first page, then the last row’s key '<date>|<id>|<metric>|<asset_class>'. A full page always has a next page; a short page is always the last. The SDK does this: iter_panel(...) / panel_all(...). Range headers and limit/offset do not page an RPC call.

For scale: 3 ids × 2 metrics × 5 years of daily quotes is ~7,500 rows, eight pages.

Notes

• Nothing is forward-filled or interpolated. A month a fund did not file is absent. Correlate pairwise-complete.
• Two shapes of “no data”. nav, quota, delinquency, yield, inflows, redemptions, quotaholders and close_return never appear with a null value; a gap is an absent row. close and volume are emitted every session and can carry a JSON null.
• close_return is adjusted across a share-count event. Where a split, grouping or bonus (DESDOBRAMENTO, GRUPAMENTO, BONIFICACAO in B3’s corporate-event history) lies between the two prints, daily or monthly (monthly: anywhere between the two month-end prints), the previous close is divided by the event’s share ratio (1 + factor/100 for a split or bonus, the factor for a grouping) before the return is taken. A close of 100.00 before a 1:4 split is 25.00, so a 26.00 close is +4%, not −74%. An event with an unreadable factor, or published twice with two factors, leaves the row absent. Dividends and JCP still move it, it is not a total return, and an event the nightly sweep has not stored yet is not seen, so treat a jump over ~40% as suspect until you have checked.
• close_return needs a predecessor. Monthly: only when the preceding row is the immediately preceding calendar month. Daily: against the previous stored session, so a session missing from the warehouse makes the return span the gap (check coverage). The first observation in a window is always absent, so request more history than you analyze.
• Windows differ per side. Quote metrics honor p_from exactly; fund metrics start from date_trunc('month', p_from) and p_to is month-truncated too, so p_to of 2026-07-01 still returns all of July for a FIDC filed at 2026-07-31.
• Derivatives carry no close_return, and their real grain is wider (COTAHIST keys them on board, termo also on term), so a within-session tie is cut deterministically (shortest term wins), not averaged. For the full grain use option_history or termo_history. There is no underlying column anywhere.
• Every cash-market BDI board is accepted; one row is kept deterministically when a ticker has multiple boards in a session.
• Statement timeout: a call that runs too long fails outright rather than returning part of a panel.
• An empty array means no observation matched: unknown ids, no filings in the window, a metric that does not apply, or an ignored metric name. Confirm with lookup and coverage.