EDI Analytics API

Programmatic access to the same engines that power the EDI platform: security resolution across the US, Brazil, Mexico and Peru; end-of-day global prices; standardized fundamentals with fiscal-period intelligence; cross-domain as-of matrices; and asynchronous batch execution. One request produces the same number in Excel, Python, the Quant Workstation, and a scheduled report — parity is structural, not tested-for.

Quickstart

Authenticate once — the session cookie carries your entitlements on every subsequent call.

# 1. Sign in (stores the session cookie)
curl -c cookies.txt -X POST https://api.edi.finance/api/auth/login \
  -H "Content-Type: application/json" \
  -d '{"email": "you@yourfirm.com", "password": "…"}'

# 2. Resolve identifiers to canonical entity ids
curl -b cookies.txt -X POST https://api.edi.finance/api/analytics/v1/resolve \
  -H "Content-Type: application/json" \
  -d '{"identifiers": [{"value": "PETR4-BR"}, {"value": "AAPL"}]}'

# 3. A cross-market comps grid in USD, one call
curl -b cookies.txt -X POST https://api.edi.finance/api/analytics/v1/matrix \
  -H "Content-Type: application/json" \
  -d '{"ids": ["AAPL-US","BBAS3-BR","BIMBOA-MX"],
       "metrics": ["close","MARKET_CAP","SECTOR"],
       "asOf": "2026-07-30", "currency": "USD"}'

Or with the Python reference client:

from edi_analytics import EdiAnalyticsClient, spill_matrix

client = EdiAnalyticsClient("https://api.edi.finance")
client.login("you@yourfirm.com", "…")

env = client.matrix(ids=["AAPL-US", "BBAS3-BR"],
                    metrics=["close", "MARKET_CAP", "SECTOR"])
grid = spill_matrix(env, ids=["AAPL-US", "BBAS3-BR"],
                    metrics=["close", "MARKET_CAP", "SECTOR"])

Service authentication — API keys

For scripts, SDKs, and server-to-server integrations, create an API key (it acts as you, with your entitlements) and send it as X-API-Key or Authorization: Bearer — no cookies involved. Keys are shown once at creation, can expire, and are revocable; managing keys always requires a signed-in session, never a key.

# Create a key (session-authenticated), then use it anywhere
curl -b cookies.txt -X POST https://api.edi.finance/api/analytics/v1/keys \
  -H "Content-Type: application/json" -d '{"label": "research pipeline", "expiresInDays": 365}'

curl -H "X-API-Key: edi_sk_…" -X POST https://api.edi.finance/api/analytics/v1/resolve \
  -H "Content-Type: application/json" -d '{"identifiers": [{"value": "AAPL"}]}'

Python: EdiAnalyticsClient("https://api.edi.finance", api_key="edi_sk_…") — skip login() entirely.

Core concepts

The envelope

Every data endpoint returns the same three-part envelope. Rows in data are flat and self-describing — they serialize identically to JSON, CSV, and Excel spill ranges.

{
  "data":   [ /* flat observation rows */ ],
  "errors": [ /* row-level failures — never a request failure */ ],
  "meta":   { "pagination": { "total": 412 }, "registryVersion": "2026.07.16.1A" }
}

Identifiers and the requestId echo

Every endpoint accepts mixed identifier types directly — market-qualified tickers (PETR4-BR, BIMBOA-MX), bare tickers, ISINs, CUSIPs, CIKs, and canonical entity ids. Every response row carries both requestId (your identifier, exactly as sent — the join key back to your spreadsheet cell or dataframe index) and entityId (our canonical key, stable across ticker changes and share classes).

Partial failure is the norm

One bad ticker must never kill a 500-cell workbook. Unresolvable identifiers, unknown metrics, and unpriceable conversions come back as entries in errors — each with a requestId, a stable code, and a human-readable detail — while every other row returns normally.

Provenance

Every response states the metric-registry version that priced it (meta.registryVersion), and rows carry the currency actually returned. Numbers are auditable across clients: the same request produces the same value everywhere, by construction.

Vocabulary

The request parameters below mean the same thing on every endpoint that accepts them.

Periodicity (fundamentals)

basis selects the restatement axis: original (default) or restated (staged — awaits vintage modeling).

Frequency (prices & series)

Price adjustment

Currency

currency defaults to LOCAL: values return in their native quote or reporting currency, stated per row — a Brazilian workbook never silently receives USD. Any ISO code (USD, BRL, MXN, EUR, …) converts server-side; see currency conversion for which rate applies where.

Security resolution

The identity layer under every other call. Send up to 2,000 identifiers of mixed types; get back canonical entities with company-vs-listing structure. Rows sharing an entityId are share classes / listings of the same issuer, enumerated under listings.

// POST /api/analytics/v1/resolve
{ "identifiers": [ {"value": "PETR4-BR"}, {"value": "US0378331005"}, {"value": "037833100", "type": "CUSIP"} ] }

// 200 — one row per candidate
{ "data": [
    { "requestId": "PETR4-BR", "entityId": "BR-9512", "ticker": "PETR4",
      "market": "BR", "matchedBy": "ticker", "candidateCount": 1,
      "listings": [ {"ticker": "PETR3", "market": "BR"}, {"ticker": "PETR4", "market": "BR"} ] } ] }

• Bare tickers resolve against your default market, else US. Qualify with a suffix (-BR, -MX, -PE, -US) to be explicit; hyphenated tickers like BRK-B are left intact.
• candidateCount > 1 marks an ambiguous identifier — data endpoints report these as row errors rather than guessing.
• Identifier shape is auto-classified; pass type (TICKER · ISIN · CUSIP · CIK · ENTITY_ID) to override.

Metric catalog

The authoritative registry of every metric the engines can price — the same registry the platform's screener, quant panels, and formula workbench execute against, including your organization's custom formula metrics. Autocomplete from here and a request can never name a metric the engines don't know.

Each row states its dataset, engine, value type, unit, supported entity types, time types (as-of / time-series), period modes, and status, and its place in the metric taxonomy: folderPath (statement lines under Financial statements, indicators under Ratios and indicators, one folder per ratio family, as-reported lines by regulator), kind, basis (standardized · as_reported · derived · observed), statement, market, industryProfiles (empty means every issuer type) and canonicalDataset, set on a copy of a code that another dataset owns. Names carry Spanish and Portuguese aliases, so ?search=deuda%20neta finds NET_DEBT.

Global prices

End-of-day OHLCV across all covered markets, with cross-listing containment built in: a São Paulo line sharing a ticker with a US issue can never splice into the US series. Duplicate observations are deduplicated to one canonical row per trading day.

// POST /api/content/global-prices/v1/prices
{ "ids": ["BBAS3-BR"], "startDate": "2026-01-01", "endDate": "2026-07-30",
  "frequency": "M", "fields": ["price", "volume"] }

// 200 — month-end observations, native currency stated per row
{ "data": [
    { "requestId": "BBAS3-BR", "entityId": "BR-1023", "date": "2026-06-30",
      "currency": "BRL", "price": 19.91, "volume": 31200400 }, … ] }

• fields: price · priceOpen · priceHigh · priceLow · volume. adjust selects the price basis.
• All eleven frequency values are supported, calendar-bucketed and start-date-anchored.
• With a currency, each observation converts at that date's spot rate; volume never converts.

Fundamentals

Standardized fiscal data with the platform's period intelligence. The server owns period semantics — year-to-date storage, fiscal-year-end inference, Q4 derivation, LTM windows — so a formula never has to.

// POST /api/content/fundamentals/v1/fundamentals
{ "ids": ["AAPL-US"], "metrics": ["REVENUE"], "periodicity": "LTM",
  "fiscalPeriod": { "start": "2025-01-01", "end": "2026-07-30" } }

// 200 — rolling last-twelve-months revenue at each fiscal quarter-end
{ "data": [
    { "requestId": "AAPL-US", "metric": "REVENUE", "periodicity": "LTM",
      "fiscalYear": 2026, "fiscalPeriod": "Q2", "fiscalEndDate": "2026-03-28",
      "currency": "USD", "value": 451442000000 }, … ] }

• fiscalPeriod.start/end are calendar dates; the server resolves them to completed fiscal periods. Default window: five years back.
• Rows report the issuer's reporting currency, or your requested currency after conversion.
• Unknown metrics are per-metric row errors; the rest of the request still returns.

Statement taxonomies, whole statements, and fiscal periods

Three discovery-and-retrieval companions to the fundamentals endpoint:

// POST /api/content/fundamentals/v1/statements — one call, whole statement
{ "ids": ["BBAS3-BR"], "statement": "FULL", "asOf": "2026-07-30", "currency": "USD" }

Ratios

POST /ratios — the ratio catalog as a series: one tab of it (group), or named ratios from any tab, with every flow leg read at one period basis. This is the web application's Ratios window over the wire — the same engine call, so a number here and a number there cannot disagree. The matrix still serves any ratio at a single as-of date; this endpoint is for the history.

// POST /api/content/fundamentals/v1/ratios
{ "ids": ["AAPL-US", "BIMBOA-MX"], "ratios": ["GROSS_MARGIN", "ROE", "PB"],
  "periodBasis": "ltm", "scale": "quarterly" }

// 200 — one row per (id, ratio, period); a flow ratio carries the basis it was read at
{ "data": [
    { "requestId": "AAPL-US", "ratio": "GROSS_MARGIN", "group": "profitability",
      "period": "FY2026 Q3", "periodEnd": "2026-06-27", "value": 0.4865,
      "valueType": "percent", "periodBasis": "ltm", "scale": "quarterly" },
    { "requestId": "AAPL-US", "ratio": "PB", "group": "valuation",
      "period": "2026-08-31", "periodEnd": "2026-08-31", "value": 43.05,
      "valueType": "multiple", "periodBasis": null, "scale": "monthly",
      "scaleClampReason": "tab_grid" }, … ],
  "errors": [] }

• period is a fiscal bucket (FY2026 Q3) on the fiscal grids and an ISO date on the price grid. periodEnd dates every row, because buckets are not comparable across issuers: Apple's FY2026 Q3 ended 2026-06-27 and a December year-end's ended 2026-09-30.
• valueType: "percent" is a fraction — a 48.65% gross margin is 0.4865, the same number the matrix and the catalog give for the same code.
• Withheld is not missing. A bank's gross margin and an industrial's loans-to-deposits come back as NO_DATA row errors that say why; every (id, ratio) you ask for yields a value or an error, never a silently absent cell.
• GET /ratios/fields lists every ratio the endpoint serves with its tab, formula, value type, grid scales and the issuer types it exists for — read it before building a query.

Point-in-time

POST /point-in-time — fundamentals as they were known on a date, for backtesting without lookahead bias (US filers). Two guarantees: a fiscal period is invisible until its filing's publication date (Apple's December quarter does not exist before January 30), and when a figure is later restated, the value that was current at pitDate is served — rows carry firstPublished, lastPublished (amendments), and valueVintage (as_known when a captured vintage applies; restatement capture accrues from 2026-08-02 forward). pitDataItems=true on the catalog lists PIT-capable metrics. FY and QTR; compose LTM client-side from PIT quarters.

Matrix

Many securities × many metrics at one as-of date, in one request — the comps-grid primitive. Metrics can mix datasets freely: prices, fundamentals, valuation ratios, derived measures, and text attributes all return through the same envelope.

// POST /api/analytics/v1/matrix
{ "ids": ["AAPL-US", "BBAS3-BR", "BIMBOA-MX"],
  "metrics": ["close", "MARKET_CAP", "PE_LTM", "SECTOR"],
  "asOf": "2026-07-30", "currency": "USD" }

// 200 — one cell per (id, metric); as-of = latest observation on or before
{ "data": [
    { "requestId": "AAPL-US",   "metric": "MARKET_CAP", "dataset": "derived",
      "valueType": "currency", "value": 4925847973597 },
    { "requestId": "BBAS3-BR",  "metric": "close",  "dataset": "prices", "value": 4.11 },
    { "requestId": "BIMBOA-MX", "metric": "SECTOR", "dataset": "entity_attributes",
      "valueType": "text", "value": "Food & Beverage" } ] }

• Metric names colliding across datasets resolve by a documented precedence — reported financials win. Pass an explicit {"metric": "…", "dataset": "…"} pair to override; GET /catalog/metrics lists which datasets serve a given name.
• Financials metrics accept periodMode: "fy" for latest-fiscal-year values; use the fundamentals endpoint for QTR/LTM/YTD series of statement lines, the ratios endpoint for ratio series at a chosen period basis, or precomputed valuation metrics (PE_LTM, EV_EBITDA_LTM, …).
• Cells that compute to null are absent, not errors — a spreadsheet fills #N/A.

Screener

Run screens on the same engine as the platform screener — conditions on financial metrics, price metrics, valuation ratios, and your organization's custom formula metrics, with sector, exchange, country, and universe filters. Companion endpoints: POST /count returns the match count without materializing rows (wire it to a live "N companies match" indicator), and GET /fields lists every screenable field.

// POST /api/content/screener/v1/search
{ "conditions": [ { "metric": "MARKET_CAP", "operator": "gt", "value": 100000000000 },
                  { "metric": "ROE", "operator": "gte", "value": 0.15 } ],
  "metrics": ["REVENUE", "PE"], "sortBy": "MARKET_CAP", "currency": "USD", "limit": 50 }

// 200 — one row per matching security; meta.pagination.total = full match count
{ "data": [
    { "ticker": "AAPL", "entityId": "0000320193", "sector": "Technology",
      "nativeCurrency": "USD", "marketCap": 4925847973597,
      "metrics": { "REVENUE": 451442000000, "PE": 43.9 } }, … ] }

• Operators: gt · gte · lt · lte · eq · neq. A company missing a conditioned metric fails the screen.
• currency defaults to USD here (not LOCAL): screening thresholds compare across issuers, which requires one common currency. Rows still state their nativeCurrency; meta.fxRateDate discloses the rate date used.
• tickers restricts the candidate universe; batch:"Y" queues large screens through batch execution.
• Delisted listings are excluded by default. Omitting excludeStatus applies the platform's reading of "not known to be dead": listings whose status is cancelled / cancelado are dropped. Thousands of delisted US issuers carry their final fundamentals (for point-in-time work), and a ratio screen would otherwise list companies dead since 2012 — a trade-recency test would not catch them, since many still print pennies over the counter. Pass excludeStatus: [] to include them, or status to select by status explicitly; tradedWithinDays remains the separate, opt-in liveness test.

Point-in-time screening

POST /api/content/screener/v1/point-in-time — run a screen as it would have run on a past date. The live screener is present-tense; running it with old thresholds tells you which of today's survivors would pass today.

// POST /api/content/screener/v1/point-in-time
{ "asOf": "2019-06-28",
  "conditions": [ { "metric": "REVENUE", "operator": "gt", "value": 20000000000 },
                   { "metric": "ROE", "operator": "gt", "value": 0.25 } ],
  "metrics": [ "REVENUE", "ROE" ], "sortBy": "REVENUE" }

// 200 — each row states the filing it was judged on
{ "data": [
  { "ticker": "ORCL", "fiscalYear": 2019,
    "fundamentalsPublished": "2019-06-21", "availabilityBasis": "sec_filed",
    "reportingCurrency": "USD", "metrics": { "REVENUE": 39506000000, "ROE": 0.51 } } ] }

• Every row carries fiscalYear, fundamentalsPeriodEnd, fundamentalsPublished and availabilityBasis, so a screen is reproducible: you can see exactly which filing each name was judged on.
• tradedWithinDays defines liveness. It has a default of 10, and widening it admits thinly-traded names a narrow window drops.
• currency defaults to USD and converts at the as-of date's spot rate, not today's — converting a 2019 screen at a 2026 rate would be a lookahead leak through the currency rather than the fundamentals. Ratios are never converted.
• An issuer whose reporting currency cannot be priced on that date is excluded and reported as an FX_UNAVAILABLE row error, never compared on a different basis.

Funds

Local-market fund data across Brazil (65,000+ registered funds), Mexico, and Peru, with the identifier conventions each market actually uses: BR funds resolve by CNPJ in any format (00.000.684/0001-21 or 00000684000121), MX and PE funds by ticker or entity id. Four endpoints share one contract:

Institutional equity holdings live at /api/content/ownership/v1: POST /holders returns who holds a security (per manager: latest filing for the target quarter wins — amendments supersede), and POST /manager-holdings returns a manager's portfolio by filer CIK. The full institutional filer universe (~9,000 managers) is covered for every fully-due quarter; because these disclosures are due 45 days after quarter end, holder lists default to the latest fully-due quarter — pass includePartial: true during a filing window to see the in-progress quarter. Values are USD; weights are percent of reported value.

// POST /api/content/funds/v1/prices
{ "ids": ["00000684000121", "+TASAD1F1"], "fields": ["nav", "netAssets"],
  "startDate": "2026-01-01", "endDate": "2026-07-30", "frequency": "M", "currency": "USD" }

• BR series are the fund's main share class; PE funds carry per-issue currency (PEN or USD), stated per row.
• currency converts monetary fields (NAV, net assets, flows) at each observation's spot rate, exactly as in global prices.

Reference data

Point-in-time identity and the filing record behind it. Both answer the same question in different registers: what did this identifier mean then — the join key any historical dataset outside our universe needs.

Identifier history

POST /api/content/reference/v1/identifier-history — every ticker, ISIN, CUSIP, CIK, exchange and MIC a security has carried, with validity intervals, back to 1926. Rows carry the corporate-action code behind each change, so a rename can be told apart from a merger.

// POST /api/content/reference/v1/identifier-history
{ "ids": [ "AAPL" ], "asOf": "2019-06-28" }

// 200 — what each identifier was on that date
{ "data": [
  { "requestId": "AAPL", "identifierType": "ISIN",
    "identifierValue": "US0378331005",
    "validFrom": null, "validTo": null, "isCurrent": true } ] }

• identifierTypes filters to TICKER · ISIN · CUSIP · CIK · EXCHANGE · MIC · MIC_SEGMENT · SECURITY_DESCRIPTION — the same words /resolve accepts, so a returned value can be sent straight back in.
• startDate/endDate select intervals open at any point in the window, not intervals that began inside it.
• Sourced from the EDI security master: US listings. Anything else returns a row error rather than an empty result, which would be indistinguishable from "never changed".

Filings index

POST /api/content/reference/v1/filings — 2.2M US regulatory filings across 20,205 registrants since 1993: form type, filing date, period end, accession number, amendment type and the document URL. The event clock for announcement-driven work, and the audit trail behind every fundamental we serve.

formTypes matches by prefix, so 10-K includes 10-K/A — an amendment is the same event refiled, and it is usually the restatement you came for.

Corporate-action adjustment factors

/corporate-actions now returns priceFactor, shareFactor, cashComponent, affectsSplitAdjusted, affectsTotalReturn and factorStatus alongside the event terms, so you can rebuild your own adjusted series and reconcile it against ours.

Trading calendars

POST /api/content/reference/v1/calendar — the sessions each market actually trades, for US, BR, MX and PE, from the 1990s through 2029. Non-trading days carry the reason; pass includeNonTrading for a full date spine.

// POST /api/content/reference/v1/calendar
{ "markets": [ "US", "MX" ], "startDate": "2026-04-01", "endDate": "2026-04-06" }

// 200 — 2 April is a session in the US and a holiday in Mexico
{ "data": [
  { "market": "US", "date": "2026-04-02", "isTradingDay": true, "origin": "scheduled" },
  { "market": "MX", "date": "2026-04-01", "isTradingDay": true, "origin": "scheduled" } ] }

• origin is observed (a session we hold prices for), scheduled (rule-derived, the only kind available for future dates) or both. A client planning a rebalance two years out can tell which it is looking at.
• Where the rules and the tape disagree, the row keeps what was observed and states the conflict. A national day of mourning is announced, not ruled — it belongs in the calendar as an observed closure, not as a rule nobody can write.
• session flags US early closes (Independence Day eve, the day after Thanksgiving, Christmas Eve). No other market publishes them reliably enough for us to claim the same.

Trading-day arithmetic

POST /api/content/reference/v1/calendar/offset — move N trading days from an anchor date. T+2 settlement, rebalance scheduling and event windows are the same question, and all silently wrong when answered with calendar days.

// T+2 from 1 April 2026, across Easter
{ "markets": [ "US", "MX" ], "dates": [ "2026-04-01" ], "offset": 2 }

// 200 — same anchor, different answers: Mexico also closes Maundy Thursday
{ "data": [
  { "market": "US", "resultDate": "2026-04-06" },
  { "market": "MX", "resultDate": "2026-04-07" } ] }

• An offset of 0 snaps to the session in force on or before the anchor — what an as-of date means when it lands on a holiday.
• Where the calendar runs out before the offset does, resultDate is null and tradingDaysAvailable says how far it reached. Clamping to the edge would return a plausible, wrong date.

Insider transactions

POST /api/content/ownership/v1/insider-transactions — 1.75M insider transaction rows since 2004, with the reporter's relationship to the issuer (director, officer, 10% owner) joined in.

Two distinctions the raw filings do not make, and that a caller would otherwise have to rebuild:

• Not every code is a trade. The filing puts open-market purchases and sales (P/S) in the same table as grants (A), option exercises (M/X) and tax withholding (F). Summing across them produces a number that reads like trading activity and is not — so rows carry an activity classification, and the endpoint defaults to open-market only.
• Derivative rows are different objects. Adding derivative and non-derivative share counts double-counts an exercise against the transaction it settles. securityBucket is explicit on every row and filterable.

Fixed income

POST /api/content/fixed-income/v1/search · /reference · /prices — Mexican paper, Peruvian corporates and Brazilian debentures. 2.2M price observations; a bond universe is discovered rather than known by ticker, so start with search.

• Durations are years everywhere. Brazil publishes dias úteis, converted at 252 — the DI curve's own divisor.
• Percent of par is served only where the source publishes it (PE, BR). Mexico carries 114 distinct par values, so a ratio derived from its master would be a normalization nobody can stand behind; parValue rides on the reference row instead.
• Nothing is FX-converted: a yield is a rate and a duration is time. Rows state their own currency.
• Ratings come back as a map keyed by agency — six rate Mexican paper and they disagree.

Fund fees

POST /api/content/funds/v1/fees — expense ratios, management and performance fees, loads and turnover. US share classes from prospectus disclosures (7.58M facts over 64,931 classes), Brazilian funds from their own regulatory fund disclosures.

• grossExpenseRatio and netExpenseRatio are never collapsed into one number: the difference is a waiver, and waivers expire.
• Every field on a row comes from one document, whose date is on the row. Reading the newest value per field independently produced fee schedules no prospectus ever stated.
• asOf selects the newest disclosure published on or before that date.

Ownership changes

POST /api/content/ownership/v1/changes — what institutions did in a security between two quarters: who entered, added, trimmed or left, ordered by what actually moved.

• Both books are assembled through the platform's institutional-holdings authority: a NEW HOLDINGS amendment is additive, not superseding, so a book is the base filing plus every additive amendment after it. Read as a replacement, one manager's book came out at $47.0bn against a true $4,042.9bn.
• Actions: NEW · ADDED · TRIMMED · EXITED · NOT_FILED · UNCHANGED, filterable.

Research

The academic research surface: point-in-time panels over survivorship-safe historical universes, pinned by citable Research Snapshots. Every dataset answer leads with its diagnostics — universe coverage, PIT withholding, empty metrics — because a panel looks equally complete at 6% and 96% coverage without them. Methods are documented in the twelve methodology chapters shipped with the academic package (identity, PIT, universes, adjustment, event studies, snapshots, and more).

Access. Same authentication as every endpoint (API key or session). While the Academia section is in staging, research endpoints additionally require the research role on your account.

Downloads — the academic package

Self-hosted and versioned with the platform: the edifinance SDK wheel, the teaching notebooks in three languages (English, Spanish, Portuguese — Excel sample workbooks included), and the twelve methodology chapters. The SDK is not on PyPI while Academia is in staging — install it straight from here (works on Colab too).

# the SDK is not on PyPI yet — install from the platform
pip install https://api.edi.finance/static/academic/edifinance-0.1.1-py3-none-any.whl

• edi-academic-package.zip — everything: SDK wheel, notebooks (en/es/pt), methodology chapters, Excel workbooks.
• edifinance-0.1.1-py3-none-any.whl — the Python SDK alone.
• notebooks/README.md — the notebook inventory; individual files under /static/academic/notebooks/.

Historical universes

Universe definitions with snapshot counts. Resolving one as of a date yields an immutable membership snapshot: status is stored as of that date (a security delisted in 2022 is active in a 2019 snapshot), and unpriced members are flagged price_coverage: none, never dropped — dropping them is the survivorship bias this engine exists to prevent.

# memberships as of a date, grid-shaped
curl -s "https://api.edi.finance/api/research/excel/universe?code=BR_ALL_EQUITY&as_of=2022-12-31" \
  -H "X-API-Key: $EDI_KEY"

Dataset builder — PIT panels

Builds a panel from a named definition (create one via POST /api/research/dataset-definitions) and returns a capped preview plus full-panel diagnostics. In PIT mode, a fact whose availability cannot be proven is withheld and counted — never served on a guess. PIT coverage is market-honest: US and Brazil are fully supported; Mexico is partial; Peru is not supported.

# SDK install: see Downloads above (not on PyPI yet)
from edifinance import ResearchClient

rc = ResearchClient(api_key="…")
df = rc.panel(universe="BR_ALL_EQUITY", metrics=["REVENUE", "NET_INCOME"],
              start="2019-12-31", end="2023-12-31", point_in_time=True)
df.attrs["diagnostics"]  # coverage, withheld facts, snapshot code

• Definitions are named and reusable; the SDK derives deterministic names so identical calls share one definition.
• Pass a snapshot_label to mint a citable Research Snapshot for the run.
• Excel: =EDI.DATASET(…) spills the same panel with a provenance line as row 1.

Research Snapshots

A snapshot (EDI-RS-<year>-<token>) pins everything needed to recreate an analysis: knowledge date, the frozen universe snapshot, metric/factor registry content hashes, currency and adjustment policies, the request payload, and an order-independent dataset hash. Published snapshots are immutable by database trigger — a citation resolves forever. Provenance for one code: GET /api/research/excel/snapshot/{code} (also =EDI.SNAPSHOT(code) in Excel).

Event studies

Runs a seeded study: market-model or market-adjusted abnormal returns, AAR/CAAR with a cross-sectional t-test, trading-day windows, event dates mapped forward to the next session (US filing events are timed by regulatory acceptance timestamps — two-thirds of filings land after the close and belong to the next session). Dropped events are counted by reason. The engine's standing validation is a placebo test: random dates must find nothing.

curl -s -X POST https://api.edi.finance/api/research/event-study \
  -H "X-API-Key: $EDI_KEY" -H "Content-Type: application/json" \
  -d '{"definition_name": "us_earnings_announcements"}'

Batch execution

Every data endpoint accepts "batch": "Y" with the identical request body — flip one flag when a workbook grows past the synchronous limits. The batch branch runs the same code path as the synchronous one, so results are identical by construction.

Jobs are owned by the submitting account and expire after 24 hours. Failed jobs deliver an EXECUTION_ERROR envelope through the same contract.

Bulk export

Batch execution scales a workbook. Bulk export is for the other shape of problem: loading a dataset into your own store, then keeping it current. One job, one compressed file, and an exact cursor to resume from.

# Seed: the full price history, one file
curl -b cookies.txt -X POST https://api.edi.finance/api/content/bulk/v1/export \
  -H "Content-Type: application/json" -d '{"dataset": "prices_eod"}'

# Every night after that: only what has been written since the last pull
curl -b cookies.txt -X POST https://api.edi.finance/api/content/bulk/v1/export \
  -H "Content-Type: application/json" \
  -d '{"dataset": "prices_eod", "changedSince": "2026-08-21T23:16:18.915379Z"}'

• Nine datasets today: prices_eod, corporate_actions, corporate_action_factors, identifier_history, filings_index, insider_transactions, fundamentals_standardized, index_levels, fund_monthly_returns.
• A date window (startDate/endDate) and changedSince are alternatives, never both: one selects a period, the other selects what has changed.
• Datasets with no source write timestamp reject changedSince by name rather than returning everything each time, which would look like a working delta feed.
• Columns are an explicit projection per dataset, so a column added internally never appears unannounced in your warehouse. Files expire after 48 hours.
• Two exports run concurrently across the platform; a third returns 429 with Retry-After.

Dataset status

GET /api/analytics/v1/datasets — for each dataset: the coverage window, when it last loaded, row and entity counts, which endpoints serve it, and a freshness verdict.

// GET /api/analytics/v1/datasets?dataset=prices_eod
{ "data": [ {
    "datasetId": "prices_eod", "cadence": "daily",
    "firstObservation": "1991-12-02", "lastObservation": "2026-08-21",
    "lastLoadedAt": "2026-08-21T23:16:18Z",
    "freshness": "FRESH", "observationLagBusinessDays": 0,
    "expectedLagBusinessDays": 2,
    "rowCount": 21768648, "rowCountIsEstimated": true,
    "computedAt": "2026-08-23T13:52:25Z" } ] }

• lastObservation is the latest business date in the data; lastLoadedAt is the latest write. They answer different questions — a loader that ran an hour ago and wrote nothing new is a different failure from one that has not run in a week — so they are separate fields, and lastLoadedScope states what the write probe measured.
• freshness is FRESH · LAGGING · STALE · UNKNOWN, judged against a per-dataset expectedLagBusinessDays. One threshold cannot serve both a daily price feed and a quarterly ownership filing, so the raw business-day lag is returned alongside the verdict for your own rule.
• rowCountIsEstimated distinguishes a planner estimate from an exact count.
• computedAt is the refresh that produced the row, so a stalled refresher reports its own age instead of passing off old coverage as current.

Currency conversion

Conversion is server-side and semantics-aware — the right rate kind applies to each value:

Cross rates compose through USD. A conversion that cannot be priced — unknown source currency, unsupported pair, no rate within the staleness window — is an FX_UNAVAILABLE row error with the affected rows dropped: you never silently receive unconverted native values.

CSV everywhere. Send Accept: text/csv on any data endpoint and the envelope's flat rows come back as a CSV file — nested maps flatten to dotted columns, row-level errors ride along as trailing # comment lines, and registryVersion is stamped at the end. The fastest path from any endpoint into a spreadsheet or dataframe.

Complete endpoint index

Every operation the service routes, 113 of them, generated from the route table itself. The sections above describe the ones most clients start with; everything here is live and callable.

Error taxonomy

Rendering rule (clients). Every row error carries a detail string with instructions. In a context entirely about one security — a scalar =EDI() cell or a single-security spill (HISTORY, FUNDAMENTALS, RETURNS, CA) — clients render the symbol plus the detail (e.g. #N/A — no match for ticker 'ZZNOSUCH' in market US), so the fix reaches the sheet. Multi-security grid cells (MATRIX) keep the bare symbol. These are strings, not native Excel error values, so ISNA/IFERROR semantics are unaffected. Programmatic consumers should key on errors[].code, never on cell text.

Request-level failures use standard HTTP statuses: 400 malformed request · 401 unauthenticated · 404 unknown/expired batch id. The add-in surfaces 4xx detail text in the cell as well.

Request limits

Clients & Excel

Python. The reference client (edi_analytics.py) wraps login, the envelope, and batch polling, and ships the spill helpers used by the Excel integration — spill_scalar, spill_series, spill_fundamentals, spill_ratios, spill_matrix, spill_fibonacci — each returning a 2-D grid ready for a dataframe or a dynamic array.

Excel. The add-in registers formulas over exactly this API. Start with the Excel installation guide, which covers both Microsoft 365 administrator deployment and individual installation. You can also download the production manifest or the demo workbook directly.

=EDI.VALUE("AAPL-US", "MARKET_CAP")                → one value
=EDI.HISTORY("BBAS3-BR", "price", "2026-01-01", "2026-07-30", "M")   → spills 7×2
=EDI.FUNDAMENTALS("AAPL-US", "REVENUE", "LTM")   → spills fiscal periods
=EDI.MATRIX(A2:A20, B1:F1)                          → spills a comps grid

Bad inputs degrade cell-by-cell (#N/A, #NAME?) — never a failed workbook refresh. The add-in batches formula requests through batch execution automatically.

Anything else. The OpenAPI 3 document drives generated clients for other stacks.

Try it

Sign in with your EDI account and run live requests against the API — every example on this page is runnable. Requests execute with your entitlements, exactly as they would from Excel or a script.

Staged capabilities

These request axes are part of the contract but deliberately rejected (HTTP 400, with an explanatory message) until their backing machinery ships — you can code against the vocabulary today without risk of silent wrong answers:

• basis: "restated" — restated financials, pending vintage modeling.
• periodicity: "SEMI" — semi-annual reporters, pending normalization.
• adjust: "UNSPLIT" / "SPLIT_SPINOFF" — pending adjustment-factor history on the serving path. The factors themselves are available today on /corporate-actions and through bulk export.
• Consensus estimates — no feed behind the dataset. Reported financials are the only source served.
• Funds and screener content domains.