Protocols

How the data was made. A Protocol is a reproducible recipe: an ordered sequence of typed steps that fetch, transform and export data. The first set below build Meridian’s open Datasets; the example Protocols further down build well-known public datasets on your own disk. Every step is a versioned operator, a SQL model, or (rarely) an opaque command, and a Protocol can gate its output against a finetype schema contract before it ships.

Running a Protocol you did not write is equivalent to running a shell script you did not read. Arcform runs a step’s command as you, with no sandbox and no container, and the finetype gate checks the data a Protocol produces, not what its steps do. What running one costs you.

This is a read-only view for browsing and discovery — authoring lives elsewhere.

EDGAR ↔ GLEIF — Company-to-LEI Crosswalk

Resolve SEC companies to their GLEIF LEI: a deterministic fund backbone unioned with a probabilistic name match, tiered by confidence.

produces library.edgar_lei
23 stepsduckdb >=1.2finetype gated

SEC EDGAR — Company Tickers

Ingest the SEC company_tickers_exchange.json into a typed, reproducible Parquet — information retrieval only.

produces sec.company_tickers_exchange
7 stepsduckdb >=1.2finetype gated

NAICS — Industry Classification

Ingest the U.S. Census 2022 NAICS descriptions workbook into a typed, reproducible Parquet.

produces census.naics_2022
7 stepsduckdb >=1.2finetype gated

GLEIF — Legal Entity Identifiers

Ingest the GLEIF golden-copy full-file CSV into a typed, reproducible Parquet — the one Protocol with an opaque fetch, by design.

produces gleif.lei
7 stepsduckdb >=1.2gate pending

Example Protocols

Recipes over well-known public datasets. Meridian hosts none of them; each downloads its data, from the address its fetch step names, when you run it.

A Brightfield dashboard of the USGS's 2025 earthquakes of magnitude 4.5 and up. The main panel maps them by longitude and latitude, shaded by how many fall in each hexagon, under the title "A fifth of 2025's M4.5+ earthquakes (1,794 of 8,502) struck off Kamchatka, 1,607 of them from the M8.8 of 29 July on". Below it, earthquakes per week; beside it, a histogram of depth, its counts on a log scale, and a count by magnitude scale.

USGS Earthquakes 2025

The earthquakes of magnitude 4.5 and up that the US Geological Survey's own network reported in 2025, from the USGS earthquake catalogue.

A Brightfield dashboard of New York's yellow taxi trips in January 2026. The main panel is a heatmap of trips by hour of pickup and day of the week, under the title "New York's yellow cabs keep two clocks: weekdays build to a 6 pm rush, weekends run late, with three times the weekday trips at 1 am". Below it, trips per day; beside it, a histogram of trip distance, its counts on a log scale, noting that 68% of the trips of 16 to 19 miles begin at an airport, a count by where the trip began, and how complete each trip record is.

NYC Yellow Taxi Trips

Every yellow taxi trip the New York City Taxi and Limousine Commission recorded as picked up in January 2026, read from the TLC's own monthly file.

A Brightfield dashboard of Epoch AI's notable AI models. The main panel plots training compute (log10 FLOP) against the year each model was published, under the title "Training compute bends upward around 2010: since then it doubles in months, not years". Below it: models published per year, how the compute is known, and counts by the kind of organisation that built the model, how it can be reached, and its first-listed domain (top eight).

Epoch AI Notable AI Models

Epoch AI's database of notable AI models: when each was published, its first-listed domain, its training compute and how that compute is known, the kind of organisation behind it, and how the model can be reached.

Two charts from the 1994 Census 'Adult' extract: the share of people earning over 50K at each level of education, and a histogram of usual hours worked per week.

Census Income (Adult)

The standard 'Adult' benchmark extract of the 1994 US Census: 32,561 person records with demographic and employment attributes and an income label, '<=50K' or '>50K'.

A Brightfield dashboard of Online Retail II. The main panel counts orders per month from December 2009 to December 2011, coloured by whether the customer was returning, new that month or had no customer ID, under the title "The regulars carry this shop: in every full month of 2011, returning customers placed 77% to 87% of orders, November's peak included". Beside it: orders by customer type, orders by country (top eight), and how sure the new and returning labels are.

Online Retail II

The invoices of a UK-based online retailer of giftware, many of whose customers are wholesalers, from 1 December 2009 to 9 December 2011, as the UCI Machine Learning Repository publishes them.

A point map of California's census districts by longitude and latitude, each point coloured by the district's median house value.

California Housing

20,640 California districts (census block groups) from the 1990 Census, each with its median income, housing age, average rooms and occupancy, population, latitude and longitude, and median house value.

Two charts of the vega-datasets films: each film's worldwide gross against its production budget, and the number of films in each genre.

Movies (vega-datasets)

3,201 films from vega-datasets, with production budget, US and worldwide gross, DVD sales, running time, distributor, director, genre and two critic scores.

A Brightfield dashboard of the Palmer Penguins. The main panel plots bill depth against bill length, coloured by species, under the title "Longer bills are shallower across all penguins (r = −0.24), but deeper within each species (r = 0.39 to 0.65): the overall trend is the species mix". Below it, a histogram of flipper length; beside it, counts by species, by island, and of complete records.

Palmer Penguins

Bill length and depth, flipper length, body mass and sex for 344 Adélie, Chinstrap and Gentoo penguins on three islands near Palmer Station, Antarctica, from 2007 to 2009.

ScienceQA

ScienceQA is a set of multiple-choice science questions with worked explanations, published by Pan Lu and colleagues with their NeurIPS paper, 'Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answering'.

A map of a 50% sample of MedMCQA's questions, one point per question, placed by a UMAP projection of its text and coloured by subject.

MedMCQA

MedMCQA is a set of multiple-choice questions from medical entrance exams, published by Pal, Umapathi and Sankarasubbu at CHIL 2022.

A map of SuperGPQA's questions, one point per question, placed by a UMAP projection of its text and coloured by discipline.

SuperGPQA

26,529 graduate-level multiple-choice questions spanning 13 disciplines, 72 fields and 285 subfields, each with four to ten answer options, the correct answer, a difficulty label and whether it requires calculation.