Bargo for Quant Researchers

Datasets nobody else has, structured for modeling.

Bargo turns unstructured speech and text into structured datasets on global equities: podcast attention, promotion, hiring and language. Every row is source‑backed and mapped to industry, company and authority. Clean enough to test in an afternoon, honest about the limits.

Every row passes the Bargo Verification Layer

Accurate · Reliable · Source‑backed

444,220
company mentions, evidence‑gated
17.8M
job postings retained, with closure dates
2006→
filings history: 13F, insiders; congress 2012→
127,000+
podcast transcripts, SHA‑256 stamped snapshots

Three plays

Not claims. Worked examples.

Edge

Features your backtest has never seen

Podcast attention per ticker over years. Promotion flags joined with filings. Hiring momentum without ghost jobs. Management language deltas per quarter.

Each series ships with the coverage caveats documented, so you know what you are testing.

■ feature_menu · illustrative
attention_seriescoverage per ticker · venue structure · years
promotion_flagspre-registered rule · append-only
hiring_momentumverified-live counts · closures dated
language_deltaQoQ guidance wording shifts · computed
Trust

Backtest-safe by construction

The classic lookahead bugs are killed at the schema level: two clocks on every row. When it happened, and when it was knowable.

And where a dataset is not yet point-in-time, the file says so, up front.

■ two_clocks · illustrative
congressional tradetraded_on 07-21 · filed_on 08-12
news itempublished 16:32 · seen 16:41
honest limitationcurrent snapshot not point-in-time · documented
Speed

Merge on day one

Every dataset keys to the same company identifiers. Your first join works across calls, mentions, jobs, filings and panels, with zero entity-matching work.

■ join_keys · illustrative
7 datasetsone ticker + CIK spine
deliveryCSV / Parquet snapshots · digest-stamped
your samplereproducible rule · you draw it, not us
Bargo Verification Layer · Accurate · Reliable · Source-backed

Every claim in your memo keeps its receipt.

A thesis built on Bargo survives the only question that kills ideas in committee: "how do you know?"

■ IC memo · short thesis · excerpt
Attention burst is manufactured, not organicfig. 01 · venue history
Company sold stock into every prior burstfig. 02 · EDGAR joins
Insider sales clustered at the topfig. 03 · Form 144s
Hiring frozen while the story acceleratesfig. 04 · live postings
Management language turned defensivefig. 05 · call diff, 23:41

Every figure is checkable without trusting us

Compliance clicks the link: the audio on the publisher's feed, the filing on EDGAR, the posting on the employer's site. Nothing rests on trusting us.

✓ The quote
Verbatim excerpt, timestamped to the second, on the source's own audio
✓ The number
Located digit-for-digit in the filing before it shipped
✓ The calculation
Computed on verified figures, never recalled from an AI's memory

How the three gates work →

Access

Runs the way you work

Ask it anything

Bargo Agent

Interrogate the catalog before you commit to a pull.

> Coverage depth for podcast mentions on Russell 2000 names?
For AI agents & terminals

Bargo MCP / CLIs

Pull snapshots straight into pandas, or wire the catalog into your research agents.

$ bargo pull mentions --all \ --format parquet
Programmatic

Bargo APIs

Every dataset as clean endpoints, joined per company.

GET /v1/podcasts/mentions?ticker=NVDA GET /v1/jobs/momentum?ticker=NVDA

The terms

Evaluate it like a data source, not a demo

SnapshotA documented evaluation snapshot with per-file SHA-256 digests: a row you check next year is provably the row published today.
Your sample, not oursA reproducible sampling rule lets you draw your own verification sample. We cannot have selected it for you.
The flag sheet, liveThe weekly promotion flag sheet as it fires, with excerpts, timestamps and EDGAR context per row.
No performance claimDetection rules were frozen before outcome data existed and flags are append-only. First graded cohort reads out late October 2026; we claim nothing before then.
Honest limitationsThe current snapshot is not point-in-time; append-only daily snapshots are being added. Backtests on today's file should assume lookahead.
We would rather show you the data than describe it.
Request an evaluation snapshot