Upcoming challenge

Company research agents · Being shaped · public sample available

Prize
$2,500 + $100 every fortnight
Deadline
60-day round
Tested on
Separate fixed test set
Build focus
Measured outcome

Norwegian public company intelligence

Build the best research view of a Norwegian public company.

Builderr supplies a large, audited universe of Norway-domiciled companies with publicly traded shares, whether they are listed in Oslo or abroad. Your product must bring together wide financial and operating evidence, keep it accurate and current, and let a user ask smart cross-company questions, save the work, and return to it. The goal is useful investment research—not a price prediction. The planned round is 60 days with $2,500 in final rewards, plus another $100 every fortnight.

Sponsored by Liltved, a Norwegian serial entrepreneur, CPO at Front End, and founder of JBox.

Rewards

$2,500 at the finish. $100 every fortnight.

Public voting is only for entries that pass the published qualification checks. It never changes the final score or ranking.

$2,000

Main challenge

The final challenge prize pool. The final split will be published before entries open.

$500

JBox bonus

A separate bonus pool for final top-three results built using the JBox platform.

$100

Fortnightly vote

One award every two weeks to the highest-voted qualified agent.

Builderr will host qualified agents for the competition voting and review period, so entrants do not pay the competition hosting cost.

What you build

Cover widely. Verify carefully. Make it simple.

01

Cover

Find the important facts, filings, announcements, outlook, risks, and operating signals across the Builderr-supplied universe.

02

Verify

Attach a source, date, reporting period, and availability state to every important claim. Never turn missing data into zero.

03

Explain

Show what improved, weakened, or remains uncertain, why it matters, and where the evidence conflicts.

04

Search and return

Let users ask cross-company questions, inspect the matching evidence, save searches and pins locally, and return to the same workspace.

Start here

See the site you are building toward.

This working mock-up shows the expected product shape: search, company profiles, source-backed research, and a refresh view. The sample has 500 dated audit records. 251 are currently exposed as Norway-qualified profiles representing 248 distinct names; the rest stay audit-only and clearly labelled. This is an Oslo-first prototype, not a claim that the worldwide Norway-domiciled universe is complete.

Open the working mock-up →
Public sample251 listed equities
251listed equities
138websites
250official filings

Registry and account fields are added only for clean issuer matches. 250 profiles have an official filing record in this snapshot; other gaps stay visible.

How it will be scored

Four things decide the result.

Builderr supplies the companies. Every entry must first capture the core financial information. After that, this is a discovery competition: find more decision-useful information across public sources without adding noise. Material factual errors, fabricated claims, or silent missing data fail the accuracy gate.

Information breadth and usefulness12: financial and operating depth beyond the baseline · 12: additional useful public signals · 6: useful source breadth and discovery automation.30
Accuracy and freshnessCorrect facts and periods, claim-level sources, honest gaps, and reproducible updates.30
Agentic research and screeningNatural-language cross-company search, useful cited insights, saved searches, pins, history, and export.25
UX and ease of useFind a company, understand the case, inspect the proof, and return to prior work.15

The hidden evaluation gives every entry the same companies, cutoff, and run budget. The mandatory baseline covers identity, reporting period, income statement, balance sheet, cash flow, debt and liquidity, margins, and core filings where available. Beyond it, breadth is competitive: unique, relevant, correctly attributed findings count—not pages, posts, or duplicated mentions. Public social posts can add context but cannot prove financial facts by themselves. The prototype stores saved searches, pins, and history in the user's browser; production can move the same workspace to authenticated accounts. Billing is not required.

Coding qualification gate

Reliable code qualifies. The product result wins.

Framework choice and complicated architecture earn no points. Before scoring, the submission must run cleanly, preserve evidence, apply screens deterministically, refresh without losing history, and handle source access safely.

01

Reproduce

One documented setup path, pinned dependencies, declared APIs and models, and one evaluator command.

02

Query exactly

Compile natural language to an inspectable structured screen. Deterministic code—not generated prose—decides company membership.

03

Preserve proof

Keep claim-level sources, periods, cutoffs, availability states, immutable snapshots, and idempotent refresh behavior.

04

Fail safely

Keep secrets server-side, reject unsafe source URLs, sanitize retrieved content, recover from corrupt local state, and test the failure paths.

The downloadable brief contains the full coding gate and minimum regression set. A setup or evaluator failure is blocked and rerun; it is not converted into a factual score.

The rules that matter

Use the best evidence you can audit.

  • Show Not available when the source has no value.
  • Keep field-level sources, snapshot dates, and refresh history.
  • Declare APIs, models, licences, caches, and hosting assumptions.
  • Login, API-key, and paid sources are allowed when their use is lawful, declared, and reproducible for the evaluator.
  • No login is required for the prototype. Persist saved searches, pins, and history in local browser storage.
  • Design the saved workspace so production can later move it behind authenticated accounts.
  • A setup failure or evaluator error is blocked and rerun, not a public zero.

Participant brief

Start with one explainable company view.

Run one lookup, prove the facts and filings, then collect dated signals and explain the trajectory. Finish with a refresh diff that a reviewer can reproduce. Use the source map to choose public inputs.

Download the starter brief →