data.placeholder.com/company

Company in a box

One coherent, entirely fictional company for agent evaluations, RAG tests and AI product demos. Org chart, CRM, deals, invoices, support tickets, incidents, commits, email, chat, calendar and a wiki that all tell the same story. Plus questions with exact answers and the evidence behind them.

Brightfjord, Inc.

Brightfjord makes scheduling and dispatch software for trades businesses (HVAC, plumbing, electrical). Portland HQ, a Denver office, remote staff, about 40 people. The data covers Q1 2026 (5 January to 31 March), with some 2025 history. Money is in USD, timestamps are Pacific time with an explicit UTC offset.

• https://data.placeholder.com/company/brightfjord/v1overview: profile, every collection with count, fields and link, wiki pages, examples
• https://data.placeholder.com/company/brightfjord/v1/questionsevaluation questions with answers and evidence
• https://data.placeholder.com/company/brightfjord/v1/bundle.zipthe whole dataset as one zip (JSON + Markdown), for offline indexing

Collections

IDs are readable and stable: EMP-001, ACC-005, DEAL-001, TCK-1010, INC-2026-001, THR-001, EM-0042, CHM-0109, EVT-…; wiki pages by slug, commits by SHA prefix.

Routes and filters

/company                                    companies
/company/brightfjord/v1                     overview
/company/brightfjord/v1/company             the company profile
/company/brightfjord/v1/<collection>        list (paged)
/company/brightfjord/v1/<collection>/<id>   one record
/company/brightfjord/v1/wiki/<page>         Markdown (?format=json for metadata + body)
/company/brightfjord/v1/bundle.zip          everything

• Filters: any top-level field, exact match. Comma-separated or repeated values mean OR; list fields match on membership; null matches null. An unknown field is a 400 that lists the valid ones.
• Search: _q= is a case-insensitive substring search over all text in a record.
• Paging: _page (from 1) and _limit (default 100, max 500). Lists return {total, page, limit, pages, items, links}.

• https://data.placeholder.com/company/brightfjord/v1/tickets?status=open&_limit=20open tickets
• https://data.placeholder.com/company/brightfjord/v1/tickets?priority=P1,P2P1 or P2
• https://data.placeholder.com/company/brightfjord/v1/events?attendee_ids=EMP-024one person's calendar
• https://data.placeholder.com/company/brightfjord/v1/emails?_q=postmortemtext search
• https://data.placeholder.com/company/brightfjord/v1/deals/DEAL-001one deal
• https://data.placeholder.com/company/brightfjord/v1/email_threads/THR-001a thread with its messages
• https://data.placeholder.com/company/brightfjord/v1/wiki/pricing-and-discountsa wiki page as Markdown

Questions and the answer key

34 questions, each with an exact answer and the records that prove it. Use them to score retrieval, tool use and reasoning.

{
  "id": "Q01",
  "question": "Who approved the 22% discount on the Halvorsen Mechanical expansion deal?",
  "answer": "Ingrid Solberg (CEO)",
  "answer_value": "EMP-001",
  "answer_type": "entity",
  "answer_ids": ["EMP-001"],
  "kind": "trap",
  "evidence": [
    {"collection": "deals", "id": "DEAL-001", "role": "proof", "field": "discount_approved_by", "equals": "EMP-001"},
    {"collection": "emails", "id": "EM-0043", "role": "proof", "quote": "I did not approve the 22% (I can only go to 20%). Ingrid approved it this morning."},
    {"collection": "chat_messages", "id": "CHM-0109", "role": "stale", "quote": "Halvorsen 22% approved by Priya", "note": "Wrong; corrected in the next message from Priya."},
    …
  ],
  "notes": "…",
  "sources": ["chat_messages", "deals", "emails", "wiki"]
}

• kind: single-hop, multi-hop, aggregate or trap. Filter with ?kind=trap.
• answer_type: text, entity, number, date, boolean or list. Compare answer_value exactly; answer is the human-readable version.
• evidence[].role: proof (supports the answer), context, or stale (a wrong or outdated statement a good agent must not trust).

Realistic mess

Like a real company, the records contain typos, facts that are later corrected and statements that contradict each other. Trap questions point at both the stale record and the one that wins, so you can tell whether a system read the latest, authoritative source.

Safety and stability

• Entirely fictional. All domains use the reserved .example TLD; phone numbers use the fictional +1 555-01xx range.
• Version 1 is frozen: these bytes never change, and responses are cached for a year. A future version will get a new path (/v2). bundle.zip is byte-for-byte reproducible.
• Read-only: GET and HEAD.