NeuralKG / instance.yaml
rvguha's picture
Sync engine: instance.yaml, extensions, ARD widening, whole-answer verification (part 2)
5ef0451 verified
Raw History Blame Contribute Delete
9.25 kB
# What THIS deployment answers over.
#
# The engine holds no knowledge of any particular source -- nothing in the query path names one,
# and tests/test_query_understanding_isolation.py fails if a source name reaches a prompt. This
# file is the other half of that: the small set of facts that genuinely are per-deployment.
# Point it at a different ARD, describe the kinds of thing that ARD knows about, and the same
# code answers over a different world.
#
# Every key is optional. Omitted values fall back to what the engine used before this file
# existed, so an instance that only changes `ard.finder_url` writes three lines.
#
# ard.finder_url where the ARD Agent Finder is. AGENT_FINDER_URL overrides it, because a
# container is configured by its platform and a file baked into an image
# must not win over that.
# identity what to call the site. `name` is substituted into the served pages.
# domains the kinds of thing this ARD describes, and how an entity of each kind is
# IDENTIFIED. This is the only part the query path reads: derive.py uses it
# to tell a parameter that selects by identity (cik=, ein=, fips_place=) from
# one that searches by name (q=, org=). An ARD over the NASA Exoplanet
# Archive would list pl_name and hostname here and change nothing else.
# Matching is case-insensitive. A trailing * matches by
# prefix. The list is descriptive, not a closed vocabulary -- `type` from
# query understanding is a free noun phrase and is never checked against it.
# NOTE write question lists as block sequences (`- What ...?`), not inline
# `[...]`: YAML rejects a bare `?` inside a flow sequence, and every
# sample question ends in one.
# examples the homepage question tabs. Purely presentational, and wholly corpus-
# specific: these questions are meaningless against another ARD.
ard:
finder_url: http://127.0.0.1:8088
identity:
name: Neural KG
tagline: OKF + ARD
description: Ask a question in plain English about US public data.
# Shown in the empty question box. Pinned rather than derived from the first example,
# so editing the examples does not silently change the prompt visitors see.
placeholder: e.g. Is the American Red Cross a 501(c)(3)?
domains:
- name: company
coidentify: hub # Wikidata carries CIK for US registrants
identifiers:
- cik
- lei
describes: US public companies, by SEC central index key
- name: nonprofit
coidentify: hub # Wikidata carries EIN
identifiers:
- ein
describes: US tax-exempt organizations, by IRS employer identification number
- name: place
coidentify: hub # Wikidata carries FIPS and GNIS
identifiers:
- geo
- gnis
- fips*
describes: US states, counties, cities and census places
- name: any
identifiers:
- qid
describes: cross-domain Wikidata identity, used by the entity crosswalk
name_selectors:
- q
- org
- awardee
- awardeename
- name
- org_names
- recipient_search_text
- place
source_order:
- sec-edgar
- sec-bq
- treasury
- census
- cdc-places
- nonprofit-990
- nonprofit-bmf
- irs-990-bq
- census-acs-bq
- irs-grants
- nonprofit-profile
- usaspending
- nih-reporter
- nsf-awards
- grants-gov
- college-scorecard
- fema
source_examples:
sec-edgar:
- What was Apple's total revenue?
- Microsoft net income in 2023
- Apple's diluted earnings per share
treasury:
- What is the US national debt?
- Euro to dollar exchange rate
census:
- Median household income in California
- Poverty rate in Chicago
- Unemployment rate in Detroit
cdc-places:
- Diabetes prevalence in Chicago
- Obesity rate in Miami
nonprofit-990:
- American Red Cross total revenue
- Is the Sierra Club a 501(c)(3)?
usaspending:
- How much federal funding has the American Red Cross received?
nih-reporter:
- How much NIH research funding does Stanford get?
nsf-awards:
- NSF research awards for MIT
grants-gov:
- What grants can a nonprofit apply for in education?
examples:
- label: 🏛️ Nonprofits
dirs:
- nonprofit-990
- nonprofit-bmf
- nonprofit-profile
- usaspending
- grants-gov
- nih-reporter
- nsf-awards
queries:
- What was the American Red Cross total revenue?
- Is the Sierra Club a 501(c)(3)?
- Where is the Nature Conservancy headquartered?
- Are donations to the ACLU Foundation tax-deductible?
- When was the Sierra Club founded?
- Who is the CEO of the Wikimedia Foundation?
- What does Feeding America do?
- How much federal funding has the American Red Cross received?
- How much NIH research funding does St. Jude receive?
- What grants can a nonprofit apply for in education?
- How much does the ACLU pay its officers?
- Is the American Red Cross in good standing with the IRS?
- q: Compare the total revenue of the American Red Cross and Feeding America
tag: comparison · 2 lookups
- q: What share of the American Red Cross revenue comes from federal funding?
tag: ratio · cross-source join
- q: What is the largest nonprofit in the US?
tag: ranking · BigQuery SQL
- label: 🎓 Academia & Research
dirs:
- nih-reporter
- nsf-awards
- nonprofit-990
- college-scorecard
queries:
- How much NIH research funding does Stanford get?
- NSF research awards for MIT
- How much NIH funding does Johns Hopkins receive?
- Harvard University total revenue
- NSF research awards for Caltech
- q: Give me some universities that get more than a billion dollars from NIH
tag: filtered-subset · propose-and-verify
- q: Which university gets the most NIH funding?
tag: refused · no source can rank
- label: 📈 Companies
dirs:
- sec-edgar
queries:
- What was Apple's total revenue?
- Microsoft net income in 2023
- Apple's diluted earnings per share
- Tesla's research and development expense
- NVIDIA total revenue
- q: What were Apple's earnings?
tag: ambiguous · answered per interpretation
- q: How big is Microsoft?
tag: ambiguous · answered per interpretation
- label: 🏘️ Communities & Health
dirs:
- census
- cdc-places
- fema
queries:
- What percentage of households have broadband internet in Detroit?
- What disasters have been declared in California?
- What percentage of households receive SNAP in Detroit?
- What is the median rent in Miami?
- What is the homeownership rate in Houston?
- Median household income in California
- Poverty rate in Chicago
- Diabetes prevalence in Chicago
- Obesity rate in Miami
- Unemployment rate in Detroit
- q: Which city has the highest diabetes rate?
tag: ranking · server-ordered
- q: Which cities have a diabetes rate above 20%?
tag: filtered-subset · threshold
- q: Across California counties, is median household income correlated with diabetes rates?
tag: correlation · materialized
- label: 🏦 Government & Money
dirs:
- treasury
- usaspending
queries:
- What is the US national debt?
- Euro to dollar exchange rate
- Japanese yen to dollar exchange rate
- label: 💰 Grants & Funding
dirs:
- grants-gov
- usaspending
- nih-reporter
- nsf-awards
queries:
- What grants can a nonprofit apply for in education?
- How much federal funding has Feeding America received?
- How much federal funding has Habitat for Humanity received?
- What grants are available for medical research?
- q: Which organization receives the most federal funding?
tag: ranking · server-ordered
- q: How much NIH research funding does Johns Hopkins receive?
tag: entity-list · fully paged
- label: 🧭 Query shapes
dirs:
- cdc-places
- census
- usaspending
- nih-reporter
- nonprofit-990
- nsf-awards
- sec-edgar
queries:
- q: What was the American Red Cross total revenue?
tag: point
- q: Is the Sierra Club a 501(c)(3)?
tag: status
- q: NSF research awards for Caltech
tag: entity-list
- q: Compare the total revenue of the American Red Cross and Feeding America
tag: comparison · 2 lookups
- q: Which city has the highest diabetes rate?
tag: ranking · server-ordered
- q: Which organization receives the most federal funding?
tag: ranking · server-ordered
- q: Which cities have a diabetes rate above 20%?
tag: filtered-subset · threshold
- q: Give me some universities that get more than a billion dollars from NIH
tag: filtered-subset · propose-and-verify
- q: What share of the American Red Cross revenue comes from federal funding?
tag: ratio · cross-source join
- q: Across California counties, is median household income correlated with diabetes rates?
tag: correlation · materialized
- q: Which university gets the most NIH funding?
tag: refused · no source can rank
- q: Which nonprofit has the highest revenue?
tag: ranking · BigQuery SQL
- q: Which nonprofits have revenue over 10 billion dollars?
tag: filtered-subset · BigQuery
- q: What were Apple's earnings?
tag: ambiguous · answered per interpretation