The money leg · where the corpora fuse
The money leg — answers no single corpus can give.
Unlike the five corpora, this leg's children are JOIN-PATTERNS: answerable questions produced by fusing the other legs through the conformed `factor` and `geography x time` pivots. A blindspot here is an UNBUILT JOIN — the highest-value gap. Most cross-arm joins are vocabulary-wired but not yet materialized. The cards below are the questions we will be able to answer once each join is built.
21% of joins materialized — 0 built 3 partial 4 unbuilt
How to read this leg
Unlike the source corpora, Integration's branches are not piles of content —
they are join-patterns: questions that become answerable only by
fusing the other legs through the conformed factor and
geography × time pivots. A blindspot here is an
unbuilt join — the highest-value gap in the company. Each card
below is a question we will be able to answer once its join is materialized.
The questions — and what they wait on
“Does what research measures match where people actually die?”
Unbuilt: THE flagship gap. fact_factor_year holds only the research arm (1,682 rows); the FARS factor arm (0 rows) is never materialized though fars_distract/drimpair and the fars_label aliases already exist. Highest value, lowest effort.
“Did handheld-phone bans cut distraction fatalities?”
Unbuilt: Both arms are now real: crash arm 2,550 rows (fact_crash_state_year), law arm 714 rows (fact_law_state_year — the full 51×14 state-law panel). The geo×time join is wired; what remains is materializing the difference-in-differences (treatment timing from the law-change timeline against the fatality series).
“Where is there strong evidence but no law — or a law with thin evidence?”
Unbuilt: No edge connects topic_tag factors to policy_type. Needs a small policy_type->factor_key crosswalk (the crosswalk table is empty).
“Which lab construct underlies each driving factor?”
Unbuilt: Partly built: dim_factor has 86 rows and parent_key is set on 25 of them, but most of the cognitive->driving hierarchy (e.g. cognitive:attention -> driving:distraction) is still unlinked. Pure vocabulary work, no new corpus.
“Which papers ground each standard — and does the standard's factor show up in crashes?”
Unbuilt: The ONE real cross edge (standard_paper_links, 2,280, 114 standards, 873 papers) lives only in research.db, embedding-confidence and one-directional — not promoted into the warehouse star schema, and dim_standard has no factor linkage.
“Distraction: from cognition -> papers -> statute -> crashes -> rate, in one view.”
Unbuilt: Only the Research arm resolves per factor; the Outcome/Law/Standard/Exposure arms per factor are missing.
“Fatalities per million miles driven, by state and year.”
Unbuilt: BUILT for the state-year total: fact_crash.fatalities_per_mvmt now computes for 2,142 state-years (1980-2022) off the real FHWA VMT denominator and validates against NHTSA. Still unbuilt: per-FACTOR rates (needs the FARS factor arm) and per-capita/per-driver rates for pre-1998 years.
Known blindspots — future work
What we do not yet have. These are explicit so the gaps stay visible and become the agenda for future DB and knowledge-integration work.
The substrate beneath the joins
The conformed dimensions, registered sources, and derived warehouse that the join-patterns above are built on.
| Dimension | Total members | By source |
|---|---|---|
| time | 110 | FARS: 50, research: 60 |
| geography | 0 | — |
| factor | 0 | — |
Derived, optional warehouse — never required; denominators and laws are placeholder fixtures, crash counts are real FARS census.
Download warehouse (sql.js .gz) → decode map (JSON) →
Honesty: real FARS crash census; placeholder denominators/laws
(); real sources
(FARS, CRSS, research, vPIC, state_laws, exposure_vmt). Built from signature
f0fca954a48fce75f7ab0e669970036593ef4aa4951eac41a15b5bb1f81ddbd6.
Source registry → Conformance coverage → Dimension explorers →