# GSS (General Social Survey) → MatrAIx 1,290 dimensions
Maps the 75,699 respondents of the raw NORC GSS 1972–2024 cumulative file
(gss7224_r3.dta, release 3) to the MatrAIx schema using deterministic rules only — no LLM or
API. GSS is a fully coded survey, so the observed (rule-based) layer is the whole extraction.
The crosswalk — persona/curation/existing_data/scripts/crosswalks/gss.py — maps 18 source variables
to their exact schema dimensions; every value is a direct survey response (provenance: observed).
Coarse / withheld / ambiguous codes resolve to null, never guessed. Richer than the thinned
submission GSS: exact integer age (65+ split resolved), educ distinguishing Primary…Doctorate, a
7-point polviews, and the fefam attitude item → att_traditional_gender_roles.
Source: NORC GSS, publicly released for research —
Download
Get gss7224_r3.dta (Stata) from the GSS Data Explorer / NORC — the 1972–2024 cross-sectional
cumulative file, release 3.
Run (rule-based, no LLM)
Read the .dta with decoded labels, then apply the crosswalk + §8 through the shared toolkit
(no bespoke engine — reuses crosswalk_engine #235 and postprocess_engine #246):
`python
import pandas as pd, gzip, json
from crosswalk_engine import apply_crosswalk # observed / rule layer (#235)
from postprocess_engine import load_schema, normalize # §8 normalization (#246)
from crosswalks.gss import CROSSWALK # this PR
order, allowed = load_schema("persona/schema/dimensions.json")
df = pd.read_stata("gss7224_r3.dta", convert_categoricals=True)
with gzip.open("gss/extraction_v1/shard_00.jsonl.gz", "wt") as f:
for i, row in df.iterrows():
observed, _, _ = apply_crosswalk(row, CROSSWALK, allowed) # 18 exact dims
fields = normalize([], order, allowed, observed=observed) # §8 → 1,290 fields
f.write(json.dumps({"user_id": f"gss-{i:06d}", "fields": fields,
"observed": observed}) + "\n")
`
(Once #267 lands this is one command: run_pipeline.py --observed-only --dataset crosswalks/gss.py.)
Validate — must be 0 errors
`bash
python persona/human_extraction/scripts/validate_extraction.py \
--input gss/extraction_v1/shard_00.jsonl.gz --schema persona/schema/dimensions.json
`
Statistics (release 3)
- Personas / unique users: 75,699 / 75,699
- Fields per persona: 1,290
- Mean grounded (
direct) dims per persona: ~14.6 - Invalid values / assignments / lengths: 0
- No LLM / API / cost
Output is published to MatrAIx2026/MatrAIx-1290-extractions/gss/ (JSONL, 5 shards + manifest).