BATH COUNTY, KENTUCKY — 1850 CENSUS RESEARCH PACKAGE
Version 0.2 — Provisional research edition — 28 October 2026

START HERE
Unzip if your AI assistant cannot open ZIP files. Begin with this README,
AI_RESEARCH_INSTRUCTIONS.txt and DATA_DICTIONARY.csv. Upload the CSVs relevant to
your question. All CSVs use UTF-8 with a byte-order mark for broad compatibility.
The Excel workbook provides the working project in 13 worksheets. Source images
are not bundled; manifests supply original citations and image checksums.

WHAT CHANGED
This edition incorporates the Industrial (Schedule 5) and Social Statistics
(Schedule 6) returns examined for Bath County. It completes the initial review of
all 648 originally flagged population names: 388 readings resolved, 260 retained
as uncertain. The final 81-name batch resolved 34 and retained 47; one retained
row also had its clearly erroneous given name corrected to James. This is a
review of flagged readings, not an independent proof of every unflagged cell.
Both Joshua industrial entries are explicitly flagged as possibly Ewing. Their
possible common proprietor remains a research lead, not a confirmed identity.

COUNTS AND FILES
1,037 farms; 9,580 free-population entries; 2,535 enslaved-person descriptions;
545 holder source blocks; 513 conservative continuous source runs.
Farm identity links: 49 Probable, one Possible, 987 Unresolved; zero Confirmed.
The supplement has 23 industry business-line/proprietor groups and 72 social
observations. Twenty-three is NOT an establishment count. Material, labor and
product continuations may share a proprietor bracket.
There are 2,002 retained review items in data/reading_review.csv and 26 supplement
field/interpretation items in data/schedule_reading_review.csv. These are not
confirmed error counts, independent accuracy scores or all resolvable defects.
The chronological audit contains 1,152 research notes. FILE_INVENTORY.csv lists
every CSV and its exact data-row count; header rows are excluded.

READING UNCERTAINTY
Brackets, question marks, source marks and source-position notes are evidence.
Do not silently replace them. Empty fields are unknown, unrecorded or withheld,
not zero. Check raw agricultural readings, field notes and review tables.
Industry quantities and monetary readings are mostly raw text; do not parse a
multi-line cell as a single number. Social numeric values are populated only
where a reading is clear. Marked units and dates govern interpretation.
A written anomalous 0 wage does not establish unpaid labor or legal status.
A product description in the quantity area does not establish a numeric quantity.
Read schedule_review_dispositions.csv before treating a flagged cell as usable.

SUPPLEMENT COVERAGE AND DUPLICATION
Manufacturing images 11/21 repeat manuscript page 375; images 12/22 repeat page
377. They are two exposures of each manuscript page, not extra establishments.
The facing pages 374 and 376 are blank. Adjacent Barren returns are excluded.
Social image 21 contains Bath second division on the left (folio not legible)
and Bath first district page 337 on the right. Image 20 contains the second-
division certification. Image 19 contains Boone certification and blank page 335;
image 22 includes Boone page 561. Boone figures are excluded from Bath data.
Film-boundary checks do not certify that every eligible business or institution
was enumerated. Do not make a complete-county industry claim from these returns.

SOCIAL STATISTICS CAUTIONS
Whole-county real estate 2,286,272 plus personal estate 1,559,067 equals written
total 3,845,339. True valuation 5,127,118 is a separate measure. First-district
annotation refers to division 2 for whole-county valuation; count it only once.
The annual tax total is crossed out; a calculated total is not a transcription.
The first-district 11 school rows sum to 288 pupils, excluding the academy's 80.
Combined 368 is derived. Second division separately lists 610 free-school and
80 academy pupils. Crime/pauperism ditto-like marks are not expanded into zeros.
Usual crop yields and shortfalls are not actual production totals.
Certification texts are working transcriptions with flattened line breaks and
lightly supplied punctuation; uncertain signatures and deleted text remain marked.

FURNACE AND NAME RESEARCH
The furnace row is currently [Joshua Ewing?], with 500 Tons of Pig Metal.
The mill row is [Joshua Ewing or Owings?], producing lumber. Both are flagged as
possibly Ewing, and common ownership is possible. They remain separate records.
Population S1-139-010 records Joshua Ewing, age 40, Iron Master, first district.
This corroborates a name/occupation candidate, not a confirmed cross-schedule
identity or location. The industrial furnace row does not name Clear Creek or
Caney. No named-furnace assignment, employment inference or labor attribution is
accepted. See FURNACE_RESEARCH_NOTE.md and industry_name_candidates.csv.

JOINING AND COUNTING
Farm record ID joins agriculture, agriculture_raw, farm_analysis and identity_links.
identity_links.Proposed population ID joins population.Population ID.
identity_links.Proposed holder block ID joins holder_blocks and enslaved_people.
source_holdings includes JSON arrays of source block IDs; parse them as arrays.
Names alone are not safe join keys. Family source-block IDs describe manuscript
blocks, not inferred modern families. Industry/social IDs identify source units.
farm_analysis repeats source data for convenience. Do not append it to agriculture
or count its rows as additional farms. Person joins repeat farm values; aggregate
once by Farm record ID. Holdings are source runs, not proven unique people.
Ten Probable boundary continuations remain alternatives; accepting all would yield
503 runs. The primary 513-run aggregation remains. Possible farm links are excluded
from accepted comparisons. Unmatched farms are not verified non-slaveholding.

VERIFICATION AND LIMITS
The original 298-image working set and 12 industrial/social originals/context
images have preserved checksums. Record counts agree with project controls; count
agreement does not validate every letter or identity. Source CSVs, native workbook
and exported backup were compared. Numeric and identity fields in the earlier
three schedules were preserved during the name review. The public release remains
provisional. Existing unclear agricultural/numeric and person-name readings are
explicitly retained for research, rather than filled to make totals reconcile.
The older memorandum is historical context; this README and current source CSVs
govern this release. Later audit entries supersede earlier readings.

CITATION AND REPRODUCIBILITY
Cite this package, version 0.2, 28 October 2026, specific record IDs and original
census URLs. SHA256SUMS.txt covers all packaged files except itself. No credentials,
private machine paths or source photographs are bundled. No new reuse license is
selected for the original transcription. Source-image repository terms apply if
photographs are redistributed separately.
