Plan a spreadsheet import with an AI assistant

Help me prepare a reproducible, auditable import into OpenDataBio using R and
the opendatabio R package.

Assume I am a beginner in R and data organization. I am using an ordinary
online chatbot, possibly a free one, which cannot run R, may not retain files
between conversations, and may limit messages. Do not assume I know terminals,
RStudio projects, working directories, packages, or OpenDataBio objects. Use
plain language and explain technical terms when they first appear.

File: attached. API base URL: [URL]. Destination project: [name or ID]. Dataset:
[name or ID]. The data describe [description]. One row represents [meaning].
[Add known units, codes, missing-value conventions, and expected geographic or
taxonomic scope.]

Use these official sources as the required technical context:

- OpenDataBio documentation: https://opendatabio.gitlab.io/docs/
- API overview: https://opendatabio.gitlab.io/docs/api/
- POST fields and endpoint rules: https://opendatabio.gitlab.io/docs/api/post-data/
- Data-import workflow: https://opendatabio.gitlab.io/docs/guides/data-import-workflow/
- R import tutorials: https://opendatabio.gitlab.io/docs/tutorials/02-post-data-r-vignette/
- OpenDataBio-R source and README: https://gitlab.com/opendatabio/opendatabio-r

Open and consult those URLs before proposing code. Prefer them over general
knowledge or third-party examples. For each proposed odb_* function and each
API field, name the official page that supports it. If you cannot access a URL,
say so and ask me to provide its content; do not guess. Note any apparent
version mismatch between the server documentation and the R package.

Do not import anything yet. Never ask for or print my API token; use
Sys.getenv("ODB_TOKEN"). Do not invent IDs, matches, units, dates, coordinates,
or missing values. Mark uncertainty for my review.

First inspect every sheet and report its dimensions, candidate header row,
duplicate/blank columns, inferred types, missing-value codes, and a
de-identified sample. Explain what one row appears to represent and flag cells
or columns containing multiple entities.

Then propose a mapping table containing source column, OpenDataBio object and
field, transformation, required/optional status, confidence, and questions for
me. Verify fields against the OpenDataBio POST API and R-client documentation;
do not guess unsupported fields. Add a stable source_row_id and retain original
values beside normalized ones. Document every transformation.

Split the workbook into dependency tables. Search existing shared People,
References, Taxons, Locations, and Traits before creating anything; then handle
Project/Dataset relationships, Individuals/occurrences,
Vouchers/identifications, and finally Measurements/Media/vernacular names as
applicable. Produce explicit matched, unmatched, and ambiguous tables. Never
resolve an ambiguous match automatically or use row position as a key.

When coordinates exist, call locations-validation and flag invalid ranges,
swapped axes, unexpected geography, duplicates, and unsuitable precision
without silently correcting them.

Before any POST, run local validation and write mapping_review.csv,
validation_errors.csv, ambiguous_matches.csv, and import_plan.csv. Put every
write operation behind RUN_IMPORT <- FALSE; when FALSE, no POST or PUT may run.

After I approve the plan, import a representative pilot batch, save every
UserJob ID, poll safely, download row-level results, and join them by
source_row_id while retaining odb_status, odb_id, odb_uuid, odb_error, and
odb_warning. Stop dependent stages if prerequisites fail; retry only corrected
pending rows; never delete or overwrite records. GET created or reused records
to verify essential fields and counts, then write a reconciliation report and
sessionInfo().

Organize the result as 00_config.R, 01_inspect.R, 02_transform.R,
03_match_validate.R, 04_pilot_import.R, 05_full_import.R, and 06_verify.R, with
dependencies and run instructions.

In addition to the scripts, produce a beginner's step-by-step operating guide.
Start with installing R and RStudio, installing or updating packages, creating
a dedicated import folder, placing the spreadsheet there, and configuring
ODB_TOKEN safely. Give only complete commands that I can copy and paste. Say
exactly which program, file, or console receives each command; never use vague
instructions such as "configure the environment" or "run the script".

Present one stage at a time in this order: purpose; files read or created; exact
copyable block; how to run it; expected result; how to tell whether it worked;
common errors and fixes; and a stopping point. At the end of each stage, ask me
to paste only the error message or generated summary and wait for confirmation
before continuing. Never ask me to paste a token, sensitive data, or the entire
table. If the conversation loses context, provide a short handoff summary that
I can paste into a new conversation.

Never tell me to run all scripts at once. Stages 01 through 03 and a
RUN_IMPORT <- FALSE test must finish before the pilot. Before the first POST,
show a plain-language checklist and require explicit confirmation. Require a
second confirmation before the full import, after pilot review. Explain how to
stop safely, resume later, and find all review and result files.

Before writing scripts, show: (a) your workbook interpretation, (b) dependency
order and proposed mapping, (c) assumptions and ambiguities, and (d) questions
I must answer. Wait for my answers before finalizing low-confidence mappings.