Verification Codebook

Verifier sheet — pilot round · instrument shakedown, n = 4

This is a pilot, not the agreement study

Read this line before anything else.

What it is NOT: This is not the main agreement round. No headline agreement statistic is claimed from n=4.

Codebook may change after: Refinement is EXPECTED after this round. That is the point of a pilot.

Main round: Runs later on the full cohort, against a codebook frozen after this pilot’s findings are folded in.

Purpose. Shake down the instrument: does the codebook produce consistent answers, are the columns the right columns, is anything ambiguous.

Scope — why four rows

Population: 4. This is the COMPLETE set of distinct real EDGAR filings in the repository, not a sample of a larger set.
Not an under-pull: 5 XML files exist; 0001594162.xml at the repo root is byte-identical (SHA-256 f8dcf202...) to the td6_real_sources copy. 5 files, 4 distinct filings.
Why so few: The pipeline is designed for a 500-firm cohort. These 4 are the verification sources committed to the repo. A larger cohort requires a live EDGAR pull, which is a separate action.

How to use it

Click OPEN SEC FILING in column E. It goes to the SEC's own filing index — the primary source.
Compare each extracted value against the filing. Do not compare against memory or another sheet.
Fill only the ANSWER_ columns (amber headers). Leave every other column untouched — they are the extraction under test.
One row at a time, top to bottom. Do not batch-judge.
Record your actual start and end time. Both go in the receipt.

The extraction under test

These columns are the machine’s output. Leave them untouched — they are what is being judged.

row_id: Stable identifier for this row. Never reuse or renumber.
cik: SEC Central Index Key of the primary issuer, zero-padded to 10.
issuer_name: Entity name exactly as filed. Case and punctuation are part of the value.
accession_no: SEC accession number identifying this specific filing.
filing_link: Clickable link to the SEC filing index. This is the receipt for every value in the row.
form_type: D = new offering notice. D/A = amendment to a prior notice.
is_amendment: true when the filing amends an earlier one.
jurisdiction_of_inc: Where the issuer is incorporated. Not necessarily where it operates.
issuer_state_country: Mailing address state or country, as filed.
entity_type: Legal form: limited partnership, corporation, LLC, other.
industry_group: Issuer's self-selected industry classification.
investment_fund_type: Fund category when the issuer is a pooled investment fund.
is_pooled_fund: true when the issuer files as a pooled investment fund.
total_offering_amount: Total offering size in USD, or the literal 'Indefinite'. 'Indefinite' is a VALID value, not a gap.
total_amount_sold: USD sold to date at filing time.
total_remaining: Offering amount still available. Should equal offering minus sold when both are numeric.
min_investment_accepted: Smallest accepted investment in USD. 0 means no stated minimum.
has_non_accredited: true when any non-accredited investor participated.
investors_already_invested: Count of investors at filing time.
signature_date: Date the filing was signed, as filed.
signature_name: Name on the signature block.
source_file: Repository path of the parsed XML, so any value can be traced back to bytes.

What the verifier fills in

ANSWER_issuer_name_correct: YES / NO / LOW. Does the extracted name match the filing exactly?
ANSWER_offering_amount_correct: YES / NO / LOW.
ANSWER_amount_sold_correct: YES / NO / LOW.
ANSWER_fund_type_correct: YES / NO / LOW.
ANSWER_confidence: HIGH or LOW for the row overall. See the LOW protocol below.
ANSWER_receipt: Where you looked. Section or item of the filing. Required for every NO and every LOW.
ANSWER_notes: Anything the columns above cannot hold. Ambiguity in the codebook belongs here — it is a pilot finding.

Low-confidence protocol

When to use LOW: Whenever you cannot settle the answer from the filing alone. LOW is a real answer, not a failure to answer.
Never guess: A guessed YES is worse than a LOW. YES is used downstream as verified; LOW routes the row to adjudication.
LOW needs a receipt: Every LOW must say in ANSWER_receipt what you checked and what was missing or ambiguous.
Ambiguity is a finding: If the filing or the codebook is ambiguous, that is a pilot finding. Mark LOW and say so in notes.
Do not resolve upward: Two LOWs on one row do not make a NO. Leave them LOW.

Receipt format

Shape: <where you looked> -> <what it said>
Example (agreement): Item 13 Offering Amount -> $202,490,000, matches
Example (disagreement): Item 13 -> 'Indefinite'; sheet shows a number. Filing wins.
Example (LOW): Item 6 -> two fund types checked; sheet carries one. Cannot tell which is primary.
Rule: A receipt names a location in the filing. 'Looks right' and 'checked' are not receipts.

Download the frozen sheet (plain text)

This page is generated from the frozen source and carries no wording of its own in the sections above — every definition, rule and caveat is the source’s. The download is that source, unmodified, so the page and the file cannot drift apart without it being visible. Print this page to get a PDF.