Tax research memos with authority citations for tax research assistants

Public primary law is easy to get: open corpora such as Pile of Law already gather statutes, regulations, agency publications and court opinions [3]. What those corpora lack is the step a tax research assistant must learn: choosing which authorities control a specific fact pattern, weighing contrary authority, and committing to a conclusion at a defensible opinion level. That work lives in firm memo files, and it is what a buyer should pay for. For the broader category of legal memos, see the owner page on licensing legal briefs and memos; this guide covers the tax-specific overlay.

What a usable tax memo record contains

A usable record splits the memo into facts, issues, authorities, analysis, conclusion and confidence level, with every citation normalized and dated. Most firms write memos in Word against an internal template, so the raw files usually share headings even when formatting drifts. The structure buyers should expect is fairly stable across practices (hypothesis): a facts section, a numbered issue list, a conclusion up front, and a discussion that walks through the Code, regulations, administrative guidance and case law.

Memo families differ in what they teach a model:

• Tax position or technical memos for a client transaction, often concluding at a stated opinion level.
• ASC 740 uncertain-tax-position documentation, the financial-reporting cousin, which records the recognition and measurement judgment behind a reserve (hypothesis: these sit in provision workpapers rather than the tax research file and may need a separate request).
• Research files for notices and examinations, which overlap with tax notice and response correspondence data.
• Internal knowledge memos written for the firm's own practitioners, usually with fewer client facts and fewer consent problems.

Illustrative example: invented to show structure; it does not describe an available dataset.

{
  "memo_id": "m-000417",
  "memo_family": "tax_position_memo",
  "law_as_of": "2025-11-30",
  "jurisdiction": "US-federal",
  "facts_redacted": "Taxpayer [ORG_1], a calendar-year C corporation, incurred costs to ...",
  "issues": ["Whether the costs are deductible under IRC 162 or must be capitalized under IRC 263(a)"],
  "authorities": [
    {"cite": "IRC 263(a)", "type": "statute", "role": "controlling"},
    {"cite": "Treas. Reg. 1.263(a)-4", "type": "final_regulation", "role": "controlling"},
    {"cite": "Rev. Rul. [redacted number]", "type": "revenue_ruling", "role": "supporting"},
    {"cite": "[case citation]", "type": "case", "role": "contrary"}
  ],
  "analysis_text": "...",
  "conclusion": "Costs must be capitalized",
  "opinion_level": "more_likely_than_not",
  "reviewer_signoff": true,
  "third_party_commentary_removed": true,
  "section_7216_basis": "no_return_information | consent_on_file",
  "privilege_review": "cleared_for_release"
}

The role field is the most valuable label in the record. A model that cites only supporting authority will look fluent and still fail review, because the weighing standard requires contrary authority to be considered.

How the authority hierarchy and opinion levels shape labels

Tax memos rank authorities on a recognized hierarchy, and the confidence language maps to penalty and practice standards, so both should be captured as structured labels. The substantial authority regulation, Treas. Reg. 1.6662-4(d)(3), lists the authorities that count, including the Code, regulations, revenue rulings and procedures, treaties, court cases and committee reports. It weighs all relevant authorities objectively, including contrary ones, and gives a revenue ruling more weight than a private letter ruling on the same issue [4].

Opinion levels such as "should," "more likely than not," "substantial authority" and "reasonable basis" connect to those penalty rules and to Circular 230 . Circular 230 Section 10.37 requires written advice to rest on reasonable factual and legal assumptions and reasonable reliance, to consider the relevant facts and law, and not to assume the position will go unaudited. The 2014 amendments that created this standard replaced the old covered-opinion rules [5].

That date matters for dataset design. Memos written before mid-2014 often carry covered-opinion disclaimers and formal fact-and-assumption sections; later memos are shaped by engagement scope. Mixing them without a drafted_under flag teaches the model two house styles at once.

Where tax memo data breaks in training and evaluation

Tax memo data fails most often on stale law, unverified citations and copied commentary, not on volume. Every one of these is detectable before signature if the buyer asks for the right fields.

For evaluation, hold out memos by law date rather than at random, so a test set never asks the model about law that changed after the training cut. Pair the memo set with a citation verifier like the one described in legal citation accuracy evaluation. For RAG training, memos map naturally to question, retrieved authorities, distractors and cited answer, as covered in retrieval-augmented fine-tuning data.

Rights checks specific to tax memos

Tax memos carry three overlays beyond ordinary copyright and confidentiality: Section 7216, practitioner privilege and third-party research content. Each one should be answered in writing per dataset, not assumed from the firm's general terms.

Section 7216. Regulations under 26 CFR 301.7216 restrict a return preparer's use or disclosure of tax return information for purposes other than preparing the return, and knowing or reckless violations are criminal [1]. A research memo that quotes return line items, filing positions or client figures from a prepared return can fall inside that rule. The deeper treatment is in tax return data and Section 7216 consent; for memos, the practical question is whether each fact section contains return information and, if so, what consent basis covers it.

Privilege. IRC 7525 extends a confidentiality privilege to tax advice from federally authorized practitioners, but it is narrower than attorney-client privilege and does not reach every setting (hypothesis: confirm scope with counsel). Licensing a memo can be argued to waive whatever protection it had, so the supplying firm, and often the client, must decide that release is acceptable. The same logic for law-firm work product is discussed in legal LLM fine-tuning data.

Third-party commentary. Practitioners often paste explanations and annotations from subscription tax research services into memos. In September 2026 the Third Circuit held Westlaw headnotes copyrightable and held that copying them to train a non-generative legal-research tool was not fair use [2]. As of October 2026, how far that reasoning reaches tax memo data is untested (hypothesis), but the safe design is to strip quoted editorial content or license it separately.

De-identification. Client names, EINs, SSNs, entity names and deal values need to be removed or replaced, and the method recorded. The SourceX guide on de-identifying legal briefs and memos covers the general approach; tax memos add identifiers such as EINs, return form references and distinctive transaction amounts that can re-identify a client.

Buyer request template for tax memo data

A good request describes memo families, fields, law dates and rights posture, not the firms you hope will supply them. Use this as a starting brief.

Illustrative example: invented to show structure; it does not describe an available dataset.

• Use case: SFT for a federal income tax research assistant plus a held-out citation-accuracy eval set.
• Memo families: tax position memos and internal knowledge memos; ASC 740 documentation as an optional second tranche.
• Scope: US federal income tax; state memos only if tagged by jurisdiction.
• Required fields: facts (redacted), issues, authorities with type and role, analysis, conclusion, opinion level, memo date, law-through date, reviewer signoff.
• Exclusions: drafts without signoff, memos quoting subscription-service commentary unless removed, any record with unconsented return information.
• Rights questions: Section 7216 basis, privilege review outcome, client consent where engagement letters require it, ownership of memos written by contractors.
• Delivery format: JSONL with one memo per line, plus the original DOCX or PDF if the license allows.

Describe the data rather than naming the firms you hope will supply it; accounting-specific needs are summarized on buyers in accounting.

How SourceX approaches tax memo requests

SourceX sources operational datasets, including documents and finance and legal workflow records, from US companies on request; it does not hold tax memos in stock, and a request does not guarantee a match. Buyers describe the data, SourceX looks for US businesses that hold it, and every release is approved by the supplying company. Each dataset is rights-reviewed for ownership and consents, personal details are removed or replaced before delivery with the method recorded and a sample checked, and no method is perfect. To start, describe your tax memo requirements to SourceX. More context on adjacent categories is in the industry-specific operational data hub.

This page is general information, not legal advice. Confirm requirements with counsel for your jurisdiction and use case.

Sourcing tax research memo datasets through SourceX

SourceX finds US businesses that hold the memo data you describe, assesses the data and licensing permissions, and agrees pricing and allowed uses in a license before anything is delivered. Nothing is contracted until a supplier agrees, and delivery runs through private, access-controlled workflows after an executed agreement. Submit your tax research memo dataset request.