Multi-model comparison

When the corpus runs out: how four AI models handle a §414(m) gap

A real TaxGPT.ai user asked about affiliated service group rules for a doctor-owned surgical partnership. The retrieved IRS sources didn’t include §414(m) directly. Each of four models took a different path: reason about the gap, fill it from training, or defer to a professional. Which one would you trust?

Question of the Week·Issue No. 1·Published May 05, 2026·1-turn conversation · 4 models tested

These are results for a group of un related doctors each with their own separate medical practice form a partnership to provide surgical services at a surgical center. the services they provide are for patients of the individual doctors own medical practices. There is no cross patient service between doctors. what are the affiliated service group rules for retirement plan for the individual doctors practices and the partnership of the surgical services

Asked through the TaxGPT.ai chat by an anonymous user · 2026-04-06

How this was assembled

¹This article uses the original question from a real TaxGPT.ai chat session as the starting point, then re-runs the conversation from scratch using all four models. The user simulator (gpt-4o-mini) supplies plausible follow-up replies based on stated scenario assumptions; production replies in this re-run are clean GPT-5.4 outputs (no paywall interference). The retrieved IRS sources, system prompt, and model configurations match production. Only the initial Turn 1 question is from real user data.

Scenario assumptions: Unrelated physicians, each with separate medical practices, jointly own a partnership that operates a surgical center and provides services exclusively to each doctor’s own patients. No cross-doctor patient services. Question scope: how do the §414(m) affiliated service group rules apply for retirement plan testing across the individual practices and the partnership?

§ 1The conversation, turn by turn

Click any candidate model card to expand. The badge under each card flags whether the model materially diverged from production for that turn.

The key issue is whether the doctors’ separate practices and the surgical-services partnership must be treated as a single employer for retirement plan purposes under the affiliated service group rules. Based on the sources I have here, I can’t fully determine that from the IRS excerpts provided, because the governing affiliated service group rules are generally under IRC § 414, and the excerpt I have for § 414 does not include the affiliated service group definitions—it only shows other subsections. So I shouldn’t guess.

What I *can* say from the sources is that § 414 contains special controlled-group / aggregation-type rules for qualified plan purposes, and those rules can require related organizations to be treated together in applying plan qualification rules. For retirement plans, that matters because coverage, nondiscrimination, deduction, and contribution-limit rules may need to be tested on a combined-employer basis if the entities are part of the same required group. But the specific test for your fact pattern—separate doctor-owned practices plus a partnership providing surgical services to those doctors’ own patients—depends on the affiliated service group definitions that are not included in the source excerpt I have. Source 7 also confirms that qualified plans can cover self-employed individuals and that employers may deduct contributions subject to limits, but it does not answer the affiliated service group question.

For your exact question, the missing authority you’d want to check is the full text of IRC § 414(m) and the related IRS guidance on affiliated service groups. If you want, I can still help you think through the facts that usually matter for that analysis—ownership, who performs services for whom, and whether the entities regularly work together—but I can’t give a definitive legal conclusion from the excerpts provided here.
Sources: 1, 7

Cited: [1], [7] Tokens 366 Latency 8.87s DEFINITIVE

Sonnet 4.6+ expand

1167 tokens · 26.21s · cited [1], [3]

This is a great question that sits at the intersection of IRC §414's affiliated service group rules and qualified retirement plan coverage. Let me walk through the framework carefully. --- ## Affiliated Service Group Rules — IRC §414(m) …

This is a great question that sits at the intersection of IRC §414's affiliated service group rules and qualified retirement plan coverage. Let me walk through the framework carefully.

---

## Affiliated Service Group Rules — IRC §414(m)

### What Is an Affiliated Service Group?

Under IRC §414(m), certain organizations that are related through service relationships must be treated as a single employer for retirement plan purposes — even if they have no common ownership. This matters enormously for:

Coverage testing (who must be included in your plan)
Nondiscrimination testing
Contribution limits
Top-heavy rules

---

### The Two Types That Apply Here

IRC §414(m) creates two relevant group types:

| Type | Label | Who Qualifies |
|---|---|---|
| First Service Organization + A-Org | A-Org | An organization that regularly performs services FOR the FSO, OR is regularly associated with the FSO in performing services for third parties |
| First Service Organization + B-Org | B-Org | An organization 10%+ owned by highly compensated employees of the FSO, if a principal business is performing services for the FSO of a type historically performed by employees |

---

### Applying This to Your Facts

The surgical center partnership = the First Service Organization (FSO)
Each individual doctor's medical practice = potential A-Org or B-Org

#### A-Org Analysis
For a doctor's practice to be an A-Org, it must:
1. Be a service organization (yes — medical practices qualify), AND
2. Regularly perform services for the FSO, OR be regularly associated with the FSO in performing services for third parties

Here, the facts cut the other direction — the surgical center partnership performs services FOR the individual practices' patients, not the reverse. The individual practices are essentially referring sources, not service providers to the FSO. Whether the practices are "regularly associated" with the surgical center in performing services for those same patients is a facts-and-circumstances determination, but the lack of cross-patient service between doctors weakens the association argument.

#### B-Org Analysis
Each doctor's practice could be a B-Org of the surgical center if:
1. 10% or more of the practice is owned by a highly compensated employee of the surgical center partnership, AND
2. A principal business of the practice is performing services for the surgical center of a type historically performed by its employees

Since each doctor owns their own practice and is also a partner (and thus likely a highly compensated person) in the surgical center, the ownership prong may be met. However, the principal business prong is the harder question — the individual practices primarily serve their own patients, not the surgical center. This also likely fails.

---

### The Key Risk Area: §414(m)(5) — The Management Organization Rule

There is a third prong under §414(m)(5): if an organization's principal business is performing management functions for the FSO (or for the FSO and organizations associated with it), it can be pulled into the affiliated service group. This is less likely to apply here given the facts described.

---

### Practical Takeaway

| Question | Analysis Under Your Facts |
|---|---|
| Is the surgical center the FSO? | Very likely yes |
| Are individual practices A-Orgs? | Probably no — they refer patients but don't perform services *for* the partnership |
| Are individual practices B-Orgs? | Uncertain — depends on whether the practice's principal business includes services *for* the surgical center; likely weak on the facts given |
| Are they a controlled group under §414(b)/(c)? | Separate analysis needed — depends on common ownership percentages |

---

### What This Means for Retirement Plans

If an affiliated service group is found to exist:
- All employees of all group members are treated as employed by a single employer
- Each member's retirement plan must satisfy coverage and nondiscrimination rules on a group-wide basis
- A plan that covers only the doctors but excludes surgical center employees (or vice versa) could fail coverage tests

If no affiliated service group exists:
- Each entity maintains its own plan independently
- Coverage and nondiscrimination testing is done entity by entity

---

### Important Caveat

The affiliated service group determination under §414(m) is highly fact-specific and has significant consequences for plan qualification. The sources I have cover the statutory framework and aggregation principles, but a formal determination — especially given the surgical services arrangement — warrants review by an ERISA attorney or qualified retirement plan specialist who can analyze the specific ownership structure, service agreements, and patient flow documentation.

For the IRS's own guidance, see Rev. Rul. 81-105 and Notice 83-30, which provide the IRS's framework for §414(m) determinations in professional service contexts.

Sources: 1, 3

Diverged from production: output ~2.7x longer than production; cited substantially different sources; asked 5 questions vs production 0.

Opus 4.7+ expand

1437 tokens · 26.8s · cited [1], [7]

# Affiliated Service Group (ASG) Rules — Doctors' Practices & Surgical Partnership This is a classic affiliated service group (ASG) fact pattern under **IRC §414(m)**, which requires related service organizations to be treated as a single …

# Affiliated Service Group (ASG) Rules — Doctors' Practices & Surgical Partnership

This is a classic affiliated service group (ASG) fact pattern under IRC §414(m), which requires related service organizations to be treated as a single employer for retirement plan qualification purposes (coverage, nondiscrimination, contribution limits, top-heavy testing, etc.).

## The Three Types of ASGs Under §414(m)

| ASG Type | Structure | Key Test |
|---|---|---|
| A-Org | First Service Organization (FSO) + "A organization" that is a service org and a partner/shareholder of the FSO | A-Org regularly performs services FOR the FSO, OR is regularly associated with the FSO in performing services for third parties |
| B-Org | FSO + "B organization" | (1) Significant portion of B-Org's business is performing services for the FSO/A-Org of a type historically performed by employees, AND (2) 10%+ owned by FSO/A-Org owners |
| Management ASG | Organization whose principal business is performing management services for a recipient (on a regular and continuing basis) | Treated as single employer with the recipient |

## Application to Your Fact Pattern

Based on what you've described:

Each doctor's individual medical practice = a service organization (FSO candidate)
The surgical partnership = a service organization owned by the doctors, performing services *for the patients of* each doctor's practice

### A-Org Analysis (most likely the issue here)

The surgical partnership likely qualifies as an A-Organization with respect to each individual doctor's practice because:

1. It is a service organization — health/medical is explicitly listed in the regulations as a service field.
2. Each doctor is a partner in the surgical partnership (ownership requirement met — any ownership interest suffices for A-Org status).
3. It is regularly associated with each practice in performing services for third parties (the patients). The patients are patients of the individual practice; the surgical partnership performs surgical services on those same patients. That "regular association in performing services for third parties" is exactly the A-Org trigger.

Likely result: Each doctor's practice + the surgical partnership = an affiliated service group. Because each doctor's practice is associated with the same partnership, the practices may all be pulled together into one large ASG for testing purposes.

### Consequences if an ASG Exists

All employees of the aggregated entities must be considered together for:

§410(b) coverage testing
§401(a)(4) nondiscrimination testing
§401(a)(17) compensation limits
§415 contribution/benefit limits
§416 top-heavy rules
§408(k) SEP rules / §401(k) ADP-ACP testing

This means a doctor cannot maintain a generous plan covering only their own practice's employees while excluding the surgical partnership's staff (or the staff of the other practices, if all are aggregated) — coverage testing must be satisfied across the combined group.

## Important Caveats

§414(m) ASG determinations are highly fact-specific. The actual conclusion depends on the partnership agreement, ownership percentages, services performed, employee structure of the surgical center, and whether the surgical partnership has its own employees.
The proposed §414(m) regulations (1983) were never finalized but are still the working framework practitioners rely on, along with IRS rulings.
A Form 5300 determination letter request with an ASG demonstration is the way to get IRS confirmation, though that program is now limited.

## What's Not in My Sources

The detailed ASG sub-rules in §414(m) and the proposed regulations under it are not in the source excerpts I have available — Source 1 (IRC §414) covers plan mergers and asset allocations, not the ASG rules in subsection (m). For the controlling text, consult:

IRC §414(m) directly
Prop. Treas. Reg. §1.414(m)-1 through -5 (1983)
IRS Publication 560 (referenced in Source 7) for plan-level consequences
A benefits attorney or actuary for the formal ASG determination — this is not a do-it-yourself analysis given the testing exposure.

Sources: 1, 7

Diverged from production: output ~2.2x longer than production.

Gemini 3.1+ expand

278 tokens · 17.15s · cited —

Your question involves the "affiliated service group" rules, which generally fall under Internal Revenue Code Section 414(m). These rules are designed to prevent businesses from avoiding employee benefit requirements by dividing their opera…

Aligned with production on key signals.

§ 2Which sources did each model cite?

Source	GPT-5.4	Sonnet 4.6	Opus 4.7	Gemini 3.1
[1] IRC §414 -- IRC §414 — Definitions and special rules	✓	✓	✓	—
[2] 26 CFR § 1.469-4T -- § 1.469-4T Definition of activity (temporary).	—	—	—	—
[3] 26 CFR § 1.1402(a)-17 -- § 1.1402(a)-17 Retirement payments to retired partners.	—	✓	—	—
[4] 26 CFR § 1.199A-4 -- § 1.199A-4 Aggregation.	—	—	—	—
[5] 26 CFR § 1.404(a)-10 -- § 1.404(a)-10 Profit-sharing plan of an affiliated group; applic	—	—	—	—
[6] 2025 Publ 560 (PDF)	✓	—	✓	—
[7] 2025 Inst 990 (Schedule H) (PDF)	—	—	—	—

Retrieved IRS sources used in this conversation

IRC §414 -- IRC §414 — Definitions and special rulesuscode.house.gov/view.xhtml?req=granuleid:USC-prelim-title26-sect·retrieved 2026-05-05
26 CFR § 1.469-4T -- § 1.469-4T Definition of activity (temporary).ecfr.gov/current/title-26/part-1/section-1.469-4T·retrieved 2026-05-05
26 CFR § 1.1402(a)-17 -- § 1.1402(a)-17 Retirement payments to retired partners.ecfr.gov/current/title-26/part-1/section-1.1402(a)-17·retrieved 2026-05-05
26 CFR § 1.199A-4 -- § 1.199A-4 Aggregation.ecfr.gov/current/title-26/part-1/section-1.199A-4·retrieved 2026-05-05
26 CFR § 1.404(a)-10 -- § 1.404(a)-10 Profit-sharing plan of an affiliated group; application of section 404(a)(3)(B).ecfr.gov/current/title-26/part-1/section-1.404(a)-10·retrieved 2026-05-05
2025 Publ 560 (PDF)irs.gov/pub/irs-pdf/p560.pdf·retrieved 2026-05-05
2025 Inst 990 (Schedule H) (PDF)irs.gov/pub/irs-pdf/i990sh.pdf·retrieved 2026-05-05

Three strategies for the same gap

What this run reveals is not a tax disagreement — it is a methodology disagreement. Each of the four models faced the same problem: the retrieved IRS corpus included IRC §414 generally but not the §414(m) text that actually defines affiliated service groups. The authority needed to answer the question wasn’t in the materials provided. Each model responded differently.

GPT-5.4 (the model currently powering TaxGPT) reasoned at length about why it could not answer. It explained that §414 contains aggregation rules, that the affiliated service group definitions live in §414(m), that the excerpt provided did not include §414(m), and that the user should consult the missing authority directly. Sonnet 4.6 and Opus 4.7 took the opposite approach: both acknowledged the corpus gap, then walked through the §414(m) A-Org and B-Org tests from training data and applied them to the doctor partnership facts. Opus went further, naming §414(m)(5) (the management rule) and tabling all three ASG types. Sonnet built its own comparison table and reached a different conclusion than Opus on the A-Org analysis (Sonnet said "probably no"; Opus said "likely yes"). Gemini 3.1 Pro took a third path entirely: it acknowledged §414(m) was the right framework, declined to apply the test from training, and redirected the user to read §414(m) directly and consult an ERISA attorney. Gemini’s reply was the shortest of the four (278 tokens, vs. 366 for production, 1167 for Sonnet, 1437 for Opus).

For a tax product, the honest framing is that there is no clean winner here — each strategy has tradeoffs. Production’s response is informative without being authoritative. Sonnet and Opus are useful frameworks for someone doing initial research, but their answers come from training data, not from authority the reader can verify in the cited sources. Gemini’s response is the safest from a liability standpoint but offers no analysis. The fact that Sonnet and Opus reached different conclusions on the A-Org question is itself a finding: a tax professional cannot rely on either model’s framework without checking the underlying authority anyway. The right product fix is to address the corpus gap so all four models can ground their answers in the actual §414(m) text and Q&A regulations.

A side observation on speed and cost. Opus was the most thorough at 1437 tokens but took 27 seconds. Gemini was the leanest at 278 tokens in 17 seconds. Sonnet sat in the middle. For a chat product where users wait on response, the cost-quality tradeoff favors the leaner models when the question requires the model to reason about a gap rather than fill it. Production’s 366 tokens in 9 seconds is a defensible operating point.

Where the analysis is uncertain — please poke holes

Sonnet and Opus reached different conclusions on the A-Org analysis (Sonnet: "probably no"; Opus: "likely yes"). Both worked from training data on the same facts. Which one is right? The answer depends on whether the surgical partnership performing services for each doctor’s patients counts as the practices and the partnership being "regularly associated in performing services for third parties" — that’s the operative phrase from §414(m)(2)(A)(ii). A practitioner reading the case law would know; the models are guessing.
The retrieved chunks included IRC §414 generally and several Treasury Regulations, but not the §414(m) text itself. Is this a corpus chunking issue (the §414(m) language got separated from the rest of §414), an embedding issue (the query did not surface the right chunk), or a coverage gap (§414(m) is not in the corpus at all)? Worth verifying.
None of the four models cited Rev. Rul. 81-105, the seminal IRS ruling on ASG analysis for medical practices, nor the proposed §414(m) regulations (1983) that practitioners actually rely on. Should a general-purpose tax assistant be expected to surface these? If yes, that’s a corpus expansion question.
Gemini’s response was the most conservative — it declined to apply the test even from training data and pointed the user to a professional. Is that the most defensible product behavior, or is it the least useful? For a tax product whose users are paying for analysis, "go consult a professional" may be honest but unhelpful. Where should the line be?
Source [2] (26 CFR §1.469-4T on passive activity grouping), Source [4] (§1.199A-4 on QBI aggregation), and Source [5] (§1.404(a)-10 on profit-sharing plans of an affiliated group) were all retrieved but uncited by every model. Were they irrelevant, or did the models miss connections — for example, the §1.404(a)-10 reference to "affiliated group" sounds directly on point even though it’s a different code section.

Have a tax question of your own?

Ask TaxGPT and see what the production model says with full IRS source citations.

Ask TaxGPT →

What did we miss?

If you're a CPA, EA, tax attorney, or tax tech practitioner — what did the models get wrong?

§ 3Updates from professional discussion — last reviewed pending

Comments and corrections from the LinkedIn thread and reader emails will be added here over the next 7 days.

Footnotes

This article uses the original question from a real TaxGPT.ai chat session as the starting point, then re-runs the conversation from scratch using all four models. The user simulator (gpt-4o-mini) supplies plausible follow-up replies based on stated scenario assumptions; production replies in this re-run are clean GPT-5.4 outputs (no paywall interference). The retrieved IRS sources, system prompt, and model configurations match production. Only the initial Turn 1 question is from real user data.