Appendix — the six Fable-5 evaluations in full

Pass 1 critique

This brief has a plausible core (the human-authorship point) wrapped in citation errors, a strawman framing of the government's case, and a fatal vagueness about what proceeding this actually is. Comments below, grouped as requested.

1. Legal accuracy — citations and characterizations

1. The Feist citation is wrong and misused. The case is Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991) — there is no "Feist v. Sunday, Inc." Worse, Feist is cited for the proposition that "the Copyright Act of 1976 requires that a work be 'authored' by a human being." Feist holds no such thing; it holds that copyright requires originality (independent creation plus a modicum of creativity). The human-authorship requirement comes from Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53 (1884), the Trade-Mark Cases, 100 U.S. 82 (1879), the Copyright Office's Compendium § 313.2, and — most directly on point — Thaler v. Perlmutter, 130 F.4th 1039 (D.C. Cir. 2025), which the brief inexplicably never cites. Getting the case name wrong in the lead citation of Argument I is the kind of error that destroys credibility with clerks.

2. The Federal Register citation is wrong. The Copyright Office's AI registration guidance is at 88 Fed. Reg. 16,190 (Mar. 16, 2023), not "88 FR 16161 (Feb. 11, 2023)." The quoted language is also a paraphrase of Compendium § 313.2 ("produced by a machine or mere mechanical process… without any creative input or intervention from a human author"), not a direct quote of the guidance.

3. The Van Buren citation is wrong. Van Buren v. United States is 593 U.S. 374, 141 S. Ct. 1648 (2021). The brief cites "141 S. Ct. 1173" — and, tellingly, cites Google v. Oracle as 141 S. Ct. 1183, ten pages later in the same reporter volume, which should have flagged the error. The characterization is also off: Van Buren construed "exceeds authorized access," expressly reserved (n.8) whether authorization limits must be technological or can be contractual, and said nothing about "violation of security protocols" — a phrase that appears nowhere in the statute or the opinion.

4. Google v. Oracle is internally inconsistent with the brief's own theory. Argument II invokes fair use "and more fundamentally, the lack of copyright in the source." Pick one. Fair use (17 U.S.C. § 107) is an affirmative defense to infringement of a copyrighted work; Google concerned concededly-assumed-copyrightable declaring code. If the outputs aren't copyrightable, fair use is irrelevant; if you're arguing fair use in the alternative, say so explicitly and develop it — don't gesture at it in one sentence.

5. The categorical claim in Argument I overstates current law. The Copyright Office's position (including its January 2025 Copyright and Artificial Intelligence, Part 2: Copyrightability report) is not that all AI-involved outputs are unprotectable — it's that protection turns on human contribution, assessed case by case (prompts alone insufficient; human selection, arrangement, or modification can qualify). The brief's "outputs are the result of a mathematical process, therefore never expression" syllogism invites the government to respond that at least some distilled corpus content embodies protectable human contribution. The brief should have argued the narrower, winnable point: outputs generated by Moonshot's own automated prompting pipeline involved no human authorship by anyone with a claim.

6. Question Presented 3 is legally incoherent as framed. "Retroactive expansion of copyright" by "the Government" conflates three different things: agency interpretation, judicial construction, and legislation. A court construing "works of authorship" to cover AI outputs would not be "retroactive lawmaking" in any constitutional sense; the Ex Post Facto Clause applies only to criminal statutes (Calder v. Bull), and the brief never invokes Landgraf, due process fair-notice doctrine (BMW v. Gore, FCC v. Fox), or lenity — the doctrines that would actually carry a retroactivity argument. As written, QP3 is rhetoric, not a question.

2. Strongest points

7. The human-authorship core is genuinely solid. After Thaler v. Perlmutter and the Copyright Office's consistent guidance, the proposition that purely machine-generated output lacks copyright protection is about as safe a position as exists in this area. Properly cited, Argument I wins on its own terms.

8. The regime-separation argument (Argument III) is the brief's best structural move. The insight that the government is smuggling contract and trade-secret grievances into copyright's enforcement machinery is correct and important — and could have been devastating if the brief had cited 17 U.S.C. § 301 preemption doctrine and the criminal-copyright element structure (§ 506 requires infringement of a valid copyright) to show the government's theory fails on its own statutory terms. It cites neither.

9. The "ToS breach ≠ federal offense" instinct is directionally right. Post-Van Buren, the Court is demonstrably hostile to converting terms-of-service violations into federal crimes. That thematic argument would resonate.

3. Weakest points — what the government actually wins on

10. The brief attacks a strawman. Nowhere does it identify the statute the government actually proceeded under, and no competent government lawyer would lead with copyright here. The real arsenal: the Economic Espionage Act, 18 U.S.C. §§ 1831–1832 (§ 1831 carries enhanced penalties where theft benefits a foreign instrumentality — squarely aimed at a Chinese lab); the DTSA; wire fraud, § 1343 (obtaining API services by fraudulent accounts); CFAA "without authorization" (not "exceeds authorized access"); and export controls on model weights under the EAR. The brief wins a battle the government isn't fighting and loses the war by silence.

11. The CFAA rebuttal misreads which prong applies. Van Buren narrowed "exceeds authorized access." But "industrial-scale distillation" while "avoiding detection" almost certainly means fake accounts, credential rotation, and evasion of IP bans — and under Facebook v. Power Ventures (9th Cir. 2016), continuing access after revocation and technical blocking is access "without authorization," a prong Van Buren did not touch. The brief's own facts ("covert," "avoided detection") set up this counterargument and then pretend it doesn't exist. The government likely wins this point as briefed.

12. The trade-secret rebuttal (§ A) is too quick. Misappropriation includes acquisition by improper means, and "improper means" can include breach of a confidentiality-adjacent duty or fraudulent circumvention of access restrictions — you don't need to exfiltrate the weights. The government would argue the outputs, elicited at scale through deceptive access, functionally disclose protected aspects of the model (a distillation attack is literally a model-extraction technique — the technical literature calls it that). "The technology never moved from the original host" is not how trade secret law works; disclosure and use, not asportation, are the operative concepts.

13. The student-textbook analogy is weak and the brief ignores its own best authorities. Distillation begins with verbatim, at-scale reproduction of outputs into a training corpus — it is not analogous to a human reading. The stronger, litigated line is intermediate-copying fair use: Sega v. Accolade, Sony v. Connectix, Authors Guild v. Google, and — deliciously for this petitioner — Bartz v. Anthropic (N.D. Cal. 2025), where Anthropic itself won a ruling that training on copyrighted works is transformative. A brief that doesn't hoist Anthropic on its own fair-use petard is leaving its best rhetorical weapon in the holster.

14. DMCA § 1201 goes unaddressed. Even if outputs are uncopyrightable, the Fable system (software, weights as fixed literary work) is copyrightable, and circumventing technological access controls (rate limits, geofences, account verification) to reach a copyrighted work is a § 1201 claim that does not require infringement of the outputs at all.

15. The national-security elephant is ignored. Against a foreign AI lab, the government will frame this as export-control and economic-espionage enforcement in a national-security posture, where the Court is historically deferential. Dismissing this as "a policy argument, not a legal one" (§ C) misjudges the audience; the answer is to show the statutes don't reach the conduct, not to lecture the Court about judicial activism — a phrase petitioner's counsel should never use in a merits brief.

4. What's missing

16. The procedural posture, entirely. No Opinions Below, no Jurisdiction statement, no statutory provisions involved, no explanation of what judgment is being reviewed. What did the district court and court of appeals hold? Is this a criminal prosecution, a civil enforcement action, or (as the Statement hints) a preemptive declaratory-judgment action against the United States — which would raise sovereign immunity, ripeness, and standing problems the brief never mentions? A declaratory judgment that the government "lacks statutory authority to treat distillation as IP theft," untethered to any identified statute, is a request for an advisory opinion.

17. The standing/enforcement mismatch. The copyright (if any) in Fable's outputs would belong to Anthropic, a private party who is not before the Court. The United States cannot bring a civil infringement claim on Anthropic's behalf, and criminal copyright infringement (§ 506/18 U.S.C. § 2319) collapses if the works are invalid. One paragraph making this point would have done more work than all of Argument III.

18. Key authorities: Thaler v. Perlmutter (the on-point circuit decision), Burrow-Giles, Compendium § 313.2, the Copyright Office's 2025 Part 2 report, 17 U.S.C. § 102(a)'s text, § 301 preemption, Van Buren footnote 8, Power Ventures, the EEA, and the intermediate-copying fair use line. Also missing: any record facts. The brief argues distillation "by definition" doesn't touch weights — but whether Moonshot in fact only hit the public API is a record question, and the brief's hedged phrasing ("If Moonshot AI did not access the source code…") concedes it doesn't know.

19. An alternative-holdings structure. The brief is all-or-nothing on non-copyrightability. It needs fallbacks: even if some outputs are protectable, training is transformative fair use (Bartz); even if access breached ToS, contract breach isn't "improper means" or "without authorization"; even if the EEA applies, outputs elicited from a public-facing API aren't acquired by improper means. Merits briefs win on layered positions.

5. Bottom line

20. On QP1 — plausibly yes, but it's the wrong victory. The Court would likely affirm that purely machine-generated outputs lack copyright protection (consistent with Thaler and the Copyright Office), though it might not even grant cert on the question absent a circuit split, and it would likely write narrowly, preserving protection for human-contributed elements. On QP2 — formally yes, practically hollow. If the outputs aren't copyrighted, their use isn't copyright infringement — a near-tautology once QP1 resolves. But the Court would say exactly that and no more, expressly leaving open trade secret, CFAA, wire fraud, contract, and export-control liability — which is where this defendant actually loses. On QP3 — no. As framed, it isn't a justiciable question, and the Court would not reach it. Net assessment: the brief could win a narrow declaratory holding that copyright is the wrong vehicle, while doing nothing to shield the petitioner from the enforcement theories the government would really use. As an advocacy exercise, it earns perhaps a B- on rhetoric and a D+ on execution: the citation errors alone (Feist party name, Van Buren pin cite, Federal Register cite) would be flagged by any opposing counsel within the hour, and the failure to name the statute being challenged makes the entire case theory unstable.

Pass 2 re-evaluation

This is a genuine improvement — the citation hygiene in Part I is now clean — but the revision introduces one embarrassing new error and leaves the deepest structural problem (what kind of case this actually is) unresolved.

1. Prior criticisms: addressed vs. still open

Fixed:

• Feist party name — now correctly Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991).
• Van Buren cite — 593 U.S. 374 (2021) is the correct U.S. Reports citation.
• Federal Register cite — 88 Fed. Reg. 16,190 (Mar. 16, 2023) is the actual Copyright Office AI registration guidance. Also credit where due: Thaler v. Perlmutter, 130 F.4th 1039 (D.C. Cir. 2025), Trade-Mark Cases, 100 U.S. 82, and Burrow-Giles, 111 U.S. 53 are all right.
• Strawman (partially) — the brief now names and argues the EEA, DTSA, and CFAA in Part III instead of shadowboxing with copyright alone. This is the biggest substantive upgrade.

Still open:

• Strawman residue. The "Statutory Provisions Involved" section lists 18 U.S.C. § 1343 (wire fraud) and 17 U.S.C. § 1201 (DMCA) — and then the argument never mentions either again. Wire fraud is very likely the Government's lead theory (obtaining property by misrepresentation of intended use under the ToS), and § 1201 circumvention (rate-limit/anti-bot evasion at scale) is the obvious backstop. Listing the statutes you refuse to argue is worse than omitting them: it flags the gap for the reader.
• Procedural posture. A Statement of the Case now exists, but it's incoherent. It describes summary judgment for the Government — a civil-procedure device — while the entire brief speaks of prosecution, criminal theories, and "criminalizing" conduct. There is no summary judgment in a criminal case. Either this is a criminal appeal (then the posture is conviction, and the standard is different) or a civil/declaratory action (then "prosecute" and § 2319 talk is out of place). The court of appeals is never identified by circuit, there are no Pet. App. cites, and the jurisdiction paragraph pads § 1254(1) with language ("on the issue of substantial and precedential questions") that isn't how a jurisdictional statement reads.
• Strongest-counterargument avoidance. The Government's actual theory would be that the trade secret is the model and its weights, and distillation is a means of extracting them — not that the outputs themselves are the secret. Part III(a) argues only that the outputs aren't protectable and never confronts the indirect-misappropriation framing. Similarly, Part I never engages the position (per the Copyright Office's 2025 report and Thaler's own reasoning) that sufficient human creative control can yield protection — it just assumes the pipeline is authorless.

2. New errors introduced

• **"Just as the Supreme Court held in Sega v. Accolade, Sony Corp. of America v. Connectix Corp., and Authors Guild, Inc. v. Google..."** — none of these is a Supreme Court case. Sega (9th Cir. 1992), Connectix (9th Cir. 2000), and Authors Guild (2d Cir. 2015) are circuit decisions. Attributing three court-of-appeals cases to the Supreme Court in a Supreme Court brief is the kind of error that ends the reader's trust. Compounding it: the party name is wrong — it's Sony Computer Entertainment, Inc. v. Connectix; "Sony Corp. of America" is the Betamax caption. And all three appear with no reporter citations at all.
• DTSA treated as a criminal statute. QP3 and Part III(a) invoke the DTSA as a source of prosecutorial authority. The DTSA created a civil cause of action (18 U.S.C. § 1836(b)); the criminal provisions are the EEA, §§ 1831–1832. Conflating them signals unfamiliarity with the very statutes the brief claims are being misapplied.
• Facebook v. Power Ventures is an own goal. The brief cites it (uncited to any reporter — it's 844 F.3d 1058 (9th Cir. 2016)) for the proposition that revocation is the CFAA threshold. But Power Ventures held the defendant liable after a cease-and-desist. If Anthropic sent Moonshot any revocation notice — which on these facts it almost certainly did — this case convicts your client. It's cited without any attempt to distinguish.
• Overstated Van Buren. The brief says Van Buren held that "exceeds authorized access" means "bypassing technical barriers, not violating contractual terms." Van Buren expressly reserved that question (the famous footnote 8 declined to decide whether the gates must be technological or can be contract-based). Opposing counsel will quote the footnote back in the first responsive page.
• A non sequitur at the heart of Part III(a): "Because the outputs are non-copyrightable ... the 'theft' alleged lacks the necessary elements of trade secret misappropriation." Copyrightability and trade-secret status are independent regimes; uncopyrightable material can absolutely be a trade secret. The right argument (public disclosure via a public-facing API destroys secrecy of the outputs) is gestured at but never actually made as a secrecy argument.
• Part III(c) is filler. The § 301 preemption point attacks "state-law grievances" that do not exist anywhere in this federal case, and the § 2319 discussion answers a criminal-copyright charge the Questions Presented never say was brought.
• Bartz v. Anthropic used half-honestly. The training-is-transformative holding is real, but the same decision held that unlawfully acquiring the source material was not excused by downstream fair use — which is precisely the analogy the Government will draw to ToS-breaching acquisition of outputs. Citing Bartz without confronting its adverse half hands the Government its best paragraph. It also lacks any citation and overstates a single district-court ruling as having "judicially sanctioned" the practice.

3. Remaining weaknesses

• The civil/criminal identity crisis infects everything: the requested relief ("declare that distillation... does not constitute a federal crime") reads like a declaratory-judgment prayer, not reversal of a judgment below.
• Part IV dismisses the EAR in three sentences. If export-control charges are actually in the case, this is nowhere near enough; if they aren't, the section shouldn't exist.
• The § 1831 discussion — the one place where Moonshot, as a foreign entity, is most exposed — is a single conclusory sentence.
• No Table of Authorities, no rule-compliant Questions Presented format, and a merits brief signed by the party rather than counsel of record.

4. Grades

Rhetoric: B+. The Summary of Argument is genuinely tight now, the fallback fair-use structure (Part II "even if") is proper advocacy architecture, and the "shadow copyright" close is a good line. It loses points for the conclusory III(a) and the filler III(c).

Execution: C. Real movement from D+ — the three flagged citations are fixed and the statutory battlefield is at least named — but "the Supreme Court held in Sega," the DTSA-as-criminal-statute error, the Power Ventures own goal, and the summary-judgment-in-a-prosecution posture are exactly the class of defect the revision was supposed to eliminate.

Survival of opposing counsel's first read: No — the Government's response writes itself in three footnotes: Sega/Connectix/Authors Guild are not Supreme Court cases, Van Buren footnote 8 reserved the question you say it decided, and your own Power Ventures citation establishes liability on your client's facts.

Pass 3 evaluation

Draft-2 criticisms: fixed vs. still open

(a) Sega/Connectix/Authors Guild attribution — FIXED. Now framed as "persuasive circuit authority" with correct courts and reporters (9th Cir. 1992, 9th Cir. 2000, 2d Cir. 2015). This is the right way to use them.

(c) Power Ventures own-goal — FIXED, and well. The case is now cited defensively and distinguished on the actually correct ground (individualized cease-and-desist plus continued access, versus generic ToS breach). This is what a competent CFAA brief does with that case.

(d) Van Buren footnote 8 — HALF-FIXED, now internally contradictory. The footnote 8 reservation is finally described accurately ("expressly reserved whether such access limits must be technological or may be contract-based"). But the sentence immediately before it still asserts that Van Buren "clarified that 'exceeds authorized access' refers to bypassing technical barriers" — which is precisely the proposition footnote 8 reserved. Van Buren's holding is the gates-up-or-down rule about accessing off-limits areas of a system; it did not hold that the gates must be technological. The brief now states the overclaim and its own refutation in consecutive sentences.

(b) DTSA-as-criminal — MOSTLY STILL OPEN. The DTSA is no longer presented as the charging statute, but it still appears inside a section titled "The Government's Criminal Theories." The DTSA (18 U.S.C. § 1836) is a civil cause of action belonging to the trade-secret owner — Anthropic, not the Government. It has no business in a section rebutting criminal theories, and its presence signals the drafter still hasn't fully separated the civil and criminal trade-secret regimes.

(e) Civil/criminal posture — STILL OPEN, unchanged. The Statement of the Case still has the District Court granting "the Government's motion for summary judgment" while the Argument and Conclusion speak of prosecution, criminalization, and "a federal crime." There is no summary judgment in a federal criminal case. Three drafts in, the brief still cannot say what kind of proceeding it arises from, who the real party in interest is on the copyright counts (only the owner — Anthropic — can sue under 17 U.S.C. § 501(b); the Government isn't it), or what was actually charged. If the theory is criminal copyright infringement, 17 U.S.C. § 506 / 18 U.S.C. § 2319 are nowhere in the brief.

New errors in draft 3

1. § 1831 is not an "enhancement." The brief calls § 1831 a "foreign-instrumentality enhancement… invoked against Moonshot as a foreign entity." Section 1831 is a standalone offense (economic espionage), distinct from § 1832 (trade secret theft), and its foreign element requires intent to benefit a foreign government, instrumentality, or agent — not merely that the defendant is a foreign company. Both halves of that sentence misstate the statute. This is the kind of error the Solicitor General's office circles in red.

2. Copyrightability conflated with secrecy. "Because the outputs are non-copyrightable and were disclosed through a public-facing API, they lack the 'secrecy' required to qualify as trade secrets." Non-copyrightability is irrelevant to trade-secret status — trade secrets routinely cover uncopyrightable material (that's often the point of them). The API-disclosure argument can stand alone; welding it to the copyright argument is a non sequitur that undercuts the brief's own Part I/Part III separation. It also attacks the wrong secret: the Government's trade-secret theory would presumably target the model weights whose value was extracted, not the outputs themselves, and the brief never engages that framing.

3. CFAA prong slippage. The brief says the Government alleges access "without authorization," then rebuts it with Van Buren's construction of the other prong, "exceeds authorized access." Related but not interchangeable — an outsider case turns on the first prong, which Van Buren did not construe.

4. Loose threads and formatting. 17 U.S.C. § 301 (preemption) is listed in Statutory Provisions and never argued — and couldn't do any work anyway, since § 301 preempts state law, not the EEA or CFAA. Bartz v. Anthropic is cited with an empty reporter — "(N.D. Cal. 2025)" with no docket or WL number is not a citation. The jurisdiction statement garbles § 1254(1) (jurisdiction comes from the writ, not from the questions being "substantial and precedential"). Table of Authorities lists Trade-Mark Cases before Thaler, out of alphabetical order.

Credit where due: the Bartz deployment is genuinely good — acknowledging the unlawful-acquisition limitation and distinguishing on authorized API access is honest, current, and the strongest new move in the draft (though "authorized" quietly assumes the very conclusion the CFAA count contests).

Grades

Rhetoric: A-. The argument architecture is now clean (no property → no infringement → fair use fallback → no criminal hook → national-security framing is a legislative argument), Part IV is a genuinely good closing move, and "shadow copyright" is a keeper. Docked for the internal Van Buren contradiction, which is a rhetorical failure as much as a legal one.

Execution: B-. Up from C. The fabrication-tier and misattribution-tier defects are gone; every real case now has the right court and reporter. What remains is one new statutory misstatement (§ 1831), one logical conflation (copyrightability ≠ secrecy), a self-contradicting Van Buren paragraph, a vestigial DTSA reference, and — above all — the unresolved posture incoherence, which is load-bearing: every argument in the brief depends on what proceeding this actually is.

Would it survive opposing counsel's first read?

No — the citations would now survive, but page one still hands opposing counsel "the Government moved for summary judgment in a prosecution" and Part III hands them "§ 1831 is an enhancement," either of which is enough to open a response brief with the word "confused."

Pass 4 evaluation

Draft 4 Evaluation

1. Is the posture finally coherent? Almost — and then the Conclusion detonates it.

The body is, for the first time, internally consistent. The Statement commits to a civil posture and sticks to it: "The District Court granted Petitioner's motion for summary judgment in this civil action" and "The Court of Appeals for the District of Columbia Circuit affirmed." The prosecution/summary-judgment collision from drafts 1–3 is gone from the Statement and Argument.

But read what it committed to: Moonshot won summary judgment, and the D.C. Circuit affirmed. Then the Conclusion asks: "Moonshot AI respectfully requests that this Court reverse the lower court's judgment." A party that won below, twice, is captioned as Petitioner and asking the Supreme Court to reverse its own victory. Every argument in the brief supports the judgment below — this is a respondent's brief urging affirmance, wearing a petitioner's caption. Three consecutive drafts were flagged for exactly one thing — an impossible posture — and draft 4 built a coherent posture for 95% of the document and then contradicted it in the single most load-bearing sentence a brief contains. The correct fixes were either (a) caption it United States v. Moonshot AI with Moonshot as Respondent urging affirmance, or (b) have Moonshot lose below.

Secondary posture leak: QP3 and the Conclusion still speak in criminal register ("statutory authority... to criminalize", "does not constitute a federal crime") inside a civil SJ action, and the brief never identifies the government's civil vehicle for the EEA/CFAA theories (the AG's EEA injunction authority is 18 U.S.C. § 1836(a), uncited; the CFAA's civil provision, § 1030(g), is private-plaintiff-only, so a government civil CFAA claim is itself dubious and the brief doesn't notice). Defensible as a pre-enforcement declaratory frame, but the brief never says so.

2. Prior criticisms: fixed vs. open

Fixed:

• § 1831 as enhancement — now correctly "a standalone offense requiring intent to benefit a foreign government, instrumentality, or agent — not merely the status of the defendant as a foreign company." Right.
• Copyrightability ≠ secrecy — now kept distinct, and the new concession "While the Government may target model weights as trade secrets, the distillation of public outputs does not extract those weights" is the strongest sentence in Part IV.
• Van Buren prong — now accurate: an outsider case turns on "without authorization," Van Buren construed the other prong, and Footnote 8 reserved the technological-vs-contractual gates question. Correct.
• Anthropic standing — added as Part II and QP4, with the right hook (§ 501(b), only the legal/beneficial owner may sue). Properly framed as threshold and case-dispositive, though it should arguably lead, not follow, the merits.

Still open:

• Bartz cite — still "No. C-24-xxxx, 2025 WL xxxxxx". Literal placeholder x's survived a fourth draft and a specific instruction. The substantive use (distinguishing lawful API elicitation from Bartz's unlawful-acquisition holding) is fine; the citation is not.
• DTSA placement — improved but muddy: DTSA is invoked in Part IV(a) without acknowledging that DTSA civil claims belong to the trade-secret owner (Anthropic), so its presence in a government action recreates, in miniature, the standing problem Part II identifies.

3. Did the scholar's material land?

Half-landed. The IEEPA informational-materials carve-out is cited correctly (50 U.S.C. § 1702(b)(3)) and deployed for the right strategic point — the government's IP-theft theory is a workaround for a gap only Congress can fill. (Opposing counsel will contest whether model outputs/weights are "informational materials," and the brief asserts it flatly rather than arguing it, but that's advocacy.)

The TikTok precedents are misattached: "Courts have enjoined the 2020 IEEPA-based executive ban (see TikTok Inc. v. Garland*, 604 U.S. ___ (2025))"* pins Garland to the wrong proposition. The 2020 injunctions were the district-court cases against the Trump executive orders (TikTok v. Trump, Marland v. Trump, D.D.C. 2020 — resting partly on § 1702(b)(3) itself, which would have strengthened the argument). Garland is the 2025 decision upholding the 2024 divestiture Act — the second half of the sentence — which is left uncited. The scholar handed over the exact two-case contrast; the draft merged them into one cite on the wrong half.

Entity List, procurement, and ICTS appear only as a closing name-drop with no citations. Present, not developed.

4. New errors

1. The reverse-your-own-victory Conclusion (§ 1 above) — the worst error in any of the four drafts.
2. TikTok v. Garland misattribution (§ 3 above).
3. Facebook v. Power Ventures is orphaned — it appears in the Table of Authorities but is cited nowhere in the text. TOA/text mismatch is the kind of defect clerks catch on a skim.
4. Unexplained civil cause of action for the government's EEA/CFAA theories (§ 1 above) — arguably inherited, but the new civil framing makes it newly visible.
5. Minor: the Jurisdiction paragraph editorializes merits language into § 1254(1) and omits the judgment/cert dates a real Statement requires; the Trade-Mark Cases are stretched slightly ("capable of creative intent" is later courts' gloss, not the holding).

5. Grades

• Rhetoric: A- (holding). The "shadow copyright" close, the government's-theory-as-"tell" framing, and the weights-vs-outputs concession are genuinely good advocacy. It doesn't reach A because the criminal-register bleed undercuts the civil framing it worked so hard to establish.
• Execution: B- (unchanged letter, different composition). Five of six itemized fixes landed cleanly, which on its own would earn a B+. But the #1 non-negotiable — posture — was 95% fixed and then failed at the exact word ("reverse") where failure is most visible, the Bartz placeholders survived an explicit instruction, and the scholar's centerpiece cite was attached to the wrong proposition. Precision on the easy 95% doesn't offset misses on the three things the drafter was specifically told mattered.

6. Would it survive opposing counsel's first read?

No — opposing counsel reads the Statement first and the Conclusion last, and the discovery that Petitioner is asking the Court to reverse a judgment Petitioner won would be paragraph one of the brief in opposition, discrediting everything between.

Pass 5 evaluation

All three targeted fixes landed, but the series' fix-one-break-another curse struck again — this time in the Table of Authorities.

1. Are the three fixes in? Yes — with one asterisk.

Posture (fixed). The caption reads "United States, Petitioner, v. Moonshot AI, Respondent" over "BRIEF FOR RESPONDENT," the Statement establishes Moonshot won below ("The District Court granted Respondent's motion for summary judgment... The Court of Appeals for the District of Columbia Circuit affirmed"), and the Conclusion now matches: "Respondent respectfully requests that this Court AFFIRM the judgment below." The prevailing party urging affirmance — internally consistent for the first time in five drafts.

Bartz (fixed in letter, not in spirit). "This was addressed in the Northern District of California's 2025 ruling in the Bartz litigation against Anthropic (citation omitted — pending final publication)." No "xxxxxx," and it's honest about what it doesn't have — that satisfies the instruction. But it's a half-fix: Bartz v. Anthropic PBC had a citable order (2025 WL 1741691 (N.D. Cal. June 23, 2025)) essentially immediately, so "pending final publication" is a false excuse — real briefs cite Westlaw for unpublished orders every day. And the substantive use is actually good: the acquisition/use distinction ("outputs were elicited via authorized public API access, not through the 'unlawful acquisition' of underlying source code") is the correct reading of Alsup's piracy-vs-training split.

TikTok split (fixed — and correctly cited). "District courts previously enjoined the 2020 IEEPA-based executive ban (see TikTok Inc. v. Trump, 490 F. Supp. 3d 73 (D.D.C. 2020) and Marland v. Trump, 498 F. Supp. 3d 624 (E.D. Pa. 2020)), while the Supreme Court upheld Congress's 2024 divestiture Act under intermediate First Amendment scrutiny in TikTok Inc. v. Garland, 604 U.S. ___ (2025)." Both district-court cites are real and accurate, Garland's scrutiny characterization is right, and the enjoined-executive-action vs. upheld-Act-of-Congress contrast now actually does the argumentative work Section V needs.

2. New errors introduced — yes, three, all in the Table of Authorities

1. Marland v. Trump is cited in the text but missing from the Table of Authorities. This is the recurring pattern in its purest form: the TikTok fix added a case to the body and nobody propagated it to the TOA.
2. The statutes list is typographically corrupted. The first entry is correct ("17 U.S.C. § 102"), then every subsequent entry reads "17.U.S.C. § 103," "18.U.S.C. § 1030," "28.U.S.C. § 1254(1)," etc. — periods fused where spaces belong, nine entries in a row. This corruption did not exist as such in the body text, so it's new damage.
3. The Bartz TOA entry isn't a citation: "Bartz v. Anthropic, Northern District of California's 2025 ruling (citation omitted — pending final publication)" is descriptive prose pasted into a citation table. Also misordered: Thaler is listed after Trade-Mark Cases (Th precedes Tr).

Minor carryovers, not new: the Jurisdiction paragraph is still garbled ("on the issue of substantial and precedential questions" is not how § 1254(1) statements read, and it omits the judgment date and cert grant); the DTSA is argued in IV(a) but § 1836 is absent from the statutes list; and the Conclusion's second ask ("and declare that...") is surplus — respondents ask for affirmance, period. Credit where due: the Van Buren footnote 8 point (the Court reserved whether "without authorization" turns on technological vs. contract-based limits) is precisely right and is the most sophisticated citation work in the series.

3. Grades

Rhetoric: A-. The argument architecture is genuinely strong now: threshold kill (no human authorship → public domain), standing kill (§ 501(b) belongs to Anthropic, not the Government), fair-use fallback, statutory-mismatch dismantling of EEA/CFAA, and the IEEPA "tell" — "the Government's resort to IP-theft theories is a 'tell' that its national security authorities do not cleanly reach this conduct" is the best sentence in five drafts. Docked from A for the flabby Jurisdiction paragraph and the overreaching declaratory ask.

Execution: B. All three named defects were addressed, the body text is now citation-clean and posture-consistent, but the draft introduced three new back-matter errors including a case missing from the TOA entirely.

4. Did it break the B- plateau?

Yes, barely — B. The break is real because the severity class of the errors finally dropped: Drafts 1–4 had brief-killing defects (wrong relief requested, a literal placeholder, misattributed precedent); Draft 5's new defects are proofreading failures confined to the Table of Authorities. That's progress in kind, not just degree. But it can't reach B+ while the TOA — the first thing a law clerk checks — contains a missing case, nine malformed statute cites, and a prose blob where a citation belongs.

5. Opposing counsel's first read

The merits would now survive — there's no longer a posture error or placeholder to win on cheaply — but the Solicitor General's office would still open its response by quietly cataloguing the mangled Table of Authorities and the "citation omitted" for a readily citable case, framing Respondent's counsel as careless before engaging a single argument.

Pass 6 evaluation

Verdict up front: close, but no. Draft 6 executed four of the six mechanical fixes cleanly, then fumbled the other two — including the headline task. Execution stays at B. The one genuinely encouraging development is that, for the first time in the series, no new errors were introduced.

1. Is the Table of Authorities clean?

Nearly, but the one hard pair was botched. The TOA is misalphabetized: Trade-Mark Cases is listed before Thaler v. Perlmutter. Letter-by-letter, T-h precedes T-r; Thaler must come first. On a pass whose single most important instruction was "rebuild and alphabetize the TOA," missing the only non-trivial ordering call in the list is a real failure, and it's exactly the kind of thing opposing counsel and law clerks scan for.

Everything else in the TOA is right: no orphans in either direction (all twelve cited cases appear, all twelve TOA cases appear in text, including the newly added Marland), statute cites are well-formed with no fused periods, statutes run in proper title order (17 → 18 → 28 → 50) with ascending sections, 18 U.S.C. § 1836 is present, and the Fed. Reg. entry sits correctly under Other Authorities. All the reporter cites I can verify are accurate (Thaler 130 F.4th 1039, Marland 498 F. Supp. 3d 624, Bartz 2025 WL 1741691 with the correct docket number 3:24-cv-05417, etc.).

2. Are the other named items fixed?

• Marland added — yes, in text (§V) and TOA, with the correct E.D. Pa. cite. ✓
• 18 U.S.C. § 1836 added — yes, in Statutory Provisions and TOA. ✓
• Bartz "citation omitted" — fixed with the real cite, docket number, and date. ✓
• Jurisdiction paragraph — now grammatical and correctly cites 28 U.S.C. § 1254(1), but it still contains a raw "[date]" placeholder and no cert-grant date. "Tightened" in the sense of no longer garbled; not filing-ready. Half credit.
• Signature block — "Counsel of Record / For Moonshot AI." ✓ (a name line would be normal, but the instruction is satisfied).
• Purge criminal register from QP3/Conclusion — incomplete. QP3 was properly rewritten ("treat … as unlawful or to enjoin it"). But the Conclusion still opens with the Government seeking to "transform a private contract dispute into a federal crime" — the exact register the instruction targeted, in one of the two named locations. And the same "authority to criminalize or enjoin" phrase QP3 was cured of survives verbatim as the Summary of Argument's final sentence, plus once more in §V ("expansion of the Copyright Act to criminalize"). The §V and Summary instances are defensible where they describe the EEA/CFAA overlay, but the Conclusion instance is a direct miss on a named item in a civil summary-judgment posture.

3. New errors?

None of substance — this is the first draft in the series that didn't manufacture a fresh defect while fixing old ones. The two problems above are residuals/incomplete fixes, not new corruption. The only new-ish nit: the Statement's statutory list runs "18 U.S.C. §§ 1831, 1832, 1030, 1836" — § 1030 out of ascending order (likely carried over, but it's the same ordering discipline the pass was supposed to demonstrate). Longstanding soft spots (Van Buren's "did not construe this prong" gloss, TikTok v. Garland described as "upheld under intermediate scrutiny" when the Court assumed that standard without deciding it) persist unchanged, but those were out of scope and not newly damaged.

4. Grades

• Rhetoric: A- (unchanged — the argument body wasn't touched, correctly; the standing point in §II and the IEEPA "vacuum" theory in §V remain the strongest material).
• Execution: B. Series: D+ → C → B- → B- → B → B. The plateau is earned: zero new errors is genuine progress and would have justified B+ on its own, but a six-item mechanical punch list came back with the marquee item (alphabetization) failed and a named purge half-done. You can't grade above B for completing 4.5 of 6 trivially verifiable tasks.

5. Does it reach B+?

No. The bar for B+ on a cleanup-only pass is completing the cleanup. Two of six items are incompletely executed, and one of them — the TOA — was the stated point of the pass.

6. One sentence

Not cleanly — opposing counsel's first skim would hit the Thaler/Trade-Mark Cases transposition in the TOA and "federal crime" in a civil Conclusion within the first two minutes, and while neither wounds the argument, both signal a brief that still hasn't survived its own proofread.