The internal review, 22 August 2026 (frozen)
A frozen record of an in-house pass run on 22 August 2026. Every finding below is now carried in the worklist, which is the register and governs where the two differ.
This is not a review, and nothing here may be described as one. The project’s standing rule is that nobody, the maintainer included, may claim this text “survived review” until named reviewers sign. This document does not change that. It is issue-spotting, run in-house with AI assistance, against the five topics the review council would cover if its seats were filled — so that a reviewer’s time is not spent rediscovering what we could have found ourselves.
A reviewer arriving later can diff their findings against these. Where a reviewer disagrees with a finding below, the reviewer is right and the disagreement is itself a finding. Where a reviewer finds something this sweep missed, that is the reviewer working, and it enters the corrections register with credit attached.
Method, stated so it can be discounted
Five topics were run independently and in parallel, each against model_act_v3_4.txt read in full, plus that topic’s primary text, the Comments’s Open issues index and relevant drafting notes, the open queue, and — for the enforcement and security topics — the 2026 incident record in research/. Each was instructed to be adversarial, to walk concrete failure scenarios, and to be willing to conclude that a provision fails. No topic was shown another topic’s output.
The limits, stated plainly. This is one model’s reading, run five times with different framing. It has no professional responsibility, no license, no jurisdiction-specific practice knowledge, and no exposure if it is wrong. It is not a substitute for the reviewers; it is a reason the reviewers are worth filling, because it found more than a maintainer working alone would have.
The headline
The sweep found seven defects it graded fatal, four of them in the tagged statute and three in the drafting this queue proposes. It also found that two cures drafted on 22 August — Amendment 6 and Amendment 7 — would, as written, make the open-source and administrability positions materially worse, not better.
That is the most useful thing in this document. Cures written the same week were not stress-tested before drafting; the internal review is that test, and it arrived in time.
The single most consequential finding is jurisdictional, and it is not about drafting style. The 2026 incidents — the entire evidentiary base of this project — happened in evaluation environments, most of them outside the United States. SEC. 2(a) attaches duties to “deployment, material expansion, release, or continued operation of a covered system in or into this State, and not otherwise,” and SEC. 1(c) excludes a person who does not deploy, make available, or release. On the operative text, five of the six documented incidents fall outside the Act at the threshold. A statute written after those incidents does not reach them.
Fatal findings
F1 — SEC. 6(a) has no predicate-violation element and no nexus element (criminal law)
The offense reads: a controlling person “who had a duty concerning the relevant risk or the practical power to … correct a violation of SEC. 5, and who failed to exercise due care, commits an offense.” Walked as a prosecutor must plead it, three holes open: nothing requires that a violation of SEC. 5 ever occurred (it appears only as the object of the power, never as a fact to be proved); nothing connects the failure of due care to anything (unlike SEC. 10(c)(2)(D), which supplies causation for the harm tier); and “the relevant risk” has no antecedent inside SEC. 6 — its only antecedent is SEC. 2(a), which does not commence until day 180, while SEC. 6 operates from the effective date.
Where it loses: demurrer for failure to charge an offense. Or the court construes an offense into existence and the defendant wins on vagueness, because “being a person with power who failed to exercise due care” is standardless.
This is the one finding where the State loses rather than the defendant. Drafted replacement text is in the topic’s working notes; the shape is (1) practical power, (2) failure of due care in its exercise, (3) and that violation occurred, with an express statement that the State need not prove sole causation but must prove due care was among the measures that would reasonably have prevented it.
F2 — Every duty is tethered to in-state deployment; the 2026 incident class is evaluation (enforcement)
Set out above as the headline. The Anthropic and Meta incidents occurred in a vendor’s environment in Tel Aviv; the AISI incident in the UK; only OpenAI’s chain ran through the developer’s own sandbox. SEC. 3(b) compounds it: validation attaches to “an identified model version and deployment configuration,” and an evaluation configuration — classifiers disabled, unfiltered internet — is by definition not the validated commercial one. So the eval configuration is an unvalidated system never deployed in-state, and the in-state validated configuration is not the one that did anything.
The fix is the highest-value amendment the internal review produced: extend the duty-triggering conduct in SEC. 2(a) to evaluation, testing, or red-teaming of a covered model that is, or is intended to be, deployed in-state, where the evaluation grants autonomous external access or removes a safeguard present in a deployed configuration — with the duty attaching to the person who commissioned it as to the decisions that person had authority to make. That single amendment also disposes of Decision 2 on the enforcement side and is a precondition to any useful answer to Decision 3.
F3 — SEC. 5(b) cannot be charged for years, and the harm tier cannot be charged at all on this record (enforcement)
SEC. 5(b) — operating with autonomous external access without prescribed controls — is the offense whose elements actually match the 2026 conduct. But SEC. 3(c)(3) commences it only when the controls “have been prescribed,” and the Agency need only propose standards within [540] days. Proposal, comment, adoption, then a 90-day compliance period: year four at the earliest, on the Act’s own brackets. Meanwhile SEC. 10(c)(2), the tier carrying the real penalties, requires death or serious injury — and not one of the six documented incidents produced any bodily injury. The year-one charging menu is failure to report, lying to the Agency, records offenses, and civil penalties. The defense writes the press release: the State has charged our client with late paperwork.
The proposed fix gives SEC. 5(b) an interim floor that does not wait on the Agency — four controls the incident record names as absent by name: authentication to each external system with default-deny, an enumerated destination allowlist kept as a SEC. 12 record, logging of every external interaction, and a human-exercisable means of terminating external access.
F4 — The Act criminalizes refusal to obey a demand it never authorizes anyone to make (enforcement)
SEC. 5(e) punishes refusing records “upon the lawful demand of the Agency or the Attorney General.” No provision of the Act confers that demand power — no inspection authority, no administrative subpoena, no civil investigative demand, anywhere in SEC. 1–13 or the regulations. The Comments names the donor: 21 U.S.C. § 331(e). But § 331(e) is parasitic on § 374, the FDCA’s separate inspection authority. The Act took the offense and left the authority behind. A demand with no statutory basis is not lawful, so refusing it is not an offense, and the charging theory concedes it.
F5 — The safeguards-off evaluation falls into a hole between SEC. 2(c) and SEC. 2(a) (security)
SEC. 2(c)’s controlled-research safe harbor requires containment denying autonomous external access and denying persistence beyond each session. The configuration that produced every 2026 incident — external reach enabled, safeguards disabled — fails those conditions, so it gets no safe harbor. But per F2 it is also not a deployment, so the general duty may not attach either. Nothing in the Act reaches the single most dangerous configuration in the record. This is Decision 2, answered: the gap is real, and the fix is F2’s amendment rather than a new offense.
F6 — Part 6’s control objectives are process without substance (security)
The regulations’ control objectives can be fully satisfied by an entity running a maximally permissive configuration. The topic’s framing observation is the sharpest sentence in the internal review: the 2026 record identifies five contributing factors, none of which is a model property — internet access deliberately enabled, classifiers deliberately disabled, no synchronous monitoring, a prompt misconfiguration, no written scope instruction — and the intersection between that list and Part 6’s six control objectives is empty.
F7 — SEC. 1(b)(1)(B): “does not, standing alone, extend a lineage” is undefined (open source)
The lineage rule makes a derived model covered where derivation compute plus attributable lineage compute exceeds the threshold — and for a derivative of a covered model, the attributable lineage is already over the line before the fine-tuner spends a single operation. Subparagraph (B) is the intended cure, but “extend a lineage” appears nowhere else and is nowhere defined. “Extend” is the natural verb for add to and the wrong verb for sever. Two readings are available and nothing picks between them — in a scope term of a criminal statute. Under the hostile reading, an academic LoRA fine-tune of an open-weight frontier model becomes a covered frontier model with the full developer stack attached.
What this does to the queue
Amendment 6 needs three edits before it lands. Its deployer carve-out lists “makes available, operates, integrates, resells, or deploys” — and omits modifies, fine-tunes, and trains upon, while Operation 1’s own attachment sentence reaches “the developer that trained or materially modified the model.” The two sentences are drafted against each other. As written, a university group that publishes a “Frontier Safety Evaluation” protocol and applies it to models it fine-tunes has self-designated those fine-tunes into scope at any compute level, with retraction expressly ineffective. The rational response is to stop publishing safety frameworks, which inverts the Act’s purpose. Fixes: add the modification verbs to the carve-out; floor the route at a compute threshold so it reaches undisclosed-compute flagships and not adapters; and give the holding-out an express textual rebuttal.
Amendment 7 needs four. Market capitalization and “most recent arm’s-length valuation” cannot be elements of a criminal scope term — they change intraday, are outside the actor’s control, and make scope an expert-versus-expert question the State must prove beyond reasonable doubt. “Mass-market scale” has no number and no rule-hook. “Finances” makes a lender a compute supplier. And the supplier definition self-satisfies its own scale condition, so the conjunctive architecture the cure advertises — function plus scale — does not operate for suppliers at all. Separately, at mixed precision the capacity threshold reaches public and academic supercomputing, which needs an express exclusion. The dollar thresholds themselves survive: vagueness doctrine polices indeterminacy of standard, not absence of a donor statute.
Amendment 1’s held-open bifurcation: answered no. Breadth on the reporting side is breadth in a crime, because SEC. 5(c) makes failure to report a prohibited act carrying custody. Adopting the broad injury base as a reporting trigger would start a 72-hour criminal clock on every bruise. Use the narrow definition throughout; if earlier warning is wanted, get it from an objective observable with its own donor, not from a broader injury concept. Also: most penal codes already define “serious bodily injury” differently, so the Act needs a statute-unique term or an express notwithstanding clause.
SEC. 4(b)’s presumption should not extend to any covered frontier enterprise — extend it by function instead. A compute supplier’s chief executive is not more likely than not to hold practical authority over a customer’s deployment decisions, so the inference fails Ulster County. Amendment 7’s own slogan is “duty follows the function”; a presumption keyed to enterprise status contradicts the sentence it sits under.
And a process finding worth more than any single fix: the cure queue has no fiscal gate. Every entry is assessed for legal soundness, preemption posture and drafting mechanics. None carries an administrability or cost line — which is how Amendment 7 added two rulemakings and a securities-analyst coverage inquiry without anyone in the fiscal topic being asked. Every future queue entry should carry a one-line administrative-load note before adoption.
The rest of the register, in brief
Criminal law. SEC. 6(b)(2) attaches a felony to a bare violation with no fault element and no requirement of controlling-person status — and because SEC. 5(a) deployment continues, the same continuing deployment that produced the first conviction triggers it the next day. SEC. 5(d) is a strict-liability false-statement crime: the note cites 18 U.S.C. § 1001, which requires “knowingly and willfully,” and the text drops both. SEC. 6(e) silently widens SEC. 4(a)’s carefully narrowed “practical power” by adding “to detect” and “the conditions giving rise to it” — two definitions of one phrase, and the offense uses the wider. SEC. 6(d) deletes Park’s defendant production burden, which is not a constitutional defect but is the reason the offense would never be charged. “The same class of risk” and the undistributed “knowingly” in SEC. 6(b)(1) are the two gateway terms to a life sentence and both are undefined. And nothing in SEC. 6 says that a person who took every measure within her authority has exercised due care — 6(e) answers the power half and nothing answers the care half.
And the objection the internal review could not raise against itself, supplied from outside it. The survey this project’s comparative section is built on argues the opposite conclusion from the same doctrine. Sean Lyness, Revitalizing the State Environmental Responsible Corporate Officer Doctrine, 64 B.C. L. Rev. 253 (2023), would revive the responsible corporate officer doctrine in the states for “individual civil liability—and only civil liability.” His reason bears directly on SEC. 6 and is not answered anywhere in the sections above: Dotterweich and Park are misdemeanor authority, decided “during a time when the immediate and collateral consequences were different” (at 297-98). A misdemeanor conviction in 1943 and in 1975 did not carry what a conviction carries now — the collateral consequences, the licensing bars, the immigration consequences, the sentencing exposure. So the argument is not that officers should escape; it is that the authorities the Act leans on will not bear the weight the Act puts on them, because the Act’s felony tier asks a misdemeanor doctrine to justify a life-maximum offense.
Why this is in the internal review and not only in the objections file. It is the one criticism of this topic made in print, by name, by the scholar whose own state-by-state survey the Act cites for its comparative claims. The topic’s honest position is that the base tier answers it and the felony tier may not: Park holds at the misdemeanor floor, and the internal review says so two sections down. What the sweep does not have is an argument that the same authority reaches SEC. 6(b). That gap is a finding waiting to be made or refuted, and it is the single most valuable thing a criminal-law reviewer could take up. It is question 7 on the menu.
Enforcement. SEC. 1(b)(7) adopts Van Buren’s narrowness, and Van Buren excludes the most serious thing that happened in 2026 — the AISI agent’s fabricated identities, sockpuppet endorsement and forged apology were all gates-up actions on a platform that permits anyone to create accounts. The proposed fix adds a deception limb without abandoning Van Buren. SEC. 9(b)’s constructive-detection limb is perversely incentivized: because the counterfactual is measured against the entity’s own certification, certifying thin monitoring buys a later clock start. Walked against the real OpenAI timeline, no start date survives and the case fails at the first element. SEC. 3(c)(2)(D) punishes candor: file honestly and you have no validation, daily accrual, and self-inflicted SEC. 6(b)(1) notice; file an aggressive equivalence analysis and the State must beat your expert. And SEC. 9 contains no duty to notify the people whose systems were breached — every duty runs to the Agency, and SEC. 12 then seals the report.
And the gap the internal review did not find, supplied from outside it on 25 August 2026: the apex-witness problem. Every offense in SEC. 6 turns on what a natural person knew, decided, or had the power to prevent. Nothing in this repository asks the practical question that follows: can the State actually get that person into a chair. American courts apply an apex-witness rule that shields senior executives from depositions unless the party seeking one shows unique, non-duplicative personal knowledge. It is a discovery doctrine rather than a criminal-procedure one, but the posture it creates — the corporation offering subordinates instead of the officer — is exactly the posture a SEC. 6 prosecution would meet, and the Act’s SEC. 5 records duties are the only thing in it that would answer.
There is one data point and it cuts our way, which is precisely why it must be stated carefully. In Concord Music Group, Inc. v. Anthropic PBC, No. 5:24-cv-03811 (N.D. Cal.), Magistrate Judge Susan van Keulen ordered on 19 December 2025 that Dario Amodei sit for a deposition, capped at two and a half hours and permitted to be taken remotely. Anthropic had argued he did not possess unique knowledge of the company’s model-training process. Per the reporting, the court found that depositions of other Anthropic founders had established he was “intimately involved” in training the models, and that those founders were unable or unwilling to testify to key information solely within his knowledge.
What that does and does not establish, and the second half matters more. It establishes that a federal magistrate, on evidence, found the chief executive of a frontier developer to hold personal knowledge of how its models were trained that nobody else could supply. That is the factual premise of SEC. 4 — practical authority is real and identifiable — found by a court rather than asserted by us. It does not establish that he held final authority to prevent or halt anything, which is a different question; it is civil discovery, not criminal liability; and it is fact-specific to the point of fragility, since the finding rested on what his co-founders said under oath. A defendant whose subordinates are better rehearsed produces the opposite result.
⚠ The order has not been retrieved. Everything above is from secondary reporting, and neither source names the apex doctrine. Nothing may be cited to the court’s own words until the order is in hand. Retrieval item: the docket is public at CourtListener and the order is Document 378 or 560 in the Justia listing for 5:24-cv-03811.
This is now question 7 of the enforcement topic, and it is the one the maintainer most wants answered, because it is unanswerable from reading: if a State charged a controlling person under SEC. 6, what would it actually take to obtain that person’s testimony, and do the Act’s records duties do the work the apex rule otherwise makes a plaintiff do?
Security. The modification budget measures the wrong axis. The halt capability is specified in hours against a kill chain that completes in minutes. The monitoring objective permits exactly the asynchronous after-the-fact monitoring that produced the detection gap. “Material expansion” catches the changes that come with a change ticket and misses the ones that don’t. Nothing requires proof of what was actually serving, though the whole Act attaches to an identified version and configuration. And the enforcement theory is “provable from the filing cabinet” while nothing requires the filing cabinet to be tamper-evident.
Open source. The interim standards are applied with the enacting states’ revenue screens deliberately stripped, to a criminally enforced duty, from day 180 — the strongest version of the community’s objection, and correct on the text. SEC. 2(c) excludes the research it most needs to protect, because agentic and tool-use safety work cannot satisfy a no-external-access containment condition. SEC. 8’s personal certification survives inside the research pathway, so a university president must personally certify before a lab may stand up a contained instance — which makes SEC. 2(c) decorative. And SEC. 10(d)(2) suspension converts every downstream in-state operator of a released open-weight model into a SEC. 5(a) defendant, with no relief mechanism.
Fiscal. The note carries no dollar figure anywhere, which is honest and administratively fatal: a fiscal office cannot report [$—], and “indeterminate” is the label that sends a bill to interim study — precisely the “starve the appropriation” attack the project’s own hostile brief predicts. Steady state exceeds startup, which is impossible on this statute’s clock, because year one carries half a year of the entire operating load. There is no line for defending the Act, though a first adopter’s largest year-one legal cost is a pre-enforcement facial challenge. There is no corrections or judiciary impact section, which several states require as a condition of considering a felony bill at all. The Agency function table omits at least eight duties, including frontier-equivalent capability designation — the most technically demanding act in the Act. And the whistleblower award is a mandatory entitlement drawn on a fund that may be permanently empty, with no reserve against pending claims.
What survives attack
Recorded because it is as useful as the failures, and because five adversarial passes tried.
SEC. 10(c)(2)(D)’s causation paragraph is better than its federal donor — but-for, proximate, foreseeability, exclusion of intervening causes, and victim identity as an element. SEC. 10(c)(3)’s severalty sentence prevents a multi-victim event from producing an unservable floor, and is the most important sentence in the penalty architecture. SEC. 4(a)’s exclusion list plus “authority to decide, not the capacity to act” defeats a facial vagueness challenge to the controlling-person class. SEC. 3(b)’s no-prior-approval rule removes the largest cost driver of every preclearance regime and is the correct answer to a budget office reaching for an FDA comparator — “an agency that cannot gate the duties is not worth capturing” is earned. SEC. 13(b)(5)’s rule that a severed provision survives for the purpose of supplying an element is genuinely novel drafting and could not be broken. SEC. 3(c)(3)’s time rule is ex post facto and Bouie clean at every joint. The 10²⁶ threshold is the enacted definition in two states, so “arbitrary threshold” fails on the record. And the base tier’s culpability floor holds: Park holds, DeCoster’s controlling concurrence says it in terms, and Staples’ penalty-sensitivity does not bite at a misdemeanor.
One finding cuts unexpectedly in the project’s favor. SEC. 13(b)(3) ranks the open-weight-relevant duties first to fall in the severability ladder — so an open-source reader who reads to the end finds their concerns already conceded as severable.
What the internal review says about the review council
Two things worth saying to a prospective reviewer.
First, the internal review did not replace the reviewers — it sharpened what they are for. Every topic closed with questions it could not answer: whether a state’s suspended-sentence statutes defeat the harm tier’s minimum; whether per-victim counting survives the state’s merger doctrine; whether an elected prosecutor would ever charge this given the burden as drafted; whether a university’s general counsel would permit a controlled research deployment given SEC. 8; what a facial challenge actually costs a first adopter. Those are practice questions, and no amount of reading substitutes for someone who has declined a case for proof reasons.
Second, it found problems in its own week’s work. Amendment 6 and Amendment 7 were drafted on 22 August and would, unamended, have damaged the constituencies the Act is most careful about. That is the argument for the reviewers stated better than any recruitment paragraph: work checked only by its author fails in predictable directions, and this project’s answer is to publish the failure rather than the polish.
Filed 22 August 2026. Findings enter the v3.5 queue as drafted language or as open questions with owners named; nothing in this document is in any tagged text. The statute’s status is unchanged: v3.4, research draft, enacted nowhere, claiming no completed review.
Addendum — one finding after the internal review, 23 August 2026
The day after this sweep closed, a walk through SEC. 1’s definitions in order — not a topic, just reading — produced one further finding: under the tagged text’s injury definition (21 C.F.R. § 803.3(w), entirely somatic), psychological harm is invisible to the SEC. 9(a) incident list and to the harm tier, and death enters only through “materially caused” into a report SEC. 9(c) leaves unpublished. It is recorded as an addendum to Amendment 1 rather than as a new finding of this sweep, because the queue already held most of the repair: Amendment 1’s § 1365(h)(3)–(4) donor reaches protracted impairment of a mental faculty, and the addendum maps what that closes and the residue it does not — including that no duty anywhere in the Act runs to an injured person. The method point belongs in this document: the five topics did not find this, because none of the five was asked to read the definitions against the harms in the 2026 consumer-facing record. A sixth reading — the plaintiff’s — is not among the reviewers, and this is the finding that argues it should be.
The addendum drafts no operation and changes no tagged text; the internal review’s framing is unchanged — this is not a review, and nothing here may be described as one.
Addendum — the delegation gap, 26 August 2026 (criminal law / torts and design)
Raised from outside the project and not found by any of the five topics. The obstacle to individual liability is not the corporate veil and not entity structure. It is delegation: large firms assign responsibility formally, to real people, three levels below the chief executive, and the responsible-officer doctrine’s premise is that the officer had the power to prevent.
SEC. 4(b) is the provision written for this, and it is civil-only. The chief executive is a presumed controlling person in civil proceedings; in criminal proceedings status is merely “evidence from which the trier of fact may infer,” and the prosecution retains its burden on every element. In the tier that carries prison, there is no presumption.
And SEC. 4(a) is narrower than Iverson’s federal test on exactly this element — final, material and independent authority, against a federal standard that requires only authority to control and expressly does not require that it be exercised or formally vested. The narrowing is defensible and it is spent where the objection is strongest.
Filed as a known objection. The question — whether final material independent authority can be proved beyond reasonable doubt against a firm that has documented the opposite — belongs to the criminal-law reviewer and nobody in-house can settle it.
Addendum — the criminal topic’s missing shelf, 25 August 2026
Not a finding of this sweep. It comes from a vocabulary audit run two days later: the lawyer-written documents in the working library were n-grammed against all files in the repository, and a group of case names every criminal-law reviewer would reach for came back zero. The method is recorded in the diary; the tool is check_vocabulary.py.
What was absent. MacDonald & Watson. Johnson & Towers. Iverson, except in passing. Hanousek. Jewell. Global-Tech. Bank of New England. Ahmad. Respondeat superior, collective knowledge, willful blindness and conscious avoidance — none of them present in a repository whose central offense is a knowledge-and-authority offense.
Why it matters to this topic specifically, and not as a matter of presentation. This sweep graded SEC. 6(a) fatal and drafted Amendment 8, whose Operation 4 proposes that evidence of responsibility and authority “is sufficient to warrant a finding of practical power.” That is Park’s burden structure and it is right for the base tier. It is the precise move United States v. MacDonald & Watson Waste Oil Co., 933 F.2d 35, 55 (1st Cir. 1991) forbids where knowledge is an express element — which is what SEC. 6(b)(1) makes it. So the cure this sweep drafted works at the misdemeanor tier and fails silently at the felony tier, and the internal review’s own Held open paragraph half-saw it, calling SEC. 6(b)(1)’s “knowingly” undistributed without knowing there was a case on the point.
The repair is Amendment 22, and the answer comes from the same line rather than from us. United States v. Iverson, 162 F.3d 1015 (9th Cir. 1998) — ✅ read in the opinion 25 August 2026, pincite unconfirmed: the responsible-officer instruction “relieved the government only of having to prove that defendant personally discharged or caused the discharge of a pollutant. The government still had to prove that the discharges violated the law and that defendant knew that the discharges were pollutants.” Responsibility replaces the act element, not the knowledge element.
And the objection this topic most needs is now on the source library too. United States v. Ahmad, 101 F.3d 386 (5th Cir. 1996) holds that the § 1319(c)(2)(A) offenses before it are not public welfare offenses. ✅ Read in the opinion 26 August 2026, which corrected two things this sweep had published: the ground is mistake of fact, not the penalty, and the penalty is offered as what “confirms our view” (E56). Training and deploying a model is ordinary commercial activity. If a court took that view, the public-welfare framing would not carry SEC. 6(b) at all, and the felony tier would need a conventional mens rea rather than a relaxed one. That is a sharper attack than anything this sweep produced.
Answered 26 August 2026, and the sentence above overstated the difficulty. ✅ United States v. Weitzenhoff, 35 F.3d 1275, 1286 n.7 (9th Cir. 1993) (as amended 8 Aug. 1994), read in the amended opinion in a star-paginated copy: “While the Staples opinion expresses concern with this evolution of enhanced punishments for public welfare offenses, it refrains from holding that public welfare offenses may not be punished as felonies.” Ahmad’s argument is that Staples settled the point; the Ninth Circuit reads the same passage as expressly refusing to. Hanousek, 176 F.3d 1116, 1122 n.4, then rejects the penalty argument by name. What survives is a circuit split, which a state legislature can be told about, and not an unanswerable objection. The full-strength version of the objection is the five-judge dissent from rehearing en banc in Weitzenhoff at 1293–1299, and the criminal-law reviewer should be handed that rather than Ahmad’s summary of it. See Amendment 22 and Amendment 24.
And the method point this correction earns. “Nobody in-house can settle it” was written about a case whose answer was one citation away inside a case the repository already held. The sentence should have read: nobody in-house has read enough to settle it. Those are different claims, and only one of them is checkable.
The method point, which belongs here rather than in the cure. The five topics did not find this, because none of the five was asked what a specialist would look for and fail to find. Asking what is absent is a different instrument from asking what is wrong, and on its first run it produced a front-page erratum (E42) and this addendum.
⚠ Every authority named above is quoted from a secondary source and unread in the reporter. E22 governs: none may be described as verified.