NYSE proposes to amend Sections 303A.00 and 303A.07 of the Listed Company Manual to extend the period in which a newly listed company must establish an internal audit function from one year to five years. This tracker classifies the comment letters filed on that proposal.
On July 31, 2026, the New York Stock Exchange (NYSE) filed a proposed rule change under Section 19(b)(1) of the Exchange Act to amend Sections 303A.00 and 303A.07 of its Listed Company Manual. The amendment would extend the transition period in which a newly listed company must establish an internal audit function from one year after listing to five years. The Commission published notice on August 13, 2026 (Release No. 34-106128; 91 Fed. Reg. 53,448).
What NYSE argues. First, that the one-year deadline is hard to meet: issuers have told NYSE they are handling competing business and regulatory obligations while upgrading accounting systems and internal controls and hiring staff, and building a capable function in that first year is difficult. Second, that waiting produces a better function rather than merely a later one — NYSE's argument is about the board, not the calendar: a newly listed company has a newly seated slate of directors, and it takes time for an audit committee to understand the business well enough to design an internal audit function worth having. Third, that other requirements cover the gap in the meantime: an audit committee of at least three independent directors under a written charter, management's assessment of internal control under Section 404(a) of the Sarbanes-Oxley Act, the auditor's attestation under 404(b), and the CEO and CFO certifications under Sections 302 and 906. NYSE notes that Sarbanes-Oxley post-dates its own internal audit requirement, and treats internal audit as a supplement to those statutory protections rather than the other way round. Finally, that Nasdaq requires no internal audit function at all, so a company could list there with none — which, on NYSE's reading, makes a five-year transition no cause for concern. NYSE solicited and received no comments before filing.
This is a self-regulatory organization rule filing, not a Commission rulemaking. It poses no numbered questions, and the Commission must within 45 days of publication — extendable to 90 — approve the change, disapprove it, or institute proceedings under Section 19(b)(2)(B) to determine whether it should be disapproved.
Sarah McVay helped significantly in building the classification scheme.
I also thank Mert Erinc for comments and suggestions.
Further feedback is welcome at zach.7@osu.edu.
The SEC's May 2026 proposal would let public companies move from quarterly Form 10-Q reporting to semiannual reporting on a new Form 10-S. The largest of these dockets by far, tracked the same way as this one, with a stance, a commenter type and a set of rationales recorded for every letter.
Open tracker →The SEC's May 19, 2026 companion proposal eases the rules for smaller and newly public companies, and drops the requirement that they have an outside auditor attest to their internal controls. Tracked with a stance on each of its eleven elements.
Open tracker →Also in May 2026, the SEC proposed rescinding its 2024 climate-related disclosure rules in their entirety. Tracked too, though the classification is simpler: a rescission is a single up-or-down stance, not a multi-element proposal.
Open tracker →The SEC's August 2026 proposal would build a new offering framework for crypto assets. Mert Erinc tracks that docket the same way as this one, with a stance on each of its seventeen elements. Comments close October 20, 2026.
Open tracker →Majority of three raters. Support = backs the proposal (five years); Oppose = keep the one-year rule, or otherwise object to the extension. A writer who asks for a shorter extension is an Oppose letter whose remedy records the period sought — see "Asks for" on the letters page.
The docket is not unanimous. One letter backs the change — #9, John Heagy, who argues the cost of the function outweighs its value at a newly listed company and adds that five years may itself be too long. 3 letters take no position on the five-year period: #1, #3, #88. 2 of them ask only for a longer comment period, and the third offers the Commission an assessment framework instead of an opinion on the transition period. Among the 123 opposing letters, 16 ask the Commission to institute proceedings under Section 19(b)(2)(B) and 13 ask for a longer comment period. Both asks appear only on letters written from scratch, never on one built from a shared template.
Entity majority call. Where a writer works in internal audit, the function decides the bucket, not the employer.
Multi-select; each code swept 1/0 on every letter. Hover a code for its definition, or open the methodology below for the full list.
Assigned mechanically, before any letter was classified. The 108 letters that match nothing are not charted.
Family A is the SEC's own designation: three letters filed as verbatim copies of one another, which the docket collapses behind a single entry — read the Type A letter on sec.gov. A pair match is two letters that are near-copies of each other but belong to no larger family; there are three such pairs on this docket, each matching at better than four fifths of its phrasing.
Family B is not designated by the SEC. Eighteen letters share a fixed opening frame — "I am an internal audit professional with [N] years in the profession, and I hold the [credential] designation. I write in my personal capacity. My views are my own and not those of my employer." — with original bodies after it. Verbatim matching misses it; whole-document TF-IDF cosine misses it. Only anchor-phrase detection finds it.
All three votes are retained on every judged field. A call is unanimous when the three raters agree, majority when two of three agree, and split when all three differ — in which case the Primary rater's call is stored and the record is flagged for review.
| Field | Calls | Unanimous | Majority | Split | Not unanimous |
|---|---|---|---|---|---|
| Position | 127 | 119 | 5 | 3 | 6% |
| Entity | 127 | 104 | 23 | 0 | 18% |
Every three-way position split — 3 of them — is a letter that never states what the Commission should do. The Literalist reads Oppose because the status quo is one year, the Charitable rater reads Support because nothing in the letter objects, and the Primary rater reads No position, which is the stored call. The 5 majority position calls are all the same 2-1: the Charitable rater alone reads the letter the other way. That is the third rater doing its job rather than a contested record. 19 of the 23 non-unanimous entity calls involve the internal audit bucket: a writer who works in internal audit but does so as a CPA, at an advisory firm, or without naming an employer, so the function and the affiliation point to different buckets. The rule is that the function wins.
Position — what the writer asks the Commission to do about the proposal. One of Support, Oppose, No position or Off-topic.
The proposal is a relaxation of a standard that already exists, so the labels run the way the proposal runs, not the way the comment volume runs: Support means backing the change to five years, and Oppose means asking the Commission to keep the existing one-year rule, or objecting to the extension on any ground. A writer whose first choice is a shorter extension is opposing five years, so the position is Oppose and the period they name is recorded under remedy.
Entity — who is writing. The scheme offers 11 buckets; 8 of them are occupied on this docket. Where a writer works in internal audit, the function decides the bucket rather than the employer: an internal audit manager at a listed issuer is an internal audit professional, not an issuer.
Arguments — which of the 15 reasons listed below the letter actually makes. A letter can carry any number of them, including none.
Recorded once, without the ensemble: the remedy and the period or schedule it names, the procedural ask, the internal-audit sub-role, and a 0–3 specificity score. Extracted rather than judged: dates, word counts, source links, and the form-letter family.
Procedural asks are coded independently of position, and only where the letter actually makes the request: PROC_EXTEND, a longer comment period, and PROC_PROCEEDINGS, proceedings under Section 19(b)(2)(B). A letter can ask for one of these and take no position at all.
Position, entity and every argument code are decided by an ensemble rather than by one reading. The same letter is read three times under three different sets of instructions, and the majority of three is the stored value.
All three votes are kept and published, not just the winner. Each call is marked unanimous when the three agree, majority when two of three do, and split when all three differ — in which case the Primary reader's call is stored and the record is flagged. No value is ever overridden by hand.
The arguments are decided by a per-code sweep: for every letter the full list is walked and each code is answered yes or no on its own, by all three readers. Codes are never free-listed from memory, which is how a reader misses the argument they were not thinking about.
The three differ by what they are willing to infer. Each field gets the trio that stresses it.
Every reader reads the whole letter. Each record's word count must equal the word count of the stored letter body, and a file that cannot be read in full is a stop rather than a partial read. Letters in this docket routinely open with a procedural request and object on the substance only later in the body.
Evidence is kept separate by what it supports. Every coded argument carries a verbatim quote from the letter — an exact substring, never a paraphrase. The position and each procedural ask carry their own quote, so the words a writer uses to state an ask are never pooled with the words used to argue the merits. A letter that argues a case but never states an ask is marked as such rather than given a stand-in quote; 23 of the 127 coded letters are in that position. Every quote and every rater vote is on the evidence page.
Form letters are clustered before anything is classified, because a shared scaffold is invisible in any single letter and only appears across the corpus. Verbatim families are reported as one entry with a signatory count; scaffold families keep their letters as separate rows, because their bodies differ.
13 letters were coded twice, in two separate runs under the same rules, with the second run given no access to the first. Position agreed on 13 of 13, entity on 13 of 13, and the remedy field on 11 of 13; the argument codes overlapped at a mean Jaccard of 0.97, comparing like with like — one code was renamed between the two runs and two were added after the first, so the rename is followed through and the later codes are held out of both sides. The disagreements are in the remedy field, where the same letter can be read as asking to keep one year or as requesting no modification at all. That is the measured reproducibility of the scheme, not a claim about its accuracy.