The method of evaluation · version 2 · 18 September 2026

Every number on this site, how it is computed, and where it comes from.

Warrant Wire measures documents. It does not judge companies. Each measurement below is computed by code that runs the same way on every filer, with a published source behind each idea. This page gives the form of every calculation — what is measured, how the parts combine, and where each idea comes from. The lines, weights and word lists themselves are the house’s and are not published. Where this page and the code disagree, the code is wrong.

What is evaluated 1 · The wire 2 · The verdict 3 · The plain-English grade 4 · The reading engine 5 · How the page decides Sources Limits

What is evaluated, and what is not

Four things are measured, in this order, and each feeds the next: whether a filing carries the language of a warrant financing (the wire); what the pattern of those filings shows (the verdict); how the document is written against the SEC’s own plain-English rule (the grade); and what the writing does to a reader — how hard it is to comprehend, how much of a reader it spends before the clause that matters, and how much of it is written to be quoted (the reading engine).

Nothing here is a finding that a company or a person did anything wrong. Raising money with warrants is lawful; writing badly is not an offence the site can find; the grade is a measurement against a standard, and the reading measures are estimates of the demand a text makes on a general reader. The one conclusion the site draws is about the reader, not the security: do not buy what you cannot understand.

1 · The wire

Is this filing a warrant financing?

Every filing the SEC publishes in the forms we watch (8-K, S-1, S-3, 424B5, 424B3, 10-Q, 10-K, DEF 14A) is searched for a list of phrases the house maintains. A phrase carries a label (what it means) and one of three weights: a heavier term, an ordinary mark of a financing, or a qualifier — a phrase that matters only when warrant language is also present.

A filing f is on the wire iff ∃ phrase p in f with weight(p) > 0 heavy(f) = |{ p in f : p is a heavier term }| marks(f) = |{ p in f : weight(p) > 0 }|

The labels a reader sees on the wire: Price reset · Reduced exercise price · Warrant inducement · Inducement agreement · Cashless exercise · Variable rate transaction · Equity line · Pre-funded warrants · ATM programme · Ownership blocker · Floor price · Most favored nation · Participation right · Placement agent · Registered direct · and, from 18 September 2026, No press release clause, the qualifier. Which phrases produce each label, and their weights, are not published.

A match is not a finding. The words are ordinary and lawful, and many of the deals are unremarkable. It means the document is worth reading. The qualifier appears in ordinary merger and proxy paper, which is why it carries no weight of its own: a filing carrying only it is not on the wire.

2 · The verdict · rules version 2

What does the pattern of filings show?

The automatic verdict runs on what the wire holds for one company: n filings, h of them heavy, dated from first to latest, each carrying its labels. Every rule is written out in plain English on the rules page, which is the authority; this is the mathematics of each. A rule that cannot run for want of a figure is reported as missing, never skipped.

W0 nothing on the wire n = 0 → no verdict; not a clean bill of health W1 how often r = n / max(1, days(first, latest) / 365.25) r < 1.5 → "occasionally" 1.5 ≤ r ≤ 4 → "repeatedly" r > 4 → "serially" W2 the heavier terms h / n ≥ 0.5 W3 the price was cut |{ f : "Reduced exercise price" ∨ "Price reset" ∈ labels(f) }| ≥ 2 W4 paid to convert early i = |{ f : "Warrant inducement" ∨ "Inducement agreement" ∈ labels(f) }| ≥ 1; i ≥ 3 → "a habit" W5 the stack max over f of |labels(f) ∩ STACK| ≥ 3, STACK = {Pre-funded, ATM, Variable rate, Equity line, Price reset} W6 arranged to sell, not hold ∃ f : "Pre-funded warrants" ∈ labels(f) ∧ "Ownership blocker" ∈ labels(f) W7 how recent days(latest, today) ≤ 90 W8 where it is incorporated state ∈ {NV, DE, TX} each with what its law permits on a reverse split W9 no press release |{ f : "No press release clause" ∈ labels(f) }| ≥ 1 once is enough

The verdict sentence is assembled from the rules that fired, the company’s name in front of the parts that take one (W1, W3, W4, W9), the rest as sentences after it. It computes; it does not conclude. Part two of the verdict (dilution, splits, overhang, repricing, spend, the parties) runs against the hand-built record where one exists and is merged on the page.

3 · The plain-English grade · 17 CFR 230.421

How is the document written, against the rule the filer was bound by?

Every filing is measured on nine parameters. Each traces to a line of the rule: 421(a) — information shall not be set forth so as to obscure; 421(b) — clear, concise and understandable, short sentences, no legalese, no repetition; 421(d) — the front sections in plain English: short sentences, everyday words, the active voice, no legal jargon, no multiple negatives. Each parameter has a good line and a bad line — the rule gives the direction, the lines give one standard applied identically to every filer — and a weight. The lines and weights are the house’s.

for each parameter p with measured value v_p, good line g_p, bad line b_p, weight w_p: score_p = clamp_[0,100] ( 100 · ( 1 − (v_p − g_p) / (b_p − g_p) ) ) 100 at the good line, 0 at the bad line S = Σ_p w_p · score_p / Σ_p w_p the grade score, 0–100 worst = argmin_p score_p reported beside S: one catastrophic parameter cannot hide behind eight good ones

Parameter	Rule	Measures
Sentence length	421(b)(1), (d)(2)	average words per sentence
Long sentences	421(b)(1)	share of sentences past the house’s length line
Legalese	421(d)(2)	legalese markers per 1,000 words
Cross-references	421(a)	cross-references per 1,000 words
Exhibit depth	421(a)	pushes to another document per 1,000 words — how many documents a reader must hold at once
Defined terms	421(a), (d)(2)	uses of capitalised defined terms per 1,000 words — uses, not definitions
Passive voice	421(d)(2)	share of sentences in the passive
Multiple negatives	421(d)(2)	share of sentences carrying two or more negatives
Repetition	421(b)	share of substantive sentences appearing more than once

The bands

S ≥ 90 → A "You can read this once and know what you are buying." S ≥ 80 → B "You can read this, but you will work for it." S ≥ 70 → C "You will need to read this twice." S ≥ 60 → D "You will have to dig for the parts that matter." S ≥ 45 → F "RED LIGHT CAUTION. This one might kill your day." S < 45 → WORD SALAD "This cannot be read."

A company’s grade is not an average. A holder is exposed to the one document that decides what happens to his shares, so the headline is the worst document graded, with prospectuses (which 421(d) reaches) and exhibits (which it does not) reported separately. Which file was graded, and its word count, travel with every grade so a number can be checked afterwards. Documents under 120 words are not graded; a document short for its form is graded and flagged.

4 · The reading engine · added 18 September 2026

What does the writing do to a reader?

Three measures, computed on the same text as the grade. Each rests on published work, named beside it. Each is an estimate of the demand the text makes on a general reader — not a measurement of any actual reader, not a clinical instrument, and not a finding about the company.

4a · Comprehensibility — the published formulas

With W words, S sentences, Y syllables and P polysyllabic words (three syllables or more):

Flesch Reading Ease (Flesch 1948) FRE = 206.835 − 1.015·(W/S) − 84.6·(Y/W) Flesch–Kincaid Grade (Kincaid et al. 1975) FK = 0.39·(W/S) + 11.8·(Y/W) − 15.59 Gunning Fog (Gunning 1952) Fog = 0.4·( W/S + 100·P/W ) SMOG (McLaughlin 1969) SMOG = 1.043·√( P · 30/S ) + 3.1291 grade level G = ( FK + Fog + SMOG ) / 3 years of schooling = round(G) FRE ≥ 60 plain · ≥ 50 fairly difficult · ≥ 30 difficult (college) · ≥ 10 very confusing (graduate) · < 10 unreadable

The caveat the finance literature supplies, printed on every answer: Loughran & McDonald (2014) show that Fog over-counts long financial words that are not hard — corporation, liabilities — on SEC filings; Li (2008) nonetheless finds Fog and length predict worse outcomes for readers of annual reports. So the level is reported as an estimate from four formulas, with the caveat, and never as a measurement of a person.

4b · Cognitive load — how much of the reader the document spends, and where

Comprehension is capacity-limited (Just & Carpenter 1992); every open clause, every carried term and every unresolved reference costs working memory (Gibson 1998), whose span is about seven items (Miller 1956); the reader’s model of the text degrades as propositions per sentence rise (Kintsch 1998); and load that comes from how material is presented rather than what it says is extraneous (Sweller 1988). Seven components are measured, each scored from a plain line to a bad line, and the heaviest weight is on the depth to the material clause — how far into the document, by words, the first term that changes what the shareholder owns appears. A clause at 88% depth arrives after the reader’s attention is spent. That is the capture.

for each component c with value v_c, plain line g_c, bad line b_c, weight w_c: load_c = clamp_[0,100] ( 100 · (v_c − g_c) / (b_c − g_c) ) 0 at the plain line, 100 at the bad line L = Σ_c w_c · load_c / Σ_c w_c a component that cannot be measured is left out of both sums capture ∈ { low, moderate, high, severe } by rising bands of L; w_depth is the largest weight

Component	Measures	Source
Nested clauses per sentence	the embedding a sentence holds open — punctuation and subordinating clauses per sentence	Just & Carpenter 1992; Gibson 1998
Terms carried per sentence	defined terms per sentence — what the reader must keep in mind	Miller 1956
Deeply nested share	share of sentences past the house’s embedding line	Kintsch 1998
Words in long sentences	share of all words inside very long sentences	Just & Carpenter 1992
Cross-references per 1,000 words	jumps to another section or document — each a context switch	Sweller 1988
Negations per 1,000 words	negative constructions; multiple negatives are named in the rule	Rule 421(d)(2)
Depth to the material clause	the share of the document a reader must get through before the first term that changes what the shareholder owns	Sweller 1988; the wire’s own list

4c · The media trigger — how much of the paper is written to be quoted

A story written on deadline lifts the line that is already loaded. The measure counts loaded language by the finance-specific sentiment lists of Loughran & McDonald (2011) — positive, negative, uncertainty, strong modal — plus a list of hype words; then finds the quotables: sentences of six to twenty-eight words carrying a loaded word, which is exactly what gets pulled. Tone in company releases moves readers (Henry 2008); media tone moves markets (Tetlock 2007); vivid language moves investor judgement (Hales, Kuang & Venkataraman 2011). Word-list subsets are used; the full lists run to thousands of words.

densities per 1,000 words: pos, neg, hype, unc, strong D = pos + neg + hype loaded density Q = |quotable sentences| / S quotable share M = α · clamp_[0,100]( 100·(D − d₀)/(d₁ − d₀) ) + (1 − α) · clamp_[0,100]( 100·(Q − q₀)/(q₁ − q₀) ) leans: (pos + hype) > κ·neg → "up" ; neg > κ·(pos + hype) → "down" ; else "mixed" band ∈ { flat, low, notable, high } by rising bands of M α, the lines d and q, κ, the quotable length window and the word lists are the house's.

The quotable lines themselves are kept with the grade, with the direction each leans, so a report can show what a story would lift.

5 · How the page decides what to say

The decision, as a function.

Everything a company page says is a function of four inputs: the wire’s answer for the ticker, the verdict computed from it, the grade profile, and the reading measures. There is no hidden step. For a company T:

wire(T) = { n, h, first, latest, labels per filing, EDGAR count since 2001, latest readable filings } verdict(T) = rules W0…W9 applied to wire(T) section 2 grade(T) = { worst document: S, band; per-form averages } section 3 reading(T) = { G, L, capture, M, leans } on each graded document; company means; the heaviest and the most quotable section 4 page(T) is then decided by these tests, in this order:

Test	Condition	What the page does
real company?	the ticker or name resolves to a CIK	If not: "type a ticker or a company name." Otherwise the page never comes up empty: it shows the EDGAR count since 2001 and the latest readable filings, each with a $20 read at 8K10Q.
on the wire?	`n` = 0	W0: "No warrant language on the wire" — and the EDGAR record beneath it. Not a clean bill of health.
open?	`T` ∈ {TOVX, MRVL, GOOG} ∨ the reader’s email holds an entitlement for `T`	Everything is shown. Otherwise: the counts, the company, the terms and the number of rules fired are free; which document, which date, and the verdict sentence are the report.
the verdict	rules fired ≠ ∅	The sentence, the rules in order, the version and date they were rendered under.
the caution	grade.worst.S < 45 (WORD SALAD) ∨ < 60 (F)	"Do not buy what you cannot understand." About the reader, not the security.
on the list	grade.worst.band ∈ {F, WORD SALAD}	"On the list" of suspected Rule 421 violations; the letter to the SEC is drawn for this company.
the reading lines	reading measured on ≥ 1 document	Three lines under the grade: years of schooling (G); cognitive load (L) with the capture band and where the material clause sits; media trigger (M) with the lean.
going concern	open ∧ the record holds a going-concern paragraph	The urgent warning at the top of the report: who said it, when, and what the shareholders put in.

No step weighs one section against another to produce a single number for the company. That is deliberate: a company is not a score. The verdict says what the record shows; the grade says how it was written; the reading engine says what that does to you; the going concern record says what the auditor said. The reader puts them together. The site does not.

Sources

Every source, and what it is used for.

17 CFR 230.421 — Securities Act Rule 421: presentation of information in prospectuses; (a) not obscure, (b) clear, concise, understandable, (d) plain English in the front sections.the grade — every parameter traces to a paragraph
SEC Release 33-7497 (28 January 1998) — Plain English Disclosure, the adopting release; and A Plain English Handbook (SEC Office of Investor Education, 1998).the grade — the 20–25 word working target; the principles named in 421(d)(2)
Flesch, R. (1948). A new readability yardstick. Journal of Applied Psychology 32(3), 221–233.4a — Flesch Reading Ease
Kincaid, J. P., Fishburne, R. P., Rogers, R. L. & Chissom, B. S. (1975). Derivation of new readability formulas for Navy enlisted personnel. Naval Technical Training Command, Research Branch Report 8-75.4a — Flesch–Kincaid Grade Level
Gunning, R. (1952). The Technique of Clear Writing. McGraw-Hill.4a — the Fog index
McLaughlin, G. H. (1969). SMOG grading — a new readability formula. Journal of Reading 12(8), 639–646.4a — SMOG
Li, F. (2008). Annual report readability, current earnings, and earnings persistence. Journal of Accounting and Economics 45(2–3), 221–247.4a — that Fog and length on filings predict worse reader outcomes
Loughran, T. & McDonald, B. (2014). Measuring readability in financial disclosures. Journal of Finance 69(4), 1643–1671.4a — the caveat: Fog over-counts long financial words that are not hard
Sweller, J. (1988). Cognitive load during problem solving: effects on learning. Cognitive Science 12(2), 257–285.4b — cognitive load theory; extraneous load from presentation; the cross-reference and depth components
Just, M. A. & Carpenter, P. A. (1992). A capacity theory of comprehension: individual differences in working memory. Psychological Review 99(1), 122–149.4b — comprehension is capacity-limited; embedding and length components
Gibson, E. (1998). Linguistic complexity: locality of syntactic dependencies. Cognition 68(1), 1–76.4b — the cost of holding an unresolved dependency across intervening words
Miller, G. A. (1956). The magical number seven, plus or minus two. Psychological Review 63(2), 81–97.4b — the span; the terms-carried component
Kintsch, W. (1998). Comprehension: A Paradigm for Cognition. Cambridge University Press.4b — the reader’s situation model degrades as propositions per sentence rise
Loughran, T. & McDonald, B. (2011). When is a liability not a liability? Textual analysis, dictionaries, and 10-Ks. Journal of Finance 66(1), 35–65.4c — the financial sentiment word lists (subsets used)
Henry, E. (2008). Are investors influenced by how earnings press releases are written? Journal of Business Communication 45(4), 363–407.4c — tone in company releases and its effect on readers
Tetlock, P. C. (2007). Giving content to investor sentiment: the role of media in the stock market. Journal of Finance 62(3), 1139–1168.4c — media tone and the market
Hales, J., Kuang, X. & Venkataraman, S. (2011). Who believes the hype? An experimental examination of how language affects investor judgments. Journal of Accounting Research 49(1), 223–255.4c — vivid language and investor judgement
SEC EDGAR — full-text search (efts.sec.gov), the submissions API (data.sec.gov), the company tickers file, and the filings themselves.1 — the wire; the EDGAR count since 2001; every document graded
Nevada Revised Statutes §78.207; Delaware General Corporation Law §242(d) as amended August 2023; Texas Business Organizations Code as amended 2025.2 — W8, what each state permits on a reverse split

Limits and corrections

What this method cannot do.

It cannot read intent. It measures words, counts and positions, applied identically to every filer; it does not know why a document was written the way it was. It cannot find a violation: the grade is a measurement against 421, and whether a rule was broken is the SEC’s to decide. The reading measures estimate demand on a general reader from published formulas; they do not measure any actual person, and the formulas are known to over-count long financial words. The wire watches a fixed list of phrases; a financing described in words not on the list is not on the wire, and the list grows only by a dated change here.

CorrectionsEvery line, weight and band behind this page is a judgement, applied identically to every filer. A challenge to any measurement, with the document and the number, is answered at research@warrantwire.com; a change to the method is dated here and on the rules page before it runs. The lines, weights and word lists are the house’s and are not published.