# ADVERSARY REPORT — Series B Brandbook (BRANDBOOK.md)

Adversarial review, 2026-07-10. Three hostile readers against `BRANDBOOK.md`,
grounded strictly in `SOURCE-DOSSIER.md` (fact base) and `SKELETON.md` (rules).
Every finding carries an exact quote. Findings that could not be grounded in a
quote were discarded.

Scope note: the document's own open items — the T4 depth dial, the [FILL] slots
on slides 8 and 11, and the ⑧ #21 derivation task — are honestly flagged in the
text and are not counted as findings. They do, however, gate print by the
document's own rule ("must close before print").

---

## PERSONA 1 — The diligence associate

Method: every factual claim checked against the dossier for (a) presence,
(b) citation, (c) estimate/belief labeling; every number checked against §2 for
value and canon-vs-backup placement; every regulatory date checked; the three
boldest claims attacked directly.

**What holds.** Almost everything. All of the following were verified exact
against the dossier, correctly cited, and correctly placed: 5.62M / +8.2% /
~6.3M annualized / 450–520k-month (§2c); 30.4M + $1.8T; 72.9M; Stripe Atlas
+130% / 100k+ / 140+; $25.2B→$47B; LegalZoom $756.0M; Delaware 2.28M+ / 66.7% /
334,461 / $1.8–1.9B; Estonia 132k+ / 5,556 / €125M; LEI 3M+; $2B registered
agents; the full payment-rails list with dates; Gartner 90% / $15T; McKinsey
$3–5T; $7.84B→$52.6B / 46.3%; Salesforce ~20% / ~$67B / $262B; Adobe +1,200% /
31%; EU AI Act Art. 50 → 2026-08-02; AB 316 → 2026-01-01; AILD withdrawal →
2025-02-11; FSB Jun 2026; NIST Mar 2026; FINRA 2026; UETA §14 / 49 states + DC /
E-SIGN 2000; Restatement §1.04 cmt. e "at present"; Wyoming DUNA 2024-07-01 /
DUNI ~$4B / AL+WV Apr 2026; *Samuels v. Lido DAO*; Project Vend, Cloudflare/
Perplexity, *Moffatt*; Skyfire $9.5M; Catena $30M + OCC; FIS Jan 2026. The
figure canon (1.2B / ~400M, founder-estimate labels) leads only in ② and slide
7 as required; the backup set (1B+ / 359M+ / 70M+) appears only as an attributed
backup line, never mixed into the canon visual. The forbidden claims in dossier
§7 (Argentina/Milei, the 20%-storefronts misquote, x402 169M, NH DAO statute,
unlabeled ClawBank) are all correctly absent or absent entirely.

**Falsification of the three boldest claims:**
1. *"Every jurisdiction has a company registry. The internet doesn't."* —
   survives: a Husam-picked verbatim hero, rhetorical absolute by design.
2. *"If Entity.ID the company disappeared tomorrow, the records would still
   exist and still verify"* — survives: grounded in dossier §4 (Ethereum
   anchoring, append-only, open naming standards).
3. *"Entity.ID is the only system already holding the answer"* — survives on
   the dossier's own logic (§5.2: every KYA competitor chains agents to a
   pre-existing entity; "none makes the agent a party"), but see NIT below —
   the "only" rests entirely on an internal research memo.

### Findings

**D1 · BLOCKER — an invented statistic on slide 2.**
> "Record formation: ~5.62M US business applications in 2025, a fifth straight record year (Census BFS)."
Why: the dossier says record volume in 2025 and "450–520k/month for five
straight years" (§2c, §5.1.1). It nowhere says each of the last five years set
a record — that is a different, stronger, checkable claim, and BFS annual
totals were not monotonic over 2021–2025. A diligence associate falsifies this
with the Census website in five minutes, on the slide whose whole job is "why
now." ⑧ #1 states the same fact correctly ("450–520k per month for five
straight years"), which makes the slide-2 morph look like carelessness.
Minimal fix: replace "a fifth straight record year" with "450–520k applications
a month for five straight years" or simply "a record year."

**D2 · SHOULD-FIX — the T4 moat passage overclaims past the dossier.**
> "The registry cannot be bought, rewritten, or shut off."
Why: the dossier supports record-level persistence ("If Entity.ID disappeared,
the records would still exist and verify," §4). "Cannot be bought" is a claim
about the company/protocol governance, not the records — the company, brand,
frontend, and (depending on upgradeability) contracts can absolutely be bought.
An associate who asks one question about contract admin keys turns this
sentence into a credibility hit inside the passage that carries the whole moat.
Minimal fix: scope it to the record — "The record cannot be rewritten or
switched off" — and let the preceding sentence carry the survivability claim.
(Passage is already flagged for Husam's dial; fold this into the dial.)

**D3 · SHOULD-FIX — slide 9 introduces a business-model element with no basis
in the fact base.**
> "Token/credit variant flagged as an option, not plan of record."
Why: no token or credit model appears anywhere in the dossier. The dossier's
rule is explicit: "If a fact is not here, do not use it." Flagging it as an
option doesn't cure the sourcing problem — it introduces the word "token" into
a Series B argument the rest of the book works hard to keep protocol-hinted,
and invites exactly the "is this a crypto deal?" question the register system
exists to control.
Minimal fix: delete the bullet, or move it to an internal appendix note
sourced to an actual plan document.

**D4 · SHOULD-FIX — a labeled belief asserted as fact in ③.**
> "Handshake equity kills more startups than competitors do; Entity.ID makes the split a recorded fact on day one instead of a fight in month eight."
Why: the dossier and ⑧ #19 both label this exact line "internal — a belief,
not a statistic." In ③ it runs as a flat declarative — statistic-shaped, no
hedge. An associate cross-reading ③ against ⑧ sees the same sentence labeled
"belief" in one chapter and asserted in another.
Minimal fix: two words in ③ — "We think handshake equity kills more startups
than competitors do" (or "In our experience, …").

**D5 · SHOULD-FIX — a naked claim in the spoken interior beats.**
> "Thirty million US businesses have zero employees; most will never incorporate."
Why: the first clause is dossier-cited (30.4M nonemployers). The second clause
— "most will never incorporate" — appears nowhere in the dossier and is doubly
loose because a nonzero share of nonemployer businesses are already
incorporated (S-corps). This is a beat meant to be spoken in a partner
meeting, i.e., the exact place a naked statistic gets challenged live.
Minimal fix: cut the second clause, or restate as the cited fact only
("Thirty million US businesses have zero employees").

**D6 · NIT — verify the Hansmann & Kraakman quotation marks before print.**
> "asset partitioning … is 'the one thing contract alone cannot replicate — it takes a registry.'"
Why: the dossier carries this in quotes sourced to Yale L.J. 2000, so it is
grounded for this review — but the phrasing ("it takes a registry") reads like
research-memo gloss, not law-review prose. If the quotation marks are printed
and the string isn't in the paper, that's a bad diligence moment on the thesis
chapter's load-bearing citation. Verify verbatim or drop to paraphrase.

**D7 · NIT — ⑧ #15's citation is not pullable.**
> "*(industry telemetry, per research dossier)*"
Why: every other Ammunition bullet names a source an associate can retrieve
(Census, SEC, GLEIF, Gartner…). "Industry telemetry" is the one citation in
the appendix that cites the dossier citing nothing in particular. Name the
telemetry (e.g., browser telemetry / Let's Encrypt stats) or relabel the
bullet as an analogy rather than a proof point.

**D8 · NIT — the "only system" close is diligence bait.**
> "Entity.ID is the only system already holding the answer."
Why: grounded (dossier §5.1.7 carries it as a positioning line, and §5.2's
competitor contrast supports it), but "only" claims get tested first. Keep it,
but make sure the §5.2 contrast (Google proves authorization, Cloudflare
proves operator, IAM proves login — nobody proves the agent is someone) is
what the speaker says immediately when challenged. No text change required.

---

## PERSONA 2 — The skeptical partner

**What holds.** The cone does not collapse: the "why this wedge" paragraph in
③ does its assigned job (largest / least-servable / cheapest-to-acquire /
seed-crop), and "the registry acquiring its entries in formation order" is a
genuinely good answer to "huge vision, tiny product." The honesty moves — no
liability shield, no tax personhood, banks still want a registered company,
bear cases carried openly in ⑧ #20, the derivation confession in #21 — are the
document's strongest credibility assets. The personhood line is walked
carefully: "Not rights for AI," "recourse, by construction," bans-as-moat —
all inside dossier §7's guardrails, with no personhood vocabulary leaking into
[site] copy. The agent chapter is the best chapter in the book.

### Findings

**P1 · SHOULD-FIX — slide 9 contradicts itself in one line.**
> "Free to form — acquisition at the moment of intent; fees sit at the moments of proof: registration and renewal, operations take-rate, verification/API."
Why: formation *is* registration everywhere else in the book ("Registration is
the product," ①). So slide 9 reads: registration is free, and fees sit at
registration. A partner catches this in real time and the business-model slide
— already the thinnest slide — loses the room. Presumably "registration and
renewal" means paid jurisdictional/incorporation events, but the slide doesn't
say so.
Minimal fix: rename the fee moment — "incorporation and renewal" or
"registration upgrades and renewal" — so the free thing and the paid thing
have different names.

**P2 · SHOULD-FIX — the round is priced on the wedge, and the wedge's revenue
logic is the one unanswered question.**
> "Wedge revenue comp: online legal services ~$25B → ~$47B by 2030."
Why: the argument structure is: product is free (③), the comp is a ~4%-CAGR
services market (⑧ #4), and the monetization slide is vague and internally
contradictory (P1). Meanwhile the ceiling story quotes $15T. A partner's
obvious question — "who pays you, when, and how much, before the agent economy
arrives?" — has no crisp sentence anywhere in the book. The skeleton says the
wedge funds the round; the book never shows the wedge earning a dollar.
Minimal fix: one sentence on slide 9 or in ③ tying a named fee to a named
moment (e.g., what the operations take-rate applies to, or what a verification
lookup costs whom), even as a labeled illustrative range. No new facts needed
— just resolve the mechanism.

**P3 · SHOULD-FIX — "standing" is never reconciled with ③'s own honesty
caveat.**
> "Entity.ID makes the honest version of that sentence true." / "until incorporation there is no liability shield and no tax personhood"
Why: ③ concedes a registered venture has no legal personhood; ④ then says
registration confers "standing" on agents and makes *Moffatt*'s "separate
legal entity" sentence "true." A partner who reads both chapters asks: standing
in what sense? Not legal standing — the book just said so. The hedges that
exist ("recourse, by construction," "not rights for machines") gesture at the
answer but never state it.
Minimal fix: one clause in ④ defining the term — standing = a checkable
registry status with a disclosed accountability chain and an agreed recourse
path, not legal personhood — and soften "makes the honest version of that
sentence true" to match (e.g., "gives that sentence a referent").

**P4 · SHOULD-FIX — the "come on" close on slide 12.**
> "Whoever holds the registry of standing holds the choke point of the agent economy … Own the category before the first systemic agent-commerce incident makes it mandatory."
Why: after eleven slides of unusually evidence-dense argument, the close
stacks an empire claim ("choke point"), a fear clause ("systemic incident"),
and a FOMO imperative ("own the category before") in two lines. This is the
single most likely passage to draw a "come on" — and it's the last thing the
partner hears. The dossier's moat logic (registries compound; two-sided
lock-in) supports a stronger, calmer version of the same close.
Minimal fix: keep "Fund the registrar. Registries compound; the second-place
registry is an export table" as the close; cut or subordinate the choke-point
and incident sentences.

**P5 · SHOULD-FIX — the traction slide leads with a number that isn't
traction.**
> "2,935,661 entities on the register, claimable (internal count; 'claimable' defined on the slide)."
Why: the product story is "we just shipped formation," and slide 8 opens with
2.9M entities. Every partner will ask where they came from, and "claimable"
(pre-populated, not user-registered) is the answer that makes the headline
number feel inflated in retrospect — the exact opposite of the e-Residency
trust move the slide invokes. The dossier permits publishing the counter; the
problem is ordering and emphasis on a slide titled "Traction."
Minimal fix: lead the slide with the [FILL] real product metrics; run the
claimable counter second, as the registry-seeding story, with "claimable"
defined in the same breath. (Come-on passage #2; #3 is D2's "cannot be
bought.")

**P6 · NIT — the endgame class never returns.**
> "**Class two: existing companies — the endgame.**" vs slide 10: "Roadmap as roadmap: incorporation partners, KYA verification API, agent-run ventures as standard entries."
Why: ② names 400M existing companies "the endgame," but the roadmap slide has
no item that serves them (computationally-binding conversion of existing
agreements never reappears after ②). A partner tracking the three classes
notices one of them has no path. Minimal fix: one roadmap phrase ("existing-
entity onboarding" or similar), or stop calling class two "the endgame" and
call it what the roadmap implies — the horizon.

**P7 · NIT — the table undercuts its own backup figure three lines early.**
> "Uncounted — the class is too young for honest external figures" … then: "70M+ autonomous economic actors — cited as the company's own figures"
Why: "no honest figures exist" followed by the company's own 70M+ implies the
company's figure is one of the dishonest ones. The intent (external vs
internal) is recoverable, but the word "honest" does friendly fire.
Minimal fix: "too young for external figures" — delete "honest."

---

## PERSONA 3 — The brand police

**What holds — verified, not assumed.** Every word count was counted:
one-liners all ≤12 (max found: 12); investor short blurb 71w, site short blurb
73w (both inside 60–75); ALT boilerplate 101w; investor pitch 87 spoken words
as stated, site pitch 73. The trimmed investor pitch carries its required
[trimmed] mark with the trim math shown. KYA is defined at its first
occurrence in the document (executive summary #5) and re-defined in ② and ④.
Zero exclamation marks. No "isn't just," no empty superlatives. Every
[site]-labeled block was scanned against the banned list: no blockchain,
on-chain, crypto, web3, Ethereum, IPFS, smart contract, wallet, DAO, immutable,
trustless, personhood, proto-entity-in-hero, or RegistryChain anywhere in
[site] copy ("the Entity.ID protocol, an open registry standard" is the
site-permitted formulation and is used correctly). CTAs verbatim in both
locations. All eight Husam-picked taglines verbatim — with one exception,
below. The ⑤ vocabulary table preserves all 16 skeleton seed rows unaltered
and extends to 33 rows without contradiction.

### Findings

**B1 · BLOCKER — a Husam-picked tagline altered without a mark, and the
alteration changes the claim.**
> Hero A variant: "On the internet, existence is a record anyone can check."
Why: the pick (brief §6, skeleton constraint 5 "use as written, do not
paraphrase") is "On the internet, existence **should be** a record anyone can
check" — and the book itself uses the correct form twice (① and the taglines
list). The variant silently drops "should be," which does two things: violates
the verbatim-picks hard rule with no [trimmed]/[variant-of] mark, and converts
an aspiration into a false present-tense assertion — on the internet today,
existence is *not* a checkable record; that is the entire thesis of the
company. One word, but it is the hard rule the constraints card exists to
enforce.
Minimal fix: restore "should be," or mark the line as a deliberate derived
variant of the pick.

**B2 · SHOULD-FIX — an investor-labeled asset breaks the ⑤ naming law.**
> "[ALT — agent-thesis investors; pure v4]" … "Built on the Entity.ID protocol, an open registry standard."
Why: ⑤, in the same document, states the law: "In investor materials the
technical layer is 'built on the RegistryChain protocol'; on the site the
layer is 'the Entity.ID protocol.'" The ALT boilerplate is explicitly an
investor asset and closes with the site formulation. Chapter ⑤'s own warning
applies: a registry that can't keep its own vocabulary straight — inside one
document — invites exactly that reading.
Minimal fix: change the ALT's final sentence to "Built on the RegistryChain
protocol, an open registry standard."

**B3 · NIT — site/press boilerplate runs long.**
> "Entity.ID is the registry where ventures begin. … Learn more at entity.id."
Why: 109 words against the ~100 target (the ALT sits at 101). Inside tilde
tolerance arguably, but it's the primary press boilerplate — the one asset
that gets pasted unedited. Trim ~9 words (e.g., "— with no lawyers and no
filing fees to start" → "— no lawyers, no filing fees to start").

**B4 · NIT — the boilerplate's investor-variant instruction doesn't parse.**
> "*[investor variant: the final two sentences read "Built on the RegistryChain protocol, an open registry standard."]*"
Why: the final two sentences are "Ventures run on the Entity.ID protocol…"
and "Learn more at entity.id." The instruction replaces two sentences with
one and doesn't say whether "Learn more at entity.id" survives. Whoever
executes the swap will guess. Minimal fix: "the final sentence pair becomes:
'Built on the RegistryChain protocol, an open registry standard. Learn more
at entity.id.'" (or explicitly drop the URL).

**B5 · NIT — antithesis density.**
> "an institution, not an app" · "a moat, not an obstacle" · "an upgrade, not a rebuild" · "a step, not a starting point" · "not the price of existing" · "Not a placeholder: a registry entry" · "inevitable, not small" (et al.)
Why: the "X, not Y" figure appears 12+ times across the book. Several are
Husam picks and untouchable ("a feature, not the front door"), and each
instance is individually fine — but at this density it reads as a template,
which is adjacent to the AI-slop pattern the register system bans. Minimal
fix: in the next editing pass, recast two or three of the non-pick instances
as plain declaratives (slide 12 and ③'s pillar 3 are the easiest hosts).

**B6 · NIT — the two figure sets share a sentence in ⑧ #21.**
> "Source the ~359M legal-entities figure externally … and publish a 1.2B derivation sketch before this document or the deck ships."
Why: constraint 4 — "Never mix the two sets in one sentence." This is a
process note about the figures rather than a claim using them, so it's
pedantry — but the constraints card says "no exceptions," and the brand
police's job is pedantry. Minimal fix: split into two sentences.

---

## VERDICT

**SHIP-WITH-FIXES.**

The document largely holds — and holds under real pressure. The fact base is
handled with unusual discipline: of the several dozen checkable figures and
dates, all but one match the dossier exactly, with correct citations,
correct estimate labels, and correct canon/backup placement; the §7 forbidden
claims are all absent; the register system survives a word-by-word scan of
every [site] block; all length limits pass an actual count. The two BLOCKERs
are both one-line fixes (a slide-2 statistic the dossier doesn't support, and
a one-word unmarked alteration of a Husam pick). The SHOULD-FIX cluster
concentrates in two places: slide 9 (the business-model slide is the weakest
artifact in the book) and the standing/traction framing a partner will probe.

Independently of this review, the document's own gates still apply before
print: close ⑧ #21 (derivation + 359M external source), fill slides 8 and 11,
and have Husam set the T4 dial.

**Counts: BLOCKER 2 · SHOULD-FIX 11 · NIT 9.**

| Severity | ID | One line |
|---|---|---|
| BLOCKER | D1 | Slide 2's "a fifth straight record year" is not in the dossier and is falsifiable against Census BFS. |
| BLOCKER | B1 | Hero A variant silently rewrites the picked tagline "should be a record" as "is a record" — unmarked, and factually wrong. |
| SHOULD-FIX | D2 | "The registry cannot be bought, rewritten, or shut off" overclaims past the dossier's record-persistence fact. |
| SHOULD-FIX | D3 | Slide 9's token/credit variant has no basis in the fact base. |
| SHOULD-FIX | D4 | "Handshake equity kills more startups than competitors do" runs as fact in ③, labeled belief in ⑧. |
| SHOULD-FIX | D5 | "most will never incorporate" is a naked claim in a spoken beat. |
| SHOULD-FIX | P1 | Slide 9: registration is free and fees sit at registration — same line. |
| SHOULD-FIX | P2 | The wedge's revenue mechanism is never stated; the round is priced on it. |
| SHOULD-FIX | P3 | Agent "standing" never reconciled with ③'s no-legal-personhood caveat. |
| SHOULD-FIX | P4 | Slide 12's choke-point/fear/FOMO close is the book's likeliest "come on." |
| SHOULD-FIX | P5 | Slide 8 leads "Traction" with 2.9M claimable — pre-seeded, not earned. |
| SHOULD-FIX | B2 | Investor-labeled ALT boilerplate uses the site-register naming ("Entity.ID protocol"), breaking ⑤'s law. |
| NIT | D6 | Verify the H&K "it takes a registry" quotation verbatim before printing quote marks. |
| NIT | D7 | ⑧ #15's "(industry telemetry)" citation isn't pullable. |
| NIT | D8 | "The only system already holding the answer" is diligence bait; know the §5.2 contrast cold. |
| NIT | P6 | Class two is "the endgame" in ② but absent from the roadmap slide. |
| NIT | P7 | "too young for honest external figures" friendly-fires the company's own 70M+ backup figure. |
| NIT | B3 | Site/press boilerplate is 109 words against ~100. |
| NIT | B4 | The investor-variant swap instruction replaces "two sentences" with one and orphans the URL. |
| NIT | B5 | "X, not Y" antithesis appears 12+ times; thin it where the lines aren't picks. |
| NIT | B6 | ⑧ #21 mixes the canon and backup figure sets in one sentence (constraint 4). |