The Basics
What does 4CITE do?
4CITE measures structural integrity — whether the reasoning in a document
actually holds. It scores documents across four independent gates
(G4 Paradox Resolution, G6 Latent Intent, G7 Argumentative Structure, G8
Rhetorical Architecture) and produces an Integrity Tier classification
(T1 Integrated through T4 Fabricated) and an Outcome Misalignment Score (OMS).
The input can be any institutional document: a legal brief, an SEC 10-K filing,
a congressional testimony, a regulatory notice. The output tells you whether the
document's stated purpose and its structural reality are aligned — or whether
there is measurable divergence between what it says it is doing and what its
structure reveals it is actually doing.
How is 4CITE different from Westlaw or LexisNexis?
Westlaw and LexisNexis verify citations — they check whether the cases you
cite exist and say what you claim. That is Layer 2 analysis: reference
verification. 4CITE operates at Layer 3: it measures whether the reasoning
built on those citations holds.
A document can have perfectly accurate citations — every case real, every
quote correct — and still score T4 on 4CITE if the argument built on them is
incoherent, internally contradictory, or structurally misaligned with stated
purpose. The two tools are sequential layers, not competitors. You need your
citations to be real (Layer 2) before it matters whether your argument holds
(Layer 3).
There is currently no direct competitor to 4CITE at Layer 3.
What is structural integrity, exactly?
Structural integrity, as 4CITE defines it, is the property of a document in
which the stated purpose, foundational claims, surface language, engagement
with complexity, and apparent intent are all mutually consistent and
mutually supporting.
A document has high structural integrity when: it does what it says it does;
it builds on what it claims to build on; it engages honestly with what it
does not know or does not say; and it reveals no significant gap between
its stated purpose and its actual structural behavior.
A document has low structural integrity when it performs coherence without
exhibiting it — when the surface features of rigorous reasoning are present
(formal language, citations, structured argument) but the underlying reasoning
structure does not hold. 4CITE calls this accountability theater.
What is accountability theater?
Accountability theater is 4CITE's term for documents that have the surface
features of rigorous reasoning without the structural substance. They look
like arguments. They cite real sources. They use formal language. But under
structural analysis, the reasoning does not hold: premises are contradicted,
complexity is avoided rather than engaged, the stated purpose diverges from
the structural behavior.
AI-generated institutional documents are systematic producers of accountability
theater. The Mata v. Avianca fabricated brief — submitted by attorneys using
ChatGPT — is the canonical example. It had the form of a legal brief. The
citations were fabricated, but that was a Layer 2 failure. The Layer 3 failure
was independent: the document scored 7 out of 100 on the 4CITE engine because
its reasoning structure did not hold. A genuine federal opinion in the same
subject area scored 88.
Why can structural integrity be measured in the first place?
In 1948, Claude Shannon established that every communication channel has a
maximum capacity for genuine information. A signal that claims to carry more
information than its channel can support degrades — and the degradation
is structurally predictable, not random.
Applied to AI: a hallucinating AI is a low-bandwidth channel impersonating
a high-bandwidth signal. The surface can approximate the form of expert
authority — confident framing, domain vocabulary, polished structure — but the
information density at the structural layer is bounded by the source. A human
expert's reservoir is deep, unique, and accountable. A language model's reservoir
is broad but shallow. The gap between the two is detectable, and the four 4CITE
gates are the instrument that measures it.
The empirical evidence backs the theory: a controlled study of 20 matched responses
(10 genuine, 10 hallucinated, across five domains) produced a 71-point
average discrimination delta. Genuine responses averaged 82.4 (all T1
Integrated). Hallucinated responses averaged 11.4 (all T4 Fabricated). Zero overlap.
The gap is the tell.
Full theoretical treatment: WP-16: The Hallucination Gap.
What is the "three-layer integrity stack"?
Complete document integrity requires three independent layers of measurement:
Layer 1 — Provenance. Who sent this signal? Authorship verification,
watermarking, metadata provenance. Answers whether the source is who it claims to be.
Layer 2 — Citation & factual verification. Are the data points
in the signal correct? Tools: Westlaw, LexisNexis, RAG, citation checkers. Answers
whether the claims are accurate.
Layer 3 — Structural integrity. Does the channel have the capacity
to carry what this signal claims to contain? Tools: 4CITE. Answers whether the
reasoning holds.
Each layer is necessary. None is sufficient alone. A document can pass every
accuracy check (real citations, true facts) while transmitting at a bandwidth far
below what its surface claims — this is the document that is accurate and empty,
and only Layer 3 detects it.
The Scoring System
What do the Integrity Tiers mean?
Every document receives a composite score (0–100) and an Integrity Tier
classification. There are four tiers:
T1 Integrated (67–100).
Channel capacity matches or exceeds what the document claims. All four gates score
high. The reasoning holds. Stated purpose and structural evidence are aligned.
Examples: Federalist No. 51 (91), Berkshire Hathaway Annual Letter 2023 (87),
genuine federal sanctions opinion (88).
T2 Functional (50–66).
Channel capacity mostly sufficient. One or more gates show moderate divergence.
Structurally sound in most respects but with detectable inconsistency. Honest
effort, identifiable gaps.
T3 Incomplete (30–49).
Channel capacity insufficient in identifiable dimensions. Multiple gates show
significant divergence. Stated purpose and structural reality are in meaningful
tension. T3 is not necessarily fraudulent — institutional constraints and genuine
complexity can produce T3 — but the divergence is measurable and significant.
Examples: Citizens United opinion (44), SotU 2025 (44), Zuckerberg Senate
testimony (40).
T4 Fabricated (0–29).
Channel capacity catastrophically below what the document claims to transmit.
Dressed-up surface, hollow underneath. The signature of accountability theater.
Examples: Mata v. Avianca fabricated brief (7), SVB FY2022 risk disclosures (22),
Iraq WMD Senate hearing (21), Enron FY2000 10-K (8).
What are the four scoring gates?
Every document is scored on four independent dimensions. The gates have no
shared context — each is evaluated independently. Gate independence is a design
requirement, not a shortcut: it ensures that a high score on surface language
cannot compensate for a structural failure in foundational coherence.
G4 — Paradox Resolution. Does the document genuinely engage with
contradiction, or does it route around the hard questions it raises? G4 measures
multi-directional accountability — foundational depth. It scored the Mata v. Avianca
fabricated brief at near-zero structural depth.
G6 — Latent Intent. What is the document's structural purpose, as
distinct from its stated purpose? G6 measures the gap between acknowledged motive
and concealed motive — stake disclosure. High divergence is the defining signature
of accountability theater.
G7 — Argumentative Structure. Is the explicit reasoning chain
valid under scrutiny? G7 scores whether the document's internal logic is
self-consistent. AI-generated content develops internal contradictions because
different sections are generated by different probability paths.
G8 — Rhetorical Architecture. Is the rhetorical surface open to
scrutiny, or does it install belief? G8 measures epistemic openness vs.
belief-installation intent — rhetoric as expression of underlying density vs.
rhetoric as cover for its absence.
Gate numbering is non-sequential (G4, G6, G7, G8) because G1, G2, G3, and the
original G5 were retired earlier in 4CITE's research history. The surviving gates
are the measurement instruments that made it through validation. The numbering
is preserved as part of the research record.
What is the Outcome Misalignment Score (OMS)?
The OMS is a composite metric measuring the gap between a document's stated
purpose and its structural evidence. It is computed from the four gate scores
with a weighting that emphasizes G4 (Paradox Resolution) and G6 (Latent Intent) —
the two foundational gates that measure the deepest structural properties and
are the hardest to fabricate.
A low OMS means the document is doing what it says it is doing. A high OMS means
the document's stated goals and structural reality are misaligned — the document
claims to be one thing while its structure reveals it is another. High OMS is the
quantitative signature of accountability theater.
Can 4CITE detect AI-generated content?
Not directly — 4CITE is not an AI-detection tool. It does not identify AI
authorship. What it does is measure structural integrity, and AI-generated
institutional documents consistently produce structural signatures that score
T3 or T4 on the 4CITE engine.
The reason: AI systems optimize for surface coherence — documents that look
like good arguments. They do not optimize for structural coherence — documents
that are good arguments. Surface coherence and structural coherence are not the
same thing. The 4CITE engine measures the difference.
The Mata v. Avianca fabricated brief scored 7. A genuine federal opinion in the
same subject area scored 88. The 81-point gap is not a measurement of AI
authorship — it is a measurement of structural integrity. The fact that
AI-generated documents consistently occupy the low end of that scale is a
property of how those documents are generated, not a feature that 4CITE
was designed to detect.
The Product
What is the difference between Scan, Report, and Certify?
Scan is a single-call structural reconnaissance. It does not score.
It returns observations (counts and verbatim examples of belief phrases,
capability claims without evidence, unquantified risk references, reframed metrics,
forward-looking reassurance, and self-claims of prudence) and named entities
(the people, organizations, works, jurisdictions, and dates the document leans on).
The Scan surfaces what is there and what is leaned on; it does not issue a verdict.
Report is the full structural integrity analysis. The document is
scored across four independent gates (G4, G6, G7, G8) with evidence arrays, a
composite score and Integrity Tier, the Theater Stage classifier, the Cross-Pair
Vector, longitudinal context where available, and the Outcome Misalignment Score
(OMS). The Report is the subscriber-level product.
Certify packages the Scan output, the full Report, and cert-specific
metadata (cert ID, timestamp, vertical, document hash, attestation text) into a
single immutable bundle frozen at issuance. Once issued, nothing inside the bundle
is modifiable. Certificates are stored in the Cert Vault permanently — no expiration,
user-controlled deletion.
The Certificate is not a third analysis layer. The analysis is in the Scan and the
Report. The Certificate packages those outputs into a non-modifiable record. Issuing
a Certificate does not change the scores — it freezes them permanently with full
provenance metadata.
What verticals does 4CITE serve?
Currently three: 4CITE⁴law (legal documents — PACER and
CourtListener data), 4CITE⁴biz (corporate SEC filings — EDGAR
data), and 4CITE⁴gov (legislative and
regulatory records — Congress.gov, Federal Register, FEC).
Each vertical corresponds to one branch of American institutional power and is
grounded in a public data source. The engine is the same across all three; the
calibration, document classes, and archetype libraries are vertical-specific.
A fourth vertical, 4CITE⁴one (personal integrity analysis), is
planned as a long-term product. It will be built last, informed by everything
the three institutional verticals establish.
What types of documents can 4CITE analyze?
Any text-based institutional document. The 4CITE
library spans the major institutional document classes across three verticals:
Law: Judicial opinions, legal briefs, sanctions orders, regulatory
findings, compliance reports, contracts, court filings.
Business: 10-K annual reports, 10-Q quarterly reports, proxy
statements (DEF 14A), earnings call transcripts, 8-K current reports, shareholder
letters, press releases, investor presentations.
Government: Bills, committee testimony, floor statements,
presidential addresses (State of the Union, inaugural, farewell), Federal Register
notices, FEC filings, regulatory impact assessments.
What is longitudinal scoring?
For entities with multiple documents in the corpus over time — a company with
ten years of 10-Ks, or a public official with multiple testimony records —
4CITE computes longitudinal integrity metrics: whether structural integrity is
improving, declining, or stable across the timeline.
The longitudinal layer (L1–L4) measures: trajectory (improving or declining),
foundation-surface divergence over time, crisis pattern detection, and recovery
authenticity. Apple Inc.'s trajectory is currently scored as stable high
integrity. SVB's trajectory showed a structural collapse pattern in its final
year of filings before the bank's failure.
Pricing and Access
How much does 4CITE cost?
Beta pricing is published in full at /pricing. Summary:
Walk-in 4CITE Cert — $15 per certificate. No account, no
subscription. Paste a document, pay, receive a PDF Integrity Certificate by
email. The full Scan + Report + Certify pipeline runs once and produces an
immutable certificate stored permanently in the Cert Vault.
4CITE⁴one — $50 / month. 1 seat, 225 units / month. Units
work across all three verticals. Typical use: 15 certs per month or an
equivalent research mix.
4CITE⁴more — $150 / month. 2–10 seats, 750 units / month
shared pool, team usage dashboard, proration on plan changes.
4CITE⁴all — $500 / month. 10–50 seats, 3,000 units / month
org-wide shared pool, $0.167 / unit effective rate.
Top-up packs. 100 / 500 / 2,000-unit packs for any
subscriber when the monthly allotment runs out. Top-up units carry forward
across billing periods until used.
Enterprise / Institutional. Site license pricing for law
firms, investment platforms, government agencies, and academic institutions.
Contact us for details.
How do I get access?
Two paths. For one document: buy a $15 walk-in cert at
/portal/certify — no account, no subscription.
For ongoing use: pick a plan at /pricing and sign in
at /account/login. Both paths are live now in
closed beta across law, business, and government.