Documentation
The adversarial-scrutiny mode,
in full.
This page documents the adversarial-scrutiny mode of Prosoche v2 in full — its nine-phase procedure, worksheets, prompts, and decision rules. It applies the eight-type contradiction taxonomy, the four-phase propagation trace, and six analytical lenses to politically charged or contested material. It is one operation in the Prosoche toolkit — the most developed one, and the right instrument for reading a hostile or self-serving record against the grain. Earlier versions developed this mode under the working names Systematic Adversarial Methodology (S.A.M.) and the Sovereign Analyst Framework; those are lineage references, not the current brand. For the wider adaptive method — the seven-stage spine this operation sits within — see the methodology overview.
The procedure
The nine-phase procedure
These nine phases are the internal procedure of the adversarial-scrutiny mode — one operation in the Prosoche v2 toolkit, not the seven-stage spine of the method itself. v2’s single Stage-4 symmetric-scrutiny pass replaced this mode’s former triple audit (the old Phases 6–8); the phases below document the mode end-to-end. Two phases (3 and 5) are conditional on the target type. For the spine, see the methodology overview.
Epistemic Self-Audit
Before the source set is opened, the analyst names priors, prior outputs on adjacent material, stake in the conclusion, and target-type bucket. The baseline against which downstream drift is audited.
Source Architecture
Source set assembled and rated against the NATO Admiralty Code (A1–F6) on directness, independence, internal consistency, methodological transparency, track record, conflict-of-interest exposure, and corroboration.
Fallacy Firewall
Two-tier fallacy taxonomy run against the source set. Tier 1 covers universal informal fallacies; Tier 2 covers political fallacies. Each detection ships with its charitable-reading companion before adverse finding.
Historical Depth Protocol
CONDITIONALConditional. Applies when the target has a relevant pre-history (prior statements, prior decisions, prior framings) the analyst can responsibly cite. Skips for discrete one-off material with no antecedent.
Contradiction Analysis
The 8-type contradiction taxonomy: SELF, INTER_DOC, TEMPORAL, EVIDENTIARY, MODALITY_SHIFT, SELECTIVE_CITATION, SCOPE_SHIFT, UNEXPLAINED_CHANGE. Each contradiction passes a charitable-interpretation field before being rated.
Propagation Trace / CASCADE
CONDITIONALConditional. Applies when the target is a propagated institutional decision inherited from an upstream source. The four-phase trace ANCHOR → INHERIT → COMPOUND → ARRIVE maps each inheritance point and any compounding step the chain failed to audit.
Analyst Steel-Man
The analyst drafts the strongest version of the case the target’s authors would make for their own position, written before the synthesis so the synthesis cannot quietly retreat from the steel-man’s strongest counter-arguments.
Adversarial Review / Council of Phronesis
Constitutional three-branch review: Legislative (seven personas — Skeptic, Adversary, Definitional Shell-Hunter, Numerical Reconciler, Steel-Manner, Reader-of-Output, Cross-Examiner), Executive (analyst drafts, resolves, signs or vetoes), Judicial (Forensic Auditor + Stage 2 Supreme review by a distinct-session LLM advisor).
Intellectual Integrity Audit
11-question audit on the final synthesis: scope adherence, evidence sufficiency, citation integrity, contradiction handling, charitable-reading discipline, fallacy exposure, framing transparency, ICD-203 confidence calibration, falsifiability, named limits, signed-off epistemic position.
Entry
Five target types
Every analysis in this mode enters under one target type, declared up front. The type keys the roster-template floor and governs which conditional phases run or skip. Templates are floors, not ceilings.
Single document
One standalone artefact — a report, an essay, a policy paper, an article. The unit of analysis is the document itself; contradictions and shifts are surfaced within its own four corners.
Multi-source corpus
A single author or institution viewed across many outputs over time. The unit of analysis is the body of work; the procedure surfaces drift, reversal, and selective citation across the corpus.
Decision propagated through a chain
Chains of authority in which a finding, judgment, or determination is inherited from a prior decision and re-applied downstream — regulatory rulings, adjudications, institutional determinations. The propagation trace (Phase 5) runs here.
Broadcast / persuasion / multimodal
Video, audio, image-led narrative, multi-modal evidence — material where rhetorical framing operates through production choices, not just text. Interviews, documentaries, livestreams, visual-led campaigns.
Contested public dispute
A live dispute with materially-affected parties on more than one side, where the task is to hold the centre line rather than argue a case. Legal-case bundles (court materials, exhibits, expert reports) are a sub-case, run through the propagated-decision and multi-source templates.
Phase 4 · within the procedure
Contradiction Analysis — eight types
Phase 4 of the mode classifies every inconsistency against an eight-type taxonomy. The taxonomy distinguishes eight structurally distinct failure modes — from a document that contradicts itself to an unexplained position reversal by the same professional. Classification drives severity scoring and complaint routing.
A single document contradicts itself—different claims on different pages that cannot both be true.
A report states "the account was closed in March" on page 3, then describes "regular monthly activity" on page 12.
Two or more documents make incompatible claims about the same event, person, or fact.
One reviewer records "the subject was cooperative"; a second reviewer, same period, records "the subject refused to engage."
Dates, sequences, or durations that are logically impossible given other established facts.
Report "summarizes" a meeting that occurred three days after report was written.
Claims presented as factual that lack supporting evidence, or cited evidence actually contradicts the claim.
"The company has a history of safety breaches"—no incidents, findings, or evidence cited anywhere in the record.
Possibility becomes probability becomes certainty—without new evidence to justify the shift.
Initial note: "the supplier may have quality issues." Final report: "the supplier’s documented history of quality failures."
Selective citation that omits context, qualifications, or contradicting passages from source material.
Quotes a source saying "I was concerned about the figures" but omits the next line: "—once the data was corrected, the concern was resolved."
Conclusions that exceed the scope of the analysis or apply findings beyond their valid domain.
A single two-hour site visit leads to a conclusion about "persistent operational deficiencies."
A party changes position without explanation, or conclusions change without new evidence.
March: "No concerns about the account." June: "Significant compliance concerns." No new findings documented in between.
Phase 5 · conditional · within the procedure
Propagation Trace — the four-phase discipline
When the target is a propagated institutional decision, Phase 5 of the mode traces the chain through four steps — ANCHOR → INHERIT → COMPOUND → ARRIVE. The trace reads institutional documents against the grain, treating every claim as requiring verification rather than acceptance. It is one conditional operation within the adversarial-scrutiny mode, formerly developed under the working name Systematic Adversarial Methodology (S.A.M., 2023).
ANCHOR
Phase 1: False Premise Identification
ANCHOR is the foundational phase of the S.A.M. methodology. It identifies the origin points where false premises first emerge in a document corpus—the moments when speculation becomes assertion, hearsay becomes evidence, or assumption becomes fact. Most institutional failures can be traced to specific anchor points: claims that entered the system without adequate verification and subsequently propagated unchecked.
INHERIT
Phase 2: Propagation Tracking
INHERIT tracks how claims propagate across institutional boundaries without independent verification. When one agency’s assessment is adopted by another without re-evaluation, errors compound and gain apparent authority. This phase maps the transmission pathways—identifying which institutions adopted claims from whom, and whether any verification occurred at each handoff.
COMPOUND
Phase 3: Authority Accumulation
COMPOUND documents how claims gain weight through repetition rather than verification—what is termed "authority laundering." A statement repeated by four professionals appears more credible than one stated by a single source, even when all four trace to the same unverified origin. This phase quantifies how apparent authority accumulates and identifies the tipping points where unverified claims become "established facts."
ARRIVE
Phase 4: Outcome Mapping
ARRIVE maps catastrophic outcomes back to their originating false premises, establishing but-for causation: would the outcome have occurred if the anchor claim had been verified? This phase connects the analytical work of phases 1–3 to real-world consequences, demonstrating how initial errors cascaded into final decisions.
Analytical Lenses
Six Lenses Within the Mode
The mode produces findings. The lenses operationalise them. Each lens applies a distinct analytical function; together they cover the full chain from identity resolution through to regulatory complaint packaging.
Entity Resolution
Canonical Identity Mapping
Creates a unified identity registry across your document corpus. Resolves aliases, tracks role changes, and maps relationships between individuals.
- Alias Detection
- Role Tracking
- Relationship Mapping
- Temporal Context
- Conflict Detection
Canonical Identity Card Canonical Name: R. Bennett Aliases: R. Bennett, Rachel Bennett, Ms Bennett, "the lead reviewer," RB Role: Lead Assessor (2019–present) Organisation: Regional Review Service Documents: 47 appearances across 23 documents First Seen: Doc 3, p.2 (14 March 2021) Last Seen: Doc 156, p.8 (9 October 2024) Relationships: Supervised by R. Williams; reviewed Subject A (12 sessions); authored reports E4.1, E4.3, E4.7 Flags: Role change: "Independent Reviewer" → "Internal Employee" at Doc 89 (no disclosure in subsequent reports)
Temporal Parser
Timeline Construction & Analysis
Constructs unified timelines from scattered date references. Identifies gaps, overlaps, and impossible sequences that indicate error or manipulation.
- Date Extraction
- Timeline Construction
- Gap Detection
- Sequence Validation
- Deadline Tracking
TEMPORAL CONTRADICTION DETECTED Event A: Assessment report authored Date: 15 March 2023 | Source: Doc 12, p.3 | Author: R. Bennett Event B: Assessment session conducted Date: 22 March 2023 | Source: Doc 14, p.1 | Participants: R. Bennett, Subject A Finding: Report (Event A) predates the assessment it describes (Event B) by 7 days. CASCADE type: TEMPORAL | Severity: HIGH Gap detected: 14 June 2023 – 9 September 2023 No recorded activity for 87 days Mandated deadline (45-day review) exceeded Last actor: Reviewer B
Argumentation
Toulmin Structure Analysis
Deconstructs claims into their logical components using the Toulmin model. Identifies unsupported assertions, missing warrants, and logical gaps.
- Claim Extraction
- Warrant Analysis
- Gap Detection
- Rebuttal Mapping
- Strength Scoring
Contradiction
Cross-Document Inconsistency Detection
Systematically detects inconsistencies across your document corpus using the CASCADE 8-type framework. Surfaces conflicts that undermine institutional conclusions.
- CASCADE Detection
- Source Comparison
- Conflict Mapping
- Confidence Scoring
- Severity Rating
Bias Detection
Statistical Imbalance Analysis
Performs statistical analysis of directional bias in reporting. Calculates omission ratios, measures framing imbalance, and quantifies selective presentation.
- Directional Scoring
- Omission Analysis
- Framing Metrics
- Selection Bias Detection
- Pattern Identification
DOCUMENT: Editorial Investigation Report SOURCE MATERIALS: 12 documents analysed OMISSION INVENTORY: Total omissions detected: 8 Pro-prosecution: 8 | Pro-defence: 0 DIRECTIONAL BIAS SCORE: +1.0 (All omissions favour prosecution narrative) STATISTICAL SIGNIFICANCE: Binomial test (H0: random omission direction) p = 0.004 (significant at p < 0.01) Probability of this pattern by chance: 0.4% FRAMING RATIO: Subject-as-suspect: 132 minutes | Subject-as-cleared: 10 minutes Ratio: 13.2:1 PATTERN: 100% prosecution-favoring omission pattern across all source materials.
Accountability
Duty Mapping & Violation Detection
Maps statutory duty violations to specific actors and regulatory bodies. Links evidence to relevant professional standards and complaint pathways.
- Duty Mapping
- Actor Attribution
- Regulatory Routing
- Breach Detection
- Evidence Packaging
ACTOR: R. Bennett | ROLE: Lead Assessor | REGULATOR: Professional Standards Body BREACH 1: Standards of Proficiency §2.7 (evidence-based practice) Finding: 4 evaluative claims with no evidential basis Evidence: Doc 23 p.7, Doc 31 p.4, Doc 31 p.12, Doc 45 p.2 Strength scores: 0.08, 0.12, 0.05, 0.15 BREACH 2: Code of Conduct §3.4 (balanced consideration) Finding: Directional bias score +0.87 across authored reports Evidence: 7 contrary findings omitted; binomial test p = 0.008 BREACH 3: Standards of Conduct §9 (duty of candour) Finding: Role change from independent reviewer to internal employee undisclosed Evidence: Doc 89 header vs Doc 92 header; no disclosure in Docs 90–156 COMPLAINT PACKAGE: 3 breaches, 11 supporting documents, 14 specific findings. ROUTE: Professional Standards, Fitness to Practise
How findings map to lenses
Propagation phase → primary lenses
- ANCHOR — Entity Resolution, Argumentation
- INHERIT — Contradiction, Bias
- COMPOUND — Bias, Accountability
- ARRIVE — Accountability, Temporal Parser
Contradiction type → primary lens
- SELF, INTER_DOC, UNEXPLAINED — Contradiction
- TEMPORAL — Temporal Parser + Contradiction
- EVIDENTIARY — Argumentation
- SELECTIVE — Bias
- SCOPE, MODALITY — Argumentation + Contradiction
Applying the mode
Applying the adversarial-scrutiny mode
These are the operator steps for running the adversarial-scrutiny mode on a corpus — not a separate model of the method. Each step has a defined scope and produces a specific deliverable; you complete one before moving to the next.
Load
Assemble your document corpus and set up your working files.
Collect all relevant documents into a single corpus. Use the intake worksheet to record sources, dates, authors, and institutional roles. At this step you are establishing the evidentiary record you will audit — not yet reading for content.
Audit
Run the mode's nine-phase procedure against your corpus.
Work through the mode’s nine phases in order — declaring the target-type bucket up front, rating sources against the Admiralty Code, running the Fallacy Firewall, classifying contradictions against the eight-type taxonomy, and (where the bucket warrants it) running the four-phase propagation trace ANCHOR → INHERIT → COMPOUND → ARRIVE. For each phase, apply the corresponding prompt set from the production library. Record findings in the claim registry, and flag every unverified anchor point before proceeding.
Cross-examine
Apply the six analytical lenses to the findings from your audit.
Run Entity Resolution first to unify your actor registry. Temporal Parser second to surface timeline anomalies. Then Argumentation and Contradiction in parallel. Bias Detection and Accountability last — they consume findings from the earlier lenses. Use the eight-type contradiction taxonomy to classify every inconsistency found.
Ship
Package and deliver your analytical output.
Compile findings into the structured output format the mode provides: a contradiction inventory, a bias score card, and (where applicable) an accountability brief with regulatory routing. The mode includes output templates calibrated to professional, legal, and journalistic contexts.
FAQ
Common Questions
Do I need an LLM provider to apply the mode?+
The worksheets and decision frameworks are standalone documents — no external dependency. The production prompt library is designed for use with any instruction-following language model (Claude, GPT-4-class, or similar). The prompts are plain text; they do not require any API integration. You paste them into your preferred interface.
Is the methodology proprietary?+
Prosoche v2 and its adversarial-scrutiny mode — the nine-phase procedure, the eight-type contradiction taxonomy, and the CASCADE propagation-trace discipline (formerly developed under the working name S.A.M.) — are documented in the public record through this reference page. The intellectual architecture is open; the value is in the structured application: the worksheets, prompts, and production library that operationalise the mode for non-specialist analysts.
What does a finished audit look like?+
A finished audit produces three structured outputs: a contradiction inventory (each finding classified by one of the eight taxonomy types — SELF, INTER_DOC, TEMPORAL, EVIDENTIARY, MODALITY_SHIFT, SELECTIVE_CITATION, SCOPE_SHIFT, UNEXPLAINED_CHANGE — with source citations and severity score), a bias scorecard (directional analysis with omission inventory and statistical note), and — where applicable — an accountability brief (breaches mapped to specific actors and regulatory bodies). The mode provides templates for each output.
See the method applied.
The mode is documented here in full. See it applied end-to-end in the published work.