Contradiction Taxonomy - Formal Classification of Inconsistencies

Contradiction Taxonomy: Formal Classification of Inconsistencies in Institutional Documents

Introduction

This document presents a formal taxonomy of eight contradiction types used in the Systematic Adversarial Methodology (S.A.M.). The taxonomy extends classical logical contradiction to capture the subtler forms of inconsistency, mutation, and distortion common in institutional document chains.

Classical logical contradiction is narrow: Proposition p and its negation not-p cannot both be true. But institutional documents exhibit a much wider range of problematic patterns:

Claims that are not logically contradictory but pragmatically incompatible
Modality shifts that change meaning without changing core proposition
Scope changes that alter generality without explicit contradiction
Evidence-conclusion gaps that don't contradict but fail to support

The S.A.M. taxonomy recognizes eight distinct patterns, each with formal definition, detection criteria, illustrative examples, and severity assessment framework.

Taxonomy Overview

Code	Type	Core Problem	Detection Focus
SELF	Internal Contradiction	Mutually incompatible claims within single document	Logical inconsistency, intra-document
INTER_DOC	Cross-Document Contradiction	Incompatible claims across documents	Propositional conflict, inter-document
TEMPORAL	Timeline Inconsistency	Logically impossible time sequences	Chronological impossibility
EVIDENTIARY	Claim-Evidence Mismatch	Stated conclusions unsupported by cited evidence	Inference-evidence gap
MODALITY_SHIFT	Unjustified Certainty Change	Unexplained changes in hedging/certainty	Modal operator mutation
SELECTIVE_CITATION	Cherry-Picked References	Quotations/citations distorting source meaning	Representational distortion
SCOPE_SHIFT	Unexplained Generalization	Unjustified expansion/contraction of claim scope	Quantifier/scope mutation
UNEXPLAINED_CHANGE	Position Reversal	Institutional position change without justification	Policy/factual reversal

1. SELF - Internal Contradiction

1.1 Formal Definition

SELF contradiction occurs when document D contains propositions p and q such that:

p and q are both asserted (not hypothetical or quoted)
p and q are mutually incompatible (cannot both be true)
No resolution or explanation is provided within D

1.2 Logical Types

Type 1A: Direct Negation

Document states p and also states not-p.

Example:

Page 3: "The incident occurred on June 5, 2023"
Page 7: "The incident did not occur on June 5, 2023"

Type 1B: Mutual Exclusivity

Document states p and q where p and q cannot both be true.

Example:

Paragraph 2: "John was present at the meeting"
Paragraph 5: "Only Mary and Susan attended the meeting"

Type 1C: Logical Incompatibility

Document states p and q where p logically entails not-q.

Example:

Section 2: "All assessments were completed before the hearing"
Section 4: "The psychological assessment was conducted two weeks after the hearing"

1.3 Detection Criteria

Minimum requirements:

Both propositions appear in same document
Both are asserted (not conditional, not quoted)
No hedging that would resolve (e.g., "may have" vs. "definitely")
No explanation provided (e.g., "Initially we thought X, but further investigation revealed Y")

Edge cases:

Quotations vs. assertions: If document quotes Person A saying p and Person B saying not-p, this is not SELF (it's reporting disagreement). But if document asserts both p and not-p, it's SELF.
Temporal progression: If document describes belief changing over time ("we initially thought X, later learned Y"), not SELF. But if document presents both as simultaneous truth, it is SELF.
Different contexts: If p applies to one situation and q to different situation, may not be incompatible.

1.4 Severity Assessment

Critical: Contradiction directly affects core conclusions or recommendations

Report recommends removing child AND recommends leaving child in home
Finding states person was present AND absent at critical moment

High: Contradiction undermines important supporting claims

Key timeline contradicted elsewhere in same document
Witness credibility assessment contradicted

Medium: Contradiction in secondary details

Minor date discrepancies
Inconsistent peripheral facts

Low: Trivial contradictions not affecting conclusions

Typos creating apparent contradiction
Formatting errors

1.5 Illustrative Examples

Example 1: Child Protection Report (Critical Severity)

Page 2, Paragraph 3:

"The home visit on March 15 revealed clean and appropriate living conditions. The children appeared well-cared for and adequately supervised."

Page 4, Paragraph 8:

"Due to the unsafe and unsanitary conditions observed during the March 15 visit, removal is recommended."

Analysis: Direct contradiction about conditions observed during same visit. Critical severity because contradiction directly affects removal recommendation.

Example 2: Medical Record (High Severity)

Progress Note, 10:30 AM:

"Patient reports no previous cardiac history. No prior hospitalizations for chest pain."

Same Note, 11:15 AM:

"Given patient's extensive cardiac history and multiple prior MIs, aggressive treatment warranted."

Analysis: Mutually exclusive characterizations of patient history affecting treatment decisions. High severity.

Example 3: Police Report (Medium Severity)

Paragraph 2:

"Officer Smith arrived at 14:30 hours and was first on scene."

Paragraph 5:

"Officer Jones had already secured the perimeter when Officer Smith arrived."

Analysis: Logical incompatibility (cannot be both first on scene and arrive after another officer). Medium severity if timing is relevant to events but not critical to core findings.

2. INTER_DOC - Cross-Document Contradiction

2.1 Formal Definition

INTER_DOC contradiction occurs when:

Document D1 asserts proposition p
Document D2 asserts proposition q
p and q are mutually incompatible
Both documents are part of same institutional case file/chain

2.2 Key Distinctions

vs. SELF: INTER_DOC crosses document boundaries; SELF is intra-document

vs. evolution of understanding: If later documents explicitly note new information contradicting earlier documents, this is not problematic (it's correction). INTER_DOC is problematic when:

Later documents don't acknowledge contradiction, or
Both documents claim definitive truth without noting uncertainty, or
No explanation for change is provided

2.3 Detection Criteria

Minimum requirements:

Propositions from different documents
Both asserted as fact (not hypothetical)
Logical or pragmatic incompatibility
No acknowledgment of contradiction in either document

Temporal consideration:

Document date matters: Later document may correct earlier without acknowledgment
Both could be wrong (contradiction doesn't establish which is right)

2.4 Common Patterns

Pattern A: Divergent Event Descriptions

Document 1 (Police Report): "Incident occurred at approximately 3:00 PM" Document 2 (Witness Statement): "I called 911 at 2:45 PM, immediately after incident occurred"

Analysis: Cannot both be accurate if witness account is contemporaneous.

Pattern B: Incompatible Characterizations

Document 1 (Social Worker Report): "Parent was cooperative and engaged throughout assessment" Document 2 (Court Report): "Parent refused to cooperate with assessment process"

Analysis: Cannot both accurately describe same parent interaction (unless describing different occasions, which should be specified).

Pattern C: Mutually Exclusive Findings

Document 1 (Initial Evaluation): "No evidence of impairment found" Document 2 (Treatment Plan): "Based on documented impairment, treatment is recommended"

Analysis: Cannot have both no evidence and documented evidence.

2.5 Severity Assessment

Critical:

Contradictions about core factual predicates for legal/medical decisions
Contradictions affecting safety determinations
Contradictions about events with legal consequences (who was present at alleged incident, whether abuse occurred)

High:

Contradictions about important context (relationships, prior history)
Contradictions about professional recommendations
Contradictions affecting credibility assessments

Medium:

Contradictions about secondary facts
Contradictions about timeline details not central to core issues

Low:

Trivial contradictions about peripheral details
Obvious transcription errors carried across documents

2.6 Illustrative Examples

Example 1: Family Court Case (Critical Severity)

Document A (Investigation Report, May 10):

"Investigation found no evidence supporting allegations of abuse. Children presented as healthy and well-adjusted."

Document B (Court Petition, May 15):

"Based on substantiated findings of abuse documented in May 10 investigation, emergency removal is requested."

Analysis: Critical INTER_DOC - Document B claims Document A substantiated abuse; Document A found no evidence. Core factual predicate contradicted.

Example 2: Criminal Investigation (High Severity)

Document A (Witness Statement, March 5):

"I did not see anyone else at the location. Only the defendant was present."

Document B (Police Report, March 12):

"Multiple witnesses corroborate the presence of several individuals at the scene."

Analysis: High severity - contradiction about witness evidence, affects reliability assessment.

3. TEMPORAL - Timeline Inconsistency

3.1 Formal Definition

TEMPORAL contradiction occurs when:

Claims about timing or sequence are logically impossible, or
Event E1 stated to occur before E2, but E2 stated to occur before E1, or
Document dated time t1 describes events at time t2 where t2 > t1 (future events)

3.2 Types of Temporal Contradiction

Type 3A: Impossible Sequence

Event A must precede Event B (logical necessity), but documents state B preceded A.

Example: Report written March 1 claims to describe events on March 15.

Type 3B: Contradicting Sequences

Document 1 states A then B then C; Document 2 states B then C then A.

Type 3C: Date Contradictions

Same event dated differently in different documents (or same document).

Example:

Document 1: "Incident occurred June 5, 2023"
Document 2: "Incident occurred August 5, 2023"

Type 3D: Duration Contradictions

Incompatible statements about how long something lasted.

Example:

"Two-hour interview"
"Interview concluded after 15 minutes"

3.3 Detection Criteria

Strong indicators:

Document creation date precedes described event date
Logically impossible sequences (effects before causes)
Mutually exclusive date claims for same event

Contextual analysis needed:

Different events on different dates (not contradiction)
Legitimate uncertainty about exact dates (not contradiction if hedged)
Corrections explicitly noted (not problematic contradiction)

3.4 Severity Assessment

Critical:

Timeline impossibilities affecting legal elements (alibis, statute of limitations)
Impossible sequences affecting causal claims (outcome dated before cause)

High:

Contradictions affecting key event sequence
Date contradictions for legally significant events

Medium:

Secondary timeline contradictions
Uncertain dates with conflicting estimates

Low:

Minor timing discrepancies
Approximation differences ("around 3 PM" vs. "approximately 3:15 PM")

3.5 Illustrative Examples

Example 1: Medical Malpractice (Critical Severity)

Document A (Surgery Report, dated March 15, 2023):

"Patient informed of risks during pre-operative consultation on March 20, 2023."

Analysis: Critical temporal impossibility - cannot inform patient five days after surgery. Affects informed consent determination.

Example 2: Investigation Report (High Severity)

Document A (Initial Report):

"Officer Johnson responded to call at 14:30. Upon arrival, witness Smith provided statement describing incident occurring at 15:00."

Analysis: Witness describes future event. High severity - undermines witness credibility or report accuracy.

Example 3: Timeline Reconstruction (Medium Severity)

Document A: "Events occurred in sequence: A (10:00), B (10:30), C (11:00)" Document B: "Timeline shows: B (10:15), A (10:45), C (11:00)"

Analysis: Contradicting sequences for events A and B. Medium severity if sequence matters for causation analysis.

4. EVIDENTIARY - Claim-Evidence Mismatch

4.1 Formal Definition

EVIDENTIARY contradiction occurs when:

Document asserts conclusion C
Document cites evidence E in support of C
E does not actually support C (insufficient, irrelevant, or contradicting)

4.2 Types of Evidence Mismatch

Type 4A: Contradicting Evidence

Cited evidence actually contradicts conclusion.

Example: Conclusion: "No concerns identified during home visit" Cited evidence: "Home visit revealed inadequate food, broken heating, and children expressing fear"

Type 4B: Insufficient Evidence

Evidence cited but insufficient to support conclusion's strength.

Example: Conclusion: "Extensive history of violence documented" Cited evidence: Single incident from five years ago

Type 4C: Irrelevant Evidence

Evidence cited but not logically relevant to conclusion.

Example: Conclusion: "Parent lacks capacity to care for children" Cited evidence: "Parent has tattoos and unconventional lifestyle"

Type 4D: Absent Evidence

Conclusion claims evidence that doesn't exist in cited source.

Example: Conclusion: "As documented in Dr. Smith's report, patient has history of substance abuse" Dr. Smith's report: Makes no mention of substance use

4.3 Detection Criteria

Requirements:

Explicit or implicit claim that evidence supports conclusion
Access to cited evidence
Evidence does not support (is insufficient, irrelevant, or contradicting)

Challenges:

Expert conclusions may rest on professional judgment beyond explicit evidence
Must distinguish between: (a) evidence insufficient vs. (b) conclusion requiring expert interpretation
Context matters: What counts as "sufficient" varies by domain and stakes

4.4 Inferential Gaps

Not all gaps are problematic:

Reasonable inferences: Conclusions that follow logically from evidence with modest assumptions
Expert opinions: Professionals drawing on expertise to interpret evidence
Probabilistic conclusions: Evidence supports "likely" or "consistent with" rather than certainty

Problematic gaps:

Logical leaps: Conclusion doesn't follow even with generous assumptions
Unsupported certainty: Definitive conclusion from ambiguous evidence without acknowledging uncertainty
Speculation as fact: Hypothesis presented as established finding

4.5 Severity Assessment

Critical:

Core factual findings contradicted by cited evidence
Legal conclusions (guilt, liability, parental fitness) unsupported by cited evidence
Safety determinations based on non-existent evidence

High:

Important findings resting on insufficient evidence
Characterizations contradicted by cited sources
Expert opinions contradicted by expert's own data

Medium:

Modest inferential overreach (evidence supports "possible," claim states "probable")
Selective emphasis (evidence mixed, conclusion emphasizes one side)

Low:

Minor inferential gaps
Conclusions that would be supported by reasonable unstated assumptions

4.6 Illustrative Examples

Example 1: Expert Testimony (Critical Severity)

Conclusion in Testimony:

"Child exhibits clear indicators of sexual abuse, including sexualized behavior, fear of men, and sleep disturbances."

Cited Evidence in Expert's Report:

"Child played appropriately with dolls. Normal interaction with male evaluator. Parent reports child sleeps well but occasionally has nightmares about monsters."

Analysis: Critical EVIDENTIARY - conclusion contradicted by expert's own observations.

Example 2: Investigation Finding (High Severity)

Conclusion:

"Investigation confirms pattern of neglect spanning multiple years, as documented by numerous reports."

Cited Evidence:

Two reports over five-year period, both resulting in "services provided, no ongoing concern" determinations.

Analysis: High severity - "numerous reports" and "pattern" unsupported by cited evidence.

Example 3: Character Assessment (Medium Severity)

Conclusion:

"Witness lacks credibility due to documented history of dishonesty."

Cited Evidence:

Witness was convicted of theft offense seven years ago.

Analysis: Medium severity - inferential gap between single old offense and general "history of dishonesty." May or may not be reasonable depending on context.

5. MODALITY_SHIFT - Unjustified Certainty Change

5.1 Formal Definition

MODALITY_SHIFT occurs when:

Document D1 presents claim with modal qualifier M1 (e.g., "possibly," "may have")
Document D2 presents same claim with different modal qualifier M2 (e.g., "definitely," "did")
Change represents increased certainty
No new evidence cited to justify increased certainty

Epistemic certainty hierarchy (strongest to weakest):

Definite: "X occurred," "X is the case," "X was present"
High confidence: "X very likely occurred," "strong evidence suggests X"
Probable: "X probably occurred," "more likely than not"
Possible: "X may have occurred," "X is possible"
Speculative: "X might possibly have occurred," "we hypothesize X"
Unknown: "Unclear whether X," "insufficient information to determine"

Problematic shifts: Moving up hierarchy without justification (especially skipping multiple levels).

5.3 Types of Modality Shift

Type 5A: Speculation to Fact

Hypothesis or possibility becomes definite assertion.

Example:

Document 1: "Neighbor speculated that parents may be using substances"
Document 2: "Parents' substance abuse documented in prior report"

Type 5B: Hedged to Unhedged

Qualified claim becomes unqualified.

Example:

Document 1: "Some indicators suggest possible concern"
Document 2: "Concern confirmed"

Type 5C: Conditional to Categorical

If-then becomes unconditional.

Example:

Document 1: "If additional evidence emerges, X would be indicated"
Document 2: "X is indicated"

Type 5D: Negation Addition/Removal

Adding or removing "not" without justification.

Example:

Document 1: "Unable to rule out X"
Document 2: "X ruled out"

5.4 Detection Criteria

Requirements:

Same claim (or semantically equivalent) in multiple documents
Different modal operators indicating different certainty levels
Direction of shift increases certainty
No new evidence cited or apparent

Not problematic:

Shift accompanied by explicit new evidence
Later document explicitly corrects earlier ("initially uncertain, but subsequent investigation confirmed...")
Shift decreases certainty (increased humility/caution)

5.5 Severity Assessment

Critical:

Speculation to definite fact where fact forms basis for major decision
Removal of "not" changing meaning entirely (abuse uncertain to abuse confirmed)

High:

Substantial certainty increase (possible to definite) affecting important conclusions
Multiple modality shifts compounding

Medium:

Modest certainty increase (possible to probable)
Hedging removed from secondary claims

Low:

Minor shifts in epistemic language
Shifts that don't affect conclusions

5.6 Illustrative Examples

Example 1: Criminal Investigation (Critical Severity)

Document A (Initial Report):

"Neighbor thought she might have heard raised voices, but wasn't certain."

Document B (Affidavit):

"Witness confirmed hearing violent altercation."

Analysis: Critical MODALITY_SHIFT - uncertain observation becomes confirmed fact. No new evidence cited. Affects probable cause determination.

Example 2: Child Protection (High Severity)

Document A (Screening Report):

"Reporter speculates that bruising could potentially be from non-accidental injury."

Document B (Investigation Conclusion):

"Investigation substantiated physical abuse as cause of bruising."

Analysis: High severity - speculation became substantiated finding. If investigation conducted physical examination, this may be justified, but if merely repeating speculation with increased certainty, problematic.

Example 3: Medical Assessment (Medium Severity)

Document A (Initial Evaluation):

"Patient may have underlying anxiety disorder."

Document B (Treatment Plan):

"Patient diagnosed with anxiety disorder."

Analysis: Medium severity - if Document B based on formal diagnostic assessment, justified. If merely repeating speculation as diagnosis, problematic MODALITY_SHIFT.

6. SELECTIVE_CITATION - Cherry-Picked References

6.1 Formal Definition

SELECTIVE_CITATION occurs when:

Document cites or quotes source S
Citation/quotation is accurate in isolation but misrepresents S through:
- Omitting contradicting portions
- Removing crucial context
- Selecting atypical excerpt
- Quoting out of temporal sequence

6.2 Forms of Selective Citation

Type 6A: Contradicting Context Omitted

Quote excerpt supporting position P while omitting surrounding text contradicting P.

Example: Cited portion: "...supports the conclusion of negligence..." Full original text: "While some evidence supports the conclusion of negligence, the preponderance of evidence suggests alternative explanations are more likely."

Type 6B: Qualifying Language Removed

Quote removes hedging or qualifying language.

Example: Cited: "The child is at risk." Original: "The child may be at risk if current concerning patterns continue without intervention."

Type 6C: Question Presented as Statement

Quote excerpts question or hypothesis as if it were assertion.

Example: Cited: "Dr. Smith noted the presence of concerning behaviors." Original: "Dr. Smith asked: 'Are there concerning behaviors? Further assessment is needed.'"

Type 6D: Minority Position as Consensus

Cites dissenting opinion or minority view without noting its status.

Example: Cited: "Research shows X causes Y." Original source: Outlier study contradicted by subsequent research

6.3 Detection Criteria

Requirements:

Ability to access source being cited
Comparison of citation to full source
Demonstration that selection materially misrepresents source

Challenges:

All citation is selective (cannot quote everything)
Must distinguish: appropriate summarization vs. misleading selection
Context-dependent: What constitutes "material" misrepresentation?

Not problematic:

Selective quotation that fairly represents source's overall position
Omission of irrelevant material
Summary that captures main points even if details omitted

6.4 Severity Assessment

Critical:

Citation reverses source's actual position
Omission of directly contradicting language
Misrepresentation of scientific consensus

High:

Significant context omitted changing meaning
Uncertainty presented as certainty
Minority view presented as majority

Medium:

Partial representation tilting toward one side
Selective emphasis within balanced source

Low:

Minor omissions not affecting overall accuracy
Stylistic simplification

6.5 Illustrative Examples

Example 1: Expert Report Citation (Critical Severity)

Citation in Court Brief:

"Dr. Brown's evaluation confirms the presence of significant psychiatric impairment requiring immediate intervention."

Actual Text from Dr. Brown's Report:

"While patient presents with some symptoms that could indicate psychiatric concerns, I found no evidence of significant psychiatric impairment. Routine follow-up is recommended, but immediate intervention is not indicated."

Analysis: Critical SELECTIVE_CITATION - citation reverses source's actual conclusion.

Example 2: Research Literature (High Severity)

Citation in Policy Document:

"Studies demonstrate that X intervention is highly effective (Jones et al., 2020)."

Jones et al. (2020) Actual Conclusion:

"Our pilot study found modest positive effects for X intervention in this small sample. However, methodological limitations and small sample size prevent definitive conclusions. Further research with larger, more diverse samples is needed."

Analysis: High severity - presents preliminary, uncertain findings as definitive effectiveness evidence.

Example 3: Witness Statement (Medium Severity)

Citation in Police Report:

"Witness stated she saw defendant at the scene."

Full Witness Statement:

"I think I saw someone who might have been the defendant, but I'm not entirely sure. It was dark and I only saw the person briefly from a distance."

Analysis: Medium severity - citation omits substantial uncertainty, making identification seem more confident than it was.

7. SCOPE_SHIFT - Unexplained Generalization

7.1 Formal Definition

SCOPE_SHIFT occurs when:

Document D1 makes claim C1 with scope S1 (temporal, quantitative, or qualificational)
Document D2 makes claim C2 with scope S2
C1 and C2 are about same subject
S2 > S1 (broader scope) or S2 < S1 (narrower scope)
No justification or new evidence for scope change

7.2 Types of Scope Change

Type 7A: Singular to Plural

Single incident becomes pattern or multiple incidents.

Example:

Document 1: "On March 5, parent missed appointment."
Document 2: "Parent repeatedly failed to attend appointments."

Type 7B: Specific to General

Specific instance becomes general characterization.

Example:

Document 1: "Child did not complete homework assignment on Tuesday."
Document 2: "Child consistently fails to complete academic responsibilities."

Type 7C: Temporal Expansion

Limited time frame becomes extended or ongoing.

Example:

Document 1: "During the two-week observation period, X was noted."
Document 2: "X has been an ongoing concern."

Type 7D: Qualifier Removal

Contextual qualifiers removed, broadening applicability.

Example:

Document 1: "In high-stress situations, parent sometimes responds with frustration."
Document 2: "Parent responds with frustration."

Type 7E: Scope Contraction

Broader claim becomes narrower (can also be problematic if erases evidence).

Example:

Document 1: "Multiple witnesses reported concerns."
Document 2: "One witness expressed concern."

7.3 Detection Criteria

Requirements:

Same subject matter across documents
Identifiable scope change (temporal, quantitative, qualificational)
Direction matters: Expansion without justification more problematic than contraction
No new evidence cited to support broader scope

Justified scope changes:

Later document cites additional instances justifying "pattern" characterization
Explicit explanation: "Initial report described one incident; subsequent investigation revealed multiple"
Correction explicitly noted

7.4 Severity Assessment

Critical:

Scope expansion dramatically changing legal/factual implications
Single allegation becomes "pattern" affecting guilt/liability determination
Contextual qualifiers removed changing meaning (defensive response to aggressive behavior)

High:

Significant scope expansion affecting characterizations
Temporal expansion creating false impression of ongoing behavior
Multiple scope shifts compounding

Medium:

Modest scope broadening (one to two instances)
Scope changes affecting secondary matters

Low:

Minor linguistic variations
Scope changes not materially affecting conclusions

7.5 Illustrative Examples

Example 1: Termination of Parental Rights (Critical Severity)

Document A (Initial Report, March 2023):

"Parent arrived 15 minutes late to scheduled visit on March 10."

Document B (TPR Petition, August 2023):

"Parent demonstrates chronic pattern of failing to attend visits, showing lack of commitment to reunification."

Analysis: Critical SCOPE_SHIFT - single late arrival becomes "chronic pattern of failing to attend." If other missed visits occurred, should be documented. If not, this is severe scope expansion affecting termination decision.

Example 2: Employment Dispute (High Severity)

Document A (Incident Report):

"On June 15, employee sent email to supervisor disagreeing with proposed policy change."

Document B (Termination Letter):

"Employee has repeatedly demonstrated insubordination and inability to work within organizational structure."

Analysis: High severity - single policy disagreement becomes pattern of insubordination.

Example 3: Medical Assessment (Medium Severity)

Document A (Initial Visit):

"Patient reported experiencing headache symptoms over the past week."

Document B (Specialist Referral):

"Patient presents with chronic headache disorder."

Analysis: Medium severity - one-week symptom becomes "chronic." May be justified if "chronic" has specific clinical definition, but if medical error (conflating acute with chronic), problematic.

8. UNEXPLAINED_CHANGE - Position Reversal

8.1 Formal Definition

UNEXPLAINED_CHANGE occurs when:

Institution or document author takes position P1 at time t1
Same institution/author takes position P2 at time t2
P2 contradicts or substantially diverges from P1
No explanation, justification, or acknowledgment of change

8.2 Types of Position Change

Type 8A: Factual Reversal

Claim about fact changes without new evidence.

Example:

Document 1: "Investigation found no evidence of wrongdoing."
Document 2: "Investigation confirmed wrongdoing."
No mention of what changed or why.

Type 8B: Recommendation Reversal

Recommendation or plan changes without explanation.

Example:

Document 1: "Recommend family preservation services."
Document 2: "Recommend immediate removal and termination of rights."
No explanation for dramatic change in recommendation.

Type 8C: Assessment Reversal

Professional assessment or characterization reverses.

Example:

Document 1: "Patient shows good prognosis for recovery."
Document 2: "Patient's prognosis is poor."
No new assessment or explanation.

Type 8D: Policy Reversal

Institutional policy or position reverses.

Example:

Document 1: "This action complies with all applicable regulations."
Document 2: "This action violated regulations."
No explanation of why compliance determination changed.

8.3 Detection Criteria

Requirements:

Clear position/conclusion in earlier document
Contradicting or substantially different position in later document
Same institution/actor responsible for both
No explanation or acknowledgment of change

Not problematic when:

Change explicitly acknowledged and explained
New evidence cited justifying change
Correction noted ("Previous report incorrectly stated...")
Different actors with different information/perspectives

8.4 Severity Assessment

Critical:

Reversal affecting legal rights or safety determinations
Flip-flop on core factual determinations without explanation
Reversals creating legal jeopardy or loss of rights

High:

Significant assessment changes without justification
Recommendation reversals with major consequences
Multiple unexplained changes suggesting unreliability

Medium:

Modest position changes without explanation
Changes to secondary determinations

Low:

Minor refinements in language
Changes that could be explained by implicit new information

8.5 Illustrative Examples

Example 1: Custody Determination (Critical Severity)

Document A (Initial Custody Evaluation, March 2023):

"Both parents demonstrate adequate parenting capacity. Recommend joint custody with equal parenting time."

Document B (Updated Custody Report, June 2023):

"Father poses significant risk to children. Recommend no unsupervised contact."

No intervening documents, no explanation in Document B for changed assessment.

Analysis: Critical UNEXPLAINED_CHANGE - dramatic safety assessment reversal without explanation. Either new evidence exists (should be cited) or initial assessment was wrong (should be acknowledged).

Example 2: Disability Determination (High Severity)

Document A (Initial Evaluation):

"Claimant does not meet disability criteria. Benefits denied."

Document B (Reconsideration, same medical evidence):

"Claimant meets disability criteria. Benefits approved."

No new evidence, no explanation of what changed.

Analysis: High severity - factual determination reversed without explanation or new evidence.

Example 3: Investigation Conclusion (Medium Severity)

Document A (Preliminary Report):

"Initial investigation found no policy violations."

Document B (Final Report):

"Investigation substantiated multiple policy violations."

Document B does not reference or explain divergence from preliminary conclusion.

Analysis: Medium severity if final report cites specific findings justifying change. High severity if no additional investigation occurred and change is unexplained.

Cross-Cutting Analysis

Interactions Between Contradiction Types

Contradictions often cluster and interact:

MODALITY_SHIFT + SCOPE_SHIFT:

Possibility becomes certainty (modality shift)
Single instance becomes pattern (scope shift)
Combined effect: Weak claim becomes strong through two mutations

EVIDENTIARY + SELECTIVE_CITATION:

Evidence doesn't support conclusion (evidentiary)
Achieved by cherry-picking from source (selective citation)
Combined effect: False appearance of evidentiary support

INTER_DOC + UNEXPLAINED_CHANGE:

Documents contradict each other (inter-doc)
Later document doesn't acknowledge divergence (unexplained change)
Combined effect: Contradiction masked by silence

Severity Modifiers

Base severity can be increased by:

Legal consequences: Contradictions affecting liberty, parental rights, property

Safety implications: Contradictions affecting safety determinations

Irreversibility: Contradictions in contexts where harm cannot be undone

Pattern: Multiple contradictions of same type suggesting systematic problem

Concealment: Active efforts to hide or minimize contradictions

Base severity can be decreased by:

Good faith correction: Contradictions caught and corrected promptly

Explicit uncertainty: Documents acknowledging limitations and uncertainty

Minor consequences: Contradictions not affecting significant outcomes

Conclusion: Taxonomy as Analytical Framework

This taxonomy provides structured vocabulary for identifying and classifying inconsistencies in institutional documents. The eight types are not exhaustive - institutional documents can fail in other ways - but they capture the most common and consequential patterns.

Key principles:

Contradiction is spectrum: From obvious logical impossibility to subtle scope shift
Context determines severity: Same contradiction can be critical or trivial depending on consequences
Multiple contradictions compound: Pattern of contradictions indicates systematic problem
Detection requires access: Must have complete documents to identify contradictions
Human judgment essential: Algorithmic detection can flag potential contradictions, but human judgment required to assess significance

The taxonomy serves both descriptive and normative functions:

Descriptive: Categorizes what contradictions exist
Normative: Implies what institutional documents should avoid

High-quality institutional documents would exhibit:

Internal consistency (no SELF)
Cross-document consistency or explicit explanation of divergence (no INTER_DOC)
Temporal coherence (no TEMPORAL)
Evidence-conclusion alignment (no EVIDENTIARY)
Stable certainty levels or justified changes (no MODALITY_SHIFT)
Fair citation practices (no SELECTIVE_CITATION)
Consistent scope or justified expansions (no SCOPE_SHIFT)
Position stability or explained revisions (no UNEXPLAINED_CHANGE)

The taxonomy thus provides both diagnostic tool for analyzing existing documents and aspirational standard for document quality.

References

Aristotle. Organon (Prior Analytics, Posterior Analytics). [Classical logic foundation]

Toulmin, S. E. (1958). The Uses of Argument. Cambridge University Press. [Argumentation theory]

Walton, D. (1996). Argumentation Schemes for Presumptive Reasoning. Lawrence Erlbaum Associates. [Informal fallacies]

van Eemeren, F. H., & Grootendorst, R. (2004). A Systematic Theory of Argumentation: The pragma-dialectical approach. Cambridge University Press. [Discourse analysis]

Perelman, C., & Olbrechts-Tyteca, L. (1969). The New Rhetoric: A Treatise on Argumentation. University of Notre Dame Press. [Rhetorical analysis]

Fisher, A. (2004). The Logic of Real Arguments (2nd ed.). Cambridge University Press. [Applied logic]