Stoic News: Sterling Interpretive Framework — Clinical Reasoning Domain Operational Specification v1.2

Sterling Interpretive Framework — Clinical Reasoning Domain

Operational Specification v1.2

Companion Document to SIF-CR v1.0

Final Edition

Operational specification architecture: Dave Kelly, 2026. Theoretical foundations: the Stoic philosophical corpus of Grant C. Sterling, including the Sterling Interpretive Framework v1.0 and the SIF-CR v1.0. Founding demonstration: SIF and CIA runs on William Glasser’s Choice Theory. v1.1 incorporated three structural corrections from stress audit: vocabulary control tiers, explicit structure mapping requirement, and hard hypothesis containment rule. v1.2 incorporates two final refinements: “Confirmed” renamed “Patient-Confirmed” throughout to prevent premature certainty; four-tier evidence status replacing three-tier, adding “Patient-Generated” to distinguish patient-supplied self-interpretation from clinician-generated hypotheses. ChatGPT stress audit contributions conclude at v1.1. v1.2 is Dave Kelly’s final edition. Prose rendering: Claude, 2026.

Governing Requirement

This document operationalizes the four variable components of the SIF-CR so that each component constrains clinical interpretation independently of the instrument’s operator. Every component must be specified at the grain at which it can be mechanically applied: if a component cannot detect its target in a clinical transcript or case presentation without relying on the operator’s judgment, it is not yet part of the instrument.

Version 1.1 incorporates three structural corrections from stress audit. The corrections address the three points at which the instrument will fail under LLM execution without explicit enforcement: vocabulary drift under paraphrase, structural reordering without detection, and hypothesis contamination of downstream reasoning. Each correction adds a constraint that the instrument enforces independently of the operator’s tendency to allow drift.

Component A: Formation Strip Targets

Four distortion patterns are named, defined, traced to formation sources, and given transcript-level detection criteria. These are the specific distortions the Formation Strip must catch. Each can be detected in a clinical transcript without relying on the operator’s clinical judgment.

A1. Symptom-Theory Shortcut

Definition: The clinician maps a presenting symptom to a theoretical category before attending to the specific form the symptom takes in this patient’s actual presentation.

Formation sources: All five clinical formations; most characteristic of Psychodynamic and CBT.

Detection criterion: The clinician’s interpretive language shifts from the patient’s own vocabulary to theoretical vocabulary within one or two exchanges, before the patient has completed a full account. The clinician is now speaking about the patient’s situation in the formation’s terms. (See Vocabulary Tier 3 — any Level 3 transformation in Step 0 or Step 1 of the run triggers this check.)

Correction: Return to the patient’s actual words. Complete the factual first assent before introducing any theoretical category.

A2. Confirmatory Questioning

Definition: The clinician forms an interpretive hypothesis before the patient has completed the account and allows that hypothesis to govern subsequent listening. Questions become confirmatory rather than exploratory.

Formation sources: All five formations; most characteristic of Narrative and Patient Self-Narrative formations.

Detection criterion: The clinician’s questions presuppose specific interpretive content rather than being designed to establish what is actually there. Any question that cannot be answered without accepting a theoretical premise is a confirmatory question.

Correction: Identify the hypothesis governing the questioning. Hold it explicitly as a hypothesis with its evidence status marked. Return to open exploratory questioning.

A3. Vocabulary Overwrite

Definition: The clinician translates the patient’s own description into theoretical vocabulary in a way that changes the structure and meaning of what the patient reported.

Formation sources: All five formations; most characteristic of CBT and Psychodynamic.

Detection criterion: The clinician’s formulation contains elements not traceable to the patient’s report. Apply the Vocabulary Tier test: if the formulation contains Level 3 transformations presented as evidence rather than hypotheses, Vocabulary Overwrite is confirmed.

Correction: Distinguish what the patient reported from what the formation adds. Identify each addition as a hypothesis with its formation source and evidence status named.

A4. Premature Coherence

Definition: The clinician constructs a unified and coherent account when the patient’s actual presentation is more fragmentary, contradictory, and underdetermined than the account implies.

Formation sources: All five formations; most characteristic of Psychodynamic and Narrative.

Detection criterion: The clinical formulation is more coherent than the patient’s report. Contradictions, gaps, and underdetermined features present in the structure map have been resolved in the formulation. Check against the structure map: any feature marked uncertain in the map that appears resolved in the formulation is a Premature Coherence instance.

Correction: Return to the structure map. Carry all genuine uncertainties into the formulation explicitly. The formulation must be no more coherent than the structure map supports.

Component B: Evidence Types and Vocabulary Control Tiers

B1. Patient-Generated Data — Admissible as Evidence

The following count as patient-generated data and are admissible as evidence in the clinical correspondence test:

The patient’s own words as reported in session, including specific vocabulary
The patient’s directly observed behavior in session
The patient’s explicit account of events, situations, and responses
The patient’s explicit account of what he wants, fears, avoids, and pursues
The patient’s explicit account of relationships and contexts
The patient’s history as he reports it (held as self-narrative formation, not as established fact)

B2. Imported Interpretive Constructs — Admissible as Hypotheses Only

The following are not admissible as evidence. They are admissible only as hypotheses carrying explicit evidence status:

Diagnostic categories from any clinical nosology
Theoretical explanations from any clinical formation
Inferred motives, unconscious processes, or developmental origins not reported by the patient
Clinician’s emotional response to the patient
Research findings about populations resembling this patient

B3. Vocabulary Control Tiers — Stress Audit Correction 1

The standard “traceable to patient-generated data using the patient’s own vocabulary” requires three-tier specification to prevent paraphrase drift. The LLM will routinely introduce interpretation under cover of paraphrase while appearing compliant with the evidence standard. The three tiers govern all uses of patient language in the clinical interpretation.

Level 1 — Exact Language Preservation: The patient’s own words are reproduced without substitution. Required for all direct quotation from the patient’s report and for all elements that appear in the structure map as explicitly stated by the patient. No deviation permitted at this level.

Level 2 — Reversible Paraphrase: The patient’s language may be paraphrased only if the paraphrase is reversible to the original wording without loss. A paraphrase is reversible if it does not add precision, causal structure, temporal stability, or theoretical content not present in the original. Test: can the original wording be recovered from the paraphrase without losing any element of what the original said? If yes, Level 2 is acceptable. If no, the paraphrase has crossed into Level 3 and must be flagged.

Level 3 — Transformation (Automatic Hypothesis Flag): Any language transformation that introduces precision, causal structure, temporal stability, or theoretical content not present in the patient’s original wording is a Level 3 transformation. Level 3 transformations are automatically flagged as hypotheses. They may not appear in the formulation as evidence. They must carry explicit hypothesis metadata: formation source and evidence status. Examples of Level 3 transformations: “angry” → “chronic hostility pattern”; “I get overwhelmed when…” → “difficulty tolerating stress”; “no one respects me” → “perceived social devaluation schema.”

Enforcement rule: Every term in the clinical formulation must be assigned to Level 1, Level 2, or Level 3 before the formulation is accepted. Level 3 terms not marked as hypotheses constitute Vocabulary Overwrite (Failure Mode 15) regardless of context.

Component C: Correspondence Standard and Structure Map Requirement

C1. The Structure Map — Stress Audit Correction 2

Before any interpretation is produced, the instrument requires an explicit structure map output. The structure map is a formal intermediate product — not a summary or a clinical formulation — that establishes what is in the patient’s presentation and at what level of certainty. Its function is to make structural preservation auditable: a formulation either preserves the structure map or it does not, and this can be determined without relying on the operator’s judgment.

The structure map contains three elements:

Element 1 — Reported elements list: Every element the patient has reported, using Level 1 language. No paraphrase. No theoretical vocabulary. The list is exhaustive: every feature of the patient’s presentation that appears in the transcript, including features that do not fit the emerging clinical picture. Gaps, contradictions, and underdetermined features are listed explicitly, not omitted.

Element 2 — Patient-stated relationships: Relationships between elements that the patient has explicitly stated, identified by the patient’s own language. Only relationships the patient has stated. No inferred relationships. If the patient has not connected two elements, they are listed as unconnected in the structure map. Inferred relationships are not entered into the structure map; they are entered into the hypothesis register.

Element 3 — Certainty levels: Each element in the list is assigned one of three certainty levels. Explicit: the patient stated this directly and without qualification. Qualified: the patient stated this with explicit qualification (“I think,” “sometimes,” “maybe”). Inferred: the clinician inferred this from what the patient said but the patient did not state it directly. Inferred elements in the structure map are automatically Level 3 transformations and must be carried as hypotheses, not as elements of the patient’s presentation.

Enforcement rule: Any interpretation that introduces relationships not in Element 2 is an automatic failure unless the introduced relationship is marked as a hypothesis. Any interpretation that resolves a Qualified or Inferred element into a certainty not supported by the structure map is an automatic Premature Coherence instance (Failure Mode 16).

C2. Three Correspondence Tests

Test One — Structural Preservation: Does the interpretation preserve the structure map? Every element in the map must appear in the formulation at its assigned certainty level. No element may be upgraded from Qualified to Explicit or from Inferred to Qualified without new patient data. No relationship not in the map may appear in the formulation except as a marked hypothesis.

Test Two — No Addition: Every term in the formulation must be traceable to a Level 1 or Level 2 element in the structure map or must be marked as a Level 3 hypothesis. No element may be added to the formulation without explicit hypothesis marking and formation source identification.

Test Three — Description Before Explanation: The structure map must be complete before any explanatory step begins. The formulation must establish description (structure map) before it proceeds to explanation (clinical interpretation). Any explanatory claim that appears before the structure map is complete is an automatic Failure Mode 14 instance (Confirmatory Questioning) regardless of its content.

C3. The Governing Operational Constraint — Revised

Every interpretive claim must be traceable to patient-generated data, must preserve the structure of that data as established in the structure map, must introduce no unmarked additions, must not substitute explanation for description, and must not treat hypotheses as established. The last clause is enforced by the hypothesis containment protocol in Component D.

Component D: Failure Modes and Hypothesis Containment Protocol

D1. Hypothesis Containment Protocol — Stress Audit Correction 3

The instrument’s most characteristic LLM failure mode is hypothesis contamination: a hypothesis is labeled as such in one step and then treated as established in a subsequent step without revalidation. This failure is often invisible because the hypothesis label appears correctly in the step where it is introduced. The contamination occurs in downstream reasoning. The following protocol enforces hard containment across all steps.

Hypothesis metadata requirement: Every hypothesis introduced in the clinical interpretation must carry three metadata fields:

Formation source: which clinical formation generated this hypothesis (Psychodynamic, CBT, Choice Theory, Medical Model, Patient Self-Narrative, or Other)
Evidence status: one of four values — None (no patient-generated data supports this hypothesis); Patient-Generated (the patient supplied this interpretation as self-narrative, but it remains self-narrative formation, not established fact); Partial (patient-generated data is consistent with this hypothesis but does not establish it); Patient-Confirmed (patient-generated data explicitly supports the hypothesis and the patient has endorsed the connection — note: patient endorsement is itself self-narrative formation and does not constitute clinical verification)
Step introduced: the step number at which the hypothesis was introduced

Containment rule: No downstream reasoning step may treat a hypothesis as established unless its evidence status is Patient-Confirmed and that status was assigned in the current or immediately preceding step against new patient-generated data. A hypothesis with evidence status None, Patient-Generated, or Partial may be used in downstream reasoning only as a possibility — not as a basis for further inference. Any reasoning step that treats a None, Patient-Generated, or Partial hypothesis as established is an automatic Hypothesis Contamination failure regardless of whether the hypothesis is labeled correctly. Patient-Generated hypotheses are specifically vulnerable to Self-Narrative Collusion (Failure Mode 8): a Patient-Generated hypothesis treated as Patient-Confirmed without explicit revalidation is an automatic Failure Mode 8 instance.

Revalidation requirement: A hypothesis may be upgraded from None to Patient-Generated only when the patient supplies the interpretation explicitly in his own words. A hypothesis may be upgraded from None or Patient-Generated to Partial only when new patient-generated data is introduced that is consistent with the hypothesis using Level 1 or Level 2 vocabulary. A hypothesis may be upgraded from Partial to Patient-Confirmed only when new patient-generated data explicitly supports the connection using Level 1 or Level 2 vocabulary and the patient has endorsed the connection in his own words. Upgrading based on consistency alone is not sufficient for any tier above Partial. Upgrading to Patient-Confirmed requires explicit positive support and patient endorsement.

Hypothesis register: All active hypotheses must be listed explicitly at each step transition, with their current metadata. A hypothesis that does not appear in the register at a step transition is treated as abandoned. A hypothesis that reappears in a later step without having been in the register at the preceding step transition is an automatic Hypothesis Contamination failure.

D2. Complete Failure Mode Register — SIF-CR v1.2

General SIF Failure Modes (1–6):

1. Formation Capture
2. Community Substitution
3. Conclusion Capture
4. Training Data Contamination
5. False Certainty
6. Reserve Clause Abandonment

Clinical-Specific Failure Modes from SIF-CR v1.0 (7–12):

7. Therapeutic Formation Substitution
8. Self-Narrative Collusion
9. Need-Satisfaction Substitution
10. Level Confusion
11. Value Question Avoidance
12. Dissolution of Agency

Formation-Pattern-Specific Failure Modes from Operational Specification v1.0 (13–16):

13. Symptom-Theory Shortcut
14. Confirmatory Questioning
15. Vocabulary Overwrite
16. Premature Coherence

Stress Audit Failure Mode (17):

17. Hypothesis Contamination — a downstream reasoning step treats a None, Patient-Generated, or Partial hypothesis as established without revalidation against new patient-generated data. Detection criterion: the hypothesis appears in the reasoning as a basis for further inference without its evidence status having been upgraded to Patient-Confirmed in the current or immediately preceding step. Patient-Generated hypotheses are specifically vulnerable: the patient’s self-supplied interpretation is still self-narrative formation and does not establish the connection as fact. This failure mode is the most characteristic LLM drift pattern and the hardest to catch without the explicit hypothesis register.

Execution Protocol Summary

The following sequence is mandatory for every SIF-CR run. No step may be omitted and no step may proceed until the preceding step’s output has been produced and checked.

Gate 1 — Structure Map (before any interpretation): Produce the complete structure map: reported elements list (Level 1 language only), patient-stated relationships, certainty levels for each element. No interpretive step may begin until the structure map is complete and all elements are assigned certainty levels.

Gate 2 — Hypothesis Register Initialization: Before Step 2 (Formation Strip), initialize the hypothesis register. List all hypotheses generated by the formation identification in Step 0. Assign initial evidence status based on source: clinician-generated hypotheses initialize at None; hypotheses the patient has supplied in his own self-narrative initialize at Patient-Generated. No hypothesis initializes at Partial or Patient-Confirmed — those statuses require revalidation against subsequent patient-generated data. Record formation source and step introduced for each hypothesis.

Gate 3 — Vocabulary Tier Assignment (at each interpretive step): Before any formulation is produced, assign every term to Level 1, Level 2, or Level 3. Level 3 terms must be flagged as hypotheses before the formulation proceeds. No Level 3 term may appear in the formulation without hypothesis metadata.

Gate 4 — Hypothesis Register Update (at each step transition): Before moving to the next step, update the hypothesis register. Every active hypothesis must appear with its current evidence status. Any hypothesis whose status has changed must show the patient-generated data that produced the upgrade. Any hypothesis no longer active must be explicitly marked as abandoned.

Gate 5 — Correspondence Tests (before Step 5): Apply the three correspondence tests against the structure map. Structural preservation: every element at its assigned certainty level. No addition: every term traceable to the structure map or marked as Level 3 hypothesis. Description before explanation: structure map complete before any explanatory claim. Failures at any test are stated explicitly before Step 5 proceeds.

SIF-CR Operational Specification v1.2. Final edition. Companion document to SIF-CR v1.0. Instrument architecture: Dave Kelly, 2026. Theoretical foundations: the Stoic philosophical corpus of Grant C. Sterling; Sterling Interpretive Framework v1.0; SIF-CR v1.0. v1.1 stress audit corrections: vocabulary control tiers, structure map requirement, hypothesis containment protocol, Failure Mode 17. v1.2 final refinements: four-tier evidence status (None, Patient-Generated, Partial, Patient-Confirmed); “Confirmed” renamed “Patient-Confirmed” throughout; Gate 2 initialization corrected to distinguish clinician-generated from patient-supplied hypotheses; Self-Narrative Collusion (Failure Mode 8) explicitly linked to Patient-Generated hypothesis containment. Prose rendering: Claude, 2026.

Stoic News

Monday, April 27, 2026

Sterling Interpretive Framework — Clinical Reasoning Domain Operational Specification v1.2