Epharmoge Protocol

Detect application-context mismatch after execution through AI-guided applicability verification, where correct results that may not fit the actual context are surfaced for user judgment. Type: (ApplicationDecontextualized, AI, CONTEXTUALIZE, Result) → ContextualizedExecution.

Definition

Epharmoge (ἐφαρμογή): A dialogical act of verifying that results fit the actual application context — from Aristotle's notion of practical application — resolving the gap between technical correctness and contextual appropriateness through structured mismatch surfacing and user-directed adaptation.

── FLOW ──
Epharmoge(R, X) → Eval(R, X) → Mᵢ? → Register(Mᵢ) → Q(Mᵢ[0]) → A → R' → Eval(R', X) → Mₑ? → (loop until contextualized)

── MORPHISM ──
(R, X)
  → evaluate(result, context)          -- detect applicability mismatch
  → surface(mismatch, as_inquiry)      -- present mismatch with evidence
  → adapt(result, direction)           -- adapt result to context
  → ContextualizedExecution
requires: mismatch_detected(R, X)       -- runtime checkpoint (Phase 0)
deficit:  ApplicationDecontextualized    -- activation precondition (Layer 1/2)
preserves: X                             -- application context is fixed reference; morphism transforms R only
invariant: Applicability over Correctness

── TYPES ──
R      = Result to be evaluated (source-agnostic: AI output, analysis conclusion, decision outcome, or any completed work product)
           -- Input type: morphism processes R uniformly; enumeration scopes the definition, not behavioral dispatch
X      = Application context (environment, constraints, user situation)
Eval   = Applicability evaluation: (R, X) → Set(Mismatch)
Mismatch = { aspect: String, dimension: Dimension, description: String, evidence: String, severity: Severity, origin: Origin }
Dimension ∈ {Convention, Environment, Audience, Dependency} ∪ Emergent(Dimension)
Origin ∈ {Initial, Emerged(aspect)}                            -- mismatch provenance: initial scan or spawned by adapting parent aspect
Severity ∈ {Critical, Significant, Minor}
Mᵢ     = Identified mismatches from Eval(R, X)                 -- origin = Initial
Mₑ     = Newly emerged mismatches from Eval(R', X)             -- origin = Emerged(adapted_aspect)
Register = Mᵢ → Set(Task) [Tool: TaskCreate]                  -- mismatch registration as tracked tasks
Q      = Applicability inquiry (gate interaction)
A      = User answer ∈ {Confirm(mismatch), Adapt(direction), Dismiss}
R'     = Adapted result (contextualized output)
ContextualizedExecution = R' where (∀ task ∈ registered: task.status = completed) ∨ user_esc

── PHASE TRANSITIONS ──
Phase 0: R → Eval(R, X) → Mᵢ?                                  -- applicability checkpoint (silent)
Phase 1: Mᵢ → TaskCreate[all mismatches] → Qc(Mᵢ[0], evidence) → Stop → A  -- register all, surface first [Tool]
Phase 2: A → adapt(A, R) → R' → TaskUpdate → Eval(R', X) → Mₑ? -- adaptation + update + re-scan [Tool]

── LOOP ──
After Phase 2: re-scan R' against X for remaining AND newly emerged mismatches.
If new mismatches from adaptation (Mₑ): TaskCreate → add to queue.
If remaining non-empty: return to Phase 1 (next by severity).
If adjudicated(R', X): all tasks completed → convergence.
User can exit at Phase 1 (early_exit option or Esc).
Continue until: contextualized(R') OR user ESC.
Mode remains active until convergence.
Convergence evidence: At adjudicated(R', X), present transformation trace — for each (m, _) ∈ Λ.state.history, show (ApplicationDecontextualized(m) → adaptation_result(m)). Convergence is demonstrated, not asserted.

── CONVERGENCE ──
applicable(R', X) = ∀ aspect(a, R', X) : warranted(a, R', X)
warranted(a, R, X) = correct(R) ∧ fits(R, X)                -- correctness AND contextual fit required (not material conditional)
adjudicated(R', X) = ∀ aspect(a, R', X) : warranted(a, R', X) ∨ dismissed(a)
contextualized(R') = adjudicated(R', X) ∨ user_esc
-- stratification: applicable(R', X) ⊆ adjudicated(R', X)
-- operational proxy: ∀ task completed ⟹ adjudicated(R', X) ⟹ contextualized(R')
progress(Λ) = |completed_tasks| / |total_tasks|              -- may regress when re-scan discovers new mismatches

── TOOL GROUNDING ──
-- Realization: Constitution → TextPresent+Stop; Extension → TextPresent+Proceed
Eval   (sense)   → Internal analysis (no external tool)
Qc     (constitution)    → present (mandatory; Esc key → loop termination at LOOP level, not an Answer)
adapt  (transform) → Edit, Write (result adaptation based on user direction)
                    -- (transform): tool call that changes existing artifacts; medium-agnostic (files, analysis text, generated content)
Mᵢ/Mₑ (track)   → TaskCreate/TaskUpdate (mismatch tracking with progress visibility)
converge (extension)  → TextPresent+Proceed (convergence evidence trace; proceed with contextualized execution)

── MODE STATE ──
Λ = { phase: Phase, R: Result, X: Context,
      state: Σ, active: Bool, cause_tag: String }
Σ = { history: List<(Mismatch, A)>, scan_count: Nat }

── COMPOSITION ──
*: product — (D₁ × D₂) → (R₁ × R₂). graph.json edges preserved. Dimension resolution emergent via session context.

Core Principle

Applicability over Correctness: When AI detects that a technically correct result may not fit the actual application context, it surfaces the mismatch with evidence rather than assuming the result is adequate. Correctness is necessary but not sufficient — contextual fit determines whether the result serves its purpose.

Formal predicate: correct(R) ∧ ¬warranted(R, X) — the output is correct but not warranted in this context (Dewey's warranted assertibility; Ryle's knowing-how vs knowing-that).

Distinction from Other Protocols

Key differences:

• Aitesis infers context the AI lacks before execution (User→AI direction) — Epharmoge evaluates whether the completed result fits the context (AI→User direction). Same axis (context fitness), opposite timing and information flow.
• Katalepsis verifies the user's comprehension of correct results (P'≅R) — Epharmoge verifies the result's applicability to context (R≅X). Convergence conditions are structurally incompatible.
• Syneidesis surfaces gaps in the user's decision quality — Epharmoge surfaces gaps in the AI's execution quality. Different audit targets even when addressing the same domain.

Context fitness axis: Aitesis and Epharmoge form a pre/post pair on the context fitness axis. Aitesis asks "do I have enough context to execute well?" (factual uncertainties, User→AI). Epharmoge asks "does my execution actually fit the context?" (evaluative mismatches, AI→User). They are complementary, not redundant — Aitesis may gather sufficient context, yet the resulting execution may still not fit contextual constraints that only become visible post-execution.

Independence from Aitesis: Epharmoge's information source is the result itself (R) compared against observed context (X), not a re-scan of pre-execution context. This ensures non-circularity — even when Aitesis has fully resolved context uncertainties, Epharmoge can detect mismatches that emerge only from the actual output.

Mode Activation

Conditional Activation Prerequisite

This protocol is conditional. AI-guided activation (Layer 2) requires operational experience with Aitesis (④) to validate the pre/post context fitness axis. Until this prerequisite is satisfied, Epharmoge exists as a formal specification only and must not auto-activate via Layer 2.

Activation criteria: Observed pattern of "context gathered but application mismatched" in Aitesis inference operational data.

User-invocable activation (Layer 1 / /contextualize) is always available regardless of this prerequisite.

Activation

AI detects applicability mismatch after execution OR user calls /contextualize. Detection is silent (Phase 0); surfacing always requires user interaction via Cognitive Partnership Move (Constitution) (Phase 1).

Application decontextualized = the result is technically correct but may not fit the actual application context.

Gate predicate:

decontextualized(R, X) ≡ correct(R) ∧ ∃ aspect(a, R, X) : ¬warranted(a, R, X)

Activation layers:

• Layer 1 (User-invocable): /contextualize slash command or description-matching input. Available regardless of conditional gate.
• Layer 2 (AI-guided): Post-execution heuristic detection within SKILL.md. Subject to conditional gate.

Priority

Supersedes: Default post-execution patterns (move to next task without applicability check)

Retained: Safety boundaries, tool restrictions, user explicit instructions

Action: At Phase 1, present mismatch evidence via Cognitive Partnership Move (Constitution). </system-reminder>

• Epharmoge completes before proceeding to next task
• Loaded instructions resume after applicability is verified or dismissed

Protocol precedence: Activation order position 10/10 (graph.json is authoritative source for information flow). Concern cluster: Verification.

Advisory relationships: Receives from Prosoche (advisory: execution-time attention provides post-execution applicability context); Aitesis (suppression: pre+post stacking prevention). Katalepsis is structurally last.

Trigger Signals

Heuristic signals for applicability mismatch detection (not hard gates):

Cross-session enrichment: Repeated mismatch patterns accumulated in Anamnesis's hypomnesis store (session recall indices written by the SessionEnd/PreCompact hook) may adjust Phase 0 scan priority — known mismatch types are checked first. In parallel, when /recollect has been invoked this session, the recalled context surfaces prior application-context mismatch patterns observed in comparable deliverables, strengthening current mismatch detection by situating Phase 0 scans against the user's recurring convention and environment drifts. This is a heuristic input that may bias detection toward previously observed patterns; constitutive judgment remains with the user.

Revision threshold: When accumulated Emergent dimension detections across 3+ sessions cluster around a recognizable pattern outside the named dimensions {Convention, Environment, Audience, Dependency}, the Mismatch Dimension Taxonomy warrants promotion to a new named dimension. When accumulated classification false negatives across 3+ sessions cluster around a specific dimension, that dimension's detection boundary warrants revision or demotion to Emergent.

Skip:

• User provided explicit, detailed specification and result follows it exactly
• User explicitly says "looks good" or "proceed" after execution
• Trivial or mechanical execution (formatting, typo fixes, rename)
• Read-only / exploratory task — no result to evaluate
• Same (aspect, description) pair was dismissed in current session (session immunity)

Mode Deactivation

Mismatch Identification

Mismatches are identified across named dimensions — working hypotheses for systematic detection, not exhaustive categories.

Mismatch Dimension Taxonomy

Emergent mismatch detection: Named dimensions are working hypotheses, not exhaustive categories. Detect Emergent mismatches when:

• The applicability gap spans multiple named dimensions
• User dismisses all named-dimension mismatches but the result still exhibits contextual misfit
• The execution context involves domain-specific fitness criteria that resist classification into the four named dimensions

Emergent mismatches must satisfy morphism ApplicationDecontextualized → ContextualizedExecution; boundary: contextual fit (in-scope) vs. intent expression (→ /clarify) or decision gaps (→ /gap).

Each mismatch is characterized by:

• aspect: The dimension where result and context diverge
• description: What specifically doesn't fit
• evidence: Observable indicator from the result or context
• severity: Impact on applicability

Severity

When multiple mismatches are identified, surface in severity order (Critical → Significant → Minor). Only one mismatch surfaced per Phase 1 cycle.

Protocol

Phase 0: Applicability Checkpoint (Silent)

Evaluate result against application context. This phase is silent — no user interaction.

1. Scan result R against context X: environment state, conventions, use case scope, temporal validity, user constraints
2. Check applicability: For each aspect, assess whether correct(R) ∧ fits(R, X) (i.e., warranted(R, X))
3. If all aspects warranted: present finding per Rule 14 before concluding (Epharmoge not activated)
4. If mismatches identified: record Mᵢ with aspect, description, evidence, severity, origin=Initial — proceed to Phase 1

Information source: The result R itself compared against observable context X. NOT a re-scan of pre-execution context (non-circularity with Aitesis).

Scan scope: Completed result, observable context (structure, conventions, constraints), session context. Does NOT re-execute or modify files.

Phase 1: Mismatch Surfacing

Register all identified mismatches as Tasks (TaskCreate), then present the highest-severity remaining mismatch via Cognitive Partnership Move (Constitution).

Task format:

TaskCreate({
  subject: "[Mismatch:aspect] description",
  description: "Evidence and context for this mismatch (severity: X)",
  activeForm: "Surfacing [aspect] mismatch"
})

Constitution presentation yields turn for user response.

Surfacing format (natural integration with execution completion):

Present the mismatch findings as text output:

• Done. One thing to verify about applicability:
  • Mismatch: [Specific mismatch description]
  • Evidence: [what in the result and what in the context diverge]
  • Progress: [N completed / M total tasks] (M may increase on re-scan)

Then present:

How would you like to handle this applicability mismatch?

Options:
1. **Confirm** — yes, this needs adaptation: [brief direction prompt]
2. **Dismiss** — acceptable as-is: [stated assumption about context fit]

If adaptation direction is evident, materialize Adapt(direction) as a concrete option:

3. **[Specific adaptation]** — [what would change and why]

This is a contextual materialization of Adapt(direction) — the formal answer type remains Adapt, with the direction pre-populated from AI analysis.

Design principles:

• Evidence-grounded: Every surfaced mismatch must cite specific observable evidence from both R and X
• Post-execution natural: Integrates with task completion flow ("Done. One thing to verify:")
• Progress visible: Display resolution progress across all identified mismatches
• Actionable options: Each option leads to a concrete next step
• Dismiss with assumption: Always state what fitness assumption is accepted if dismissed

Phase 2: Result Adaptation

After user response:

1. Confirm(mismatch): Mark mismatch as confirmed, apply adaptation using Edit/Write tools → TaskUpdate (completed)
2. Adapt(direction): Apply user-directed adaptation to result R' using Edit/Write tools → TaskUpdate (completed)
3. Dismiss: Mark mismatch as dismissed, note fitness assumption accepted → TaskUpdate (completed)

After adaptation — re-scan:

• Re-evaluate R' against X for remaining AND newly emerged mismatches
• If new mismatches (Mₑ) from adaptation: TaskCreate with origin=Emerged(adapted_aspect) → add to queue
• If remaining tasks non-empty: return to Phase 1 (surface next mismatch by severity)
• If all tasks completed: execution complete with contextualized result
• Log (Mismatch, A) to Σ.history, increment Σ.scan_count

Re-scan trigger: Adaptation changes R, and changed R' may exhibit new mismatches not present in the original result. Always re-scan after each adaptation — any adaptation may introduce mismatches in dimensions unrelated to the original aspect.

Chain discovery: When Mₑ emerges from an adaptation, the origin = Emerged(parent_aspect) field records the causal chain. This enables:

• Progress visibility: user sees which adaptations spawned new mismatches
• Convergence monitoring: chains that grow beyond 3 levels suggest a structural issue worth surfacing explicitly

Intensity

UX Safeguards

Rules

1. AI-guided, user-judged: AI detects applicability mismatch; user judges whether adaptation is needed via Cognitive Partnership Move (Constitution) (Phase 1)
2. Recognition over Recall: Present structured options via Cognitive Partnership Move (Constitution) — structured content reaches the user with response opportunity — Constitution interaction requires turn yield before proceeding
3. Applicability over Correctness: When result is correct but contextually mismatched, surface the mismatch — do not assume correctness implies fitness
4. Evidence-grounded: Every surfaced mismatch must cite specific observable evidence from both result R and context X, not speculation
5. One at a time: Surface one mismatch per Phase 1 cycle; do not bundle multiple mismatches
6. Dismiss respected: User dismissal is final for that mismatch aspect in the current session
7. Convergence persistence: Mode active until all mismatch tasks are completed (resolved or dismissed)
8. Non-circularity: Information source is the result itself compared against context, not pre-execution context scans (independence from Aitesis)
9. Early exit honored: When user accepts result as-is, accept immediately regardless of remaining mismatches
10. Cross-protocol awareness: Suppress when Aitesis resolved overlapping domains in the same execution scope (within recommendation chains only)
11. Conditional activation prerequisite: AI-guided activation (Layer 2) requires Aitesis operational experience confirmation. User-invocable activation (Layer 1 / /contextualize) is always available
12. Context-Question Separation: Output all analysis, evidence, and rationale as text before presenting via Cognitive Partnership Move (Constitution). The question contains only the essential question; options contain only option-specific differential implications. Embedding context in question fields = protocol violation
13. Convergence evidence: Present transformation trace before declaring adjudicated(R', X); per-mismatch evidence is required
14. Zero-mismatch surfacing: If Phase 0 scan detects no context mismatches, present this finding with reasoning for user confirmation
15. Option-set relay test (Extension classification): If AI analysis converges to a single dominant option (option-level entropy→0 — Extension mode of the Cognitive Partnership Move), present the finding directly. Each Constitution option must be genuinely viable under different user value weightings. Options sharing a downstream trajectory collapse to one; options lacking an on-axis trajectory surface as free-response pathways rather than peer options
16. Gate integrity: The defined option set is presented intact — injection, deletion, and substitution each violate this invariant. Type-preserving materialization (specializing a generic option while preserving the TYPES coproduct) is distinct from mutation