Method profile 07Comparative implementation profileTwo distinct mechanisms

Hermeneutic-didactic

Interpretation that develops understanding, and interpretation that turns a recorded run into a teachable, governed artifact

This profile compares two related but non-equivalent implementations. SPINE DIALECTIC uses model-backed iterative interpretation, scratchpad rehydration, synthesis checks and RigVedan findings. ChangePlane performs a deterministic whole-parts-reinterpreted-whole pass over first-class run records, derives one lesson, emits a derived didactic projection and drafts a memory-admission candidate.

Two systems, two jobs

The shared hermeneutic vocabulary does not make the implementations interchangeable.

Model-backed interpretation

SPINE DIALECTIC

Develops understanding of a subject through repeated encounter, accumulated horizon, synthesis comparison and bounded stopping.

Deterministic run pedagogy

ChangePlane Hermeneutic Looped Didactic

Reads a recorded development run through selected parts, derives one source-grounded lesson and produces governed downstream artifacts.

SPINE iterative interpretation

Drag and inspect the horizon, model calls, typed findings, iteration record and stopping surface.

InterpretationWorking memoryEvidenceStopping boundary

Model control signals are not truth proofs.Similarity, has_converged and is_complete govern iteration. They do not establish that the resulting interpretation is correct.

ChangePlane six-step didactic pass

Inspect the deterministic whole-parts-reinterpretation arc and its governed outputs.

The current rereader reaches a stable fixpoint on pass two when records are unchanged.This demonstrates deterministic stopping behavior. It does not prove that the rereading is semantically adequate.

Comparative mechanism map

Filter the comparison by implementation, memory or evidence boundary.

Layer

Bounded convergence laboratory

Switch implementation and inspect the source-derived stopping order. The result is browser-local.

CONTINUE

Stopping result

Didactic projection and memory-admission boundary

Construct a synthetic ChangePlane output package without writing a record or memory.

PROPOSAL ONLY

Derived output package

A candidate verdict is not promotion.The runner persists a draft/generated candidate. Durable memory admission remains a separate governed act.

Provenance, projection and seal boundary

Follow the source records through the loop, derived projection, memory proposal and protocol trace.

Structural and source receipts

Source structure and feature-specific test additions remain separate from outcome efficacy.

Source structure

Mechanism dimensions

Counts are not quality scores.

Source-reported, mixed scopes

Feature test landmarks

The increments cover different feature bundles and were not rerun here.

SPINE phases

Encounter, interpretation, adjustment and synthesis.

ChangePlane steps

Whole through memory candidate.

SPINE default bound

Maximum cycles.

ChangePlane default bound

Maximum reread passes.

ChangePlane outputs

Loop record, projection and candidate.

Model calls after first pass

Encounter plus synthesis in SPINE.

Derived lesson

Per ChangePlane pass.

Independent comparative studies

Identified in inspected sources.

Evidence classification

The table separates implementation, behavioral evidence, operational authority and unproven outcomes.

Claim	Evidence class	Status
SPINE has an implemented model-backed hermeneutic loop	Source implementation	Supported at pinned commit
ChangePlane has an implemented six-step hermeneutic pass	Source implementation and schema	Supported at pinned commit
ChangePlane emits a derived didactic projection and draft candidate	Implementation and committed tests	Supported; tests not rerun here
Protocol trace tampering is detectable within seal scope	Committed tamper tests	Supported within documented unkeyed-hash boundary
Either convergence mechanism proves correct understanding	Epistemic outcome	Not established
Either implementation improves future-agent performance	Comparative efficacy	Not established

Evaluation interpretation and boundaries

The mechanisms make interpretation and teaching more inspectable. Their real effects require separate measurement.

Problem definition

AI work often loses how understanding evolved and what future work should learn from a completed run. A result may survive while the interpretive path, unresolved questions and teachable lesson disappear.

Mechanism hypothesis

Iterative horizon accumulation should make evolving interpretation inspectable, while deterministic whole-parts rereading plus governed projection should make run-derived lessons more reproducible and auditable. These are hypotheses.

Possible evaluation

Measure correction rate, unresolved-question recall, reviewer agreement, lesson source coverage, unsupported-memory admission, future-run rework, time to reorientation and resistance to misleading convergence signals.

Failure modes

Model loops can converge around a wrong reading. Deterministic rereading can be stable but shallow. Selected parts can omit decisive evidence. Didactic output can oversimplify, and candidates can be promoted without adequate review outside these mechanisms.

Explicit non-claims

The implementations are not equivalent and are not merged into one protocol.
SPINE similarity, synthesis and completion signals do not prove truth.
ChangePlane fixpoint convergence does not prove semantic completeness.
A didactic projection is derived and cannot outrank its source records.
A memory candidate is not a durable-memory write.
A tamper-evident seal does not prove the original reasoning was truthful.
Committed tests do not establish real-world learning or decision quality.
No independent comparative user study was identified.

Sources and reproducibility

Claims are bounded to the source artefacts and commits named below.

SPINE source pinfbratten/spine @ 8fa62c1ebfc0c7b65680ef2d89b9953a519e4138

SPINE loopspine/patterns/hermeneutic_loop.py

SPINE working memoryspine/memory/scratchpad.py

SPINE grammarspine/grammar/ebnf_rv.py

ChangePlane source pinfbratten/changeplane @ 32cb7840915df7f0d527044e37242cd6123801e7

ChangePlane runnersrc/changeplane/protocols/hermeneutic.py

ChangePlane modelsrc/changeplane/protocols/models.py

ChangePlane schemaschemas/changeplane.protocol.hermeneutic.v1.schema.json

ChangePlane teststests/test_protocols_mvp11.py

Publicationfbratten.github.io/methods/hermeneutic-didactic/

This page performs no model call, source retrieval, repository write, memory write, protocol insertion, projection generation or seal verification. All examples are synthetic and browser-local.