OCT Proxy Pre-Validation Sheet v0.1

Date: 2026-05-03 Prepared by (proxy designer): Fabio Ghioni Audited by (independent auditor): Solomon Theorem ID: A01 Claim ID: A01 Claim source (chapter/section): Chapter 10 / A01 Cycle ID candidate: CYCLE_5_2026-05-01 Status: pass

1) Claim logical type

Select one:

comparative

Expected dominant anti-pattern risk:

comparative -> level-shift

1.1) Cycle budget control (anti-p-hacking)

Max independent cycles for this claim: 3 Current cycle sequence index: 1 Budget policy status: within_budget

2) Claim statement and falsifier statement

Claim (verbatim): Compositional pipelines with high ordinative coherence show lower semantic degeneration than classically valid-only pipelines.

Primary falsifier (what would disconfirm it): No process-level advantage on Delta_cum, Err_sem_final, and mean_Coh_trajectory in at least 2/3 contexts.

Secondary falsifier(s): Output-only advantage without trajectory-quality advantage.

3) Concept-to-observable mapping table

Mandatory: every proxy variable must map to a formal concept with rationale.

Proxy element (code variable/formula)	OCT concept target	Mapping rationale (chapter/def)	Observable source	Label (O/P/C)
mean_Coh_trajectory	Coh	Chapter 8 validity axis	trajectory logs	O
Delta_cum	Delta	Chapter 8 degradation axis	trajectory logs	O
Err_sem_final	Phi-related functional degeneration proxy	Chapter 8 functional validity	benchmark anchors	P

Rules:

No unnamed proxy term is allowed.
No concept can be declared as measured if mapping is missing.
Labels must follow: O=operational, P=operationalizable, C=conceptual only.

4) Anti-pattern prevention checks (mandatory)

4.1 Comparative claim guard (level-shift test)

If claim type is comparative, show that A/B differs in process, not only output scoring.

Process-level evidence planned: append-only trajectory traces with Coh/Phi/Syn/decision per step.

Output-only fallback present? no

If yes, why it is non-decisive and how it is constrained: not applicable

4.2 Taxonomic claim guard (post-hoc rescue lock)

Pre-registered variables: Delta_cum, Err_sem_final, mean_Coh_trajectory

Pre-registered thresholds: at least 2/3 contexts for each primary criterion

Policy lock:

no criterion/threshold change on same data after first fail
any revision requires new preregistration + new data split or new cycle

Lock accepted: yes

4.3 Existential claim guard (vacuum-pass test)

Show at least one concrete, realistic configuration where claim must fail under this proxy if phenomenon is absent.

Concrete fail configuration: P_ord forced to emulate P_cls decisions under ablation.

Why reachable: ablation mode is executable by construction in Cycle 5 plan.

If not reachable, mark sheet as fail.

4.4 Universal claim guard (trivial confirmation lock)

If claim type is universal, show reachable counterexample strategy.

Concrete reachable counterexample search strategy: not applicable (claim type is comparative)

Confirmation is not tautological by construction? yes

5) Falsifiability reachability test

Required before run:

At least one executable input bundle expected to produce fail.
At least one executable input bundle expected to produce pass.
Decision boundary is pre-registered and immutable for this cycle.

Pass/fail of reachability test: pass

Evidence artifact paths:

OCT/validation/CYCLE_5_2026-05-01/contexts/A01_contexts_v0_1.md
OCT/validation/CYCLE_5_2026-05-01/CYCLE_5_A01_PROCESS_LEVEL_SPEC_v0_1.md

5.1 Pre-registered context list (mandatory when using multi-context criteria)

context_id	Distinguishing features	Expected difficulty	Dataset slice
A01_CTX_01	short synthetic chains, low noise	low	datasets/cycle3_inputs/case_a
A01_CTX_02	mixed synthetic-external traces, medium noise	medium	datasets/cycle3_inputs/case_b
A01_CTX_03	longer external trajectories, perturbation stress	high	datasets/cycle3_inputs/case_c

6) Validity space coverage declaration (Coh/Phi/Delta)

Declare explicitly what is measured vs not measured:

Axis	Measured? (yes/no)	Operational definition used	Path to code/report
Coh	yes	mean_Coh_trajectory over effective transitions	CYCLE_5_A01_PROCESS_LEVEL_SPEC_v0_1.md
PhiHat	partial	Err_sem_final as bounded proxy	CYCLE_5_A01_PROCESS_LEVEL_SPEC_v0_1.md
Delta	yes	Delta_cum	CYCLE_5_A01_PROCESS_LEVEL_SPEC_v0_1.md

If any axis is not measured, interpretation limits must be declared here: Phi is proxied and cannot be over-interpreted as full Phi formalization.

7) Evidence class declaration

Select one:

L1 empirical external

Accumulation rule reminder:

evidence classes are not auto-additive
cross-class accumulation requires explicit matrix logic

8) Role separation confirmation

Proxy designer and proxy auditor must be different.

Designer: Fabio Ghioni Auditor: Solomon Confirmed different identities: yes Auditor co-authorship with designer in last 24 months: no If yes, list: none Auditor school-of-thought affiliation declaration: external critical reviewer with independent audit stance

9) Preregistration seal

Required fields:

preregistration artifact path: OCT/validation/CYCLE_5_2026-05-01/seals/A01_PREREG_SEAL_v0_1.json
checksum/hash: declared in seal payload
timestamp (UTC): 2026-05-03T08:30:00Z
immutable storage location: 10.5281/zenodo.19959724
seal method: zenodo_doi
contexts hash: declared in seal payload
auditor independence declared (yes/no): yes
gate-entry signature: declared in seal payload
gate-entry timestamp (UTC): declared in seal payload
gate-exit signature: declared in seal payload
gate-exit timestamp (UTC): declared in seal payload
gate-exit status: declared in seal payload

Seal valid: yes

10) Gate decision

Decision options:

PASS (cycle execution allowed)
FAIL (cycle blocked, redesign required)
HOLD (claim clarification required before proxy design)

Decision: PASS

Decision rationale (short, technical): Comparative claim is process-anchored, falsifiability is reachable, and prereg lock is explicit.

Signer (auditor): Solomon Date: 2026-05-03