soul-frame/research/instrument-scoping-b3-psychological-continuity.md

# Instrument Scoping — Constitutive Integration (B3) and Psychological Continuity

**Workplan:** SOUL-WP-0010-T02, resolved 2026-09-28
**Follows:** `research/metrics-scoping-note.md` (SOUL-WP-0009-T04)
**Purpose:** scope — not build — candidate operational approaches for the two
dimensions the metrics note marked **No existing instrument** while carrying
the most argumentative weight across the thought-experiment catalog
(TE-02, TE-04, TE-05, TE-08).

Scoping, in this note's sense: name a candidate operational approach, say
what it would measure and what it would miss, and judge whether it is
viable enough to justify a build effort — or state explicitly that none is.

---

## 1. B3 — Constitutive integration

**Target claim to operationalize:** the system is one *integrated
organization*, not loosely coupled components (decisive in TE-05 The
Collective: "at which level does self-individuating organization close?";
load-bearing in TE-04 for what survives gradual replacement).

### Candidate approaches

| # | Approach | What it measures | What it misses |
|---|---|---|---|
| 1 | **Constraint-closure ratio** (operationalizing Montévil & Mossio 2015): over a dependency graph of the system's processes and the constraints acting on them, the fraction of constraints that are themselves produced/maintained by processes inside the candidate boundary | Degree of organizational closure at a chosen boundary; directly comparable across candidate levels (member vs. collective — exactly TE-05's question) | Graph construction is modeler-dependent; no canonical granularity for "a process" |
| 2 | **Integration/segregation measures from network science** (participation coefficient, modularity Q, global efficiency) applied to the component-interaction graph | Whether interaction structure is one integrated module or several weakly coupled ones | Measures topology, not constitution — a message bus scores "integrated" without any self-maintenance |
| 3 | **Perturbation-propagation profiling**: ablate or degrade one component; measure how widely and how fast compensatory adjustment propagates | Functional (not just topological) integration; cheap to run on artificial systems | Conflates integration with fragility unless compensation is distinguished from cascade failure |
| 4 | IIT-style integrated-information approximations (Φ proxies) | Principled "more than the sum of parts" formalization | Computationally intractable beyond toy systems; imports contested consciousness commitments Soul Frame deliberately avoids (P7 boundary) |

### Judgment

**Viable candidate exists.** The recommended composite is **1 + 3**:
constraint-closure ratio as the structural instrument, validated against
perturbation-propagation profiling as the behavioral check (a genuinely
integrated organization should show closure *and* distributed
compensation). Approach 2 is a cheap screening proxy only; approach 4 is
**rejected** for Soul Frame use — cost and consciousness entanglement.

For artificial systems the raw material (dependency graphs, service
telemetry, chaos-engineering-style perturbation runs) already exists in
production tooling; the invention needed is the mapping and the closure
criterion, not the data collection. This parallels the metrics note's B4
finding ("needs mapping, not invention") more closely than its flat "No"
suggested — B3 is *buildable*, not measurable-today.

---

## 2. Psychological continuity

**Target claim to operationalize:** "the same adaptive/modeling process"
persists over time — as distinct from *similar outputs* (the metrics
note's explicit objection to output-based proxies). Load-bearing in TE-02
(is the restored process the same one?), TE-04 (H6 channel under gradual
replacement), TE-08 (degraded continuity when the autobiographical channel
fails).

### Candidate approaches

| # | Approach | What it measures | What it misses |
|---|---|---|---|
| 1 | **Representational continuity**: compare internal state-space *structure* across time slices (representational similarity analysis between model snapshots — geometry of the representation space, not I/O behavior) | Whether the modeling process itself evolved continuously vs. was swapped for a behavioral look-alike; directly answers the "same process, not similar outputs" requirement | Requires white-box access; no established threshold for "continuous enough"; substrate-specific (embedding spaces vs. neural recordings don't compare) |
| 2 | **Overlapping-chains connectedness** (Parfitian chains operationalized): connectedness of the retrieval graph over the autobiographical store — do current states causally draw on states that drew on earlier states, in overlapping spans? | Graded continuity that degrades the way TE-08 needs (severe amnesia → sparse graph → DEGRADED, not binary death) | Measures the memory channel only (C4/C5-adjacent); a process could stay "the same" while its store is pruned |
| 3 | **Longitudinal disposition profiling**: repeated CAPS-style if-then situation×response probes (SF-R094), tracking profile drift over time | Behavioral-level continuity of character/valuation (P4, P12) without white-box access | It is exactly an output-similarity measure — the thing the metrics note warns is *adjacent to*, not identical with, process continuity |

### Judgment

**Viable candidate exists, with a caveat.** **Approach 1** is the only
candidate that measures the continuity claim itself rather than a
correlate; it should be primary. Approach 2 is a complementary channel
instrument (autobiographical continuity specifically), and approach 3 is
**admissible only as corroborating evidence**, never as the operational
definition — otherwise the instrument re-imports the output-similarity
error the metrics note flagged.

Caveat: approach 1 is currently substrate-specific. A cross-substrate
psychological-continuity metric remains out of reach; what is scoped here
is an instrument *for artificial systems with inspectable internals*.

---

## 3. Overall conclusion

Both dimensions clear the scoping bar set by SOUL-WP-0010-T02's done
condition: at least one candidate operational approach each —
**constraint-closure ratio + perturbation profiling** for B3;
**representational continuity** (with overlapping-chains as channel
instrument) for psychological continuity. Neither is viable *today* as a
validated metric; both are viable *build targets* with no missing
theoretical prerequisite.

Whether to actually build/pilot either instrument is a founder-gated
prioritization question, recorded as a hub decision (origin: residual,
SOUL-WP-0010) rather than opened as a workplan here.