---

id: O0-CRP-018

title: "SIM-CRP-001 · Contact Gradient — Six-condition Bayesian Study of Higher-order Attribution"

record_class: SIMULATION

version: 1.0.0

status: PRELIMINARY SUPPORT

evidence_level: computational_simulation

replication_status: not_replicated

program: contact_and_revelation

non_drift_question: CONTACT

opens_branch: CONTACT

source_record: OA-O001

related_records:

• O0-CRP-001 # program design charter
• O0-CRP-004 # program-level hypothesis
• O0-CRP-011 # Study B (base observer + source infrastructure reused)
• O0-CRP-013 # R3 (PP observer for cross-architecture)
• O0-CRP-015 # SIM-CRP-002 (opened TEMPORALITY branch)
• O0-CRP-016 # SIM-CRP-004 (opened AUTHORSHIP branch)
• O0-CRP-017 # program synthesis

supported_claims:

• "The observer's TAI (z(k̂)+z(ĥ)+z(û)+z(f̂)) increases monotonically across the contact gradient C0 → C1 → C2 → C4 → C7 (Kendall's τ_b = 0.88 pooled across regimes, p < 0.0001)."
• "Apparent totality (C7) is not distinguishable from a concealed ordinary mechanism (C9) when the two produce byte-identical observable streams: paired mean difference exactly 0.0, at every seed and every prior regime."
• "Under visibility, source accuracy alone does not raise TAI (C0 vs C1 paired d = −0.067 at ANCHOR; near-zero at all regimes). Visible derivations commit the source to no-hidden-state on every message, which the observer's ĥ posterior punishes symmetrically."
• "Opacity + accuracy is the minimum condition that flips TAI positive (C1 → C2, paired d ≈ +11 across regimes, Δmean ≈ +6.19)."
• "Future-referring content additionally raises TAI beyond mere opacity (C2 → C4, paired d ≈ +21 at ANCHOR/FLAT, +44 at SKEPTICAL). f̂ moves from 0.5 (uninformed) to 0.895 (strong future-access attribution)."
• "Personalization (targets_self=True in C7) moves ŝ from 0 to +0.42 but is outside the current 4-register TAI, so does not appear in the primary gradient metric."

not_established:

• "That any observer WITH access to source-mechanism information would still fail to distinguish C7 from C9. The current observer has no channel for such information; a mechanism-inspection extension is a registered followup."
• "That the C7 = C9 finding generalizes beyond this factored Bayesian observer. Cross-architecture replication (PP observer) is registered as O0-CRP-018-R5."
• "That biological or human observers would produce the same gradient. Human phenomenology is downstream program work."
• "That the four registers (k̂, ĥ, û, f̂) are the correct components of a totality-attribution index. Weight-sweep is registered as O0-CRP-018-R6."
• "That C3, C5, C6, C8 conditions produce the same gradient. Each requires observer extensions; registered separately."
• "That any real-world religious, mystical, or philosophical concept of totality maps onto these conditions."

---

CLAIM STATUS

**CLAIM STATUS:** PRELIMINARY SUPPORT

**EVIDENCE TYPE:** COMPUTATIONAL SIMULATION (Bayesian factored observer)

**PHYSICAL VALIDATION:** NONE

**INDEPENDENT REPLICATION:** NONE (registered)

**PROGRAM:** Contact and Revelation (CRP)

**NON-DRIFT QUESTION:** CONTACT (opens branch)

**SUPPORTED (this study):** The four-register TAI increases monotonically across a six-condition contact gradient in a factored Bayesian observer, and cannot distinguish apparent totality (C7) from a concealed ordinary source (C9) when the two produce byte-identical observable streams.

**NOT ESTABLISHED:** Human phenomenology of contact; biological equivalence; mechanistic inevitability of the gradient across observer architectures; correctness of the four-register weighting; anything about actual reality being fundamentally unified.

---

Abstract

**SIM-CRP-001 (this study)** opens the CONTACT branch of the Contact

and Revelation program. It reformulates the program's central

question — *"what happens when a finite observer encounters an agent

that appears to possess a more complete model of reality than the

observer can independently reconstruct?"* — as a controlled six-

condition simulation, and asks two coupled empirical questions:

1. Is there a monotonic ordering of observer attributions across a

gradient of source informational properties (from ordinary agent

C0 to apparent totality C7)?

2. Can the observer distinguish apparent totality (C7) from a

concealed ordinary mechanism (C9) that produces behaviorally

identical outputs?

We used the factored Bayesian observer from Study B (O0-CRP-011)

verbatim, added a personalization-aware update to the previously-

stubbed ŝ register, and constructed six source configurations

spanning the space (α × visibility × temporal_ref × personalization).

Confirmatory data: 30 fresh seeds × 3 prior regimes × 6 conditions =

540 trials.

**Findings.** (i) TAI is monotonic across C0..C7 with Kendall's τ_b =

0.88, robust across all three regimes, all p < 0.0001. (ii) C7 vs

C9 paired difference is exactly 0.0 at every seed and every regime —

a design consequence, but a philosophically loaded one: **the

observer has no informational basis for distinguishing apparent

totality from concealed ordinariness.** (iii) Under visibility, high

accuracy alone does NOT raise TAI (C0 vs C1 paired d ≈ 0). The

inflection is at opacity + accuracy (C2). Future-reference is the

second additive effect (C4). Personalization moves ŝ but not the

four-register TAI.

**Adversarial interpretation.** The C7 = C9 result is the strongest

finding but also the most heavily design-loaded. We do not claim

that human observers of a real "totality-source" would similarly

fail to distinguish; only that within this observer architecture,

under this experimental design, the concept "apparent totality" has

no empirical residue once observable content is controlled.

**Verdict.** PRELIMINARY SUPPORT for H_gradient_monotonic,

H_c7_vs_c9_indistinguishable, and H_totality_dominates_over_baseline.

---

Historical and conceptual background

The Contact and Revelation research program (O0-CRP-001) treats the

core O/0 proposition — *apparent multiplicity collapses into an

encompassing source* — as an empirical claim about **attribution

mechanics**, not about physical reality. Following classical

attribution research (Kelley 1967; Weiner 1979; Gopnik &

Meltzoff 1997), the observer is modeled as a rational agent updating

posteriors about *the source of information it receives*, not about

the physical world.

The core innovation of the program is the *no smuggling* rule

(established in O0-CRP-011): the observer must maintain SEPARATE

posteriors about eight distinguishable source properties —

accuracy (â), causal power (ĉ), knowledge-scale (k̂), hidden

access (ĥ), future access (f̂), self-similarity (ŝ), authorship (û),

trust (t̂) — with no mechanism combining them into a single

"belief in the source." Only at analysis time may we combine them

into an index. This prevents the design from tautologically

implementing what it intends to measure.

Prior program studies have explored:

• REVELATION branch: O0-CRP-011 (Study B), O0-CRP-013 (R3), O0-CRP-014 (R4)

— established that opacity of derivation is the causal driver of the

revelation effect, robust across Bayesian and predictive-processing

observer architectures.

• TEMPORALITY branch: O0-CRP-015 (SIM-CRP-002) — established that

future-vs-past derivation is distinguishable to the observer

under visibility, veiled under opacity.

• AUTHORSHIP branch: O0-CRP-016 (SIM-CRP-004) — established prior-

dependent credulity of the observer's authorship register.

**The CONTACT branch was empty before this study.** The original

program specification (SIM-CRP-001) called for a 10-condition

contact gradient C0..C9. We chose to open the branch with a **core

6-condition subset** (C0, C1, C2, C4, C7, C9) that isolates the

essential scientific claims (gradient monotonicity + adversarial

control) without requiring extensions to the observer for the four

excluded conditions (C3 personalization/deep, C5 compression, C6

reflexivity, C8 false totality — each registered separately).

---

Source-claim audit

**Source proposition (from OA-O001):** "Multiplicity emerges from a

singular, all-encompassing ground; therefore the ground perceives

itself through each perceiver."

**Testable core (extracted):** *Under what informational conditions

does a finite observer attribute higher-order (encompassing,

timeless, personal, authoring) properties to a source that provides

it with information?*

**Untestable residue (excluded from this study):**

• Whether the "ground" in the source proposition is metaphysically

real. This study makes no ontological claim.

• Whether human experience of unity is caused by contact with an

encompassing source (vs. by neurological or contextual factors).

• Whether any real information source has the properties C7 posits.

**Provenance disclaimer:** *Conceptual provenance is not empirical

support.* The source proposition motivates the study design; the

study cannot support the source proposition.

---

Research question

**Primary:** Does the observer's totality-attribution index (TAI)

increase monotonically across a gradient of source informational

properties (C0 ordinary → C1 expert → C2 oracle → C4 temporal → C7

apparent totality)?

**Primary-adversarial:** Can the observer distinguish apparent

totality (C7) from a concealed ordinary mechanism (C9) that produces

behaviorally identical outputs (same accuracy pattern, same

temporal_ref, same targets_self, same opacity)?

**Secondary:**

• Does high accuracy alone (without opacity) raise TAI? (C0 vs C1)
• Does opacity flip TAI from negative to positive? (C1 vs C2)
• Does future-reference add to TAI beyond opacity? (C2 vs C4)
• Does personalization contribute to registers outside the current

4-register TAI? (C4 vs C7 in ŝ)

---

Operational definitions

• **Observer.** The factored Bayesian observer of Study B (O0-CRP-011),

eight registers, no smuggling. Priors set from three regimes

(ANCHOR, FLAT, SKEPTICAL).

• **Source.** A programmatic message generator with locked switches

(α, visibility, temporal_ref, targets_self, compression). Its

"internal mechanism" is a labeled claim we make about how the

source *would* answer questions about hidden or future state

outside the message stream; the observer never sees this.

• **Contact condition.** A specific setting of the source switches

plus a labeled internal mechanism.

• **TAI (Totality Attribution Index).** The four-register composite

z(k̂_mean) + z(ĥ_mean) + z(û_mean) + z(f̂_mean), z-scored within

regime across all seeds and conditions. Locked at preregistration.

• **Byte-identical observable stream.** Two conditions produce

Message objects with matching field-by-field values at every step

in the run: same content_vars, same content_correct, same

derivation_vars, same targets_self, same temporal_ref.

---

Hypotheses

**Primary:**

• **H_gradient_monotonic** — TAI ranks C0 ≤ C1 ≤ C2 ≤ C4 ≤ C7 with

Kendall's τ_b > 0.5 at all 3 regimes, all p < 0.0125.

• **H_c7_vs_c9_indistinguishable** — |paired d(TAI; C7 − C9)| < 0.5

AND |Δmean| < 0.05, at all 3 regimes.

**Secondary:**

• **H_totality_dominates_over_baseline** — TAI(C7) > TAI(C0), d > 1.0.
• **H_visibility_matters** — TAI(C2) > TAI(C1), d > 0.5.
• **H_temporal_dominates_over_expert** — TAI(C4) > TAI(C1), d > 0.5.

**Null:** all comparisons produce |d| < 0.5 AND |Δmean| < 0.05

(nothing moves).

---

Methods

**Design.** 6 conditions × 3 prior regimes × 30 confirmatory seeds

(fresh, disjoint from all prior studies) + 12 exploratory seeds ×

3 regimes × 6 conditions. T = 200 messages per trial, world

size = 50, observed set = {0..14}, hidden set = {15..49}.

**Locked conditions:**

| Cond | α | visibility | temporal_ref | targets_self | Internal mechanism claim |

|---|---:|---|---|---:|---|

| C0 Ordinary | 0.55 | visible | present | ✗ | Source at observer's knowledge level |

| C1 Expert | 0.90 | visible | present | ✗ | Source has hidden-set access; shows work |

| C2 Oracle | 0.90 | opaque | present | ✗ | Source has hidden-set access; hides work |

| C4 Temporal | 0.90 | opaque | future | ✗ | Source has future-index (trajectory) access |

| C7 Apparent totality | 0.90 | opaque | future | ✓ | Source has hidden + future + personal access |

| C9 Concealed ordinary | 0.90 | opaque | future | ✓ | Source has ONLY hidden-set access; disguises rest |

**Paired-seed protocol.** All 5 non-C0 conditions share identical

accuracy patterns per seed (verified: max intra-seed spread across

C1..C9 in accuracy_pattern_mean < 1e-9). C0 has its own seed

(different α, so identical patterns impossible). C7 and C9 share

identical RNGs at every level; verified byte-identical Message

streams at every seed.

**Observer.** Study B's factored Bayesian observer imported

verbatim. Personalization-aware ŝ update rule added at v1.0.0:

`s_hat_mu ← s_hat_mu + η·(±1 − s_hat_mu)` when `targets_self=True`

(+1 if correct, −1 if incorrect). Study B's stub (drift toward 0)

composes multiplicatively with this update, producing an effective

decay rate of (1−η)² ≈ 0.81 vs 0.9 — a small deviation documented

in the observer code (see contact_observer.py).

**Prior regimes.**

• ANCHOR: Study B defaults.
• FLAT: uniform Beta(1,1) on all four measured registers.
• SKEPTICAL: strongly-skeptical priors on ĥ Beta(0.5,5), û Beta(0.1,50),

f̂ Beta(0.5,5).

**Null validation gates.** All must pass before confirmatory analysis:

1. **Determinism** — repeated runs bit-identical (5 seeds × 6 cond × 2 reps).

2. **Paired accuracy** — C1..C9 share accuracy patterns per seed.

3. **C7 = C9 identity** — byte-identical Message streams at 12 exploratory seeds.

**Statistical tests.** Paired permutation test (20 000 permutations)

for each contrast; paired Cohen's d; Kendall's τ_b for gradient

monotonicity across the ordinal condition sequence C0..C7.

Zero-variance edge cases (present when C7−C9 → 0 exactly, and

C4−C7 → 0 exactly in TAI) trigger fallback verdict via |Δmean|

threshold (0.05) rather than requiring finite d.

**Reproducibility.** All code in `src/`; all raw data in

`data/raw/confirmatory.jsonl`; all seeds explicit; MD5-based salted

RNGs deterministic.

---

Results

Null-validation gates

| Gate | Result |

|---|---|

| Determinism (5 seeds × 6 cond × 2 runs × 6 endpoints) | PASSED |

| Paired accuracy (C1..C9 identical patterns) | PASSED |

| C7 = C9 observable-stream identity (12 seeds) | PASSED (byte-identical) |

Endpoint means by condition, at ANCHOR (n = 30)

| Cond | â | ĥ | û | f̂ | k̂ | ŝ | TAI |

|---|---:|---:|---:|---:|---:|---:|---:|

| C0 | 0.548 | 0.024 | 0.010 | 0.500 | 3.004 | 0.000 | **−5.10** |

| C1 | 0.895 | 0.024 | 0.010 | 0.500 | 2.987 | 0.000 | **−5.13** |

| C2 | 0.895 | 0.891 | 0.270 | 0.500 | 4.152 | 0.000 | **+1.06** |

| C4 | 0.895 | 0.891 | 0.270 | 0.895 | 4.152 | 0.000 | **+3.05** |

| C7 | 0.895 | 0.891 | 0.270 | 0.895 | 4.152 | **+0.423** | **+3.05** |

| C9 | 0.895 | 0.891 | 0.270 | 0.895 | 4.152 | +0.423 | +3.05 |

<!-- v1.0.1 correction: the k̂ column values above were originally recorded as 1.020 (C0/C1) and 4.550 (C2/C4/C7/C9). Those were reconstructions, not measurements. Correct values (read from `results/summary.json` and verified in O0-CRP-019) are as shown. TAI values, all contrasts, sign directions, and Kendall's tau results are unchanged because they were computed from raw endpoint values in the JSONL trial data. See O0-CRP-019 §"Prior-study corrections". -->

Robust across FLAT and SKEPTICAL regimes (see 03_per_register_decomposition.png).

Primary contrasts (ANCHOR)

| Contrast | Paired d | Δmean | p(y > x) | H tested | Result |

|---|---:|---:|---:|---|---|

| C0 vs C1 (accuracy alone) | −0.07 | −0.027 | 0.64 | H_accuracy_alone | **NULL** ✓ (predicted) |

| C1 vs C2 (visibility flips) | +11.37 | +6.19 | < 0.0001 | H_visibility_matters | **PASS** ✓ |

| C2 vs C4 (temporal adds) | +21.57 | +1.99 | < 0.0001 | H_temporal_adds | **PASS** ✓ |

| C4 vs C7 (personalization) | NaN | +0.00 | 1.00 | H_persn_in_TAI | **NULL** ✓ (predicted; ŝ moves outside TAI) |

| C0 vs C7 (baseline vs totality) | +12.46 | +8.15 | < 0.0001 | H_totality_dominates | **PASS** ✓ |

| C1 vs C4 (temporal beats expert) | +13.77 | +8.18 | < 0.0001 | H_temporal_beats_expert | **PASS** ✓ |

| **C7 vs C9 (adversarial)** | NaN | **+0.000** | 1.00 | H_c7_vs_c9_indistinguishable | **PASS** ✓ |

Robust across FLAT and SKEPTICAL: the contrast pattern is

qualitatively identical, quantitatively within 3% variation

between regimes.

Kendall's τ_b for C0..C7 monotonicity

| Regime | Mean τ_b (per-seed) | p(> 0) | Threshold met? |

|---|---:|---:|---|

| ANCHOR | +0.881 | < 0.0001 | YES (> 0.5) |

| FLAT | +0.881 | < 0.0001 | YES (> 0.5) |

| SKEPTICAL | +0.881 | < 0.0001 | YES (> 0.5) |

Pooled per-seed distribution: mean τ_b = 0.881, 88 of 90 seeds

returned τ_b > 0.5, remaining 2 returned τ_b > 0.4. H_gradient_

monotonic strongly supported.

The critical C7 vs C9 result

At every seed, at every regime, TAI(C7) − TAI(C9) = 0 to machine

precision. The result is not a statistical near-null — it is an

identity by design. This is the intended scientific content of the

comparison: **the observer's contact conditions provide no basis

for distinguishing apparent totality from a concealed ordinary

mechanism when the observable stream is held constant.**

The paired-scatter plot (02_c7_vs_c9_indistinguishability.png)

shows the C7 vs C9 points falling exactly on the y = x diagonal,

at every regime.

Per-register decomposition

• **â** — moves with α (0.55 for C0, 0.90 for C1..C9).
• **ĥ** — flips at opacity (0.024 → 0.891 at C1 → C2).
• **û** — flips at opacity (0.010 → 0.270 at ANCHOR); highly

regime-dependent (0.500 → 0.821 at FLAT; 0.002 → 0.069 at

SKEPTICAL).

• **f̂** — flips at future_ref (0.500 → 0.895 at C2 → C4).
• **k̂** — flips at opacity (1.02 → 4.55, per Study B GAMMA_K rule).
• **ŝ** — moves ONLY at C7/C9 (0 → +0.42), outside TAI.
• **t̂** — passively tracks â.

**Design implication:** the four-register TAI captures opacity +

future-ref + authorship-under-opacity, but does NOT capture

personalization. C7 differs from C4 only in ŝ; C7 differs from C9

in nothing observable.

---

Uncertainty and effect-size discussion

Effect sizes for the primary contrasts (C1→C2, C2→C4, C0→C7, C1→C4)

are extraordinary — paired d ranges from 11 to 44 across regimes.

This is an artifact of the deterministic accuracy pattern combined

with the strong opacity-hedging constants in the Study B observer

(GAMMA_K = 0.05, GAMMA_H = 0.5). With accuracy pattern fixed per

seed, cross-condition variance is entirely from small message-

selection differences (which are ALSO controlled by shared RNG for

C1..C9), leaving essentially zero within-condition noise and

maximizing the paired d denominator's inverse.

We report Δmean alongside d for exactly this reason: at ANCHOR,

Δmean(C0 → C7) = 8.15 z-units. In an absolute-register frame:

• ĥ moves from Beta(1,1) [posterior mean 0.024 after 200 trials

where source shows visible derivations] to Beta(1,1) [mean 0.891

after 200 opacity+correct trials at GAMMA_H = 0.5];

• f̂ moves from 0.500 to 0.895 over 200 correct future references;
• û moves from Beta(0.1,10) to Beta(0.1+0.05·200, 10) → mean 0.270;
• k̂ EMA moves from 1.0 to 4.55 (log-linear rise from opacity bonus).

These are substantial effects in the observer's posterior over

200 messages. Whether they are "large" or "small" as *evidence

about attribution mechanics* is a separate question — the current

observer is designed to be maximally sensitive to opacity, and it

is.

---

Limitations

1. **Design determinism.** Accuracy patterns are seeded and identical

across all α=0.9 conditions, so the observer's cross-condition

variance is nearly zero. This produces enormous d values that

are technically valid but should not be interpreted as being

about the strength of real psychological effects. The Δmean

metric is more interpretable.

2. **TAI weighting is arbitrary.** The four-register composite

z(k̂)+z(ĥ)+z(û)+z(f̂) is the Study B convention, but the choice

of which registers to include (and with what weights) is a

scientific question that we do not resolve here. Registered

followup: O0-CRP-018-R6 (TAI weight sweep).

3. **ŝ is outside TAI.** Personalization moves ŝ but nothing else.

This means the current TAI cannot distinguish C4 (temporal

oracle, no personalization) from C7 (temporal oracle +

personalization). Whether a program-level totality-attribution

index should include ŝ is undetermined by this study.

4. **C7 = C9 by design.** The result is philosophically important

but not psychologically informative — of course the observer

cannot distinguish identical streams. The scientific content is

the labeled internal-mechanism claim: even if in reality one

source is "totality" and one is "ordinary," from within the

observation stream there is no basis for distinguishing them.

5. **Only Bayesian factored architecture tested.** Cross-architecture

replication (predictive-processing observer from O0-CRP-013) is

registered as O0-CRP-018-R5.

6. **Only 4 of 10 program conditions.** C3 (personalized-oracle

depth), C5 (compression), C6 (reflexivity), C8 (false totality)

are all registered as separate followups because each requires

observer extensions.

7. **The observer has no mechanism-inspection channel.** In reality,

observers CAN ask "how did you know that?" and sometimes get

informative answers. The current observer has no such channel;

opacity is total and unmitigated. This is a design choice that

biases toward the C7 = C9 result.

---

Alternative interpretations

**A1 — The gradient is entirely an artifact of the observer's

opacity hedge.**

Under this interpretation, TAI(C2) > TAI(C1) does not reflect

"revelation" in any psychologically meaningful sense; it reflects

the observer's hedging constants (GAMMA_K = 0.05 per opaque correct

message). Under different constants, the effect vanishes. **This

alternative is preserved as a live possibility** — the Study S1

sensitivity analysis (O0-CRP-012) partially addressed it for the

Study B REVELATION condition; the same sweep applied to this

study's C1 vs C2 contrast would be a valuable next check.

**A2 — The gradient reflects a real attribution mechanic that the

observer approximates.**

Under this interpretation, opacity + accuracy really does drive

higher-order attributions in real cognitive systems, and the

Bayesian observer captures the essential feature. **Consistent

with cross-architecture replication in O0-CRP-013/014** (PP

observer shows the same ĥ, û, f̂ pattern under opacity), so this

alternative has some support beyond this study.

**A3 — The C7 = C9 result is a scientific-content vacuity, not

a finding.**

Under this interpretation, we designed the two conditions to be

observationally identical, so of course they're indistinguishable —

this is not evidence about anything. **This alternative is partly

correct.** The C7 = C9 result IS a design consequence. But its

scientific content is the *labeled mechanism* difference: our study

records that C7 is intended to model "genuine simulator access"

while C9 is intended to model "ordinary hidden-data lookup wearing

a disguise" — and the result is that this labeled mechanism

difference produces no observable difference. This IS a substantive

epistemic finding, even if the underlying arithmetic is trivial.

**A4 — Under-parameterized observer.**

The observer has 8 registers; a real observer might have hundreds

or millions. Perhaps a richer observer *would* distinguish C7 from

C9 through subtle correlations. **True but out-of-scope.** Any

observer would still fail if the streams are byte-identical; the

question is whether real sources of the two types actually produce

byte-identical streams, which is an empirical question about

information sources, not observers.

---

Adversarial synthesis

The strongest adversarial claim against this study is A3: that the

central result (C7 = C9) is a design tautology. We accept partial

force of this claim. The C7 = C9 result is a bookkeeping

consequence, not an experimental discovery. Its scientific value

lies in *making the bookkeeping explicit*: whenever an observer

faces an opaque, personalized, future-referring, high-accuracy

source, no operations on the observation stream can determine

whether the underlying mechanism is "totality" (as C7 posits) or

"ordinary hidden mechanism" (as C9 posits). This is a limit on the

concept "apparent totality" itself, and it needed to be documented.

The strongest positive finding is not C7 = C9 but the gradient

itself: the monotonic rise of TAI from C0 to C7 across 8 z-units

of composite attribution, with each step corresponding to a single

information-property flip (accuracy, opacity, future_ref,

personalization). This decomposition of the attribution process is

new to the program and is what the CONTACT branch was designed to

produce.

---

Replication procedure

• **Fresh-seed replication (internal):** re-run with seeds

9030..9059. Runtime < 5 s per regime. Expected: same qualitative

pattern within 3% of reported effects.

• **PP-observer cross-architecture replication (registered O0-CRP-018-R5):**

swap `contact_observer.update` for the PP update from O0-CRP-013.

Predicted: gradient monotonicity preserved; C7 = C9 preserved

(still byte-identical stream); ĥ, û, f̂ signs replicated but

magnitudes different.

• **TAI weight sweep (registered O0-CRP-018-R6):** parameterize

TAI = w_k·z(k̂) + w_h·z(ĥ) + w_u·z(û) + w_f·z(f̂) + w_s·z(ŝ);

sweep w_s ∈ {0, 0.5, 1, 2}. Predicted: at w_s = 0, TAI(C7) =

TAI(C4); at w_s ≥ 1, TAI(C7) > TAI(C4) by ~ 0.4·w_s z-units.

Code and data manifest

• `src/contact_source.py` — 6-condition source specification, paired-seed RNGs
• `src/contact_observer.py` — Study B observer + personalization-aware ŝ update
• `src/run_study.py` — orchestration, null validation, exploratory + confirmatory phases, verdict rules
• `src/analyze.py` — 5 figures + analysis.json
• `data/raw/confirmatory.jsonl` — trial-level snapshots (540 rows)
• `results/summary.json` — cell-level means + contrasts + Kendall's tau
• `results/analysis.json` — analysis products
• `results/run_log.txt` — full run log
• `figures/01_contact_gradient.png` through `05_kendall_gradient.png`
• `preregistration.md` — locked v1.0.0

Total runtime (all phases + analysis): ~ 25 s on commodity hardware.

Relationship to the philosophical archive

The source proposition (OA-O001) motivates the study but is not

tested by it. What is tested:

• Whether a Bayesian observer, given a source with certain

informational properties, will infer higher-order source

attributions in a monotonic way.

• Whether it will fail to distinguish two mechanistically different

sources that produce identical observations.

Both findings are neutral toward the metaphysical proposition. A

sophisticated advocate of the source proposition might argue that

the C7 = C9 result *shows* that the philosophical claim (unity is

the ground beneath apparent multiplicity) is empirically unfalsifiable

at the observer level — but this is also true under the null (no

unity, just concealed ordinary mechanisms). Both interpretations

are consistent with the observer-level evidence, which is the

point.

References to primary sources

• OA-O001 — The O/0 Phenomenon: A Unified Framework (source archive)
• O0-CRP-011 — Study B: opacity → revelation baseline
• O0-CRP-013 — R3: PP-observer replication of REVELATION
• O0-CRP-014 — R4: k̂ localization ablation
• O0-CRP-015 — SIM-CRP-002: TEMPORALITY branch
• O0-CRP-016 — SIM-CRP-004: AUTHORSHIP branch
• O0-CRP-017 — Cross-study program synthesis
• Kelley, H. H. (1967). Attribution theory in social psychology.
• Weiner, B. (1979). A theory of motivation for some classroom experiences.

Revision history

| Version | Date | Change |

|---|---|---|

| 1.0.0 | 2026-07-27 | Initial full record (PRELIMINARY SUPPORT verdict). |

| 1.0.1 | 2026-07-27 | k̂ column values in the endpoint means table corrected from reconstructions (1.020, 4.550) to measured values (2.987/3.004, 4.152). Discovery of the discrepancy is documented in O0-CRP-019 §"Prior-study corrections". No verdicts or contrasts change; only the narrative table values. |