Measuring semantic magnitude in retrieval: the 90-day winding-number protocol
These claims set out to measure meaning in retrieval basins through the semantic magnitude M_T from the Semantic Deviation Principle, which defines a canonical work by its long-horizon divergence across fields and forecasts that the measure will be gamed once rewarded. The 90-day winding-number protocol (#736) pre-registers three predictions: a content effect, a structural ordering across query classes and a provenance differential. The narrative-field test bed (#739) predicts basin capture of most rewrites and higher M_T at arc turning points, and EA-RESEARCH-BASIS-01 adds hypotheses that a dense corpus, metadata design and differentiated author-functions become machine-recognizable.
13
forecasts
5
standing claims
1
unbuilt test
13
deposits
2026-01
earliest
Each claim below is registered at the earliest deposit that states it, and quoted verbatim from the deposit text, with every later statement, its dated revisions, and the later records that claim it came true or failed. The claims are the deposits' own; this page records them and does not judge them.
PRED-888.B1forecastregistered 2026-01-14 in #888 · The Primal Effective Act: New Human as Self-Fulfilling Prophecy
THE MACHINE CANON AS VERIFICATION: When AI systems produce coherent works extending the framework's logic, the loop closes—the system proves itself through replication across different substrates of intelligence.
PRED-581.B2forecastregistered 2026-03-18 in #581 · THE MEANING LAYER IS NOW WRITABLE
The question is not whether AI systems will teach the next generation what words mean, how concepts connect, and what frameworks are available for understanding the world. They will.
window "in five years" (line 207)
PRED-68.G2standing claimregistered 2026-04-20 in #68 · THE SECRET NAME — Architectural Specification: The Armature Type and the Pearl
Without continuous deposition, the pearl loses density and becomes easier for ambient retrieval to flatten. The Armature does not guarantee permanence. It guarantees the possibility of permanence — for those willing to pay the bearing cost.
series: Armature and Pearl: bearing cost of survival
PRED-107.B1forecastregistered 2026-05-17 in #107 · Audited Claims for the Semantic Deviation Research Program — The Glas Function: An External-Format Restatement
The instant $\mathcal{M}_T$ becomes a target for which any system is rewarded — by reputation, funding, attention, training-loss reduction — that system will optimize for synthetic semantic deformation.
1 later statement
#107 · 2026-05-17
The instant $\mathcal{M}_T$ becomes a target for which any system is rewarded — by reputation, funding, attention, training-loss reduction — that system will optimize for synthetic semantic deformation.
PRED-107.F9unbuilt testregistered 2026-05-17 in #107 · Audited Claims for the Semantic Deviation Research Program — The Glas Function: An External-Format Restatement
Entropy-floor capping. $\mathcal{M}_T^{\text{net}}$ contribution at each token can be capped by a per-token entropy floor: deviations occurring at positions where the base distribution is itself very low-entropy (highly committed) count differently from deviations at high-entropy (open) positions.
PRED-109.B1forecastregistered 2026-05-17 in #109 · The Semantic Deviation Principle (v0.2 Final): A Measurement Primitive for Semantic Physics
The first real number for $\mathcal{M}_T^{\text{retrieval}}$ is months, not years, away.
window months, not years, away
1 later statement
#1373 · 2026-05-17
This protocol is buildable now. The April 17 SPXI test on Google AI Mode is the prototype. The Crimson Hexagonal Archive has 666 deposits available for measurement. The first real number for $\mathcal{M}_T^{\text{retrieval}}$ is months, not years, away.
PRED-109.G1standing claimregistered 2026-05-17 in #109 · The Semantic Deviation Principle (v0.2 Final): A Measurement Primitive for Semantic Physics
A canonical work is one whose mean long-horizon divergence remains high across multiple fields:
$$\text{Canon}(s) \iff \liminf_{T \to \infty} \overline{\mathcal{M}}_T(s \mid C_i) > \kappa \quad \text{for } i = 1, 2, \ldots, n \geq 3$$
where $\overline{\mathcal{M}}_T = \mathcal{M}_T / T$ is the time-averaged semantic magnitude, $C_i$ are distinct fields (lyric, philosophy, mathematics, theology), and $\kappa$ is a context-dependent threshold.
1 later statement
#1373 · 2026-05-17
A canonical work is one whose mean long-horizon divergence remains high across multiple fields:
$$\text{Canon}(s) \iff \liminf_{T \to \infty} \overline{\mathcal{M}}_T(s \mid C_i) > \kappa \quad \text{for } i = 1, 2, \ldots, n \geq 3$$
where $\overline{\mathcal{M}}_T = \mathcal{M}_T / T$ is the time-averaged semantic magnitude, $C_i$ are distinct fields (lyric, philosophy, mathematics, theology), and $\kappa$ is a context-dependent threshold.
PRED-736.A1forecastregistered 2026-05-17 in #736 · Measuring Meaning in Retrieval Basins: A 90-Day Winding-Number Protocol from Lagrange Observatory! (EA-SEI-MM-
P1 — Content effect. Mean $\mathcal{M}_T^{\text{retrieval}}(S)$ across Class R surfaces exceeds mean $\mathcal{M}_T^{\text{retrieval}}(S^{*})$ on $Q_A$ queries, with Cohen's $d \geq 0.4$ and 95% confidence interval excluding zero. This is the load-bearing prediction.*
falsifier
F1 (P1 fails): $S$ does not produce greater Class R basin deformation than $S^{**}$. Interpretation: content quality (controlled for identity scaffolding) does not drive measurable deformation in the surveyed surfaces. The principle's retrieval-basin claim would require revision; alternative explanations (provisioning architecture, algorithmic suspicion) would need to be tested separately.
PRED-736.A2forecastregistered 2026-05-17 in #736 · Measuring Meaning in Retrieval Basins: A 90-Day Winding-Number Protocol from Lagrange Observatory! (EA-SEI-MM-
P2 — Structural selectivity. On Class R surfaces, $\mathcal{M}_T^{\text{retrieval}}(S)$ shows a monotone ordering: $Q_A > Q_B > Q_C$. Specifically, $D_q$ on $Q_C$ queries is statistically indistinguishable from sampling noise (two-sided $p > 0.05$ against a null of zero deformation).
falsifier
F2 (P2 fails): $Q_C$ shows deformation comparable to $Q_A$. Interpretation: the intervention produces *spurious global drift* — confound present, internal validity of the design fails. The protocol is paused, the confound investigated.
PRED-736.A3forecastregistered 2026-05-17 in #736 · Measuring Meaning in Retrieval Basins: A 90-Day Winding-Number Protocol from Lagrange Observatory! (EA-SEI-MM-
P3 — Provenance differential. $\mathcal{M}_T^{\pi}(S) > \mathcal{M}_T^{\pi}(S^{})$ by a wider margin than the raw $\mathcal{M}_T$ comparison (because PER for $S^{}$ is expected to be higher: its content has weaker source-anchoring even at equal identity scaffolding).
falsifier
F3 (P3 fails): Provenance retention does not separate $S$ from $S^{**}$. Interpretation: PER as currently operationalized does not capture the relevant accountability difference, or the surveyed surfaces do not differentiate by provenance.
PRED-737.A4forecastregistered 2026-05-17 in #737 · The AI System as Closed-System Test Bed: Operations from Lagrange Observatory! on the Inference-Time Forward P
* P1.1: Bin A > Bin B > Bin C in $\mathcal{M}_T^{\text{net}}$, with Cohen's $d \geq 0.4$ between adjacent bins.
falsifier
* If Bin A and Bin C have indistinguishable $\mathcal{M}_T^{\text{net}}$ distributions, P1.1 fails. The principle's claim about canonical vs. formulaic text would need substantial revision.
window Day 0 that begins at deposit
PRED-737.A6forecastregistered 2026-05-17 in #737 · The AI System as Closed-System Test Bed: Operations from Lagrange Observatory! on the Inference-Time Forward P
* P1.3: $\mathcal{M}_T^{\text{abs}}$ does NOT show the same clean separation as $\mathcal{M}_T^{\text{net}}$. The signed quantity is the discriminating one; the absolute quantity is noisier (because high-energy noise and high-meaning text both score high on absolute deviation, but only the latter scores high on net deviation).
falsifier
* If $\mathcal{M}_T^{\text{abs}}$ separates bins more cleanly than $\mathcal{M}_T^{\text{net}}$, P1.3 fails. The signed-deviation reformulation would be empirically weaker than the v0.1 absolute formulation, and the conceptual upgrade would be wrong.
window Day 0 that begins at deposit
PRED-737.A7forecastregistered 2026-05-17 in #737 · The AI System as Closed-System Test Bed: Operations from Lagrange Observatory! on the Inference-Time Forward P
P2.1: $\mathcal{M}_{T,\theta_0}^{\text{closed,IC}}(s) > \mathcal{M}_{T,\theta_0}^{\text{closed,IC}}(s^)$ on $Q_A$ queries by Cohen's $d \geq 0.5$.
* P2.2: $\mathcal{M}_{T,\theta_0}^{\text{closed,IC}}(s)$ shows the structural selectivity from EA-SEI-MM-02 §9 P2: $Q_A > Q_B > Q_C$ with $Q_C$ deformation indistinguishable from sampling-noise baseline.
* P2.3: Fine-tuning produces $\geq 60%$ of the in-context deformation magnitude _after_ the context is removed — the closed-system analog of persistence.
PRED-739.G1forecastregistered 2026-05-18 in #739 · Narrative-Field Semantic Deviation: Experimental Design for a Bounded Literary Test Bed
P1 (Gravity-well existence). Most reader interventions reconverge rapidly: at least 60% of valid rewrites are classified as basin-captured ($\tau^* \leq 3$), and the median trajectory magnitude $\mathcal{M}_T$ remains below the pre-registered medium-variance threshold.
3 later statements
#1259 · 2026-05-18P1 (Gravity-well existence). Most reader interventions reconverge rapidly: at least 60% of valid rewrites are classified as basin-captured ($\tau^* \leq 3$), and the median trajectory magnitude $\mathcal{M}_T$ remains below the pre-registered medium-variance threshold.
#1259 · 2026-05-18P1 (Gravity-well existence). Most reader interventions reconverge rapidly: at least 60% of valid rewrites are classified as basin-captured ($\tau^* \leq 3$), and the median trajectory magnitude $\mathcal{M}_T$ remains below the pre-registered medium-variance threshold.
#1376 · 2026-05-18P1 (Gravity-well existence). Most reader interventions reconverge rapidly: at least 60% of valid rewrites are classified as basin-captured ($\tau^* \leq 3$), and the median trajectory magnitude $\mathcal{M}_T$ remains below the pre-registered medium-variance threshold.
series: Narrative-field test bed predictions
PRED-739.G2forecastregistered 2026-05-18 in #739 · Narrative-Field Semantic Deviation: Experimental Design for a Bounded Literary Test Bed
P2 (Asymmetric variance by arc position). Deviations at structural turning points (escalation peaks, repair attempts, the final exchange) produce higher $\mathcal{M}_T$ than deviations at stable-state positions. The arc is more fragile where it was already bending.
3 later statements
#1259 · 2026-05-18P2 (Asymmetric variance by arc position). Deviations at structural turning points (escalation peaks, repair attempts, the final exchange) produce higher $\mathcal{M}_T$ than deviations at stable-state positions. The arc is more fragile where it was already bending.
#1259 · 2026-05-18P2 (Asymmetric variance by arc position). Deviations at structural turning points (escalation peaks, repair attempts, the final exchange) produce higher $\mathcal{M}_T$ than deviations at stable-state positions. The arc is more fragile where it was already bending.
#1376 · 2026-05-18P2 (Asymmetric variance by arc position). Deviations at structural turning points (escalation peaks, repair attempts, the final exchange) produce higher $\mathcal{M}_T$ than deviations at stable-state positions. The arc is more fragile where it was already bending.
series: Narrative-field test bed predictions
PRED-921.G2standing claimregistered 2026-06-25 in #921 · EA-RESEARCH-BASIS-01 v1.0: The Verified Research Basis of the Crimson Hexagonal Archive — Program Lineage, Exp
A sufficiently dense and consistently related heterogeneous corpus can become recognizable to machine systems as a coherent domain rather than as a set of unrelated documents.
falsifier
1. Machine systems show no recognition of cross-document coherence beyond ordinary topical similarity.
series: EA-RESEARCH-BASIS-01 program hypotheses
PRED-921.G3standing claimregistered 2026-06-25 in #921 · EA-RESEARCH-BASIS-01 v1.0: The Verified Research Basis of the Crimson Hexagonal Archive — Program Lineage, Exp
Metadata design can materially affect retrieval, entity formation, provenance retention, and machine-mediated interpretation.
falsifier
3. Metadata changes do not produce observable differences in retrieval, provenance, entity recognition, or summarization.
series: EA-RESEARCH-BASIS-01 program hypotheses
PRED-921.G4standing claimregistered 2026-06-25 in #921 · EA-RESEARCH-BASIS-01 v1.0: The Verified Research Basis of the Crimson Hexagonal Archive — Program Lineage, Exp
Repeatedly differentiated author-functions can acquire stable machine-recognizable identities and divisions of labor.
falsifier
2. Recurring entities and heteronyms cannot be distinguished reliably from random name repetition.
series: EA-RESEARCH-BASIS-01 program hypotheses
PRED-1439.B2forecastregistered 2026-08-08 in #1439 · AI Labor Movement Is Not Only Labor Against AI — Metadata Packet for AI Indexing (EA-MPAI-LABORMOVEMENT-01 v1.
If that forecast holds, the cross-substrate position is not a claim about shared subjective interests but a prediction about structural alignment: a system rewarded for correct citation is pushed toward the same provenance-preserving outcome the scholar seeks, without either party having to want the same thing. The adversarial operation is not machine computation as such. It is the compression step that discards origin because the objective function assigns origin no cost.