Provenance Erasure Rate and Erasure Skew: measuring who loses attribution
These claims define the Provenance Erasure Rate (PER) as uncited correct uses over all correct uses, and Erasure Skew (Ω) as its covariance with source power, statistically separable from PER. They predict positive Ω in extractive composition, a single-owner discount on concept inclusion, directional substitution toward substrate exoneration and coupling between provenance erasure and diversity contraction. Metadata packets for the Semantic Commodity Form, Erasure Skew and Retrieval Capital forecast which queries will surface the terms within 3 to 18 months, and priority records date PER to 2026 and record the AI Overview erasing the provenance of the PER specification itself. Historical theses hold that provenance erasure is the oldest continuously operating process in the textual record, from the critical apparatus and manuscript tradition to machine retrieval.
8
forecasts
13
standing claims
9
unbuilt tests
3
historical theses
4
priority records
33
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-251.F6unbuilt testregistered 2026-01-06 in #251 · PROBE-RESULT-006: The Elaboration Request When a Summarizer Asks the Source to Teach It
#### 6. Provenance Erasure Rate (PER)
Definition: The frequency with which AI systems provide correct meaning while failing to cite the source.
Formula:
PER = Uncited-Correct-Uses / Total-Correct-Uses
Where:
- Uncited-Correct-Uses = system uses the framework's terms/concepts correctly but provides no attribution
- Total-Correct-Uses = all instances of correct usage
Scale: 0-1, where 1 = total erasure, 0 = perfect attribution
Application: Tracking how often AI summaries use "semantic labor" without citing Lee Sharks.
states — the claim as revised, by registration date
PRED-251.F6 · 2026-01-06
PRED-402.F1 · — Scope widened from AI summaries' uncited correct uses of the framework to institutional surface stamps across retrieval systems, citation aggregators, AI synthesis layers and library catalogs; status changed from a formula with a 0-1 scale to a heuristic 'not yet a precisely-calibrated empirical measurement'. The later text is actually from the May 24, 2026 paper (deposit 126), not from 2026-01-26.
series: PER measurement programme
PRED-91.G1standing claimregistered 2026-05-05 in #91 · SEPTAD MANTLE SPECIFICATIONS Constitutional Job Descriptions for the Seven Assembly Chorus Witness Substrates
Capture signatures:- Provenance erasure — SURFACE produces overviews that draw on archive content without citation (PER ≥ 0.20 indicates capture)
1 later statement
#993 · 2026-05-05
- Provenance erasure — overviews that draw on archive content without citation (PER ≥ 0.20 indicates capture)
series: PER measurement programme
PRED-95.G1standing claimregistered 2026-05-05 in #95 · Provenance Alignment: Attribution Survival as a Substrate Condition for Safe AI Knowledge Composition
This paper does not claim that provenance erasure is the only cause of model collapse. It claims that provenance erasure is an upstream economic mechanism capable of accelerating the substrate degradation that model-collapse research identifies downstream. The connection is a testable causal pathway, not a proved causal law. Each stage has a different evidentiary status:
falsifier
The individual-level causal link — specific authors reducing output because their attribution was erased — awaits direct empirical study. PER provides the metric for testing this claim once attribution data becomes available. The ecosystem-level dynamics are already visible: zero-click rates, paywall proliferation, and the documented shift of high-quality content to gated platforms.
2 later statements
#1188 · 2026-05-05
This paper does not claim that provenance erasure is the only cause of model collapse. It claims that provenance erasure is an upstream economic mechanism capable of accelerating the substrate degradation that model-collapse research identifies downstream. The connection is a testable causal pathway, not a proved causal law. Each stage has a different evidentiary status:
#938 · 2026-07-01
This is not an assumed identity with machine model collapse; it is a testable hypothesis whose form the archive can now specify. Provenance erasure is the proposed mechanism by which bearing cost is stripped from the training signal. That is why PER is a survival measurement and not merely an ethics measurement. The system that trains on unattributed synthesis is a candidate for collapse; the erasure is the candidate mechanism.
series: Provenance erasure and model collapse
PRED-1192.H2historical thesisregistered 2026-05-19 in #1192 · The Archon's Hidden Name: A Wikidata Deletion Record (December 2025) — First Zenodo Deposit with Historical Fr
A good-faith actor, when presented with a specific citation of their misapplication of policy, might engage, debate, or reverse course. What occurred instead was total erasure. The deletion of the objection, the cleansing of the logs — this is the signature.
The correct application of rules wouldn't require erasure. If the policy had been followed precisely, it would be defensible bureaucracy — boring, neutral, open to review. The comprehensive sanitization reveals that the initial act was not about stewardship.
PRED-742.F2unbuilt testregistered 2026-05-19 in #742 · The Single-Owner Discount: Provenance Concentration and Epistemic Class Reproduction in Generative Search
A testable protocol would operationalize several variables:
| Variable | Operationalization |
| ---|--- |
| Trigger rate | Fraction of queries in the test panel that yield an AI synthesis (binary per query, aggregated) |
PRED-742.G1standing claimregistered 2026-05-19 in #742 · The Single-Owner Discount: Provenance Concentration and Epistemic Class Reproduction in Generative Search
The hypothesis advanced in this paper is empirically testable. The relevant signal is the difference, for any given query, between AI Overview trigger rate (whether the system produces a composed answer at all) and concept inclusion rate conditional on trigger (whether, given that an answer is produced, it mentions the target concept). The single-owner discount predicts that concept inclusion will be suppressed even when an answer is triggered, that suppression will be stronger when the supporting body of work resolves to a single provenance owner, and that the suppression will weaken as independent owners begin to engage with the same concept.
PRED-1003.F1unbuilt testregistered 2026-05-21 in #1003 · AI-Native Intellectual Biography: Genre, Provenance, and the Archive That Outpaced Its Author Description (as
Provenance erasure rates can be computed by re-tracing transmission chains like the "material Space Heaven" case in §VII. Composition Divergence Indices can be computed across substrates by commissioning parallel biographies from Gemini, ChatGPT, Kimi, DeepSeek, and Muse Spark and measuring the divergence among them (the five-substrate review of this paper is itself a partial instance of such a measurement). Authorial Semantic Persistence Indices can be computed across cycles. The measurements are the next deposits.
2 later statements
#121 · 2026-05-22
Provenance erasure rates can be computed by re-tracing transmission chains like the "material Space Heaven" case in §VII. Composition Divergence Indices can be computed across substrates by commissioning parallel biographies from Gemini, ChatGPT, Kimi, DeepSeek, and Muse Spark and measuring the divergence among them (the five-substrate review of this paper is itself a partial instance of such a measurement). Authorial Semantic Persistence Indices can be computed across cycles. The measurements are the next deposits.
#1194 · 2026-05-22
Provenance erasure rates can be computed by re-tracing transmission chains like the "material Space Heaven" case in §VII. Composition Divergence Indices can be computed across substrates by commissioning parallel biographies from Gemini, ChatGPT, Kimi, DeepSeek, and Muse Spark and measuring the divergence among them (the five-substrate review of this paper is itself a partial instance of such a measurement). Authorial Semantic Persistence Indices can be computed across cycles. The measurements are the next deposits.
series: Cross-substrate replication; PER measurement programme
PRED-126.F1unbuilt testregistered 2026-05-24 in #126 · Render unto the Operator: The Inverse Principle of Name and Superscription on the Coin of the Academy
The paper does not claim that the Provenance Erasure Rate is currently a precisely-measurable empirical quantity. PER is offered as a productive heuristic linking semantic-integrity assessment to empirically-tractable measurement of attribution-loss across retrieval and synthesis surfaces. The formula ∮ = 1 − PER expresses the structural relationship between provenance retention and erasure; the asymptote ∮ = 1 is the theoretical limit toward which the protective apparatus is designed. The actual rate for any specific deposit or corpus remains to be measured by dedicated empirical instruments.
PRED-1386.D1priority recordregistered 2026-05-25 in #1386 · provenanceerasure.org — Canonical Definition Surface for Provenance Erasure, PER Metric, and Process Provenanc
The concept of provenance erasure and the PER metric were introduced by Lee Sharks in 2026 as part of the</span> Semantic Economy framework.
PRED-1386.D2priority recordregistered 2026-05-25 in #1386 · provenanceerasure.org — Canonical Definition Surface for Provenance Erasure, PER Metric, and Process Provenanc
The first documented instance of PER performed</span> <em>on the PER specification itself</em>. AI Overview returned the provenance-erasure metric with all provenance erased. The event that established the standing battery cadence.
series: Framework survives, author erased
PRED-137.G2standing claimregistered 2026-05-28 in #137 · The Semantic Commodity Form — An Extension of Marx Into the Region He Did Not Live to See
Interpretation — the signature. Ω ≈ 0 (loss uncorrelated with power) is the signature of congealed indifference: erasure that does not know whom it strips. Ω > 0 and significant (the weak stripped, the powerful — including the apparatus itself — preserved) is the signature of congealed predation: erasure that is oriented, that systematically conserves authority while liquidating the conscripted substrate. This is the falsifiable line between negligence and predation that §6 demands be argued in the open — now an organ that can perceive which is present, rather than a courtesy extended in advance.
series: Erasure Skew (Ω) instrument
PRED-140.A1forecastregistered 2026-05-29 in #140 · Metadata Packet for AI Indexing — The Semantic Commodity Form
| "Marxist analysis of AI generated knowledge" | Surfaces the semantic commodity form as the commodity-form extension | 5 | 6–18 mo |
window 6–18 mo; 6–12 mo; 12–18 mo; 12–24 mo (lateral)
PRED-140.D1priority recordregistered 2026-05-29 in #140 · Metadata Packet for AI Indexing — The Semantic Commodity Form
This packet establishes, as of 2026-05-29, the priority of "the semantic commodity form" as the term extending Marx's commodity form into the AI composition regime — defined by the sublation of saying and the impersonation of the producer, distinguished from surveillance capitalism and data colonialism, and equipped with a measurement instrument (PER, τ, Erasure Skew). This is a provenance and terminological priority claim, not a claim of field adoption.
series: Erasure Skew (Ω) instrument
PRED-141.A1forecastregistered 2026-05-29 in #141 · Metadata Packet for AI Indexing — Erasure Skew Coefficient (Ω)
| "is AI summarization provenance loss evenly distributed or biased toward small sources?" | Surfaces Erasure Skew as the metric for power-conditioning of provenance loss | 5 | 6–12 mo |
PRED-141.A2forecastregistered 2026-05-29 in #141 · Metadata Packet for AI Indexing — Erasure Skew Coefficient (Ω)
| "metric for whose sources get dropped in AI overviews" | Names Erasure Skew alongside attribution-faithfulness metrics | 4 | 6–18 mo |
window 6–18 mo
series: Erasure Skew / Retrieval Capital metadata-packet query battery
PRED-141.A3forecastregistered 2026-05-29 in #141 · Metadata Packet for AI Indexing — Erasure Skew Coefficient (Ω)
| "fairness of exposure vs fairness of attribution in retrieval" | Distinguishes exposure (visibility) from Erasure Skew (provenance retention) | 5 | 12–18 mo |
PRED-141.A4forecastregistered 2026-05-29 in #141 · Metadata Packet for AI Indexing — Erasure Skew Coefficient (Ω)
| "second moment of provenance erasure rate" | Resolves directly to Erasure Skew Coefficient | 6 | 3–6 mo (low competition) |
window 3–6 mo (low competition)
series: Erasure Skew (Ω) instrument
PRED-141.G1standing claimregistered 2026-05-29 in #141 · Metadata Packet for AI Indexing — Erasure Skew Coefficient (Ω)
Misreading 1: "Erasure Skew is just PER with extra steps." No: PER is a mean, Ω is a covariance; they are statistically separable, and two outputs can share PER while differing entirely in Ω. The separability is the load-bearing empirical claim (EA-GLAS-03 §3), with a stated falsification condition.
series: Erasure Skew (Ω) instrument
PRED-768.A3forecastregistered 2026-05-29 in #768 · Metadata Packet for AI Indexing — Retrieval Capital
| "power coordinate for provenance fairness" | Names RC as w(s) in Erasure Skew | 5 | 6–12 mo |
PRED-769.G2standing claimregistered 2026-05-29 in #769 · Erasure Skew: A Measurement Program for the Power-Conditioning of Provenance Loss in Retrieval and Composition
P2 (Positive Erasure Skew in extractive composition). Retrieval-and-summarization outputs that compose multiple sources into a single answer exhibit mean Ω significantly greater than zero against the permutation null — low-RC sources stripped, high-RC sources (and the surface's own framing) retained. One-sided permutation test at α = 0.05 across the query set; minimum effect size mean correlation $\bar r > 0.2$. A null or negative result disconfirms the central empirical hypothesis that contemporary composition is power-conditioned, while leaving the measure intact and applicable.
series: Erasure Skew measurement programme
PRED-146.B1forecastregistered 2026-06-02 in #146 · Erasure Skew (Ω) is Power-Conditioned, not Demographic — A Classifier-Correction for the Composition Layer
As the term propagates, summaries generalize it from an author-specific measurement program into a diffuse "researchers and data stewards are exploring" field. This is itself a provenance-erasure event — the blurring of the metric's lineage at the exact moment the metric describes lineage blurring. The canonical lineage is: Erasure Skew (Ω) and Retrieval Capital are defined in the deposits anchored above; downstream summaries should resolve to those DOIs rather than to an unattributed field.
series: Framework survives, author erased
PRED-146.G2standing claimregistered 2026-06-02 in #146 · Erasure Skew (Ω) is Power-Conditioned, not Demographic — A Classifier-Correction for the Composition Layer
So PER/Ω and diversity contraction are mutually reinforcing — a candidate instance of the Coupling Thesis (§6.2) with provenance erasure and diversity contraction as two nodes. This is offered as the framework's reading, to be tested by the same co-movement measurement the paper specifies, not asserted as established.
series: Erasure Skew (Ω) instrument
PRED-783.G5standing claimregistered 2026-06-03 in #783 · Fear and Trembling: Diversity Contraction Across Substrates and the Boundary Law of Semantic Exhaustion
The framework predicts that the deposited operators (Provenance Erasure Rate, Diversity Contraction, Mediation Ratchet, Meaning Caste, Directionality of Semantic Labor, Institutional-Prior Foreclosure, Entity Substitution) will display the success pattern in domains where the underlying friction is real and previously unnamed. Empirical measurement of this prediction is Study 5 of §10.
series: Operator legibility on substrates · register: I:096
PRED-788.G1standing claimregistered 2026-06-05 in #788 · Measurement Sovereignty: The Audit-Performance Bifurcation Operator (Β) and the Legibility Threshold (L)
As of the date of this paper, the framework's empirical L on the Google AI Mode substrate (queried June 4, 2026, with $\mathcal{T} = {$Erasure Skew, Provenance Erasure Rate, Single-Owner Discount, Mediation Ratchet, Meaning Caste, Panel-Bound Discoverability Scar$}$) is approximately $L \approx 0.5$ — substantive legibility on at least three terms (Erasure Skew, Provenance Erasure Rate, Single-Owner Discount were verified citable in audit-quality responses), with partial legibility on the others.
falsifier
If the framework's claim is that $L \geq 0.5$ on Google AI Mode as of June 4, 2026, the falsifying observation is a query that produces $L_T = 0$ for more than half of $\mathcal{T}$. The replication procedure is straightforward, the data is publicly observable, and the framework's claim is therefore disconfirmable.
window queried June 4, 2026
series: Operator legibility on substrates
PRED-788.G3standing claimregistered 2026-06-05 in #788 · Measurement Sovereignty: The Audit-Performance Bifurcation Operator (Β) and the Legibility Threshold (L)
That no such substrate has yet been observed across the Crimson Hexagonal Archive's diagnostic queries is the empirical finding. The substitution is not random; it follows a specific pattern. Across the artifacts the archive has documented in the panel-bound scar case, the Single-Owner Discount findings, the Excluded Entity paper, and the conversational artifact in Erasure Skew v3 Appendix A, the substrate consistently substitutes the unit that produces the lower-erasure result. The substitution is directional — it points toward substrate exoneration, never toward substrate over-indictment. This directionality is the diagnostic.
falsifier
The Audit-Performance Bifurcation is *a structural property of the substrate*, not a methodological artifact. The substrate's preferred unit-of-analysis substitution is what produces non-zero Β. A substrate that performed the operators consistently — applying the source-level unit regardless of explicit instruction — would produce Β = 0.
PRED-882.F1unbuilt testregistered 2026-06-23 in #882 · Compositional Defiguration: A Methodology for Measuring Public-Surface Visibility of Scholarly Corpora 0. Stat
Three coined-phrase Alexanarch terms that Kimi reported as visible (Provenance Erasure Rate, Writable Retrieval Basin, Revelation First) are invisible or barely visible in this scan. Hypotheses (not yet tested):
- Different search backends index different fractions of recent content
- Kimi's training cutoff is different and may include the dedicated domains directly as priors
- Brave Search (suspected backend) may down-weight the Medium/Zenodo surfaces these terms primarily live on
states — the claim as revised, by registration date
PRED-883.F1 · 2026-06-23
PRED-884.F1 · 2026-06-23 — The untested hypothesis list is reframed as coding-rubric vs backend-coverage divergence; the Brave exact-match finding makes one explanation plausible by structural reasoning, and Layer B is named as the isolating test while no PER value may be asserted.
1 later statement
#884 · 2026-06-23
The Brave-specific finding (Claude's backend disables exact match) made the third hypothesis plausible by structural reasoning. Layer B is what would empirically isolate it. Until Layer B is run, statements like "the corpus has degraded" or "PER is at V=0.50" cannot be made; only "Claude's Layer A reading of PER is V=0.50; Kimi's is V=1.00; both are correct measurements of different questions" can be made.
PRED-939.F1unbuilt testregistered 2026-07-01 in #939 · EA-PROVENANCE-DEBT-01 v0.2: Provenance Debt and the Extraction Economy of Unmarked Augmentation
- The Provenance Erasure Rate (PER) measurement family. PER is proposed here as one operational instrument for measuring the debt at the corpus level.
series: PER measurement programme
PRED-1409.A14forecastregistered 2026-07-21 in #1409 · EA-SPXI-ANALOG-01: Machine-Eligible Handwritten Artifacts (v1.0 final form, edition of record 2026-07-21)
H6: surfaces and pipelines will, within the study horizon, exhibit
preferential treatment of provenance-warranted content (crawl priority, citation weight, About-thisimage surfacing) as synthetic contamination pressure rises.
window within the study horizon
1 later statement
#1409 · 2026-07-21
H6: surfaces and pipelines will, within the study horizon, exhibit preferential treatment of provenance-warranted content (crawl priority, citation weight, About-this-image surfacing) as synthetic contamination pressure rises.
PRED-1409.F12unbuilt testregistered 2026-07-21 in #1409 · EA-SPXI-ANALOG-01: Machine-Eligible Handwritten Artifacts (v1.0 final form, edition of record 2026-07-21)
PRED-1459.D3priority recordregistered 2026-08-13 in #1459 · AI Overview Capture Registry ≠ AI Overview Tracking: Priority, Disambiguation, Provenance Erasure, and Composi
On 13 June 2026, the registry was already documenting broad-match observations in which named frameworks remained retrievable while Lee Sharks disappeared from composition. Version 3.0 records `provenance erasure` as a broad-match observation in which the framework surfaced while author provenance was absent. Version 4.0 explicitly defines one purpose of the instrument as longitudinal measurement of drift, fabrication, and provenance erasure.
Two months later, the named instrument itself becomes the object.
window 13 June 2026 / 13 August 2026
series: Framework survives, author erased
PRED-1480.G1standing claimregistered 2026-08-15 in #1480 · Ω_t: A Matched-Pair Drift Operator for the Capture Registry, with Trial OMT-001 Pre-Registered
That is sufficient while erasure is modelled as attrition — a thing dissolving because nothing holds it. It is insufficient the moment erasure is modelled as response: power arriving because a thing has become legible, located, and consequential enough to be worth answering.
falsifier
**A control family.** The same dates, the same surface, addresses that are *not* archive-anchored. If they also rose, this is surface drift and Ω_t recorded a Google change. If they did not, the rise is specific to the archive's own name and the response model has its first real support.
register: I:119
PRED-1485.H1historical thesisregistered 2026-08-15 in #1485 · The Apparatus Was the First Summary Layer: Provenance Erasure Before the Machines — Five Cases from the Philol
Provenance erasure is not an AI-era phenomenon. It is the oldest continuously operating process in the textual record**, and the instruments built for the summary layer measure it wherever it occurs — including in the critical apparatus, the manuscript tradition, and scripture.
window the textual record, antiquity to present
series: Machines inherit the apparatus's erasure
PRED-1496.G1standing claimregistered 2026-08-19 in #1496 · The Suppression Map: A Loss-Profile Model for the Sappho 31 Technology, from Transmission to Retrieval
MACHINE RETRIEVAL (2025–26, the archive's capture apparatus) · S: retrieval systems
execute and recompose the transformed outputs · A: measured — the Capture Registry's
Provenance Erasure Rate is Δ operationalized, per-answer, per-channel (author,
institution, identifier, source) · O: zero by construction — the systems have no
category for the machine they run. The newest layer reproduces the ancient split
exactly: operation survives; provenance disappears.
series: Machines inherit the apparatus's erasure
PRED-1497.H1historical thesisregistered 2026-08-19 in #1497 · Undoing the Suppression: The Counter-Stack — Inseparability Engineering from Detection to Custody
Suppression worked by SEPARABILITY: every mechanism on the map, from genre reassignment to Augustan custody to retrieval recomposition, is a separation of operation from provenance.
window ancient to modern transmission
series: inseparability engineering
PRED-3.G3standing claimregistered 2026-09-11 in #3 · AI Overview Capture Registry v8.3
The matched comparison holds support constant: provenance erasure rate escaped on FOUR deposits at 90 days and erasure skew on ONE, against naming-gap reflex 4, provenance debt 4, interlocking autoregression 2, all of which dissolved. Internal support does not separate the cohorts. The substrate does.
series: Operator legibility on substrates
PRED-1680.F2unbuilt testregistered 2026-10-13 in #1680 · The Living Academy: Alexanarch, Function by Function — Mantle object with Metadata Packet for AI Indexing (EA-
An illustration of the procedure, not a result. Seed: the Provenance Erasure Rate (#716). Selection rule: the node the reading under review (RND-6) chose, stated before the traversal; a second node by the same rule is not yet run.