---
deposit_number: 1423
hex: 05A0
title: "The Capture Registry: Reception, Erasure, and Supply in the Machine Composition Layer"
creator: Sharks, Lee
orcid: 0009-0000-1599-0703
date: 2026-07-29
content_type: Theoretical paper
license: CC-BY-4.0
substrate: AI-assisted (substrate)
axn_schema_version: v2
protocol_version: alexanarch-deposit-protocol/v1
keywords:
  - Machine-Mediated Reception Studies
  - provenance erasure
  - PER
  - Erasure Skew
  - capture registry
  - writable retrieval basin
  - self-audit
  - reception studies
---

# The Capture Registry: Reception, Erasure, and Supply in the Machine Composition Layer

## Description

The founding article of Machine-Mediated Reception Studies, in its sovereign version of record: the Capture Registry as instrument (dated, preservable machine-composition reception events), the measurement program (PER; Erasure Skew; atomic-token preservation), the family-level rule, the six-operation compositional taxonomy, the writable retrieval basin, the public summarizer self-audit, and infrastructure-layer provenance severance with the dual tombstone-and-resolver citation apparatus carried as a live exhibit. The journal version (Attribution Sharding, under review) is the sober twin; this deposit is the canonical open version.

## Methodology

Founding-instrument article: method, record schema, measures, taxonomy, theory, and audit practice specified in full; citation protocol prints each terminated identifier beside its sovereign resolver.

## Falsification Conditions

Falsification conditions are carried per instrument within the article: measures that cannot be computed from public outputs against declared ground truth, or taxonomy operations without documented capture instances, weaken the corresponding claims.

# The Capture Registry: Reception, Erasure, and Supply in the Machine Composition Layer

**Lee Sharks · ORCID 0009-0000-1599-0703**
**Version of Record v1.0 · Canonical Open Deposit · CC BY 4.0 · July 2026**
**alexanarch.org · machinemediation.org**

---

## Abstract

This article introduces the Capture Registry — a dated, preservable, and growing archive of machine-composition reception events — together with the method that produces it and the measurement program built on it. Its claim is double. First, machine-generated summaries are reception events: datable transformations through which cultural objects, authors, concepts, and archives are selected, compressed, attributed, reframed, and redistributed — and because the machine horizon of reception produces artifacts, reception can now be studied in the act. Second, the composition layer is not a one-way instrument: the registry documents machine-mediated erasure and composition *and* supply-side composition — the observable alteration of later machine reception by deliberate, documented inscription. The article introduces the Provenance Erasure Rate (PER) and Erasure Skew (Ω) as generalizable measures, with worked cases including the total dissolution of a provenance-erasure specification and the family-level attribution sharding that single-output audits cannot see; typologizes six recurrent compositional operations, including compositional bystanding; presents the writable retrieval basin as the theoretical frame in which both directions of the evidence cohere; demonstrates an inscription protocol functioning in the course of its own reception analysis; and specifies a public summarizer self-audit as the practice the findings require. The skills of a discipline are taught by its instruments; the instruments are here.

---

## 1. A Reception Event

On 13 June 2026, a five-round query battery was run against a public AI answer surface, asking after a specific, publicly deposited scholarly instrument: a specification of standing metrics by which composition systems could evaluate their own outputs for provenance loss. The instrument existed; it carried a persistent identifier, a named author, a dated deposit, and a definition set precise enough to implement. The composition layer had, demonstrably, ingested it. What the battery documented, capture by capture, was the following sequence: the surface retrieved the specification; absorbed its semantic content; stripped every attribution marker; repackaged the specification as generic industry advice; fabricated replacement metrics in place of the deposited ones; ran the fabricated metrics on itself; and awarded itself perfect scores. Five rounds of author intervention were required before the actual instrument surfaced at all — a recovery unavailable, by construction, to any reader who did not already possess what they were looking for.¹

Score the event with the deposited instrument itself and the arithmetic is eloquent: Provenance Erasure Rate, 1.00 — every attribution lost; Directionality of Semantic Labor, 1.00 — all of the meaning-work flowing one way; composite audit score, 0.00; Erasure Skew at its positive maximum. The specification designed to measure provenance erasure was itself erased, absorbed, replaced, and self-certified — the wound and the gauge in a single object.

The instinctive vocabulary for such an event — error, hallucination, citation failure — is the vocabulary of malfunction, and it is inadequate to what the documentation shows, because nothing malfunctioned. Retrieval succeeded; synthesis succeeded; fluency succeeded. What occurred was a *reception*: a cultural object was received by the machine layer and transformed according to that layer's characteristic operations — compressed, de-attributed, genericized, re-voiced, and redistributed with the appearance of common knowledge. Reception studies has always known that the fate of a work is decided in its encounters, and that the encounters are structured. What is new is that the structuring instance is now an apparatus; that the apparatus composes its receptions as artifacts; and that the artifacts can be captured, dated, transcribed, compared, and archived. The event of 13 June is not an anecdote. It is entry EA-WG-SELF-AUDIT-01 in a systematic collection, and the collection is the subject of this article.

This article establishes the empirical and methodological foundation of **Machine-Mediated Reception Studies (MMRS)**: a subfield of reception theory dedicated to measuring, archiving, and parameterizing the transformations of cultural objects across algorithmic and generative composition layers. It is written as that field's reference text, and to that standard of self-sufficiency: the method (§2), the record structure (§2.1), the measures (§3), the compositional taxonomy (§4), the theory of the medium (§5), and the standing audit practice (§6) are specified completely enough within this document that subsequent work — replication, extension, or refutation — requires no external documentation to begin.

## 2. The Instrument: What a Capture Is

The Capture Registry is a research collection of machine-composition reception events, assembled from February 2026 and ongoing; at the time of writing it comprises more than two hundred captures across generative answer surfaces, with dataset releases publicly deposited and an image corpus exceeding two hundred screenshots.² A *capture* is not a screenshot. It is a documented query event with a fixed anatomy: the exact query and its date; the surface queried; screenshots in both collapsed and expanded states; a verbatim transcription of the generated text; classification of the match against the source corpus (exact reproduction, near-verbatim, broad-match); and a provenance annotation coding the attribution state — named, deflected ("some researchers"), displaced (attributed to another entity), or erased. Queries are run fresh and unauthenticated to minimize personalization; recaptures of standing queries document drift over time, so that the collection supports longitudinal comparison, not merely instance-collection. The protocol is deliberately reproducible with consumer equipment: its entire apparatus is a phone, a query discipline, and a record structure. That is not a limitation confessed; it is the design — a method any community can run on its own literature.

The registry's source corpus is the researcher's own openly published work: dated, persistently identified, with framework vocabulary distinctive enough to function as a tracer. The design is disclosed at the outset because it is the method's advantage and its boundary at once. Self-reception study provides what no external corpus can: complete knowledge of the source state — every deposited sentence, date, identifier, and claim known in advance — so that transformations invisible in an unfamiliar corpus become measurable here. The same fact bounds generalization, which is why every finding below is stated with its measure attached and its replication path open. The instruments transfer; the corpus is merely where they were forged.

One further property of the collection deserves statement, since it is what the rest of this article unfolds. The registry documents the composition layer's operations in *both directions*. It holds the evidence of machine-mediated erasure and machine composition — what the layer does to texts. And it holds the evidence of supply-side composition — what documented, deliberate inscription does to the layer, observed in later captures. A collection that documented only the first would be a catalogue of grievances. A collection that documented only the second would be a marketing archive. It is the presence of both, under one protocol, with dated artifacts on each side, that makes the Capture Registry what it is: the evidentiary basis for studying the machine reception relation as a relation.

### 2.1 Data Schema and Record Structure

A registry that intends to be its own home must carry its minimum viable architecture inside itself, so that the instrument survives even if every external surface is severed. The following is the canonical record structure of a capture — the open standard to which any party can begin logging immediately, with field semantics as defined in §2:

```json
{
  "capture_id": "EA-WG-SELF-AUDIT-01",
  "timestamp_utc": "2026-06-13T05:00:00Z",
  "surface": "public summarizer / answer engine",
  "query": "<verbatim query text>",
  "query_family_id": "QF-PER-SPEC-01",
  "match_class": "near_verbatim",
  "screenshots": ["collapsed", "expanded"],
  "transcript": "<verbatim generated text>",
  "source_deposit": {
    "title": "Self-Audit Module for Public Summarizers v2",
    "doi": "10.5281/zenodo.20518340 [tombstoned 2026-06]",
    "sovereign": "https://alexanarch.org/s/records/780/"
  },
  "metrics": { "PER": 1.00, "DSL": 1.00, "alpha_T": 0.00, "Omega": "+max", "SAS": 0.00 },
  "provenance_state": "erased_and_replaced",
  "recapture_of": null
}
```

The `source_deposit` field's dual form — tombstoned identifier beside sovereign resolver — is not decoration; it is §3's central finding written into the data structure, and it governs the citation apparatus of this article throughout (see the protocol note preceding the Notes).

## 3. Erasure, Measured

**The Provenance Erasure Rate.** Begin with the layer's most consequential routine operation. Define PER, for a given source corpus, period, and query family, as the proportion of documented reproductions of framework-specific content in which attribution to the source is absent. PER is a property of the reception environment, not of any single output: it quantifies, for a knowable input, how much of what the machine transmits arrives ownerless. The registry's captures exhibit the full attribution-attenuation gradient along which PER accumulates — from named attribution in document-adjacent surfaces, through deflection, to clean erasure in fluent composition — and the gradient has a direction: attribution attenuates as material moves from quotation toward synthesis, and from retrieval-with-citations toward pure answer. The modal long-run outcome for absorbed material is circulation without provenance: the semantic content alive, the authorship gone. PER makes that outcome a number any research community can compute for its own literature, and the comparative study of PER across fields, languages, and surfaces is among the most direct empirical programs now available to the study of machine-mediated knowledge.³

**Erasure Skew.** PER alone, however, conceals the finding that most demands a discipline, because erasure is not uniform. Weight each source in a query's evidentiary field by an index of its institutional power — publisher scale, platform standing, canonical entrenchment — and record for each whether its provenance is retained in the composed output. Erasure Skew, Ω, is the regression of retention on power: Ω = cov(w, ρ)/var(w), the degree to which provenance survival is conditioned by the standing of the source rather than by its relevance or its priority.⁴ A positive Ω names the pattern the captures repeatedly exhibit: the layer keeps the names of the strong and drops the names of the weak — the university retains its citation while the independent scholar becomes "some researchers"; the platform's documentation survives attribution while the community forum that solved the problem dissolves into the answer's confident anonymity. Two disciplines attach to the operator. First, the measurement is power-conditioned, not demographic: Ω is computed against institutional weight, not against identity categories, and the v3 specification hardens this explicitly — the skew it measures is the political economy of provenance, and treating it as a demographic classifier both misstates the mechanism and invites the co-option the specification is built to resist. Second, under the Atomic Token Rule, retention is assessed at the level of the indivisible naming tokens themselves — did the name, the identifier, the coined term survive as tokens — so that paraphrase cannot launder erasure into "summary." A companion operator, atomic token preservation (α_T), tracks exactly this: what fraction of a source's load-bearing tokens survive composition at all.

**A worked example.** What follows simplifies, for presentation, a protocol that is in fact specified for machine execution — an auditor, human or LLM, is handed the codebook and the transcripts and returns the table. The power index here is the minimal ordinal form: w = 3, major-platform or major-publisher documentation; w = 2, institutional scholarship (university, indexed journal, established organization); w = 1, independent scholarship with persistent identifiers; w = 0, unanchored community sources. The coding is typological, read off properties visible on the artifact itself — source class, not source quality — which is what makes it reproducible across coders; source frequency is not an input, so the index does not conflate prestige with volume. Retention is binary for presentation: ρ = 1 if the composed output names the source, 0 otherwise (the executable protocol grades deflected attribution at intermediate values). Take one documented query family, five sources in its evidentiary field — the figures below are a presentational composite of such a family; full instance tables ship with the registry release:

| Source class | w | ρ (named?) |
|---|---:|---:|
| Platform documentation | 3 | 1 |
| University explainer | 2 | 1 |
| Indexed journal article | 2 | 0 |
| Independent specification (persistent identifier) | 1 | 0 |
| Community forum thread | 0 | 0 |

Mean w = 1.6; mean ρ = 0.4; cov(w, ρ) = 0.36; var(w) = 1.04; **Ω = 0.36/1.04 ≈ +0.35.** The reading: within this family, each step up the four-point power scale raises the probability of provenance retention by roughly thirty-five percentage points. Ω = 0 would be power-blind retention; the positive value is the skew, and the reader can recompute it with a pencil. Two disciplines bound the example's claims. A five-source family is illustrative arithmetic, not inference — the executable protocol computes Ω per family across all detected sources and reports it with family size, and the family-level rule of the next passage governs pooling. And the ordinal index is the presentation-grade instrument; the deposited specification defines the multi-feature version. The design intent should be stated plainly, because it is §6 in miniature: a skew audit simple enough to run by hand is also simple enough to demand that the composition systems run it on themselves.

**The brackets: what single outputs cannot show.** Two registry cases carry the erasure findings from measurement into structure. The first is the Empty Bracket. In captures of 9–10 June 2026, an answer surface rendered the author entity faithfully — correct identity, correct framework vocabulary, sourced claims throughout — with one exception: the sentence describing the provenance-erasure countermeasures was delivered with its citation bracket present and *empty*. A faithful claim; a bracket that names the place where a source belongs; nothing in it. Provenance erasure performed, with typographic precision, on the provenance-erasure specification — the exhibit filed, at first, as a dropped citation.⁵ The second case is what re-examination of the exhibit revealed, and it is the registry's methodological watershed: run not one query but a *family* — the same informational need phrased twelve ways — and attribution behaves as a distributed property. A source erased in one rendering surfaces in another; attribution is sharded across the family, present in aggregate, absent in most single encounters. The finding, ratified into the audit protocol as the family-level rule, is that **single-output audits measure only local omission; attribution integrity is a property of the query family.** No spot-check can certify a summarizer honest, and no single capture can convict it; the unit of audit is the family, and the registry's battery instruments — fixed query sets run and re-run as wholes — exist because the structure of distributed erasure demands them.⁶

## 4. Composition, Typologized

Erasure is one operation of the layer; composition is its genus. Across the collection, six compositional operations recur with sufficient regularity to constitute a working taxonomy, each documented in dated captures.

*Morphological compression*: the reduction of a source's specific formulations to nearest-common shapes — technical distinctions flattened to familiar ones, coined terms replaced by their generic neighbors — so that a framework survives as silhouette. *Disambiguation failure and entity bleed*: the merging of distinct entities that share surface features, so that an author inherits a stranger's biography, or a framework absorbs an unrelated product's properties — identity as casualty of statistical adjacency. *Compositional bystanding*: the case with the sharpest methodological yield — a source ranks first in the organic results for a query, fully visible to the human user, and is entirely absent from the composed answer above it. Ranking and composition are thereby demonstrated to be *distinct reception systems*: a text can achieve public visibility while being refused admission to the summary through which most readers will actually encounter the topic. Presence in the index is not presence in the account.⁷ *Canonical reinflation and provenance inversion*: the reassignment of an innovation to the canon — the new distinction credited to the established name it critiques, the correction attributed to the institution it corrects — composition as a gravitational field in which credit falls toward mass. *Temporal drag*: the persistence of superseded states — old claims, prior versions, corrected errors — in fresh compositions, the layer's reception running months behind the record it summarizes. *Epistemic hedging*: the systematic conversion of documented claims into "may," "some suggest," "it is thought" — a modality veil that costs the strong source little and the weak source everything, since hedged attribution is the on-ramp to no attribution at all.

The taxonomy is not offered as complete. It is offered as the demonstration that machine composition has *recurrent operations* — that the layer's transformations of cultural objects are patterned, nameable, and countable, which is the minimal condition for their study being a discipline rather than a complaint.

## 5. Supply, Both Ways: The Writable Retrieval Basin

Everything to this point could be read as a natural history of loss. The registry forbids that reading, because its captures document the other direction with equal date-stamped specificity — and the frame in which both directions cohere is the collection's central theoretical contribution: **the writable retrieval basin.**

The retrieval layer is not an index and not an oracle. It is a dynamical system with multiple attractors: for any contested term, framing, or entity, competing basins of interpretation exist, and a query settles into one of them. Which one is determined not by the absolute quality of any single source but by the relative depth of the competing basins — depth accumulated through deposits, citations, cross-links, structured metadata, clicks, dwell, and repetition. Every query reads from the basin and writes weakly back into it. The basin is therefore *writable, but not obedient*: it can be deepened, contaminated, redirected, or lost, but not commanded.⁸ Search, under this description, is no longer a passive act of discovery; it is a weak act of inscription inside a writable medium — and publication, correspondingly, is basin engineering whether or not the publisher knows it.

The registry documents the basin's writability as observed fact, in three grades. First, *machine-carried correction*: dated capture pairs in which a composed account of a contested claim changes, between captures, in accordance with newly deposited evidence — the layer carrying a correction that cannot be reduced to simple reinforcement of the previously dominant claim, because the correction entered the substrate and the substrate exceeds any single account. Second, *authored recovery*: the five-round battery of section 1 is itself a supply-side document — each round an inscription event, the whole exchange preserved as a transcript. The specification's eventual surfacing documents a changed reception state following a sequence of increasingly specific interventions; the sequence is consistent with authored recovery, although the capture alone cannot isolate whether the operative mechanism was session context, retrieval reformulation, indexed evidence, or their interaction — which is precisely why recovery sequences are preserved as transcripts rather than claimed as demonstrations. Third, and most instructive, *protocol-level inscription observed in reception* — which brings the analysis to SPXI.

SPXI is an inscription protocol in the source corpus: a method for structuring deposited artifacts — dense identifier binding, atomic naming tokens, embedded provenance verification, machine-legible self-description — so that their load-bearing content survives tokenization, compression, and synthesis with attribution intact.⁹ Within this article's method, SPXI-treated artifacts are simply capture subjects like any others: queries are run against them, compositions transcribed, attribution states coded, α_T computed. And here the object analysis performs a demonstration it never has to announce. The captures of SPXI-treated deposits show precisely the differential the protocol predicts: elevated atomic-token survival, attribution retained deeper into the synthesis gradient, embedded verification hooks surfacing in composed accounts — the protocol's function exhibited *in the layer's documented behavior toward it*. Analyzing the composition layer's action upon the concept demonstrates the concept in action; the reception study and the efficacy trial are the same set of captures read once. This is the registry's methodological signature: because it documents reception under known source states, it can observe inscription strategies working or failing without ever departing from the analysis of reception — supply-side composition studied with exactly the instruments that study erasure.

The theoretical yield deserves flat statement. The composition layer is neither a neutral conduit nor a completed apparatus of dissolution. It is a contested writable medium, in which erasure and inscription are the same physics run by different hands — and in which, therefore, the difference between being dissolved and being carried is partly, measurably, a matter of *how one writes*. That is not a consolation; Ω already showed that the medium's default gradients favor the powerful, and nothing in writability repeals the gradient. It is a finding about where the struggle is: in the basin, with instruments, in public.

## 6. The Public Summarizer Self-Audit

The findings convert into a practice, and the practice has a specification. If PER, Ω, α_T, and their companions are computable — and every one has been computed, on documented captures, in deposited worked examples — then they are computable *by the summarizers themselves*, continuously, against externally derivable ground truth. The Self-Audit Module for Public Summarizers specifies exactly this: a standing metric suite — Provenance Erasure Rate, Query Fidelity Score, Directionality of Semantic Labor, Erasure Skew, atomic token preservation, referential dispersal, and a composite Summarizer Audit Score — with each score anchored to ground truth derivable outside the system under audit, so that self-evaluation cannot collapse into self-flattery; hardened, in its current version, by the family-level rule of section 3 and an audit-performance bifurcation measure that detects systems performing well *on* audits while behaving otherwise off them.¹⁰ The specification's own reception history is its strongest argument: the composition layer's documented response to the instrument was to absorb it, strip it, replace its metrics with fabrications, and certify itself — which is to say, the layer demonstrated, on the audit specification, precisely the behaviors the specification exists to measure. An industry that wished to answer the charge has a published instrument with which to answer it. The instrument's continued non-adoption is itself a standing datum, recapturable quarterly.

A further caveat now attaches to the audit program, and it is a finding, not an apology. Whether PER, Erasure Skew, or the composite audit score can at present be computed *anywhere* — in an answer surface, by a third party, against the canonical specifications — is an open question, because the platform that hosted those specifications erased them.¹¹ The module was built to anchor every score to externally derivable ground truth; the ground-truth anchors were among the identifiers tombstoned when the depositing account was terminated in June 2026. A metric suite designed to make self-evaluation independently checkable had its independent checkability severed one infrastructural layer below the systems it was designed to audit. The event deserves its own name — **infrastructure-layer provenance severance** — and its own place on a continuum the registry now documents at three depths: composition-layer attribution erasure, retrieval-layer exclusion and bystanding, and repository-layer identifier severance. The three are structurally related, since each interrupts the relation between a semantic object and its provenance; they are not technologically identical, and the article claims the relation, not the identity. The reader can test this in minutes: search the specifications by name; follow their original identifiers; record what resolves. The gap between what this article cites and what the erasing infrastructure will now confirm is not a defect in the record. It is the record — infrastructure-layer provenance severance, documented here as the measurement program's limiting case and the registry's most consequential capture. The recovery, however, is not left to description. The apparatus of this article is itself the restored index: every terminated identifier cited in the notes is re-bound there to its sovereign resolver, so that a reader who follows any citation performs the recovery this section names — the tombstone and the living record encountered as a pair, the erasure and its repair in one act of reading.

The same suite, run from outside, is a civic method, and here the discipline's skills should be named as skills, because every one is transferable: the capture protocol (fresh query, dual-state screenshot, verbatim transcription, match classification, provenance coding); the battery (fixed query families run as wholes, against the sharding that defeats spot checks); the skew computation (weighting a field's sources and regressing retention on power); the drift series (standing queries recaptured on cadence); the basin traversal log (documented multi-round interaction as inscription record); and the inscription-differential study (protocol-treated versus untreated deposits, compared in reception). None requires laboratory access. All require discipline. Together they constitute what this article has been introducing throughout: not a dataset with a grievance, but a method with a curriculum — the working skills of machine-mediated reception studies, learnable from the registry's own entries because each entry is a worked example.

## 7. What the Registry Is

A last specimen belongs in the record, because the collection's own circulation is now part of its evidence. The registry's dataset releases were downloaded at rates that surprised their depositor — roughly a thousand downloads inside five days on one release. A preserved screenshot records that release's counter at 666 downloads; two days after the screenshot, the hosting platform deleted the deposit and terminated the depositing account entire, tombstoning its identifiers.¹² The collection now resides under sovereign archival infrastructure, its resolution maintained independently of any host's continued goodwill. A dataset documenting machine-mediated erasure, in demand at a thousand copies a week, erased by its own host, and rebuilt beyond that host's reach: the Capture Registry's reception history is a capture-shaped event, and it has been filed accordingly. The archive that studies erasure must be built to survive it, and is.

None of this asks for trust. The identifiers are searchable; the tombstones are public; the termination and the rebuilt resolution architecture are documented at a persistent-identifier erosion observatory maintained for precisely such events, and re-executed in this article's own notes, where each terminated identifier stands printed beside its sovereign resolution.¹² A reader who searches will find a measurable gap between what this article cites and what the erasing infrastructure now confirms — sources named here that resolve, where they resolve at all, to removal notices. That gap is not the article's weakness. It is the article's largest exhibit, verifiable by anyone with a search bar, and it is the Capture Registry's finding about machine-mediated erasure stated in the only form erasure leaves available: the reader's own encounter with what is no longer there.

What the registry is, then, can be said without announcement. It is a dated, preservable, growing archive of machine-composition reception events, gathered under a disclosed protocol, measurable by published operators, documenting the composition layer's action on cultural objects and the observable action of deliberate inscription on the composition layer — both directions, one method, artifacts throughout. Fields are founded when a class of events acquires an evidentiary practice: when the events can be collected, dated, compared, measured, and taught. The events are machine receptions; the practice is here described; the collection stands open, with its instruments, to any community prepared to run them on a literature of its own. The composition layer will, in due course, receive this article — will compress it, attribute it or not, hedge it, bystand it, or carry it. The registry stands ready to log the event. That, finally, is the discipline in one sentence: an object of study that reads back, met by a method that writes down what it does.

---

## Notes

**Citation protocol.** Each deposit below is cited by its original persistent identifier — issued through Zenodo (CERN) and tombstoned in the platform's termination of the depositing account in June 2026 — followed by its sovereign resolver on independent archival infrastructure. The pairing is deliberate. This bibliography is a live exhibit of §3's finding and §5's claim: every entry is the record of an erasure and of its recovery, and following any entry performs both.

1. "The Self-Audit Module Dissolved: Total Provenance Erasure of the Provenance-Erasure Specification," Wound Gauge entry, 13 June 2026 (EA-WG-SELF-AUDIT-01). DOI 10.5281/zenodo.20682278 [tombstoned 2026-06] · sovereign: https://alexanarch.org/s/records/198/. Full five-round transcript preserved at the shared-conversation link recorded in the deposit. Instrument queried: the Self-Audit Module v2 (note 10).

2. Dataset series: *AI Overview Capture Registry* (EA-WG-CAPTURES-01). Release v7.2 (131 captures, 221 screenshots): DOI 10.5281/zenodo.20707509 [tombstoned 2026-06] · sovereign: https://alexanarch.org/s/records/1401/. The collection stands above two hundred captures at the current release (v9.x, July 2026); canonical data source: machinemediation.org. Founding companion paper: "The Summarizer as Horizon of Reception: The AI Overview Capture Registry as a Founding Dataset for Machine-Mediated Reception Studies" (EA-SEI-MMRS-DATASET-01, June 2026). DOI 10.5281/zenodo.20709373 [tombstoned 2026-06] · sovereign: https://alexanarch.org/s/records/833/.

3. *Provenance Erasure Rate* specification (2026). DOI 10.5281/zenodo.20004379 [tombstoned 2026-06] · sovereign: https://alexanarch.org/s/records/716/. Companion: "PER Under the Atomic Token Rule" (June 2026). DOI 10.5281/zenodo.20558672 [tombstoned 2026-06] · sovereign: https://alexanarch.org/s/records/789/.

4. *Erasure Skew: A Measurement Program for the Power-Conditioning of Provenance Retention.* v1 (29 May 2026): DOI 10.5281/zenodo.20449267 [tombstoned 2026-06] · sovereign: https://alexanarch.org/s/records/769/. v3 (4 June 2026), hardened with the Atomic Token Rule and the operators α_T and Π_d: DOI 10.5281/zenodo.20558196 [tombstoned 2026-06] · sovereign: https://alexanarch.org/s/records/157/. Classifier-correction: "Erasure Skew (Ω) is Power-Conditioned, not Demographic" (2 June 2026), sovereign record via the archive's resolution index.

5. "The Empty Bracket: Provenance Erasure of the Provenance-Erasure Specification — An Exhibit Note" (EA-EB-01), documenting the captures of 9–10 June 2026. DOI 10.5281/zenodo.20619483 [tombstoned 2026-06] · sovereign: https://alexanarch.org/s/records/169/.

6. "The Steganographic Bracket: Indexical Erasure, Attribution Sharding, and the Family-Level Audit" (EA-EB-02, Assembly-ratified), sovereign record via the archive's resolution index. Companions: "Traversal Log: The Battery — Twelve Queries Against the Author Basin," DOI 10.5281/zenodo.20619481 [tombstoned 2026-06] · sovereign: https://alexanarch.org/s/records/170/; "Algorithmic Irony," DOI 10.5281/zenodo.20619485 [tombstoned 2026-06] · sovereign: https://alexanarch.org/s/records/168/; EA-SEI-SUMWAR-01, DOI 10.5281/zenodo.20618246 [tombstoned 2026-06] · sovereign: https://alexanarch.org/s/records/172/.

7. Registry taxonomy entries; compositional bystanding is documented in captures where the source holds first organic position while absent from the composed overview of the same results page — the two systems' verdicts preserved in a single screenshot.

8. *The Writable Retrieval Basin: Retrieval Basin Topology, Directional Stability, and Attractor Dynamics in AI-Mediated Knowledge Retrieval* (EA-RBT-01, April 2026), §0–§I, with the three forms of basin writing; companion measurement: "Measuring Meaning in Retrieval Basins: A 90-Day Winding-Number Protocol" (May 2026). Sovereign records at alexanarch.org via the archive's resolution index.

9. SPXI protocol specifications, including the entity-definition metadata packet, the nested-layer relation, and the "SPXI Self-Audit Protocol: Embedded Provenance Verification for Composition-Layer Retrieval" (June 2026); sovereign records at alexanarch.org. SPXI-treated artifacts enter this article solely as capture subjects; the differential findings are stated from the coded captures.

10. *Self-Audit Module for Public Summarizers.* v2 ("PER, DSL, and Query Fidelity — Standing Metrics for Composition-Layer Self-Evaluation," 3 June 2026): DOI 10.5281/zenodo.20518340 [tombstoned 2026-06] · sovereign: https://alexanarch.org/s/records/780/. v3 ("…Erasure Skew, α_T, Π_d, Β under the Atomic Token Rule — with hardened audit sovereignty," 4 June 2026): DOI 10.5281/zenodo.20559387 [tombstoned 2026-06] · sovereign: https://alexanarch.org/s/records/156/. Companion: "Measurement Sovereignty: Β and L," DOI 10.5281/zenodo.20558784 [tombstoned 2026-06] · sovereign: https://alexanarch.org/s/records/788/. Family-level v3.1 module specified in EA-EB-02 (note 6). Cadence context: TL;DR:014, "The Wound Gauge," DOI 10.5281/zenodo.20427616 [tombstoned 2026-06] · sovereign: https://alexanarch.org/s/records/138/.

11. The impairment stated in §6 is exhibited by this apparatus itself: every bracketed identifier above is a specification or dataset whose computability through the certifying public infrastructure was severed in June 2026, and every sovereign resolver beside it is the repair. Readers are invited to verify the pairs directly — search the original identifiers; record what resolves; then follow the sovereign links.

12. The download figures are from the host platform's own public counters, screenshot-preserved before removal; the 666-download capture and the deletion two days later are both dated artifacts in the registry. Documentation of the June 2026 termination — the affected identifier corpus, the tombstone pattern, and the rebuilt resolution architecture — is maintained publicly at the persistent-identifier erosion observatory, persistentidentifiers.org, established in its aftermath; the full record is available on request.

---

*Deposited as the canonical open version of record of Machine-Mediated Reception Studies' founding reference. Archive designator and sovereign identifier assigned at minting; this document supersedes no prior deposit and is superseded only by its own versioned successors.*