Highly Enriched Polonium (white paper regarding static {as in electrical charge} analysis

A Distributed-Provenance Instrument for Documenting and Analyzing Public Positions on Known, Legally-Sanctioned Activity

**A White Paper**

*Draft for discussion. Subject to revision.*

 Abstract

Many categories of activity in public life are legally sanctioned and broadly known to exist, yet are not, as a matter of convention, disclosed or foregrounded by the public figures who participate in or position themselves around them. Examples include certain procurement relationships, advisory affiliations, intelligence- and security-adjacent authorities, and policy commitments that are exercised in practice but rarely campaigned on. The result is an accountability gap that is not secrecy in the legal sense — nothing is hidden that is required to be revealed — but is opacity in the practical sense: a public cannot easily understand where its officials and prominent public actors *stand* on activity it has a legitimate interest in.

This paper specifies an instrument that closes the practical gap without manufacturing a surveillance one. The system documents **publicly attested positions** taken by public figures on a defined category of known, sanctioned activity, structures those attestations into a queryable corpus with cryptographically verifiable provenance, and provides an analytical layer for understanding how a network of public actors positions itself over time. The design commits a tamper-evident, takedown-resistant record of *what was said and where*, not an inferential profile of *what a person has not disclosed*. The distinction is load-bearing and is enforced in the architecture, not merely asserted in the framing.

---

## 1. Purpose and Scope

### 1.1 What the system is for

The instrument exists to answer a specific, civically legitimate question: **for a defined set of public figures, on a defined category of known and legally-sanctioned activity, what positions have they publicly attested to, and how does the resulting network of positions look and change over time?**

This is ordinary accountability work — the kind performed by oversight researchers and political journalists who map, for instance, which legislators quietly support a surveillance authority they never mention on the campaign trail. The contribution here is infrastructural: a verifiable, durable, analyzable corpus that survives pressure, resists quiet alteration, and ties every claim to a primary source.

### 1.2 The unit of analysis

The atomic object of the system is an **attestation**: a single instance of a public figure expressing a position on the activity-category, bound to its source. The unit is *a position attested to a source*, not *a fact extracted about a person*. This choice is the foundation of everything that follows. A system whose records are attested public statements is auditable and contestable; a system whose records are inferences about undisclosed conduct is a dossier. The two have similar surface descriptions and opposite consequences.

### 1.3 Subjects in scope

- **Public officials acting in an official or candidate capacity.** Documented on positions relevant to their public role. This is the core and least contestable category.

- **Public figures who have voluntarily made themselves public on the specific issue.** Documented *only on their public statements on that issue* — never on their affiliations, associates, or network ties.

### 1.4 Subjects and operations explicitly out of scope

- Private individuals.

- Affiliation- or association-mapping of non-officials, including from public data. (See §6 for why this is excluded even though it is technically feasible.)

- Any subject for the purpose of cataloging who they are connected to rather than what they have publicly said.

---

## 2. The "Peripheral" Question, Stated Precisely

A motivating observation behind this work is that a public figure's *real* position on a sensitive-but-sanctioned activity often surfaces incidentally — in an aside, a reply, a procedural vote, a stray remark — rather than in their headline messaging. The system is explicitly designed to capture this incidental material. But "incidental" must be defined carefully, because two readings diverge sharply:

**Reading A — incidentally-expressed positions (in scope).** A public figure expresses a stance on the known activity-category in a low-salience context: an offhand remark, a committee aside, a vote with no accompanying statement. The *subject* is still a position the figure has publicly taken; it is simply not foregrounded. Capturing this — and flagging it as low-salience so it is not dressed up as a manifesto — produces a *more* honest corpus than one limited to polished statements.

**Reading B — accumulation of tangential traces to establish undisclosed involvement (out of scope).** Here the mechanism is the assembly of many individually-innocuous peripheral signals into an inferential picture of a person's relationship to activity they have *not* disclosed. This is the mosaic method of profiling. Automated, applied to named individuals, and committed to a permanent ledger, it is an automated dossier engine — the precise apparatus that accountability work of this kind ought to be auditing, not building.

**The system implements Reading A and explicitly forecloses Reading B.** Concretely:

- An attestation must bind to a passage in which the subject *expresses a position*. Material that merely associates a subject with the activity-category, without an expressed position, is **not** an attestation and is not retained as one.

- The salience flag (§3) records *how* a position was expressed (deliberate vs. incidental). It is not a mechanism for upgrading non-positional association into an inferred position.

- The analysis layer (§5) operates over attested positions and over-the-public-record action. It does **not** aggregate non-positional peripheral signals into inferred-conduct profiles.

This is the single most important design commitment in the paper, and §6 expands on why it is enforced structurally rather than left to operator discretion.

---

## 3. Data Model: The Attestation Object

Each attestation is a structured record with the following fields:

| Field | Description |

|---|---|

| `subject` | The public figure, resolved to a canonical entity (§4). |

| `activity_category` | The defined, known, legally-sanctioned activity being positioned on. |

| `stance` | The position expressed. Source wording is **paraphrased by default**; any direct quotation is minimal and used only where exact wording is materially significant. |

| `stance_classification` | A normalized classification (e.g., supports / opposes / conditional / declines-to-state), with the classifier version recorded. |

| `provenance` | An archival capture of the source: rendered snapshot, content hash, capture timestamp, and capturing-agent identity. (§4.2) |

| `context` | Sufficient surrounding material that an incidental remark is not distorted by isolation. The capture links to, rather than replaces, the source. |

| `salience` | An explicit flag: was this a deliberate, foregrounded statement, or an incidental/low-salience one? Required, not optional. |

| `subject_qualification` | The recorded basis on which this subject met the in-scope criteria (§4.1) at capture time. |

| `signature` | A cryptographic signature over the above by the capturing agent. (§4.4) |

The `salience` and `context` fields are what make incidental material usable without making it misleading. A corpus that silently flattens the difference between a floor speech and a stray reply is analytically worse, not merely ethically worse, because downstream analysis inherits the distortion.

---

## 4. Technical Architecture

### 4.1 Entity resolution and scope gating

Mentions, handles, and references are resolved to canonical public-figure entities. Critically, resolution is **gated by an explicit, auditable scope criterion**: before any subject is admitted to the corpus, the system records *why* they qualify (official capacity; or voluntary public engagement on the specific issue). The criterion is versioned and logged, so scope cannot quietly creep from officials toward private individuals over time. A subject failing the criterion is rejected at ingestion; there is no "capture now, classify later" path that would allow out-of-scope accumulation.

### 4.2 Ingestion and archival capture

Ingestion captures **rendered source snapshots**, not merely scraped text. Each capture is hashed (e.g., SHA-256) at the moment of acquisition, with the hash, a trusted timestamp, and the capturing agent's identity bound to the record. This makes each attestation independently verifiable: anyone can confirm that a stored capture is the artifact it claims to be and has not been altered since capture.

### 4.3 Position extraction

A natural-language layer identifies passages in which a subject expresses a stance on the activity-category — including incidental ones (Reading A) — and produces the `stance` and `stance_classification` fields. Two hard rules constrain it:

1. **Link, don't replace.** Extraction always links to the captured source and preserves context; the paraphrased stance never stands in for the primary material.

2. **Position-bound only.** The extractor emits an attestation only where a position is expressed. Passages that associate a subject with the activity-category without an expressed position do not become attestations. This is the code-level enforcement of the Reading A / Reading B boundary.

### 4.4 Provenance and distributed verifiability

Captures are committed to **content-addressed storage** (e.g., an IPFS-class system), and attestations are **cryptographically signed**. This yields three properties that genuinely strengthen accountability:

- **Tamper-evidence** — a record cannot be quietly altered without detection.

- **Takedown-resistance** — the corpus cannot be memory-holed by a subject or a host under pressure.

- **Verifiable provenance** — any third party can independently confirm a capture's integrity and origin.

These are the properties that a centralized database cannot guarantee, and they are the legitimate substance of the "distributed ledger" idea. The ledger records *signed, sourced attestations of public statements*. It is **not** a market: positions are not minted as priced, tradeable assets. Pricing claims about named individuals rewards whoever produces the most damaging or salacious assertion regardless of accuracy, and a market keyed to real people's conduct carries incentive structures around what happens to those people that have no place in an accountability instrument. The corpus is the asset; it is not a tradeable one. (See §6.3.)

A note on permanence: tamper-evidence and takedown-resistance are virtues for *sourced public statements* and liabilities for *inferential claims about undisclosed conduct*. This asymmetry is a further reason the corpus admits only the former. Permanence is safe precisely because every record is a verifiable, contestable, primary-sourced statement rather than an uncorrectable inference.

### 4.5 Analysis layer

Over the attested corpus and over public-record action, the analysis layer supports:

- **Position network graphs** — how a defined set of public actors clusters around a stance.

- **Temporal drift** — how positions shift over time, including divergence between earlier and later attestations.

- **Contradiction detection** — gaps between attested positions and other *documented public action*: votes, public filings, prior public statements. Contradiction is computed against the public record, not against inferred conduct.

### 4.6 Query interface and correction mechanism

Every surfaced claim resolves to its primary source. The interface includes a **correction/reply mechanism**: a documented subject can contest a mischaracterization, and the contestation is bound to the record. A tool that documents people and offers them no path to dispute is not accountability; it is exposure. Because the corpus is composed of sourced public statements rather than inferences, corrections operate meaningfully — they attach to a verifiable artifact rather than to an unfalsifiable profile.

---

## 5. Analytical Outputs

The intended product is not a list of gotchas but a structured, verifiable map of how a network of public actors positions itself around a known category of activity — including the positions it does not foreground. Representative outputs:

- A network view of where officials cluster on the activity-category, with each node's stance traceable to attested statements.

- Over-time trajectories for individual subjects and for the network.

- Flagged divergences between foregrounded messaging and incidentally-expressed positions, each side sourced.

- Flagged divergences between attested positions and public-record action.

Every output is a function of attested, sourced positions and public-record facts. No output is a function of accumulated non-positional peripheral signal.

---

## 6. Design Principles and Foreclosed Capabilities

This section states, as documented design commitments, the things the system deliberately does **not** do — including capabilities that are technically straightforward and were considered.

### 6.1 No inferential accumulation of undisclosed conduct

The system does not assemble tangential, non-positional signals into a profile of an individual's relationship to activity they have not disclosed (Reading B, §2). This is foreclosed at the extraction layer (§4.3), not merely discouraged by policy. The rationale: such accumulation is the mosaic method of dossier construction; automated and committed to a permanent ledger it produces uncorrectable inferential exposure of named individuals; and it degrades accountability by rewarding the most damaging inference over the most defensible documentation.

### 6.2 No affiliation- or network-mapping of non-officials

For voluntarily-public non-official figures, the system documents *public statements on the issue* only — never associations, associates, or ties. Affiliation-mapping of non-officials, even from public data, is structurally a movement map, and movement maps are used for intimidation far more reliably than for accountability.

### 6.3 No tradeable market in claims about individuals

The distributed layer provides verifiability, not monetization. Positions are not minted as priced or tradeable assets. (§4.4.)

### 6.4 Scope cannot silently creep

The subject-qualification criterion is explicit, versioned, logged, and enforced at ingestion (§4.1), so the corpus cannot drift from officials-in-official-capacity toward private individuals without a visible, auditable change to the criterion.

### 6.5 Contestability is structural

The correction mechanism (§4.6) is part of the architecture, not an afterthought, and is meaningful precisely because records are sourced public statements rather than inferences.

---

## 7. Legal and Ethical Posture

The instrument documents public statements by public figures on matters of public concern — the heartland of protected, legitimate civic speech and oversight. Its risk surface is governed by the design commitments in §6. Two points warrant emphasis:

- **Defamation / false-light exposure** is minimized by binding every claim to a verifiable primary source, paraphrasing rather than re-publishing at length, classifying stances conservatively, recording salience, and providing contestation. The system asserts "here is what was said, and where," not "here is what this person secretly does."

- **The intellectual-property and consent problems** of wholesale re-publication are mitigated by capturing for provenance and verification while linking to, and minimally quoting from, primary sources rather than reproducing them.

This paper does not constitute legal advice; any deployment should be reviewed by counsel in the relevant jurisdiction.

---

## 8. Summary

The accountability gap this instrument addresses is real: known, legally-sanctioned activity around which public figures take positions they do not foreground. The instrument closes that gap by documenting **attested public positions**, with cryptographically verifiable and takedown-resistant provenance, and by providing network and temporal analysis over the resulting corpus — while structurally foreclosing the inferential-accumulation, affiliation-mapping, and market mechanisms that would convert an accountability tool into the surveillance apparatus it ought to be auditing. The boundary between documenting what was publicly said and inferring what was undisclosed is the design's central commitment, enforced in the data model and the extraction layer rather than asserted in the preamble.