AURiX WHITEPAPER: Record
================================================================================

Title: Record — Universal Data Egress: Anyone's Anything from Anywhere While Possible, Observation Layer When Not

Author: AURiX Protocol | 2026-04-21

Status: Desktop infrastructure, 6-layer ladder. Layer 1 shipping live (cron 04:08 Pacific, writing to /AURiX/cowork_archive/). Layers 2-6 scoped, not yet shipped. Track-two framework for the universal data egress thesis (see memory: project_data_egress_thesis).

================================================================================

ABSTRACT

Record is the egress infrastructure for the AURiX protocol. It answers the operational question: when a platform, service, or tool holds data that belongs to the user (or that the user has agency over), how does the user get that data out — reliably, durably, and without depending on the platform's continued cooperation?

The answer is a six-layer ladder. The bottom layers retrieve what is already the user's (conservative, clearly consented). The top layers observe what is ambient in the user's environment (aggressive, requires careful permission and user agency). Every layer declares its scope, consent model, and ethical gradient. The ladder is the architecture; the user chooses how far up they climb.

THE THESIS

The AURiX north-star framing is: *anyone's anything from anywhere while possible; observation layer when not.* Platforms export or don't. When they export, retrieve. When they don't, observe structurally. The user retains the record either way.

This thesis lives in two tracks inside AURiX:
  - Track one: protocol-level observation (AURiX itself — make structural conditions of the page visible).
  - Track two: egress infrastructure (Record — retrieve what's retrievable, observe what isn't).

Record is track two. This paper specifies it.

THE LADDER — SIX LAYERS, CONSERVATIVE TO AGGRESSIVE

LAYER 1 — COWORK SESSION AUTO-ARCHIVE (LIVE)

Problem: Cowork session transcripts (Aaron's current primary working context with Claude) are not included in the standard Claude.ai export. Leaving them only in Cowork's own database = single point of failure, zero egress.

Solution: Scheduled task fires 04:08 Pacific daily. Reads each session's transcript via Cowork's session-info tool. Writes JSON + clean text to /AURiX/cowork_archive/<date>_extraction_v*/. Ledger of runs lives alongside.

Status: LIVE since mid-April 2026. Verified by multiple archive dates already on disk.

Consent model: self-retrieval of the user's own working sessions. Fully consented (the user runs it).

LAYER 2 — CLAUDE.AI WATCH-FOLDER AUTO-RECLAIM (SCOPED)

Problem: Claude.ai exports contain the user's own chat history but arrive as ZIP archives that are hard to search, cross-reference, or index.

Solution: A watch folder monitors for new Claude.ai export ZIPs. On detection, pipes through reClaim to produce structured searchable text files. Result lands in the corpus, queryable by Answer.

Status: scoped, not shipped. reClaim already exists; the watch-folder wrapper is the remaining piece.

Consent model: self-retrieval of the user's own exports.

LAYER 3 — FILE-MANIFEST JSONL OVER MOUNTED FOLDERS (SCOPED)

Problem: Files scattered across the user's drives (Downloads, Desktop, project folders) lose metadata and cross-reference over time. What was edited when, what lives where, what was the previous version.

Solution: Walk mounted folders on a schedule, emit JSONL with path / hash / size / mtime / declared type / extracted keywords. Build a local searchable index without uploading anything. Feeds Answer directly.

Status: scoped.

Consent model: self-observation of the user's own file system.

LAYER 4 — BROWSER-EXTENSION AGGREGATION (SCOPED)

Problem: The browser instruments (Gauge, Scope, Trace, Ledger, Receipt) each hold their own observations in chrome.storage.local per extension. Cross-extension queries and long-term retention require aggregation.

Solution: Ledger's Exterminator aggregator pattern generalized. Each extension can push observation events to a central Record endpoint (file on disk via native messaging, or browser-based aggregation). The user gets a unified record across instruments, exportable as a corpus.

Status: scoped. The Exterminator error-aggregation pattern (already shipped) is the proof-of-concept.

Consent model: self-aggregation of instrument outputs the user already has.

LAYER 5 — CLIPBOARD CAPTURE (SCOPED, ETHICAL GRADIENT STARTS)

Problem: Much of what passes through the user's attention passes through the clipboard. Research notes, quoted passages, URLs, code fragments. None of it is retained after the next copy.

Solution: Opt-in clipboard watcher that writes each clipboard event to a local, encrypted ring buffer. User browses their own clipboard history, exports what matters.

Status: scoped. Marked as the start of the ethical gradient because clipboards can contain sensitive transient data (passwords briefly in transit, financial details pasted once). The design must make the recording visible, the storage encrypted, the deletion trivial, and the opt-in explicit.

Consent model: explicit opt-in, user controls retention.

LAYER 6 — SCREEN / AUDIO CONTINUOUS (SCOPED, SELFiX-ONLY UNTIL COMPASSION-CONSTRAINT IS IRONCLAD)

Problem: The ambient user environment contains signal that no explicit export can reach — spoken conversations, on-screen content not tied to any exportable source, moments of realization that happen and vanish.

Solution: Continuous screen and audio capture with on-device processing. No network layer. No cloud. User-held data.

Status: scoped; flagged SELFiX-only. Layer 6 cannot ship without the compassion constraint holding — the property that surfaces without shaming, enables return without punishment. If Layer 6 data can be weaponized (even against the user themselves, e.g. shame spirals over replayed moments), the design has failed. The invariants hold, but the design layer above them must ensure the tool supports agency, not self-surveillance-as-coercion.

Consent model: SELFiX explicit opt-in, user-only access, compassion-constraint-enforced design.

THE ETHICAL GRADIENT

Layers 1-4 retrieve data the user already has rights to. The only question is how reliably the retrieval works.

Layers 5-6 cross into ambient observation. This requires careful design beyond technical consent. Explicit opt-in is necessary but not sufficient. The compassion constraint enters here: even user-to-self observation can become punitive if the design invites replay-as-regret or surveillance-as-self-control. Layers 5-6 must support agency and return, not punishment.

THE ARCHITECTURE LOGIC

Why a ladder, not a single tool? Because consent and ethics don't scale linearly with capability. Layer 1 is uncontroversial (self-retrieval of explicit working artifacts). Layer 6 is delicate (continuous self-observation with replay capability). The same codebase shouldn't handle both without distinguishing the consent model, storage constraints, and design posture at each rung.

The ladder makes the gradient declared, not implicit. Invariant 4 (observable or declared) at the architecture layer.

AURIX-NATIVE PROPERTIES

Record honors the 8 invariants:
  1. Mechanism only. Each layer retrieves or observes; none judges what it finds.
  2. No interpretation imposed. Layer outputs are raw records; interpretation is a separate step (Answer, AUDiT).
  3. Separation of layers. Literal — six declared layers with distinct consent models.
  4. Observable or declared. Every capture enters a local record; the layer boundary is declared.
  5. Explicit origin. Every record cites its source layer and retrieval timestamp.
  6. No silent transformation. Captures are raw; anonymization (via Anonymize) is a separate declared step.
  7. Reproducibility boundary. Layer 1 runs on cron and produces the same output given the same inputs.
  8. No undeclared meaning. The ladder structure itself declares the consent and ethical posture of each layer.

THE STRATEGIC FRAME

Record exists because:
  1. The universal data egress thesis requires infrastructure, not just philosophy.
  2. Platform export gaps are already causing friction (see the Cowork export gap memory: standard Claude export does NOT include Cowork session transcripts). Record is the live fix.
  3. Grant pipeline (NLnet, Shuttleworth, OTF) wants concrete demonstrations. "The application you are reading was retrieved by the tool it describes" is the line that lands.
  4. Personal survival: if the user loses access to any platform (account suspended, service discontinued, hardware failure, death), Record has already moved the data to local, user-held, inspectable form.

CONCLUSION

Record is not a backup tool. It is the egress infrastructure that makes the AURiX thesis operational. "Anyone's anything from anywhere while possible; observation layer when not" requires a concrete ladder of mechanisms that implements both halves. Layer 1 is live; Layers 2-6 are scoped against the same invariants. Every user with Record configured is one Aaron-death away from complete data continuity, not a platform's cooperation away.

================================================================================
END OF WHITEPAPER
================================================================================