PROVENANCE
Request Access

MEMORY INTEGRITY INFRASTRUCTURE

Your AI has been thinking with contaminated memory.

Project Provenance imports your ChatGPT memory export, runs multi-layer detection across factual errors, stale assumptions, and decision-level drift — then generates a clean seed file you can trust.

V1 scope: ChatGPT memory export audit. Early access only.

MEM-DRIFT-001

Stale assumption

Context from 14 months ago still active

MEM-CONT-047

Factual contradiction

Two incompatible beliefs held simultaneously

MEM-SCAF-012

Downstream scaffolding

Bad frame propagated to 8 dependent decisions

THE PROBLEM

Corrupt the foundation, corrupt the output.

AI memory accumulates over months. Facts become outdated. Beliefs contradict. Frameworks built on early assumptions persist long after the assumptions failed. Your AI keeps thinking — and advising — from a compromised ground truth.

01

Factual errors and stale assumptions

A company you mentioned closed two years ago. A role you held changed. A market you described evolved. The AI still treats all of it as current.

02

Decision-level contamination

One bad frame from an early strategy session propagated into a dozen later recommendations. You can not see the lineage. The damage compounds.

03

Overconfident beliefs and classification errors

High-confidence memories that are directionally wrong are more dangerous than acknowledged unknowns. Your AI does not surface its own uncertainty.

HOW IT WORKS

A complete chain-of-custody for your AI memory.

01ChatGPT Export

Import

Export your ChatGPT memory file. Upload it to Provenance. We parse and normalize every memory record into a structured, auditable format.

028 Detection Layers

Audit

Multi-layer detection runs across your memory: factual accuracy, staleness, confidence calibration, contradiction, scaffolding effects, and reset candidates.

03Human Review

Quarantine

Review flagged memories with full context. Mark contaminated, stale, or ambiguous records for removal. You control every decision.

04Clean Seed File

Export

Generate a clean memory seed file ready to re-import into your AI system. Your model starts its next session from verified ground truth.

DETECTION LAYERS

Not a grammar check. A full audit.

FACTUAL LAYER

Factual error detection

Identifies memories that contradict verifiable facts or that have become outdated since they were recorded.

TEMPORAL LAYER

Staleness detection

Flags memories by age and context drift. Assumptions from 12 months ago are scored differently than last week's.

SCAFFOLDING LAYER

Downstream contamination mapping

Traces how a single bad memory frame propagated into dependent beliefs and decisions across your history.

CONFIDENCE LAYER

Overconfidence scoring

Detects memories that are held at high confidence but have low source reliability or significant contradiction signals.

CLASSIFICATION LAYER

Category and role error detection

Surfaces memories where the AI has miscategorized you, your work, your context, or your relationships.

PROVENANCE LAYER

Source tracing

Every memory record is linked to its origin conversation where possible, so you know what produced each belief.

EARLY ACCESS

How confident are you in what your AI remembers about you?

Project Provenance is in private early access. Join the list to be notified when audit slots open. V1 targets ChatGPT memory exports.