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.
Stale assumption
Context from 14 months ago still active
Factual contradiction
Two incompatible beliefs held simultaneously
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 Facts become outdated.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.
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.
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.
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.
Import
Export your ChatGPT memory file. Upload it to Provenance. We parse and normalize every memory record into a structured, auditable format.
Audit
Multi-layer detection runs across your memory: factual accuracy, staleness, confidence calibration, contradiction, scaffolding effects, and reset candidates.
Quarantine
Review flagged memories with full context. Mark contaminated, stale, or ambiguous records for removal. You control every decision.
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Identifies memories that contradict verifiable facts or that have become outdated since they were recorded.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.