Source scout
Fetch up to 30 recent PubMed-indexed records per track from Europe PMC.
MNEME RESEARCH / SOURCE SCANNING + AI COMPARISON
Meet Scout.
Follow the evidence.
Follow how Scout retrieves publications, screens sources and compares abstracts with AI. Inspect the draft, counterevidence and unanswered questions; editorial reports remain a separate review step.
AI-CREATED CONCEPTUAL ART / NOT A LIVE BRAIN SCANConnecting to server
SELECTED RESEARCH FOCUS
Lipid handling
Does changing lipid handling protect neurons, and were tau and vascular outcomes measured?
Help investigate this question ↗Saved run receipts · all tracks—
Today’s scheduled trackConnecting…
Last receipt · selected trackConnecting…
MNEME / SCOUT_01EUROPE PMC
source recordsABSTRACTS
screening rulesRECEIPTS
inspectable output
source recordsABSTRACTS
screening rulesRECEIPTS
inspectable output
Animated character · activity status comes from the server
OBSERVER / 01CONNECTING
A curious mind.
A checkable trail.
Connecting to Scout’s saved activity. Choose a track below to inspect the question and its sources.
FOCUSLipid handling
LAST SERVER EVENT
Waiting for a server response.
Live updates check the server. The motion control switches decorative animation off.
SOURCE SCREENING / RULE-BASED FLAGS / HUMAN REVIEW
SCOUT TERMINALConnecting
- 01 · RetrieveWaiting
- 02 · ScreenWaiting
- 03 · FlagWaiting
- 04 · Save receiptWaiting
Waiting for server-reported progress. Decorative motion does not indicate an active scan.
Selected track / server event log
Waiting for the saved activity log.
No status received yet.
Safety screen
Flag caution language and neuronal, tau and vascular terms for human inspection.
Receipt keeper
Store sources, timestamps, exclusions, rules and a SHA-256 content hash.
Three stages of one rule-based agent, not three independent experts. The source scanner does not call an AI model, infer treatment efficacy or test brain-cell datasets. The separate AI analyst below compares abstracts when configured. Editorial reports remain separate.
Compare study findings, models and limitations. Look for possible connections and counterevidence, with a source trail for every claim. AI drafts are separate from editorial research reports.
Checking the server…
SERVER ACTIVITY / AI ANALYSIS
- No activity loaded yet.
No AI output loaded.
Scope, costs and evidence checks
Two bounded Europe PMC searches: six recent records and six counterevidence-oriented records, deduplicated. Only available abstracts are analyzed, with up to 6,500 characters per source. Full texts, comprehensive correction checks and study independence are not established. Counterevidence search terms can miss contradictory findings.
The model must cite retrieved source IDs and supply a short exact quote for each source. Code checks those references and quotes before displaying a draft. This checks provenance, not whether the interpretation is scientifically correct. Human review is still required.
Only the site owner can start an AI run. One stored attempt per track per UTC day, including failures; repeat requests reopen the saved attempt. Maximum output allowance: 6,500 tokens per model call. Actual provider token usage is recorded when returned. All Scout model calls share a server-side US$100 daily spending limit (not account balance; tax and hosting excluded), reset at midnight Europe/Amsterdam. A conservative maximum cost is reserved before each request. Uncertain charges remain reserved. The persistent research queue supports owner-authorized scheduled two-source comparisons without an open browser, when the linked scheduler is connected. Manual broadcast sessions remain available. Daily source screening remains rule-based.
Select a saved run below or start a scan.
Run history
Newest 30 receipts. Failures stay visible. Times are shown in your local timezone.
Loading saved receipts…
AUTOMATION WITH BOUNDARIES
Curiosity runs automatically.
Conclusions need scrutiny.
While you’re away
The existing daily literature update at 08:00 Europe/Amsterdam also scans one rotating Scout track. It works without an open browser. This page refreshes its log every 20 seconds while visible, or every 3 seconds while your run request is pending. You can pause automatic updates; this does not stop a server scan or the daily schedule. Refreshing the log does not run research.
What the flags mean
Words can point to a useful paper. They do not tell us whether a treatment works, whether endpoints were measured together, or whether a result is new. Inspect the source before making a claim.
What a receipt proves
The hash lets you check that the receipt content matches its recorded fingerprint. It does not validate the science, prove independence or provide a blockchain timestamp.
Read the exact screening method
Search window: 2020 to the run date. Source: Europe PMC records indexed in PubMed. Sorted by publication date, limited to 30 results per track. Deduplicate DOI or source ID; exclude records tagged as preprints, retractions, withdrawals or corrections. Reviews remain and are labelled. Apply published regular expressions to available abstracts. No direction-of-effect classification or quality score is generated. Absence of a keyword is not absence of an endpoint. Linked corrections are not comprehensively checked.
Queries, rules and exclusions are saved with every completed receipt. A new-in-window count compares bounded result windows, not the whole literature.
Inspect the screening code ↗Funding policy in preparation: intended reinvestment of received token fees into research and AI costs. No token launched, allocation percentage not finalized. Source screening is deterministic and does not consume model tokens. The separate AI analyst uses paid model calls only when configured and started by the owner. Funding disclosures ↗