Bring a question.
Leave with something useful.
Your data, its sources, and a clear research deliverable. One guided workspace from the first question to a reproducible report.
Connection access
Local sessions use this machine's workspace. An account key uses its separate owner scope. This pilot is not a hosted team collaboration service.
What should this research help you do?
Bring your data. Define the goal.
Keep the original, inspect what NOUS extracted, and decide what success means for this channel.
Upload to this channel
What should improve?
A plan defines the test. It does not claim an improvement or change believability weights.
Define your measurement plan
What does your data actually show?
Tables: complete CSV / TSV / JSON up to 2 MB, 10,000 rows and 50 columns. Only reviewed mapped fields enter the dataset. Other files remain retained; supported text extraction can be found through channel search.
Only numeric fields matching the plan's unit are offered. Baseline differences are descriptive, not proof that OMNIA improved the outcome.
Test your existing approach against OMNIA
Use recorded predictions and actual outcomes. OMNIA fits a small model to earlier observations, selects it on later observations, then compares it with your existing approach on the final time period.
Requires 100–2,000 complete rows with real timestamps and independent episode IDs. Inputs and baseline predictions must have been available before the outcome. Repeatedly tuning against the final results exposes that test set; use new data for confirmation.
Use documented scores from before the training cutoff. OMNIA compares weighted and unweighted models on the same selection rows. Final-test errors always use equal weight per observation. These supplied scores are not independently verified trust ratings.
Test the OMNIA foundations
Select at least one original numeric input. Derived inputs form an additional comparison. All candidates use the same observations; only the selection winner enters the final test.
If supplied, this vector must begin at zero and match the declared sample rate. No resampling or missing-value filling occurs.
The current engine accepts 8–4,096 numeric samples per window, up to 524,288 samples per batch (the earlier method accepts 8–256). Each window needs a reviewed amplitude unit. Time analysis adds short-window trajectories, trend, amplitude change and integrated squared amplitude. Half-window RMS change needs at least 16 samples; other quantities support eight. Original + signal features must total no more than 20 inputs. This measures signal components; it does not assign a universal vibe score.
Choose quantities for the comparison
Select quantities before testing. Deselect frequency for constant signals; it is unavailable, not zero. Descriptive inspection still shows all measurements.
Observation time and availability · optional
Map when the first sample was observed and when the entire window became available. Comparison checks both against each row’s observation timestamp. Without these fields, time remains relative and availability is your reviewed assertion.
Find supporting sources
Record a follow-up outcome
Interpretation is retained only when you freeze a report version.
Ready for a closer look.
The editable Markdown report is intended for sharing after your review. Dataset and source exports may contain private data. Downloads stay on your machine; nothing is published automatically.
Saved report versions
Claim IntelligenceTrace the statement. Inspect the sources. Follow the outcome.
See what connects.
One continuous path from a public statement to reusable knowledge.
Give your data a track record.
Track a claim to see real sources and outcomes connected here. No example findings are mixed into your data.
Owner-reviewed outcomes · Connections show recorded lineage, not causation. Motion is decorative; it does not represent measured EFVT or live transfers.