Confidence tiers

How research items get a 0–1 score, and how that score changes over time.

Every item in Research carries a confidence score between 0 and 1. The score is the system’s best guess at how seriously a thoughtful reader should take the claims in the piece.

The score, briefly

Score	Tier	Color	What it means
0.80 – 1.00	High	Bright green	Strong author track record, specific testable claims, often corroborated
0.60 – 0.80	Mid-High	Light green	Solid author, reasonable specificity, worth reading carefully
0.40 – 0.60	Medium	Yellow / amber	Mixed signals — read with skepticism
0.20 – 0.40	Mid-Low	Light red	Either weak author or vague/promotional claims
0.00 – 0.20	Low	Bright red	Skip unless you have a specific reason

The badge on each card shows the score. The default sort on the browse page is Newest, but you can sort by confidence if you want signal-first.

What goes into the score

Roughly:

• Author reliability — historical track record on prior calls (where measurable). Substack writers and X accounts are scored separately.
• Claim specificity — “NVDA is a buy” scores worse than “NVDA’s data-center revenue will accelerate to 50% Q/Q because of X, with X being a falsifiable claim about contract pipeline.”
• Corroboration — does another source independently make a similar claim? Multiple confirms raise the score.
• Source quality — paid newsletters with a track record score above free, anonymous threads.
• Hedging language — pieces full of “might,” “could,” “potentially” score lower than confident specific claims.

Lifecycle and staleness

Confidence isn’t static. A piece written 6 months ago about Q4 earnings is stale by Q1 — the claims either resolved or didn’t, and the score collapses.

Three lifecycle states:

State	Display	Meaning
Fresh	Full opacity	Claims are live, score is current
Aging	Slight fade	Old enough that some claims may be resolving
Stale	Heavy fade (~55% opacity)	Old enough that the piece is mostly historical

The fade is visible in the feed — older pieces literally dim. Your eye learns to treat them as reference, not signal.

What confidence is not

• Not a return prediction — high-confidence calls can be wrong. The score reflects how well-reasoned the case is, not the future.
• Not censorship — low-confidence pieces still appear in the feed (just with the badge); you decide what to read.
• Not personalization — the score doesn’t reflect your views or holdings. Same score for everyone.

How to use it

• Filter to High when time-constrained — the most efficient first pass through a busy day’s content
• Sort by Confidence on a researched name — the highest-scored takes on $XYZ are the ones that earn the read
• Don’t blindly trust — a high-confidence call from a strong author still gets your skepticism. The score is a starting point, not a verdict.