What Prescient promises. And what it won't.
This page is a public, immutable commitment. Changing it requires 90 days of public notice posted at /manifesto and emailed to all signed-in users with no opt-out. We consider this our deepest competitive moat.
- 1.No surveillance advertisingNo ad inventory, no programmatic ads, no third-party trackers, no user-data syndication to ad networks. Period.
- 2.No loot boxes or randomized reward purchasesWe never sell randomized outcomes.
- 3.No purchasable in-product currencyInsight Points, Lede Tokens, and Streak Shards are earned through use. Subscription unlocks features, never advantages.
- 4.No tier shortcutsEvery tier — Apprentice through Oracle and beyond — is reachable on the free plan at the same rate as paid.
- 5.No energy refillsNo artificial scarcity that resets when you pay.
- 6.No engagement-bait push notificationsBanned: streak-at-risk panic pings, FOMO breaking-news at unusual hours, empty 'you have notifications' pushes.
- 7.No infinite scroll without an exitEvery feed has a 'you're caught up' endpoint card.
- 8.No false progress barsNo 99%-stuck loaders. Real numbers only.
- 9.No streak shameA broken streak surfaces in the Streak Museum as a chapter complete, never with a punishing '0' or guilt copy.
- 10.No dark-pattern paywallsFree-tier limits are stated upfront. No surprise popups mid-article. No nag walls.
- 11.No data sale, everReading history, predictions, calibration data, game scores, and notes are not sold, licensed, or shared without explicit per-feature opt-in.
- 12.No disabling Doomscroll ShieldAfter 45 minutes of continuous reading, gamification pauses and a break is prompted. Always on.
- 1.
Lose revenue rather than break this charter.
- 2.
Make tier progress measurable and earnable — no hidden multipliers.
- 3.
Surface 'Why this story' on every recommendation — no black-box algorithm.
- 4.
Calibrate against ourselves: Cockpit shows our own historical Brier score, hallucination rate, and fact-check coverage — visible to every user.
- 5.
Honor declared breaks: vacation, bereavement, work crunch — streaks pause cleanly without losing tier progress.
- 6.
Treat under-18 readers with extra care: Teen mode disables leaderboards, rivals, public profiles, and any streak-shame surface.
- 7.
Respect declared interests: the blind-spot radar surfaces what you're missing without forcing it on you.
What is calibration, and why does it matter?
Calibrated is not the same as accurate. A weather forecaster is accurate if they're right. They're calibrated if the 70%-chance-of-rain forecasts actually produce rain 70% of the time. Accuracy without calibration lets an AI sound confident while being systematically wrong. Calibration is a checkable discipline.
Prescient uses the Brier score (Glenn Brier, 1950) — the mean squared error between a forecast probability and the realized outcome. The formula is simple:
Brier Skill Score (BSS)
A raw Brier score only becomes meaningful when compared to a climatology baseline — the score a naive predictor would achieve by always forecasting the historical average outcome rate. BSS normalizes for this:
Prescient reports BSS per surface. A surface at BSS 0.20 is 20% better than the baseline predictor for that type of forecast — that improvement is the value the engine is delivering.
12 calibrated forecast surfaces
Every prediction Prescient makes is routed through one of these 12 falsifiable surfaces. Each surface has a stated horizon and a climatology baseline. At horizon resolution the prediction is scored, the Brier and BSS are recorded, and the running per-surface aggregate is updated. This table is generated from the live engine spec — it stays in sync with the code automatically.
Predicted personal-relevance score (0-100) of a topic 24 hours from now — climatology baseline: 0.100
Predicted personal-relevance score (0-100) of a topic 72 hours from now — climatology baseline: 0.120
Predicted momentum class (accelerating / steady / fading) at +24h — climatology baseline: 0.667
Predicted %-decay-per-day of a topic — realized vs predicted at +24h — climatology baseline: 0.080
Predicted %-decay-per-day of a topic — realized vs predicted at +72h — climatology baseline: 0.100
Did a "breaking_topic" catalyst remain >200% velocity in next 24h? — climatology baseline: 0.210
Did a "cluster_forming" catalyst reach ≥4 articles + ≥3 sources in 24h? — climatology baseline: 0.240
Did a "topic_fading" catalyst lose ≥50% relevance in 72h? — climatology baseline: 0.210
Did a "momentum_peak_predicted" catalyst show importance↑ in next 24h? — climatology baseline: 0.240
Did the user read a "now"-urgency top-read recommendation within 4h? — climatology baseline: 0.210
Did the user read a "today"-urgency top-read recommendation within 14h? — climatology baseline: 0.250
Predicted reader trajectory class (expanding / narrowing / stable) at +7d — climatology baseline: 0.667
The engine predicts which stories will matter to you in 24-72 hours. It gets things wrong. A Brier score of 0.0 is the theoretical maximum; we're nowhere near it. The calibration zone in the Cockpit shows you exactly how wrong we are, per surface.
Each of the 12 forecast surfaces starts with a seeded climatology baseline derived from theoretical priors, not measured data. We need at least 30 resolved claims per surface before the calibration numbers are statistically meaningful. The Cockpit surfaces this count explicitly.
A positive Brier Skill Score means the engine is beating the baseline predictor. It does not mean the stories we surface are the most important stories in the world — it means we're better than naive at predicting which ones matter to you specifically.
When the prescience methodology changes (e.g., new surface thresholds, model retraining), we version the methodology hash. Old predictions resolve against the methodology that issued them. We do not retroactively restate historical predictions.
This trust page can be strengthened immediately. It can only be weakened with 90 days of public notice, an email to all signed-in users with no opt-out, and a posted notice at /manifesto. Weakening includes any change that expands permitted monetization, reduces Doomscroll Shield enforcement, allows tier shortcuts, or removes a charter clause.
We sell features, not advantages. Pricing is transparent and listed on /pricing.
3 AI takes/day · 24h prescience · basic Cockpit
Cockpit advanced panels · unlimited AI takes · 72h prescience · audio briefings
API access · bulk RSS export · calibration analytics export · white-label briefings
One partner cap per day · never interleaved with editorial
The most-used news platforms have converged on engagement metrics and surveillance advertising. Prescient is built on a different premise: that readers will pay for a product that respects them. This charter is a bet on that premise. We are willing to lose money to keep the bet honest.