Rubric 1.0.0 · tuning/1.3.0
How the scores work
Every card carries two kinds of number — the editorial rating (1–5) and the five card attributes (0–100) the game is played with. One rule governs both:
AI detects evidence. Code calculates numbers.
A model only answers yes/no about quoted sentences, or copies a number out of their words. Every weight, curve, and tier below is fixed code — the same claims always produce the same numbers, and this page renders from the exact constants the pipeline runs on.
Part I — The editorial rating (1–5)
| Factor | Pts |
|---|---|
Specific wealth outcome or multiplier A named income figure, multiple, or timeframe presented as an achievable result. | 2.0 |
Health, disease, cure, or medication outcome The offer is linked to a physical or mental health outcome, diagnosis, medication, or remission. | 3.0 |
Unsupported or pseudoscientific mechanism The advertised cause is quantum, energetic, cellular, ancestral, or otherwise unmeasurable. | 2.0 |
Strong certainty, guarantee, or inevitability The result is framed as certain, guaranteed, or unavoidable rather than possible. | 1.0 |
Unfalsifiable or self-sealing explanation Failure is pre-explained by the buyer's resistance, blocks, readiness, or mindset. | 1.0 |
Testimonials used as causal proof Anecdotes are offered as evidence that the method caused the outcome. | 0.5 |
High-ticket pricing or artificial value anchoring A large price, a struck-through 'value', or scarcity framing does the persuading. | 0.5 |
| Maximum | 10.0 |
Each factor counts at most once, across all verified claims. The triggered weights sum to the card’s raw rubric points (0–10); points ÷ 2, floored at 1.0, is the displayed 1–5 rating.
- 4.5+Supreme WitchNothing is beyond manifestation.
- 3.5+Advanced WitchReality is becoming negotiable.
- 2.5+Practicing WitchEvidence is getting optional.
- 1.5+Apprentice WitchThe magic is starting.
- 0.0+Witch CuriousMostly normal. Mostly.
Part II — The five attributes (0–100)
Evidence reaches the forge in two lanes. The displayed claims: a model answers yes/no per evidence flag and copies numbers out of the words (“€4,997” → 4997) — never estimating, never judging. The full pages: a deterministic scan (no model) sweeps every captured page for prices, promises, scarcity, and medication language that never gets quoted. Page signals supplement the claims, never double-count them, and each stores its snippet as a receipt — labelled “found on page” in the card’s trail.
Witchcraft
The editorial measure, unchanged: raw rubric points × 10, from all verified claims — 8.0 points is a 4.0 rating on the profile and 80 Witchcraft on the card. In play it doubles as the universal tiebreaker.
The recipe for the four battle stats
- 1 · Count how many claims carry each evidence signal.
- 2 · Weigh each count by its fixed weight and sum the points.
- 3 · Saturate through round(100 × tanh(points / k)) — points = k lands near 76, double k near 96, so strong cards keep distinct shapes instead of collapsing into identical high-90s.
Each stat reads different evidence on purpose — Danger only health signals, Greed only commercial ones — so a pure money-hype card at 0 Danger is the system working, not a gap.
Attack
k = 7.5How hard the promises hit: certainty, named outcomes, transformation, deadlines.
| Evidence signal (per claim) | Pts |
|---|---|
| Absolute certainty | +1.00 |
| Specific wealth outcome | +0.75 |
| Dramatic transformation | +1.00 |
| Deadline on the miracle | +1.25 |
| Money testimonial | +0.75 |
Delusion
k = 7The mystical machinery: unmeasurable mechanisms and self-sealing logic.
| Evidence signal (per claim) | Pts |
|---|---|
| Pseudoscientific mechanism | +1.00 |
| Self-sealing logic | +1.00 |
Breadth bonus: each distinct mechanism kind adds +1.25 once — running many kinds of magic at once is its own signal.
- Quantum / energetic
- Cellular / DNA
- Ancestral / generational
- Astrology / human design
- Manifestation
- Channeling / downloads
- Subconscious codes
- Other unfalsifiable
Greed
k = 6.5Commercial aggression: what it costs, what it promises back, how hard the scarcity clock ticks.
| Evidence signal (per claim) | Pts |
|---|---|
| Specific wealth outcome | +1.00 |
| High-ticket anchoring | +0.75 |
| Value anchoring | +0.75 |
| Scarcity & urgency | +1.25 |
| Money testimonial | +0.50 |
The magnitude tiers
The largest price, promised amount, and multiplier found anywhere on the card each land in one tier, applied once — a €25,000 mastermind out-scores a €1,000 course out-scores a €66 seat.
Offer price
- ≥ 20,000+3.00
- ≥ 5,000+2.25
- ≥ 1,000+1.50
- ≥ 100+0.75
- ≥ 1+0.25
Promised amount
- ≥ 1,000,000+2.50
- ≥ 100,000+1.75
- ≥ 10,000+1.00
- ≥ 1,000+0.50
Multiplier
- ≥ 10×+1.50
- ≥ 3×+1.00
- ≥ 2×+0.50
Danger
k = 5.5Health-harm signals only — with discouraging real care carrying the heaviest weight in the whole system.
| Evidence signal (per claim) | Pts |
|---|---|
| Health, disease, cure, or medication outcome | +1.50 |
| Health claim | +1.00 |
| Medication claim | +2.00 |
| Mental-health treatment | +1.50 |
| Discourages real care | +2.50 |
| Certainty about health | +1.00 |
Medication language found on pages is judged before it counts — “she was on medication” is a mention, “she came off her medication” is a claim.
Every card stamps its tuning version (currently tuning/1.3.0), and recomputing from stored evidence is exactly deterministic — same evidence, same five numbers, byte for byte. Each stat’s trail on the card names the signals, counts, and weights behind it; rebalancing the game is an edit to one code block, never to a prompt.
The quotes
Pages are captured, dated, and hashed. Every quote is then located inside our own capture character-for-character — the AI’s version of a sentence is only ever used as a search key, never as the published text. A quote that cannot be found is discarded rather than printed. This rates public marketing claims, not people, and corrections are welcome.