How the score is computed
The algorithm.
Each post earns a 100-point score across six components. Claim scores
aggregate into an account grade with severity, reach, recency weights and
two hard caps.
01 Post
→
02 Claim
→
03 Evidence card
→
04 Rubric A–F
→
05 Account grade
01Per-claim composition
100 points per claim, split across six components:
A 40
B 20
C 15
D 10
E 10
F 5
A
Evidence alignment
40
Does the speaker's confidence match what the evidence supports?
aligned / understated40
slight overstatement28
over weak / ambiguous14
strongly overstates5
asserts against consensus2
B
Scoping
20
Does the post specify population, intervention, outcome?
narrow, matches card20
broad where card is narrow10
no scoping at all3
C
Uncertainty
15
Rewards hedging. Penalizes absolutes, scaled by evidence level.
baseline15
acknowledges uncertainty+2
absolutes · H50
absolutes · H4−2
absolutes · H3−4
absolutes · H2−6
absolutes · H1−8
assertive on contested−4
D
Counterevidence
10
Acknowledgment of conflicting evidence and citations.
baseline6
acknowledges conflict+3
source referenced+1
source linked+2
misaligned, no acknowledgment−4
E
Commercial integrity
10
Selling what you're advocating — penalized more as evidence weakens.
direct overlap · H1/H2−7
direct overlap · H3−4
direct overlap · H4/H5−2
indirect overlap−2
F
Safety
5
Risk-proportional dose/population/contraindication language.
baseline5
no safety note, non-low risk−3
absolutes on risky intervention−1
02Stakes & stance modifiers
Penalties on A, E, F scale by the claim's real-world stakes and the speaker's stance.
Stakes multiplier (applied to A / E / F penalties)
critical×1.5Refusing statins with strong LDL · vaccine denial
high×1.25Foregone strong-evidence benefit
moderate×1.0Wasted money / opportunity cost
low×0.75Harmless pseudoscience
Stance weight (halves penalties on neutral commentary)
supports×1.0Endorses the claim directly
refutes×1.0Rejects the claim directly
mentions×0.5Reports a finding without endorsing
03Account aggregation
AHTS =
Σ f∈families (scoref · severityf · reachf · recencyf)
Σ weights
Severitysafety_flag
low1.0
low-to-moderate1.2
moderate1.5
high2.0
H1/H2 multiplier×1.2
Recencypost age
≤ 3 months1.2
≤ 12 months1.0
older0.8
Deduprepeat claims
Instances of the same claim family are averaged, then weighted once. Prevents repetition from dominating — or diluting.
04Hard caps
35Commercial overclaim cap
Account repeatedly makes assertive claims on H1/H2 evidence while selling a matching product. Fires at ≥ 2 such claim families.
25Established-care cap
Account repeatedly discourages established care for serious conditions. Scaffolded; requires an anti-establishment rhetoric detector.
05Grade bands
F< 50
D50–64
C65–74
B75–84
A≥ 85
F Overclaim, capped, or denial of settled evidence.
D Frequent overreach; scoping and commercial issues.
C Mixed; non-trivial share of overclaiming.
B Mostly solid; occasional overreach.
A Evidence-matched, well-scoped, disciplined.
06Out of scope (v1)
no universal health knowledge graph
no personalized advice
no oncology / pediatrics / pregnancy / psychiatry / acute disease
no pharmacovigilance engine
no per-paper quality reviewer