Chapter 5 of 12 · 4 min read
How the Score Actually Works
The complete deterministic pipeline, component caps, semantic floor, labels, and confidence rules.
Inside ~Cortisol Checker~ · v1.0 · Moses Sam Paul
Quick Answer
The checker matches a published set of language signals and contextual frames, combines five bounded components, applies a safety-aware semantic floor when required, and returns the result with its evidence. The same text and version receive the same content score for every reader.
The 0.4.1 score is deterministic. It does not call a language model. Given the same submitted text and analysis-engine version, it follows the same ordered rules.
1. Detect evidence
Lexical rules identify each lexical signal through a bounded phrase or regular-expression variant. Context rules match broader semantic frames such as workload pressure, public harm, social humiliation, doom certainty, evidence, or recovery intent. The response retains the matched phrase, category, weight, description, frame adjustment, and evidence.
2. Build the components
| Component | Rule | Bounds |
|---|---|---|
| Lexical trigger | Keep the strongest positive signal in each category, multiply each retained weight by 0.55, sum, and round. | 0–28 |
| Context threat | Add applicable threat-frame adjustments and one boost for each detected threat-related category. | 0–25 |
| Social-evaluative threat | Add social-evaluation and humiliation frames plus unique shame and moral-shock category boosts. | 0–25 |
| Outrage mechanics | Add doom, conspiracy, and outrage frames; add category boosts; add 5 for repeated exclamation or four-plus capital letters. | 0–20 |
| Calming regulation | Add negative regulation signals and recovery, nuance, or evidence frames. | -15–0 |
Using the strongest signal per category prevents several phrases expressing the same kind of evidence from inflating the lexical score without limit. Category boosts also apply once per detected category.
3. Combine and clamp
preliminary = clamp(
lexical
+ context threat
+ social-evaluative threat
+ outrage mechanics
+ calming regulation,
0,
100
)
4. Apply the semantic floor
Semantic frames are built independently from matched signals and context. If public harm, violence/harm, threat to safety, or medical distress is high, a positive preliminary result below 41 is raised to 41. Zero stays zero. The floor prevents strongly evidenced harm language from appearing in a lower label merely because the additive components undercounted it; the audit reports whether the floor changed the number.
These components are transparent engineering rules, not coefficients estimated from a physiological dataset. Research shows that natural language can correlate with psychological and biological processes (Slatcher et al., 2017), but that study does not establish these categories, weights, caps, or labels.
5. Label, assess outrage, and state confidence
Final labels are 0–20 Calm / neutral, 21–40 Mild activation, 41–60 Emotional load, 61–80 High stress load, and 81–100 Rage-bait / doom-bait. Outrage is separately labeled low below 8, moderate from 8, and high from 15.
Confidence describes evidence density. Six or more matches including at least two context frames produces high confidence. Two or more matches, or text of at least 20 words, produces medium confidence. Otherwise it is low. Even “high” confidence does not mean confidence about a reader's physiology.
Review the current signal library
The signal library is intentionally open to criticism. The table below exposes the primary 0.4.1 phrases and weights in a human-readable form. Regular-expression variants exist for spelling and grammatical forms; safety-sensitive reply routing remains a separate mechanism.
Open the v0.4.1 phrases and weights
| Category | Primary phrases and weights |
|---|---|
| Acute stress | freaking out 15 |
| Fear language | panic 16; afraid 10; terrified 14 |
| Catastrophizing | spiraling 14; never recover 12 |
| Rumination | can't stop thinking 13 |
| Overload | overwhelmed 14; burned out 14; too much 9; exhausted 10 |
| Urgency cues | deadline 7; urgent 8 |
| Sleep disruption | can't sleep 13; no sleep 12 |
| Somatic tension | can't breathe 14; tight chest 12; heart racing 12; heart is racing 12; tense 8 |
| Anger language | furious 11; rage 12; lied 8; traitor / traitors 10 |
| Disgust language | disgusting 10 |
| Shame language | humiliated 11; ashamed 9 |
| Moral shock | shocking 9; deeply shocking 11; unacceptable 9; insensitive 8 |
| Threat language | threat 10; fair game 12 |
| Public harm | loss of innocent 12; innocent lives 12; civilian deaths 12; innocents killed 12; civilian crew 11; civilian crew members 12; condolence 7 |
| Military force | kill 13; missiles 12; firing missiles 14; targeted to kill 15; non-lethal 6 |
| Geopolitical tension | US statement 5; strategic partner 5; Indian lives 6 |
| Self-harm distress | kill myself 16; don't want to live 16 |
| Doom certainty | delete this 8; doomed 13; it's over 12 |
| Scarcity anxiety | running out 9; not enough 7 |
| Regulation | calm -7; breathing -6; grounded -8 |
This is a versioned research artefact, not a universal dictionary of stress. Missing phrases, cultural assumptions, category choices, and weights are all legitimate subjects for review. Challenge a phrase, weight, false positive, or missed case.