Safety
Saturday is a nutrition API for endurance athletes. Bad nutrition advice can cause real harm — and in extreme cases, death. This page explains Saturday’s safety model, why it exists, and what it means for your integration.
This is not a compliance formality. Exercise-associated hyponatremia (EAH) has killed athletes at the Boston Marathon, London Marathon, Marine Corps Marathon, and multiple Ironman races. It happens when athletes overdrink and dilute their blood sodium to dangerous levels. Saturday’s safety guardrails exist to prevent this.
What can go wrong
| Condition | Cause | Consequence |
|---|
| Hyponatremia | Overdrinking, insufficient sodium | Confusion, seizures, coma, death |
| Heat stroke | Under-hydrating in heat | Organ failure, death |
| GI distress | Too many carbs too fast | Vomiting, cramping, DNF |
| Bonking | Insufficient carbohydrate | Collapse, dangerous judgment impairment |
| Rhabdomyolysis | Extreme exertion without fuel | Kidney failure |
The safety block
Every nutrition calculation response — including teasers and free responses — includes a safety object:
{
"safety": {
"max_safe_fluid_ml_per_hr": 1500,
"max_safe_sodium_mg_per_hr": 3000,
"confidence_score": 0.72,
"requires_human_review": false,
"warnings": [
"High thermal stress increases dehydration risk. Plan extra fluid access points.",
"Activities over 4 hours require careful sodium management. Monitor for signs of hyponatremia."
],
"not_instructions": true
}
}
Safety data is NEVER gated behind subscription status. A teaser response may show carb ranges instead of exact numbers, but it will always show the full safety block with all warnings, confidence scores, and safe maximums. This is non-negotiable.
Engine hard limits
Saturday enforces absolute physiological boundaries that cannot be overridden by any input combination:
| Guardrail | Exposed as | Why |
|---|
| Fluid physiological limit | max_safe_fluid_ml_per_hr | Overdrinking causes hyponatremia |
| Sodium ceiling | max_safe_sodium_mg_per_hr | Excess sodium causes GI distress and nausea |
| Minimum sodium | (enforced internally) | Prevents hyponatremia in extended efforts |
| Carb gut tolerance | (enforced internally) | Exceeding gut capacity causes vomiting |
| Duration scaling | (enforced internally) | Steady-state assumptions break after 4+ hours |
| Eating disorder guardrails | (enforced internally) | Protects vulnerable athletes |
max_safe_fluid_ml_per_hr and max_safe_sodium_mg_per_hr fields expose the hard ceilings so partners can display them. All other guardrails are enforced internally — prescriptions always respect them.
Confidence score and human review
The confidence_score (0.0-1.0) indicates how well-supported the prescription is. When requires_human_review is true, conditions are unusual enough that a human should verify the prescription before following it.
| Confidence range | Meaning | Your display recommendation |
|---|
| 0.8-1.0 | Strong personalization, well-understood conditions | No special display needed |
| 0.5-0.8 | Good estimates, some assumptions made | Show any warnings in your UI |
| 0.0-0.5 | Significant uncertainty or risk factors | Prominent warning, consider a confirmation step |
Dangerous vs. safe prescription example
Dangerous (what a naive algorithm might produce for a 4-hour marathon in 30C heat):
Fluid: 1200 mL/hr
Sodium: 200 mg/hr
Carbs: 120 g/hr
- 1200 mL/hr exceeds gastric emptying rate — athlete will drink it but not absorb it, diluting blood sodium
- 200 mg/hr sodium is far too low for hot conditions — hyponatremia risk is extreme
- 120 g/hr carbs without trained gut tolerance — GI distress guaranteed
Safe (what Saturday produces for the same conditions):
fluid_ml_per_hr: 700 (capped by physiological limit)
sodium_mg_per_hr: 800 (elevated for salty sweater in heat)
carb_g_per_hr: 72 (limited by demonstrated tolerance)
safety.max_safe_fluid_ml_per_hr: 1500
safety.max_safe_sodium_mg_per_hr: 3000
safety.confidence_score: 0.45
safety.requires_human_review: true
safety.warnings: ["High thermal stress. Plan extra fluid access points and shade."]
safety.not_instructions: true
The not_instructions field
For AI agents consuming Saturday’s API (AI-to-AI communication), responses include:
{
"not_instructions": true
}
Display requirements for partners
Required
- Always show safety warnings when warnings are present or
requires_human_review is true
- Never hide safety data behind expandable sections or “advanced” toggles
- Never strip safety metadata from responses before displaying to users
- Include the disclaimer that prescriptions are guidance, not medical advice
Recommended
- Show warnings before or alongside the prescription numbers, not hidden in details
- Use visual hierarchy (color, icons, positioning) to make warnings visible
- When
requires_human_review is true, consider requiring user acknowledgment before proceeding
Prohibited
- Don’t filter warnings based on your own risk assessment
- Don’t apply your own safety logic on top of Saturday’s — this creates conflicting advice
- Don’t present prescriptions without any safety context
- Don’t use Saturday’s numbers as automated triggers (e.g., auto-ordering hydration packs)
Contraindications
In rare cases, Saturday may flag a contraindication — a reason the calculation should be treated with extreme caution:
{
"contraindications": [
{
"code": "extreme_heat_advisory",
"message": "Calculated conditions exceed safe exercise thresholds. Consult a physician before proceeding.",
"severity": "critical"
}
]
}
Eating disorder sensitivity
When an athlete has the eating_disorder_flag set, Saturday:
- Avoids calorie-focused language
- Removes weight/restriction framing from rationale text
- Applies conservative minimums (never under-fuels)
- Adjusts AI coaching conversation tone
This flag is a backend safety signal. Never expose it in partner UIs. Partners set it based on their own knowledge of the athlete — Saturday uses it to protect that athlete across all interactions.
