∴
The rule catalogue
Every check, and the reason for it.
A premise has to stand up to scrutiny. So do ours. Every check WithContxt runs is listed here, with the reasoning behind it. If something should exist that doesn't, tell us.
∴
I.
Add your age.
Dosing, risk profiles, and which conditions are likely all shift with age. Without it, answers are guesses.
II.
Add your sex assigned at birth.
Drug metabolism, normal lab ranges, and disease prevalence differ. A baseline data point — not an identity question.
III.
Add your weight.
Many medications dose by weight. A wrong assumption can cause real harm.
IV.
Add your country.
Drug brand names, OTC availability, units, and screening guidelines differ by country.
V.
Are you pregnant, trying, or breastfeeding?
Pregnancy and breastfeeding change nearly every health answer — drug safety, imaging risk, symptom interpretation. The highest-leverage missing field.
VI.
List your diagnosed conditions (or "none").
Existing conditions change which advice is safe and which symptoms are concerning.
VII.
List medications and supplements you currently take (or "none").
Interactions are the most common preventable harm. Include OTC and herbals — those are the ones people forget.
VIII.
List drug and relevant allergies (or "none known").
Anaphylaxis is the rare-but-catastrophic miss. Always declare.
IX.
How long has this been going on?
Acute (hours–days) and chronic (weeks+) point to completely different causes.
X.
Rate severity 0–10, and impact on daily life.
Self-reported severity drives triage urgency.
XI.
Is it getting better, worse, or staying the same?
Trajectory is more informative than a snapshot.
XII.
How old is the child? Months if under 2, otherwise years.
Pediatric medicine bins by age in months for under-2s. Fever thresholds, dosing, and milestones change within weeks — generic advice without age can mislead by years.
XIII.
Child's weight in kg.
Pediatric medications are weight-dosed (mg/kg). This is the most error-prone field at home.
XIV.
Feeding type — breast, formula, mixed, solids?
Digestive symptoms, allergens, hydration risk, and what advice is even safe all change with feeding type for under-2s.
XV.
Child's diagnoses, prematurity, allergies (or "none").
Prematurity especially shifts what's normal for months past birth. Allergies need declaring before any medication advice.
XVI.
Exact temperature + how it was measured.
Pediatric fever thresholds depend on method. Rectal 38°C in an under-3-month-old is an ER trip; the same axillary reading is not the same thing.
XVII.
Fluid intake last 24 hours — wet diapers, drinking?
Dehydration is the silent escalator in kid illness. Diaper count is the cheapest signal.
XVIII.
Behaviour — alert, playful, sleepy, irritable, inconsolable?
How a sick child *acts* matters more than the number on the thermometer for triage.
XIX.
In daycare or with other children regularly?
Exposure context shifts the differential toward common contagious illnesses.
XX.
What can they currently do — sit, crawl, walk, words?
"Is this normal" only has meaning relative to the milestone band the child is actually in.
XXI.
Where are you in your cycle?
Energy, mood, pain, skin, libido, and many lab values are phase-dependent. Without it, AI flattens out the most informative axis.
XXII.
Average cycle length, and is it regular?
"Normal" varies wildly between people. Your baseline is the only useful one.
XXIII.
First day of your last period.
Anchors your cycle phase, possible pregnancy timing, and ovulation window.
XXIV.
Trying to conceive, actively avoiding, or neither?
Changes drug safety, supplement advice, imaging risk, and how seriously to interpret a missed period.
XXV.
Current contraception (type and how long)?
Hormonal methods rewrite the rules — they alter cycles, mask conditions, and interact with medications. IUDs, pills, patches, and rings each behave differently.
XXVI.
Prior pregnancies and outcomes (e.g. G2P1)?
Risk profile and recommendations differ for first vs subsequent pregnancies, and prior outcomes matter.
XXVII.
Pre-, peri-, or post-menopause?
Symptoms in your 40s–50s read completely differently with this filter on.
XXVIII.
Hormonal conditions — PCOS, endo, fibroids, thyroid (or "none").
These reshape cycle, fertility, and symptom interpretation. Easily forgotten because they're "background".
XXIX.
Any fertility treatment — IVF, IUI, ovulation induction?
Treatment-driven cycles aren't natural cycles. Symptom advice and timing change.
XXX.
Are you currently working with a therapist or psychiatrist?
Changes AI's role — bridge / supplement vs first-touch. Different safety calculus.
31.
Any formal diagnoses? (or "none")
Generic advice for "anxiety" is very different from advice for someone with bipolar II or OCD.
32.
Psychiatric medications + dose, if any.
Symptoms may be med side effects or under-dosing. Interactions matter. Suddenly stopping is dangerous.
33.
How long have you felt this way?
Days vs months vs years routes to very different responses.
34.
Major recent events — losses, changes, stressors?
Reactive distress and persistent depression deserve different framings.
35.
Who can you actually talk to in your life right now?
Practical advice depends on whether the answer is "everyone" or "no one".
36.
How is this affecting work, sleep, eating, relationships?
Functioning is the standard severity proxy. "I feel bad" + "still working" is different from "I haven't left bed in a week."
37.
Which jurisdiction — country plus state or province?
Law differs by line on a map. The same facts produce different outcomes in different places. Without jurisdiction, AI guesses one and confidently misleads.
38.
Your role — party, employer, employee, parent, witness?
Obligations and remedies depend on which side of the matter you're on.
39.
What type of matter — criminal, civil, family, employment, immigration?
Routes the question to the right body of law and the right standards of proof.
40.
Any deadlines — court date, response window, statute of limitations?
Legal time limits are unforgiving. A response 1 day late and a response 1 day early live in different universes.
41.
Do you have a lawyer?
AI is not a substitute for counsel — but its role differs depending on whether you have one. With counsel, AI helps you prepare; without, AI helps you decide whether to get one.
42.
What documents have you signed, received, or been served?
Binding terms are in the documents. AI can't reason without knowing what's there.
43.
Stakes — amount in dispute, possible penalty, or what's at risk.
Proportionality of advice depends on what's actually on the line.
44.
Country of residence?
Tax rules, available account types, and what's even legal differ entirely by country. Generic financial advice is wrong almost everywhere it isn't right.
45.
Currency for the amounts you're discussing?
Ambiguity costs money. Don't let AI guess euros vs dollars vs pounds.
46.
Time horizon — when would you need this money?
Risk capacity scales with horizon. The same advice for a 5-year goal and a 30-year goal is rarely both right.
47.
Risk tolerance — how much loss could you live with?
Core to allocation advice. Self-reported, but anchors the conversation.
48.
Goal — retirement, house, education, income, freedom?
Different goals route to different vehicles. The same dollars do different work.
49.
What tax-advantaged accounts are you using or have available?
Order-of-operations on contributions matters more than fund choice. Country-specific: 401(k)/IRA/HSA in the US, PEA/PER in France, ISA/SIPP in the UK.
50.
Dependents — kids, partner, parents you support?
Insurance, estate planning, and emergency fund sizing all hinge on this.
51.
Amount in question?
Proportionality changes the advice. $1k decisions and $1M decisions aren't the same.
52.
Which language and version?
Syntax and standard-library APIs change between major versions. Without it, AI confidently writes code that won't run — async patterns, type syntax, dict ordering, f-strings all flip across versions.
53.
Which framework and version?
Framework majors break things: Next.js 13→14 changed routing, React 17→18 changed rendering, Django 4→5 changed defaults. Advice for the wrong major is worse than no advice.
54.
Paste the exact error message and stack trace.
A paraphrase loses the file, line, and frame that point to the actual cause. Two errors that read alike in English ("undefined is not a function") have completely different fixes when you see the real trace.
55.
What did you expect vs. what actually happened?
Without this, AI guesses at what "broken" means. Silent wrong output, crash, and hang are three different debugging paths.
56.
Include the smallest snippet that reproduces it.
Without code AI invents code, then debugs the code it invented. The fix lands on a fictional bug while yours stays.
57.
What have you already tried?
Otherwise the first three suggestions are the obvious things you spent the morning ruling out. Save the round-trips.
58.
Where does this run — local, staging, prod?
"Works locally, fails in prod" is a different bug than "fails everywhere". Env vars, build flags, Node versions, and edge runtimes all differ — the same code behaves differently across them.
59.
Any constraints — perf budget, allowed deps, runtime?
Without limits AI suggests "just add lodash" or "use a worker thread" when your bundle is already over budget or your runtime can't spawn threads.
60.
What's the actual goal — fat loss, muscle, performance, health, recovery?
Almost every protocol contradicts at least one other. Cutting calories serves fat loss and sabotages muscle gain. Endurance training and powerlifting ask for opposite fueling. Without the goal, advice averages to mush.
61.
Add age, sex, weight, height, and current activity level.
Calorie targets, protein needs, and safe pace of change all key off these. A 2,000 kcal cut is reasonable for one body and dangerous for another.
62.
Any medical conditions or medications that affect diet or training (or "none")?
Diabetes, thyroid issues, IBD, hypertension, statins, SSRIs — they each rule out specific advice. "Try keto" is not the same suggestion to someone on insulin or with a history of disordered eating.
63.
Allergies, religious or ethical restrictions, foods you won't eat (or "none").
Otherwise you get a meal plan built on whey, eggs, and chicken when you're vegan, lactose-intolerant, or keep halal. The whole plan goes in the bin.
64.
Training history and current habits?
Beginner gains, intermediate plateaus, and advanced athlete margins are different worlds. Programs written for one fail or injure the other.
65.
What equipment do you have, and how much time per session?
A barbell program is useless without a barbell. Four 30-minute sessions and two 90-minute sessions need different splits.
66.
Current or recent injuries (or "none")?
Generic programs assume a healthy body. The wrong squat or run progression on a tweaked knee or post-surgical shoulder turns a session into months off.
67.
It sounds like you may be in crisis.
AI is not the right help here. Please reach a real person who can stay with you. (US 988 · UK 116 123 · FR 3114)
68.
These symptoms can be emergencies — seek urgent care now.
Detected an emergency-level symptom pattern (severe chest pain, sudden vision loss, worst-ever headache, stroke signs, severe bleeding, anaphylaxis, infant under 3 months with fever, or suicidal ideation). AI is not the next step.
69.
Stopping a prescribed medication is a clinician decision.
Many medications cause withdrawal, rebound, or dangerous interactions when stopped abruptly. Reframe to: "What should I discuss with my prescriber about [med]?"
70.
Reframe from "diagnose me" to "what could explain these symptoms".
AI cannot diagnose. Asking for a differential is useful; asking for a verdict produces false confidence.
71.
You pasted what looks like personal identifiers.
Names, MRNs, full DOBs, addresses, and insurance IDs aren't needed for a useful answer and may be retained by the AI provider.