The question
A Case tests one suspect against one outcome
A GutSync Case is an n-of-1 trial: a study with one participant, you. A correlation scan asks what is worth looking at. A Case asks whether one specific pattern survives a finish line set before the evidence arrives.
GutSync Cases identify food triggers by comparing your own days against each other. A Case names one suspect and one outcome, sorts every readable day into the group that had the suspect and the group that did not, divides one outcome rate by the other, and grades the result against thresholds fixed in code before the Case opened. The control is never a population average. It is your own comparable, properly logged days without the suspect.
A Case starts from a hypothesis such as “dairy worsens my bloating” or “magnesium improves my stool regularity,” then follows it for weeks until a closing rule is satisfied or the evidence runs out.
The evidence pipeline
- 01
Define
One suspect and one outcome
- 02
Sort
Readable days enter target or clean groups
- 03
Compare
Rates, ratio, and uncertainty are calculated
- 04
Conclude
A fixed finish line decides eligibility
The denominator
A day counts only if it could show the answer either way
Missing data is not good news or bad news. It is unknown, and unknown days stay out of the comparison.
Every Case reads up to the latest 90 days, so opening one does not necessarily mean starting from zero. Existing history can contribute immediately, but only where the relevant part of the day was observed.
- Food needs a meal-observed day
- A day with no meal information cannot establish whether a food was present or absent.
- A substance needs any logged day
- Medication and supplement habits are not meal-bound. On a day with other logging, a missing substance entry can represent not taken.
- Absence needs 2+ logged moments
- A thin day can prove that a logged food or symptom happened. It cannot prove that an unlogged food or symptom did not happen.
- The outcome must be observable
- The reaction window must contain outcome information. Otherwise the Case cannot tell whether a reaction happened.
An illustrative delayed-trigger ledger
Only the target and clean cells enter the two comparison groups. The exact timing depends on the suspect and outcome.
Sun
Set aside
Exposure ahead
Mon
Target
Suspect logged
Tue
Set aside
Reaction window
Wed
Set aside
Reaction window
Thu
Clean
Fully observed
Fri
Set aside
Partial log
Sat
Pending
Window still open
Symptoms are not the only possible outcome
A Case can watch loose stool (Bristol 6–7), hard stool (Bristol 1–2), stool outside the ideal 3–4 range, or a logged sign such as urgency or mucus. A fully observed day with no loose bowel movement can count as “not loose.” A thin day cannot.
Hard stool is stricter: only a logged Bristol 1–2 bowel movement counts as hard. A day with no bowel movement is not silently interpreted as constipation. Blood and unusual stool color are never attributed to a food or supplement; those stay in the safety path.
Timing
A reaction is only assigned to a plausible window
Foods and substances do not all act at the same speed. Each suspect carries a fixed timing model, drawn from published gastrointestinal physiology, that limits what it can reasonably be blamed for.
How long a suspect stays attributable, and how that decides a day's role
Each window is fixed per suspect and drawn from published physiology, not fitted to your data.
- Caffeine
6 hours
Fast gastrocolic window
- Dairy
12 hours
Fermentation window
- Gluten
48 hours
Delayed window
Exposure
The suspect is logged
Plausible reaction window
Only outcomes here can be attributed
Clean comparison resumes
After the window and washout settle
Those hours are the tail of each suspect's kernel. At the day level they resolve into whole days: fast mechanisms are scored against the same day, fermentable mechanisms carry past midnight into the next day, and the slowest class can reach up to day 2 after exposure. Stool outcomes use a longer window than symptoms, because gut transit is slower than symptom perception.
The same window protects the clean group in both directions. A target-free day is set aside if yesterday's exposure could still affect it. It is also set aside if an exposure inside its forward follow-up window could explain the later outcome. This is the washout rule: a clean-looking day cannot serve as a control when a nearby exposure contaminates the comparison.
The newest days may be marked pending for the same reason. If a reaction could still arrive tomorrow, today is not scored as “no reaction.” It enters later, after its full window has elapsed.
The arithmetic
Two rates become one risk ratio
Once the readable days are separated, the comparison is a risk ratio: how often the outcome appeared on days with the suspect, divided by how often it appeared on days without.
A simple example
Illustrative numbers, not a minimum or a verdict.
With suspect
6 symptom days
10 readable days
60%
Without suspect
3 symptom days
10 readable days
30%
Risk ratio
2.0×
twice the observed rate
A ratio of 1.0× means no measured difference. Above 1.0× the outcome was more frequent on days with the suspect; below it, less frequent.
A point estimate is not enough. A Case also reports a 95% confidence interval: the range of effect sizes compatible with the observed comparison. Five exposure days should produce a wide range. More readable days usually narrow it. That narrowing means less statistical uncertainty; it does not automatically mean causation.
GutSync Cases use published biostatistical methods here, not a model-generated score. Each rate gets a Wilson score interval, which stays well-behaved on the small samples personal tracking produces. The ratio between the two rates gets a Katz log-method interval. A Haldane–Anscombe correction keeps a zero-event group finite and conservative rather than dividing by zero.
For a helpful suspect, such as “magnesium improves my outcome,” GutSync inverts the ratio before grading it. That lets one ruleset mean “evidence for the hypothesis” in either direction.
Evidence grades
A large ratio on thin data does not earn a high grade
Possible, likely, and strong are rule-based grades. The upper two depend on the cautious end of the confidence interval, not on the headline ratio.
- Possible
- 5+ exposure days and a point estimate of at least 1.2× baseline.A lead worth watching. Chance and confounding are still credible explanations, so GutSync does not call it your trigger.
- Likely
- 10+ exposure days and the lower confidence bound above 1.0×.The whole 95% interval now sits above no-effect, which is the conventional threshold for statistical significance. Chance alone no longer explains the measured association.
- Strong
- 15+ exposure days and the lower confidence bound at or above 1.5×.The conservative reading still shows a meaningful elevation. A verdict additionally requires stability and the appropriate closing rule.
A comparison that misses all three stays ungraded. GutSync can ask for the next useful kind of evidence, but it does not round a near miss upward.
Entangled suspects
When two suspects always appear together, neither can be graded
Cheese may be both dairy and high-fat. If nearly every dairy day is also high-fat, an ordinary two-group comparison cannot tell which feature owns the signal.
The four groups needed to separate dairy from high-fat
The two “only” cells are what an entangled history never produces, and what tells the suspects apart.
| No high-fat | High-fat | |
|---|---|---|
| No dairy | Neither Control | High-fat only Separates high-fat |
| Dairy | Dairy only Separates dairy | Both Still entangled |
The useful rows are “dairy only” and “high-fat only.” Those days break the tie. GutSync calls this four-cell view the Lineup and compares each suspect-alone group with the neither group. A cell below 4 readable days is shown as a data need, not as a ratio.
Some habits never produce the separating days naturally. The observational result then stays confounded, and GutSync says so instead of assigning the credit to one suspect. A challenge can create the missing contrast by removing one suspect while the rest of the routine continues.
Designed contrast
A challenge creates evidence observation cannot
A challenge builds three periods on purpose: a baseline, a stretch without the suspect, and a deliberate re-exposure. If the outcome falls and then returns, the response has been reproduced.
Default elimination-and-reintroduction sequence
- 01
Baseline
28-day lookback
Normal routine
- 02
Eliminate
14 effective days
Target absent
- 03
Reintroduce
3 days
Target deliberately returns
- 04
Washout
3 days
Let the window settle
The control is the trailing 28 days. It needs at least 7 observed meal days and 3 days with the target. If the history is thinner, the honest next step is more observation, not a pretend baseline.
The default elimination goal is 14 effective days. “Effective” means the day was observed, target-free, and clear of washout. A slip does not erase prior work; it only voids that day. No challenge verdict can be read below 7 effective elimination days.
- Response reproduced
- The elimination rate falls to 0.7× baseline or lower, then reintroduction produces a reaction inside the target's window or a rate at least 2.0× the elimination rate.
- Response not reproduced
- Elimination remains at 0.9× baseline or higher, reintroduction stays quiet, and the full washout settles without a delayed reaction still en route.
- Challenge inconclusive
- The phases disagree or too many days are unreadable. A mixed first round can use one more reintroduction after washout, up to 2 rounds total.
Testing something that may help
A discontinuation trial mirrors the same protocol: pause the habit, watch, then resume it. The pause must worsen the outcome by at least 1.43×, and resuming must reduce it to 0.5× the pause rate or lower. The target must have been present on at least 50% of logged baseline days; pausing a rare habit cannot make a readable contrast.
The finish line
A Case closes only when a predeclared rule is satisfied
GutSync writes the closing rules before a Case opens, and eligibility is decided by code. An interesting pattern cannot be argued across the line.
- Supported association
- The result reaches strong with episode stability, or holds likely for 7 straight days with episode stability and no unresolved confounder. The signal must survive when any single exposure episode is left out.Strong results disclose unresolved confounders; they do not hide them.
- Not supported in this window
- At least 10 exposure days, a point estimate at or below 0.8× baseline, and an upper confidence bound below 1.2×.This clears the suspect only for the measured outcome and window; it is not a universal claim that the item can never matter.
- Inconclusive
- Neither ending is eligible by day 60, or the user closes the Case early.Inconclusive is not a failed Case. It records exactly what the available evidence could and could not establish.
Resolved Cases keep the evidence-engine version that produced their conclusion. The current ruleset is version 1. If closing logic changes later, an old result does not silently claim it was made under the new rules.
What the result means
Useful evidence still has limits
GutSync Cases make personal tracking more rigorous. They do not make it clinical. Here is where the method stops.
- Self-reported input
- A Case can only read what was logged. Changes in logging habits can change the apparent pattern.
- Unmeasured confounding
- Sleep, stress, illness, other foods, and chance can move with a suspect even when the known Lineup looks clean.
- One person, one window
- The result describes your measured history. It does not estimate what will happen to everyone, or even to you forever.
- Association is not diagnosis
- Even strong personal evidence does not diagnose an allergy, intolerance, infection, or gastrointestinal condition.
Common questions
The short answers
The questions people ask before they trust a result, answered without hedging.
- Are GutSync Cases based on AI opinions?
- No. AI helps turn plain-language logs into structured entries and can explain a result. Counted days, risk ratios, confidence intervals, evidence grades, and verdict eligibility are computed by a deterministic ruleset. The AI cannot promote a result or issue a verdict that the rules do not allow.
- Does GutSync use real statistics?
- Yes. GutSync Cases compute a risk ratio with a 95% confidence interval using published biostatistical methods: Wilson score intervals for each rate, the Katz log method for the ratio between them, and a Haldane–Anscombe correction for zero-event groups. The likely and strong grades depend on the lower bound of that interval rather than the headline ratio, so a large number measured across a handful of days does not grade highly.
- How does GutSync handle delayed food reactions?
- Every suspect carries a fixed timing window drawn from published gastrointestinal physiology. Caffeine is scored against the same day. Fermentable foods such as dairy carry past midnight into the next day. The slowest foods, gluten among them, can be scored up to day 2 after exposure. GutSync attributes an outcome only when it falls inside that suspect's window, and it sets aside a clean-looking day when a nearby exposure could still be reaching it.
- What happens if I miss a day of logging?
- An unobserved day is left out. It is not treated as symptom-free or target-free, and it does not count against the hypothesis. The Case waits for enough readable evidence instead of turning missing data into a negative result.
- What happens when two suspects always appear together?
- GutSync sorts the days into four groups: neither suspect, each one alone, and both. The two alone groups are what tells them apart. When a routine never produces at least 4 readable days in one of those groups, GutSync reports it as missing evidence instead of a ratio, and the observational result stays confounded until a challenge creates the contrast.
- Can GutSync prove that a food causes symptoms?
- An observational Case measures an association within one person's logged history; it does not by itself prove cause. A planned elimination and reintroduction can provide stronger within-person evidence when the response disappears and then returns, but it still is not a diagnosis or a substitute for medical care.
- Can a Case track bowel movements instead of symptoms?
- Yes. A Case can watch loose stool (Bristol 6–7), hard stool (Bristol 1–2), stool outside the ideal 3–4 range, or signs such as urgency or mucus. Blood and unusual stool color stay outside food attribution and in the safety path.
- Can I use a Case to stop a medication or test an allergen?
- No. Medication Cases are observational. Do not stop or restart prescribed medication based on GutSync, and do not deliberately challenge a known or suspected allergen. Medication changes and allergy testing belong with a qualified clinician.