Every wellness app on earth will tell you to sleep more, walk more, drink water. Not one of them can tell you whether any of that works on you. The Repair Bay is the missing half of a maintenance program: the record of which repairs hold — and which well-meant ones quietly cost you.
The first is the flight log — where it went, how hard it flew, what it hit. Hindsight Daily has kept that book for you since day one: damage, crack growth, the inspection interval.
The second book is the maintenance record: every repair ever made, and — this is the part that matters — whether it held. A fleet that keeps the first book without the second fixes the same part the same wrong way forever.
Mechanics even have a name for the dark side of that book: hangar rash — the dents the maintenance itself puts in. The ladder scraped against the wing. The panel over-torqued until the thread stripped. Care, sincerely applied, that reads afterward as damage. People do this constantly: the supplement stack, the punishing "recovery" regimen, the self-improvement project that leaves you worse — maintenance-induced damage, MID for short.
The Repair Bay is the second book. One tag on repairs you already log — what it was, its dose, and whether you did it to recover. Nothing new is asked of you at night.
What qualifies for the book? One filter, no exceptions: can you put it on a calendar? A walk can go on a calendar. "Being more positive" cannot — that's an instrument reading, and it stays on the gauges. And every procedure must state a dose. "Water" is not a procedure. "16 oz before noon, daily" is. If you can't say the dose, you don't have a procedure yet — you have a hope.
Here is the trap every habit tracker falls into. Good days come in runs. If Tuesday was good, Wednesday tends to be good — not because of anything you did Wednesday morning, but because moods have momentum, the way a knocked marble takes time to settle. Score your morning walk on raw feel-good numbers and the walk gets credit for Tuesday's momentum. That's not measurement; that's flattery.
So the Repair Bay refuses to score the day. It scores the surprise. The Modal Survey already fitted how your days settle — how much of today is predictable from yesterday alone. The engine subtracts that prediction first, and a procedure earns credit only for the part left over: the part momentum can't explain.
Two more honesty rules, learned the hard way on the founder's own thirteen-year record:
Your hard year is not on trial. Days on a procedure are compared only against days from the same era of your record — within about three months. Compare a practice from your worst year against days from your best one and the verdict grades the year, not the procedure. The engine doesn't allow it.
The timing is named in plain words. An effect is checked on the days you did it, the day after, three days after, and across the week after — and the verdict tells you which. One pattern gets flagged automatically: an effect that shows only same-day and vanishes by morning may be something you do on good days, not something that makes days good. The flag comes with its remedy — run a coupon test.
And seasons get their own reading. Run the same study in different years and it pools into one all-time number — but underneath, the bay keeps a by-period breakdown: each stretch of practice (split wherever there's a 90-day gap) scored on its own, against its own era. When two periods flatly disagree, the verdict says so — periods disagree — because an average of a good season and a bad one is an average of different things. That disagreement is the seasonal signal, delivered instead of hidden.
Verdicts wait for five occurrences. Below that, the bay says INSUFFICIENT RECORD and refuses to guess.
The bay keeps exactly three lists, each with a hard cap:
CERTIFIED — 3 slots. It earned its place: dose stated, interval stated, a positive read in your record, and a survived coupon test. These are your working repair manual — the 𝑖SRM entries written by your own data.
GROUNDED — 3 slots. Things you did to recover that your record reads negative. The hangar rash list. Both halves of the flag are required — negative association and recovery intent — because a vice reading badly is not news; a remedy reading badly is.
IN TEST — 5 slots. Candidates, ranked, waiting on evidence.
A full list never silently takes a fourth item — it asks who gets displaced, every time. The caps are the whole design: a repair manual with forty procedures is a manual nobody opens. Nothing is lost to the caps, though — everything ever tried keeps its verdict in the Service History archive, including the "inert" list: things the record can't tell from zero.
Bars are the effect in W (your 0–10 daily score); whiskers are the 95% range. A whisker crossing zero means the record isn't sure yet. Illustrative values — your bay draws this from your own record.
That chart — the effect ladder — sits at the top of the tab under THE BRIEF: your top three that read good, your top three reading as maintenance-induced damage. Size and certainty, one glance, no digging. A toggle flips the Brief between the lists (curated) and the whole record — every behavior ever scored, printed with its caveat, because unvetted extremes are mostly timing artifacts. Rankings go to certainty first: a large, blurry effect loses to a smaller, sharp one, and anything prescribed prints set aside — a finding for the prescriber, never ranked.
When engineers need to know what a metal can take, they don't philosophize about the wing. They cut a small strip from the same sheet — a coupon — and break it on a bench, on purpose, under known loads. Cheap, controlled, and the answer belongs to that material.
The bay's coupon test is the same move, run on you, by you. You declare it before the run: the procedure, the dose, which direction you expect, the schedule — at least 14 days on and 14 off, or alternating blocks — and one sentence in your own words: the falsifier. What result would make you drop this?
And it doesn't assume you were perfect. Every ON day sits on the coupon card as a dot. Missed one? Tap it — ideally within the week, while memory is honest. Missed days are excluded from the comparison, not counted as rest days, and adherence prints beside the verdict; a low-adherence run says so on its face instead of quietly certifying a dose you never took. Nothing is asked nightly — the schedule is the record, and you only touch it when reality deviated.
At close, the bay reports the comparison you declared — and nothing else. No shopping through the data afterward for a window that flatters. You wrote the test before you knew the answer; that's what makes the answer worth having.
Survive the coupon and the certification gate opens. Fail it, and the disposition is one honest click — recorded, reversible, and off the list.
One rule in the bay is a program rule, not a design taste: results recompute only when your inspection interval comes due — the same date the Trends panel already shows. Between reads, the lists sit frozen under NEXT READ DUE IN n DAYS.
Why? Because continuous self-inspection couples with the thing it inspects. Check the gauge hourly and you stop flying the airplane and start flying the gauge. Aircraft engineers call the failure mode flutter; it has a task number in this program. So inspection happens at the interval, and not otherwise.
You can read early — the bay won't stop you. It will tell you plainly what the habit costs, and it will write the early read into the log. Your record, your call, on the record.
Every verdict reads associated with, in your record. One person's observational data can rank candidates and retire failures; it cannot prove mechanisms. The candidate list runs a false-discovery correction so a list of maybes can't masquerade as a list of findings — and the honest path from "associated" toward "trust it" runs through a declared coupon test.
Units you can feel, on your own scale, or it doesn't print.
Medications that appear in your history are recorded as findings — the bay never grounds, certifies, or doses them. That conversation belongs to a professional.
If what turns up involves persistent depression, crisis, or harm, that finding routes outside this program — to a person. In the US: 988, by call or text.
The Repair Bay adds no task numbers to the 𝑖SIP — it executes eight that were already on the books. That's the point: a maintenance program that needed new paperwork to record its repairs wasn't finished.
| What the bay does | Task | In the standard |
|---|---|---|
| Recompute on your logged data — the engine | i5.5.3 | 𝑖DADTA updates |
| Tag the repairs you already log | i5.5.11 | Repairs |
| Coupon tests — declare, run, close | i5.2.13 | Development tests |
| Disposition every finding — certify, ground, drop | i5.3.8 | Resolution of findings |
| Accumulate certified procedures per material | i5.5.10 | 𝑖SRM updates |
| Record the schema change | i5.5.4 | Sensing & tracking updates |
| Read at the interval, not otherwise | i5.2.10 | Feedback coupling |
| Instrumented run windows | i5.3.3 | Flight test |