An aircraft does not wait for the calendar when its g-meter shows an overload. The overloads are counted, enough of them call an inspection, the work is done in the depot, and the aircraft is timed back into service. This page reads one person's nightly number the same way — and shows the three gauges that came out of it.
Hard days stack up, and most of us only notice a bad stretch once we are two weeks into it.
The same day knocks us down every week, and it feels like a personal failing instead of a repeating load.
Inside a dip it feels like it will never end — and afterwards there is no way to tell whether the break we took actually helped.
Hindsight Daily asks for one number a night: how did the structure hold up, 0 to 10. We call it W. The founder of iSight Labs has logged it for seven and a half years. Version 5.0 of the Daily adds three gauges that work on those three problems, using nothing but that number. This page is where they came from — his own record, shown as counts and averages — and exactly how far the evidence goes.
The honest place to start is with a failed test. We asked whether the past month could forecast the next one. It could not. The average W over the past 30 days barely predicts the next 30 (a correlation of 0.22, where 1.0 would be a perfect forecast and 0 is none). How swingy the past month was adds nothing (0.06).
So the Daily makes no month-ahead claims. But an aircraft's maintenance program does not forecast either. It counts what has already happened and acts on the count. Read that way — as a load log, not a crystal ball — the same record has three things in it that are solid enough to build on, and they are all a few days out, not a month.
Call any day with W of 4 or below a hard day — the human version of an hard-day count (over-g), a load past the normal limit. Over the whole record, 27% of days are hard. If hard days landed at random, yesterday would tell you nothing about today. They do not land at random.
Look back five days and count how many were hard. The more there were, the likelier the next day is hard too.
After five clean days, about one day in nine is hard. After three hard days in five, it is more than four in ten. After five in a row, two in three. Only stretches where all five days were logged are counted (2,479 days). The rust bars are where the counter trips.
That is exactly how an hard-day count (over-g counter) works: one overload is noted, a cluster calls an inspection. The Daily's rule is three hard days in the last five. When that happens it raises one heads-up and offers one thing — a planned break (depot block), which is a planned break with a named job, on the calendar inside 48 hours. A trip does not mean something is wrong with you. It means a chain has started, and it is easier to break in the next two days than in the next two weeks.
Think of a smoke alarm you have already acknowledged. It does not need to keep sounding for the same smoke. Without a rule like that, a ten-day hard stretch would trip the counter on day 3, again on day 4, again on day 5 — the same news every day. So after one trip the counter goes quiet: the count still shows, but no new heads-up is raised until 14 days have passed without three hard days in five.
With that rule, the counter would have tripped 38 times in seven and a half years — three to seven times a year, roughly once every two months. Rare enough to stay meaningful.
The line is a 30-day average, so it moves slowly; it ranges from 2.7 to 8.9. Each rust mark is a day the counter would have tripped.
In aircraft work the ground-air-ground cycle — sit on the ground, take off, fly, land — is often the single most wearing cycle an airframe sees, because it swings through the whole load range once per flight. This record has one too, and it is the working week.
W climbs from Monday to Friday, holds on Saturday, and dips again on Sunday into Monday. Monday is the landing load.
This is not one bad year showing up in an average. Monday comes in lower than Friday in every full year of the record:
| Year | Monday | Friday | Drop |
|---|---|---|---|
| 2019 (Apr–Sep only) | 6.48 | 6.29 | +0.19 — Monday higher |
| 2020 | 4.79 | 5.82 | 1.04 |
| 2021 | 5.16 | 5.69 | 0.54 |
| 2022 | 5.65 | 6.90 | 1.25 |
| 2023 | 6.12 | 6.63 | 0.52 |
| 2024 | 4.89 | 6.77 | 1.88 |
| 2025 | 4.90 | 6.67 | 1.77 |
| 2026 | 5.92 | 7.57 | 1.65 |
Across 355 weeks with both days logged, the average fall from Friday to the following Monday is 1.16 points, and 60% of weeks have one.
The first gauge works on a chain once it has started. This one works on what may be starting it. If Monday is the load that begins most chains, then a softer Monday is preventive maintenance. The Daily's weekly-cycle panel has one optional button that sets up the experiment: two weeks of softer Mondays, two normal, then both again. You say what "soft" means, and what result would make you drop the idea. An average shows a pattern, not a cause — the test is how you find out whether changing Monday changes anything.
A dip starts on the first hard day after a day that was not hard. Recovery time after a hard stretch (Return to service) is how many days it takes for W to get back to the typical value — here 7, the middle value of the whole record. There were 273 dips that could be timed. Half were back within 3 days; the average was 4.1; 19% took a week or more.
2026 so far is the fastest year in the record. A dip that ran into an unlogged day is left out, so a gap can never make the number look better.
This gauge describes; it does not predict. That is its strength. It is the number a planned break (depot block) should be judged by: if a named job works, dips that got it should come back faster than dips that did not. The Daily marks them and shows the two side by side.
The Daily's Bounce-back check (Modal Survey) already reports a bounce-back time (ring-down) — about 1.8 days on this record. The two are related and they are not the same. Bounce-back time is a fitted estimate of how fast an ordinary knock fades, read from every pair of days. Recovery time after a hard stretch (Return to service) is counted: real hard days, and the real days until W was all the way back at 7. Bounce-back time (Ring-down) describes the spring. Recovery time after a hard stretch (Return to service) times the actual recoveries.
Each gauge does one job. Together they close a loop the Daily did not have before.
The Hard-day count (Over-G counter) sees the chain start and raises one heads-up.
A planned break (depot block) puts one planned break with a named job on the calendar. It is logged as a repair action in the Daily's Repair Bay.
Recovery time after a hard stretch (Return to service) shows whether dips that got the job came back faster than dips that did not.
Your weekly rhythm (weekly cycle) sits beside the loop, working on the cause instead of the symptom. That is the difference between a logbook and a maintenance program: the logbook records what happened; the program says when planned work is due, and afterwards whether it helped.
| Piece | Where it lives · what it asks of you |
|---|---|
| Hard-day watch | Today — one folding panel: hard days in the last five, and your usual time back. Automatic — it reads the number you already log. |
| Hard-day count (Over-G counter) i5.5.2.1 | Trends — the chain chart from your own record, the quiet period, the trip history. Automatic; the heads-up can be switched off. |
| Your weekly rhythm (weekly cycle) i5.5.1.1 | Trends — weekday bars and the Friday-to-Monday drop. Automatic; the soft-Monday test is optional. |
| Recovery time after a hard stretch (Return to service) i5.5.11.1 | Trends — the trend, the latest dips, with and without a planned break (depot block). Automatic. |
| Planned break (Depot block) i5.5.11 | Offered by the heads-up; listed with the trips. Optional — plan it, say "not now", or ignore it. It never asks twice for the same stretch. |
Your own numbers replace the founder's wherever the app shows them: the chain chart appears once your record reaches about 200 days, the counter switches on after five logged days in a row, and recovery time after a hard stretch (return to service) waits for 30. A planned planned break (depot block) is counted as held once its day has passed, unless you say it didn't happen — there is no nightly question. Every threshold on this page is a setting on the Daily's Data tab.
The data. One person's nightly W, 2,494 logged days from 19 April 2019 to 7 October 2026 — 91.4% of calendar days, with gaps in late 2019 and mid-2021. Only W is used. Nothing on this page comes from notes, and no individual day is shown.
Hard day. W at or below 4. Typical W. The median of the record, 7.
Chains. For every day whose five previous days were all logged, count the hard days among those five, then record whether the day itself was hard. Trips. A day trips the counter when at least three of the last five logged days, counting that day, were hard, and no such day occurred in the 14 days before it. Trips by year — 2019: 3 · 2020: 4 · 2021: 4 · 2022: 5 · 2023: 3 · 2024: 6 · 2025: 7 · 2026: 6 (2019 and 2026 are partial years).
The week. Average W by weekday over all logged days. The Monday drop is the previous Friday's W minus Monday's, for each week with both logged.
Recovery time after a hard stretch (Return to service). From the first hard day after a logged day that was not hard, count days until W is at or above 7. Dips that reach an unlogged day first, or run past 60 days, are not timed.
The forecast check. For each week from October 2021, compare the average W of the previous 30 days with the average of the next 30 (253 windows).