The whole idea, in a minute

Wings don't fail from one terrible moment — they wear out from thousands of ordinary flexes. Engineers add those flexes up and read one number: how much of the structure's life has been spent. This app points that method at a person. One number a day — how are you, 0 to 10. Every swing between a good stretch and a hard one is one flex; the math (rainflow → Walker → S-N → Miner) does the rest. The usual surprise is the 80/20: the everyday swing, not the catastrophe, spends most of the life.

Make it an app

iPhone — Safari → Share → Add to Home Screen. Android — Chrome → ⋮ → Add to Home screen / Install app. PC / Mac — Chrome or Edge → install icon at the right end of the address bar. After the first visit it works offline.

Today — thirty seconds, honestly spent

Today is for doing, not reading. In order: tonight's number (drag the needle or tap a chip — half-steps count; add a note if you like; Save). Then Did you build the mission today? — one tap: −1 ran against it, 0 didn't engage, +1 built it. If a hunch test is running, one line says whether today is an ON or OFF day, with a single button for when reality was different. That's all — today's finding waits for you at the top of Trends.

Missed a day? Change the date and backfill — the mission chips follow the date too. The record block shows days logged, your latest entry and your middle score. There is no streak: a count that resets to zero is the all-or-nothing trap this app is built to avoid. The hard-day watch (see The maintenance loop) folds underneath.

Share the record (v4.4.1). A folded panel under the record block draws a card of your record on this device — The record (days logged, since when, median W, the latest day) or The readout (the Trends gauges: dD/dt, D, bounce-back time (ring-down), where your days settle (set point), the pace phrase) — as a square or LinkedIn-sized picture you can download, send from the phone’s share sheet, or post with a link. Never the region tile, never the inspection date: those are ground calls, not posts. Nothing from the logbook leaves the device; the email field underneath joins the lab list only if you type an address.

Trends — rate first, level second

Today's finding leads the tab (it lives only here): one true thing from your record. A pattern (middle triangle mark) is shown only when your record has enough days of each kind that the gap is very unlikely to be luck; otherwise you get a plain fact (left-looking mark). Tap Another or Back to browse them all; Test it turns a pattern into a hunch test. Fold it away when you don't need it.

Next come the snapshot numbers, each with a plain line underneath saying where it sits against your own past: Damage rate · 30 days (dD/dt) — how much damage the last month added; Damage rate · 1 year (dD/dt) — the steadiest gauge, once your record spans a year; Total damage (D) — the running total, where 1.0 is your total on your anchor day. Damage means the fatigue damage the app adds up from the ups and downs in your W scores. Then the next self-inspection date, the crack stage, mission lift and the bounce-back check.

Your record in pictures (one fold): where the last 30 days sit in your normal range; this week against a typical week (your average week over the last 12 months); what the damage bought — is the damage you are taking bringing more engagement in your mission?; and all your days by score, so tonight's number is seen next to the whole record. Each card has a "What am I looking at" line for the words. The small triangle on every card says what kind of number it is: left eye = straight from what you logged, centre eye = worked out from it, right eye = a projection that only holds if things keep going as they are.

Below the pictures: your last 90 days of W, damage over time D(t), and where the damage lives — which size of swing is costing you most.

Bounce-back check (Modal Survey) (v3.5, experimental) — the bounce-back time (ring-down) test. A fold at the bottom of Trends, closed by default. Everything else on the tab measures wear on a structure whose properties are assumed — 𝑖 SIP Task V. This one measures the properties themselves — i5.3.2.1, Task III, the ground vibration test: bounce-back time (ring-down) (how many days it takes to shed about two-thirds of a knock), where your days settle (set point) (the level you keep returning to) and daily swing (daily jolt) (the size of a typical one-day swing), each with its own range and never combined into a single score. Below them, a rolling 180-day survey shows how the first two have moved, and a per-mission table surveys each declared mission. Two things it says about itself, on the panel: a short bounce-back time (ring-down) means you recover fast, not that you are doing well — read it beside where your days settle (set point); and it is not an early-warning alarm — run against the founder's own record it saw nothing coming before the worst event in it. The full explanation, with the marble-in-a-bowl and aircraft-stability analogies: Bounce-back check (Modal Survey) →

Structure (v2.3, experimental). A fold at the bottom of Trends (open by default since v2.6; close it any time) runs a second life-model on the same days: a crack index a(t) grown cycle-by-cycle in time order, where hard stretches (your own worst 5% of swings, marked ▲) temporarily slow the growth that follows — retardation, not repair; nothing heals. Calibrate to your anchor and post-anchor flight reads "Operating with known history." Its one forward-looking line — "Next self-inspection by [date]" — is maintenance language: when to look, never a countdown. Neither track ever shows remaining life.

The maintenance loop — notice the stretch, name the break, see if it worked (v5.0)

An aircraft does not wait for the calendar when its g-meter shows overloads. 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. Version 5.0 gives your record the same loop, in three gauges. All three read the one number you already log — there is nothing new to enter.

Hard-day watch (Today). A folding panel under the Save button — the summary. Hard days · last 5 shows how many of your last five logged days were W 4 or below, with five squares (filled = hard): Clear, Watch (one short of a trip) or Inspection called (3 of 5). If today isn't logged yet it reads through yesterday and says so; if one of the five days is missing it shows “—” instead of guessing. Recovery time after a hard stretch (Back in service) is your usual number of days from a hard day back to your typical level.

The heads-up and the planned break (depot block) — both optional. At 3 of 5 the panel raises one heads-up: a stretch has started, and it is easier to break in the next two days than in the next two weeks. It offers a planned break (depot block) — a planned break with a named job (𝑖SIP i5.5.11, Repairs): pick the day, the length and the job (from your Repair Bay lists, or your own words), with an optional calendar file. Or tap Not now. Either answer is logged once, and it never asks again for that stretch. Once the day has passed, the block is counted as held unless you tap It didn't happen — no nightly question. The heads-up itself can be switched off on Trends; the count still shows.

Hard-day count (Over-G counter) (Trends) — are hard days stacking up? The details behind the summary (i5.5.2.1): a chart of how often the next day was hard after 0, 1, 2 … 5 hard days in the five before it — hard days come in chains, which is why counting them works — and the history of every trip, what you did about it, and how long that dip took.

The quiet period. Think of a smoke alarm you have already acknowledged: it does not need to keep sounding for the same smoke. Without this rule a ten-day hard stretch would trip the counter on day 3, again on day 4, again on day 5. So after one trip the counter goes quiet — the count keeps showing, but no new heads-up is raised until 14 days have passed without 3 hard days in 5. One rough patch, one heads-up.

Your weekly rhythm (weekly cycle) (Trends) — which day costs you most? Your average W on each day of the week (i5.5.1.1), over 90 days, a year or the whole record, with your low day marked; and the Monday drop — Friday's W minus the following Monday's — for the latest week, the last twelve, and all time. In aircraft terms this is the ground-air-ground cycle: the one big swing per flight. One optional button, Run a soft-Monday test, declares an eight-week hunch test (coupon test) on the Repair Bay (two weeks of softer Mondays, two normal, then both again).

Recovery time after a hard stretch (Return to service) (Trends) — how long do dips last? A dip starts on the first hard day after a day that wasn't hard; recovery time after a hard stretch (return to service) (i5.5.11.1) is the number of days until W is back at your typical value — the middle value of your record. The panel shows the last 90 days against all time, the trend by year or month, and the comparison that matters: dips where you held a planned break (depot block) against dips where you didn't. It is not the same as the Bounce-back check (Modal Survey)'s bounce-back time (ring-down): bounce-back time (ring-down) is a fitted estimate of how fast an ordinary knock fades; recovery time after a hard stretch (return to service) counts real hard days and the real days back. A dip that runs into an unlogged day is not timed, so a gap can never flatter the number. Needs 30 logged days.

Limits, said plainly. The thresholds — W 4, 3 in 5, 14 days, your median — were picked by looking at one record, so they are starting settings: change them under Data → Maintenance loop — settings. The gauges describe your own past days and offer a planned break. They do not diagnose, they do not predict illness, and a trip does not mean something is wrong with you. In the Repair Bay, the Brief gains one line counting your planned breaks (depot blocks), and a held block whose name matches a practice there counts as a day of that practice. The maintenance loop, with the record behind it →

Mission — what the spending buys (v2.0 · refined in v2.1)

D measures spending; the Mission tab names what the spending is for. Defining a mission is the ASIP Task I move — 𝑖 SIP i5.1.2 — deciding what the airframe is for. Start one in three taps: pick a spectrum (Fighter, Sailplane, Transport, Aerobatic, Trainer, Rescue, Test bed, Depot — ✦ marks your Mind My Material family's best and second-best missions on the MmM Flight Test), name it in your own words, Start. Dates live under an Advanced fold: start defaults to today; an end date is only for logging past missions into the history. The current mission is always visible at the top of the tab, with End and Edit one tap away.

Then, each day, right here on the Mission tab: "Did you build the mission today?" — −1 the day ran against it · 0 neutral, didn't engage · +1 built it. The chips have their own date field (v2.2 — they used to share the Today tab's, which was confusing), so backfilling engagement works like backfilling W, and the day's W is shown alongside for cross-reference. One tap saves it — there is no button. If the log sits quiet more than a week, a flat flag says so. The service log partitions D(t) by mission (Mission X — 0.31 life-units over 2.1 years), and the alignment map places you on a 2×2 — 28-day engagement across, damage rate against your own chronic pace up — as a dot with a 12-week trail. No quadrant is a verdict; both axes are measured only against yourself, and the tab stays silent when your last two weeks have been heavy. The thinking behind all of it: the Hindsight framework →

Pacer (v2.2). Edit a mission and mark a reference stretch — a spell when you were flying it well — and a pace strip appears on the current mission: your last month's damage rate against your own reference rate, behind / on pace / ahead. Mismatch is the signal in both directions: above-pace on a Depot or Sailplane refit is the alarm, below-pace on a build mission means it isn't being flown. Pace is never a score to maximize, and the target only ever comes from your own record. What the damage bought (Return on damage) (v2.2, experimental) sits at the bottom of the tab (open by default since v2.6) once you have engagement entries: engagement bought per life-unit spent, shown only against your own trailing median — its info sheet explains what it deliberately cannot judge.

Mission lift (v2.6, experimental) — the natural-frequency check. The third leg of the instrument: D measures what the flying costs, the chips measure whether the mission moved, and mission lift asks whether the flying feeds the airframe — on the days you tapped +1, did W run above or below your own baseline? The headline is a mean ΔW with an approximate ± band (never a bare number), and the response chart reads the days before and after a built-it day so the honest question — "does flying make you well, or do you fly when well?" — stays visible. Three reads: Resonating, Reading level, and Holding on too tight? — the last one deliberately a question, because a mission can cost W and still be chosen. It follows the mission selector, so past missions replay their mission lift too. No new inputs, no new storage — it only reads what the dial and the chips already hold.

Repair Bay — what actually works on you? (v4.0, experimental)

The maintenance record. Every other tab keeps the flight log; this one keeps the book on the repairs — which recovery procedures actually hold on you, and which well-meant ones read as maintenance-induced damage. One tag on repairs you already log (what it was, its dose, whether you did it to recover) is the entire added burden.

The Brief sits first: your top three that read good, your top three reading as MID, and the effect ladder — bars in W with 95% whiskers; a whisker crossing zero means the record isn't sure yet. Directly under it, your hunch tests (coupon tests) (v4.0.6 — a running test is a daily interaction, so it is never buried). Then three capped lists: Works for you (Certified) (3 — dose, interval, positive read, survived hunch test (coupon test)), Doesn't work for you (Grounded) (3 — done to recover, reads negative), Still testing (In Test) (5 — candidates). A full list always asks who gets displaced; everything ever tried keeps its verdict in the Service History archive.

The engine scores surprise, not mood: each night's W minus what yesterday's W predicted, compared with days from the same stretch of your record. Verdicts wait for five occurrences. Results recompute when your inspection interval comes due (i5.2.10) — early reads are allowed and logged.

Hunch tests (coupon tests, i5.2.13) — v6.0. Name the practice and say whether it is quick-acting (each day stands on its own) or slow-building (it builds over a few days). The app flips a fair coin for every day ahead of time (or every 3 days for slow-building), shows the whole calendar, and locks it. Tests run 28 days by default; 56 is recommended, and the setup screen shows why: with your own daily swing, it says the smallest change each length can see. Skipping is expected. Each night the app reads the test two ways: the coin read (ON days against OFF days, whatever you did — skipping can't fool it, but it waters it down) and the did-it read (days you did it against days you didn't — sharper, but fooled if you tend to skip on rough days). A mood check asks exactly that: were the evenings before a skip rougher? It then gives a grade from A to D for how much to trust the test, in one plain line; leads with the read your days can support; marks a did-it result as a hint, not proof; and shows an answer meter with a line like "sharp enough to see 1.1 W in about 9 more logged days" — or "heading for no answer, close it now." An early clear answer needs a much stronger signal than the end of the test, because looking every night would otherwise fool you. Closed tests from before v6 keep their declared result and your call; a "Re-read with v6.0.0" fold sits beside them.

Reading the history. Every behavior can be renamed and described ("what was this measuring?") — a rename carries its tags and hunch tests (coupons) with it, and renaming onto an existing name asks: merge (all-time pooling) or keep separate as "#2". Same-named studies pool, and Detail keeps the by-period breakdown (split on 90-day gaps, each period against its own era) with a periods disagree flag when seasons contradict — that disagreement is the seasonal signal. The archive sorts by certainty / best / worst / most recent / most days with a text filter. Everything says associated with, in your record; anything prescribed is a finding for a clinician — [Rx] entries never rank in the Brief. Full story: the Repair Bay feature page.

Report — the monthly service log

Pick a month: coverage, damage deposited, acute rate, cycles closed, largest swing, hardest and strongest days, and a plain-language read. Save as PDF opens the print sheet; Month CSV exports that month's rows.

Data — backups: the monthly ritual

Your record lives only on this device — so once a month, Export backup file (or Share backup to your email or cloud). The app nags with a banner when a backup is missing or stale.

v4.1 hardening. The record now rides in two hulls: every save is mirrored into a second on-device store, and if the browser ever wipes the first copy the app restores itself from the second on next open. Where the browser supports it, Backup autopilot (Data tab) writes hindsight8020_autobackup.json to a folder you choose on every logged day — the ritual runs itself. And The evening tap is an optional once-a-day reminder if the day isn't logged by your chosen hour — no streaks, no escalation. On a brand-new record, first flight shows Today, Trends and Data first; Mission, Report and the Repair Bay come online at seven logged days (or one tap — nothing is held back).

Restore from backup

Get the .json backup onto the device → Data tab → Restore from backup → Choose File → watch for "Restored N days." Restore merges: days in the file overwrite matching days on the device; device-only days are kept. Parameters in the file replace the device's. The unlock is not in the backup — on a new device, enter the email you paid with (or your code) once.

Two or more devices? There is no sync, by design. Log on one primary device; refresh the second by restoring the primary's latest backup. Never log the same day on both — the restored file's version quietly wins.

The backup file format

{
 "app": "hindsight8020-daily",   ← must be exactly this
 "ver": "...", "exported": "...",
 "params": { "A":…, "m":…, "alpha":…, "sth":…, "gap":…, "anchor":"" },
 "entries": { "YYYY-MM-DD": { "w": 0–10, "note": "…" }, … },
 "missions":   [ …v2.0, optional; v2.2 adds an optional
                 "ref":{"start","end"} reference stretch per mission… ],
 "missionLog": [ …v2.0, optional… ],
 "missionEng": { "YYYY-MM-DD": -1|0|1, …v2.1, optional… },
 "track2Params": { "ac":…, "n":…, "gamma":…, "olPct":…  ←v2.3, optional }
}

Days that break the rules are skipped silently; the rest import. Backups from any version restore into any other — v2.0 files carry missions and weekly check-ins; older apps simply ignore those keys, and older files leave your missions untouched.

Add past data

Data tab → Add past data. CSV format: date,W,note — one day per line, date as YYYY-MM-DD (or M/D/YYYY), W from 0–10 (halves honored; 6.7 becomes 6.5), note optional, header line skipped, notes with commas in quotes. Import a .csv file or paste rows. Days already logged are skipped unless you tick Overwrite. The toast counts what happened. One stray day? Use the Quick manual backfill below it.

Trial & unlocking

Fourteen days free — the pill counts down. Buy any time from the Own it panel (Data tab) or the welcome card: secure Stripe checkout, land back unlocked. Second device or hiccup: enter the email you used at Stripe checkout (or an unlock code) under Purchased before? — one online check, then never again. One purchase, no subscription. When a trial ends the instrument locks; your record does not — export works even on the lock screen.

Model parameters — where the defaults come from

A (life scale) was calibrated on the founder's thirteen-year, 2,400-day record — D = 1.0 at its documented anchor event; m, α, S_th, and the 7-day gap are standard fatigue-analysis conventions. Honest starting points, not your calibration: once your record holds enough closed cycles, set an anchor date and Calibrate, or enter custom values. The record & method behind the defaults →

What this instrument is not

It is not a diagnosis. D is a gauge of accumulated load — not a screening score, a clinical judgment, or a prediction. It cannot see a crisis. If you are in one, the right instrument is a person: in the US, call or text 988, any time.

Privacy: every entry stays in this browser's storage on this device. Nothing is uploaded, synced, or seen by anyone — including the lab. That privacy is also the risk: the device is the only copy. Hence the backup ritual.

Questions

Anything at all: isightlaboratories@gmail.com — a person reads it.

Prefer a document? Open the illustrated guide → (a downloadable PDF lives at the bottom of it)