The Long Watch — Monitoring and Logging Over Time

Section Overview

When no lever compresses the window — the fault tied to weather, the third day, or nothing yet spotted — the campaign changes posture: instruments stand watch so people do not have to (why-intermittents-are-the-hardest-faults). The sentries are matched to the fault's shape. Min/max recording holds the extremes a signal visited, a logger holds its trend over hours, the scope's tripwire and persistence hold the fast and the rare, and an event counter holds the rate — chosen by whether the expected fault is a drift, a dip, or a flash (capturing-transients-and-single-shot-events). The rigging is layered. A coarse trend and a fine trigger cover each other's blind spots, the device runs its real workload rather than idling politely, and data logging extends to the environment — temperature and humidity recorded alongside the electrical channels, because the correlation is often the diagnosis (thermal-provocation-forcing-heat-and-cold-dependent-faults). Timestamps become the second evidence. Timestamp correlation — synchronized clocks, a diary of real-world events, the environment log — turns a capture's time into a refined window: what happened, and what else was happening (documenting-and-reasoning-about-faults). The log is read honestly. Drift is separated from discrete events, artifacts like probe bumps are recognised before they become findings, and silence is weighed against the fault's own interval. And the watch feeds the campaign. A refined window aims the next provocation, and when the fault will not visit the bench, the watch goes home with the device. Match the sentry, layer the rig, correlate the clocks, read with honesty, and feed the findings forward — and the faults that outwait patience stop outwaiting the instruments.

Why This Matters

The time-and-environment class is where repairs stall for weeks and reputations for thoroughness are made or lost — and it is the class the bench most often surrenders to (why-intermittents-are-the-hardest-faults). This matters because human watching does not scale: nobody attends a screen for three days, but a logger does — and a fault that strikes at four in the morning meets the same armed witness as one that strikes at noon (capturing-transients-and-single-shot-events). This matters because the correlation is frequently the whole answer: a reset log that lines up with the compressor's cycle, the evening humidity, or the office laser printer has named its provocation without a single probe moving. It matters because silence has a value only bookkeeping can give it: a quiet watch acquits a node in proportion to its length against the fault's interval, so an honest log turns even nothing-happened into progress (documenting-and-reasoning-about-faults). It matters because unread instruments lie by omission: min/max holds two numbers with no times, loggers sample too slowly for narrow events, and a bench's own bumps write artifacts into every channel — the reader who does not know each sentry's blind spot trusts the wrong silence. And it matters because some faults never visit the bench: a field watch — a logger sent home with the device — catches the fault in its own habitat and often ends a saga the bench never could (thermal-provocation-forcing-heat-and-cold-dependent-faults). Learn to rig and read the long watch, and the slowest faults on the shelf become bookkeeping instead of purgatory.

Required Prerequisites

Why Intermittents Are the Hardest Faults — Section 9.1 built the failure window, the recipe, and the instrumented observable; this section extends the observable across hours and days.
Capturing Transients and Single-Shot Events — Section 8.5 taught the scope's standing watches — tripwires, persistence, peak-detect; this section surrounds them with the slower, longer sentries and the correlation discipline a multi-day watch needs.

Recommended Consumables

A notebook that becomes the event diary — to record every real-world event with its time while the watch runs (documenting-and-reasoning-about-faults)
Labels for every instrument and channel — to keep a multi-sentry rig readable when the log is reviewed days later
Tape and strain relief for probe and sensor leads — to keep a long watch's connections from becoming its artifacts
A cheap clock or synchronized phone per station — to stamp the diary from the same time base as the instruments
A checklist card for the watch's start and end — to arm everything in order and close everything with its readings saved

Recommended Practice Hardware

A multimeter with min/max recording — to practise the cheapest sentry on a live rail (capturing-transients-and-single-shot-events)
A data logger or logging meter — to turn hours of rail behaviour into a reviewable trend
A temperature and humidity logger — to run the environment channel every serious watch carries (thermal-provocation-forcing-heat-and-cold-dependent-faults)
A scope with single mode for the fine layer — to practise layering a tripwire under a trend
A device that can run its real workload unattended — to watch behaviour under true conditions, not idle ones
A folder of past logs, real or sample — to practise reading drift, events, and artifacts before a live case depends on it (documenting-and-reasoning-about-faults)

Real-World Applications

The long watch closes the cases that every faster method bounced off. A technician with a till that crashes "some mornings" rigs a logger on the rail and a humidity logger beside it, and three days later the correlation is plain: crashes follow the damp nights — a moisture window no bench session would have met (thermal-provocation-forcing-heat-and-cold-dependent-faults). A repairer chasing a resetting router leaves a min/max meter on the 12-volt input overnight and finds a 9-volt minimum recorded by morning — the extreme happened, timestamp unknown, and the next night's data logger turns it into "2 a.m., nightly" (capturing-transients-and-single-shot-events). Someone watching an amplifier that dies weekly layers the rig — logger for the trend, tripwire staked on the rail for the moment itself — and when the week pays off, the trend shows the slow sag and the capture shows the collapse, cause and effect in one night's files. A shop with a fault that only happens at the customer's site sends a logger home with the device and an event-diary card for the owner, and the returned pair — log plus diary — names the neighbour's arc welder (documenting-and-reasoning-about-faults). And a bench closing a silent watch files the duration against the fault's known interval — a quiet three weeks against a twice-weekly fault — and issues the acquittal with its arithmetic attached (why-intermittents-are-the-hardest-faults). The failures this prevents: a three-day fault surrendered after one quiet afternoon, an extreme caught without its timestamp and wasted, a field-only fault chased forever on the bench, and a silent watch filed as proof when it was too short to prove anything.

Common Challenges

Every sentry has a blind spot. Min/max keeps no times, loggers sample too slowly for narrow events, tripwires see one node — the difficulty is knowing each instrument's blindness and layering the rig so another sentry covers it (capturing-transients-and-single-shot-events).
The watch outlives attention. Leads sag, batteries die, storage fills, and someone borrows the meter on day two — the difficulty is rigging for duration: secured connections, checked capacities, and a labelled do-not-touch station (documenting-and-reasoning-about-faults).
Correlation invites false stories. Given enough channels, something always lines up — the difficulty is treating a correlation as a hypothesis to provoke and confirm, not a verdict, because coincidence scales with watch length (why-intermittents-are-the-hardest-faults).

Safety Notes

Risk Level: Medium. A long watch is unattended powered operation of suspected-faulty equipment, sustained for days — the chapter's standing rule at its largest scale.

Days of unattended powered operation — protect the supply, clear the area, and rig for absence

A watch means a possibly faulty device running powered with nobody present, so the supply is current-limited or fused for the device's real draw, the area is cleared of flammables, smoke or heat around the station gets a detector where one exists, and a suspected power-stage fault does not run overnight in an unoccupied space — a fault that trips protection ends the watch, it does not burn the shop (why-intermittents-are-the-hardest-faults). All probe and grounding discipline holds for the duration: grounds on verified circuit ground, an isolation transformer on any mains-referenced circuit under test — never on the scope — with the board treated as live, and leads secured and strain-relieved so days of gravity cannot drag a probe into a short (capturing-transients-and-single-shot-events). The station is labelled — device, hazard, owner, and do-not-touch — because the person who moves a meter on day two does not know what they are ending. Field watches carry the rule home: a logger sent with the device is chosen and connected so it cannot create a hazard in an owner's house, with low-voltage sensing only and no improvised mains connections. The takeaway: rig every watch as if nobody will look at it again until it is over — because that is the plan.

Professional Tips Before Starting

Estimate the fault's interval first. The history says how often it strikes — plan the watch's length to beat that interval, or its silence will prove nothing (why-intermittents-are-the-hardest-faults).
Synchronize every clock before arming. Correlation lives and dies on timestamps — set the loggers, the scope, and the diary phone to one time base at the start (documenting-and-reasoning-about-faults).
Photograph the rig when it is armed. Days later, "which channel was the rail?" is a real question — one photo of the labelled station answers it.

The Long Watch — Sentries, Rigging, Correlation, and the Honest Log

Recap and Frame

Three provocations compress windows; this section serves the windows that will not compress, and its method is delegated patience — instruments that wait better than people do (why-intermittents-are-the-hardest-faults). The class is time and circumstance. Faults tied to weather, season, duty hours, or unidentified events resist every lever because their provocation is not on the bench — so the bench watches until the provocation arrives. The fast sentries are already learned. The tripwire, persistence, and peak-detect from the capture chapter are the fine end of the watch — precise, fast, and node-specific — and this section builds the slower, wider layers around them (capturing-transients-and-single-shot-events). The new sentries trade speed for span. Min/max recording and data loggers see nothing quickly but hold hours and days, which is exactly the trade a time-class fault demands. Correlation is the second product. A watch produces captures and timestamps, and the timestamps — held against a diary and an environment log — often carry more diagnosis than the waveforms (documenting-and-reasoning-about-faults). And honesty rules the bookkeeping. Every sentry has a blind spot, every bench writes artifacts, and silence has value only in proportion to the fault's interval — the log is read with all three in mind. Hold the frame — delegated patience, layered sentries, correlated timestamps, honest bookkeeping — and the slowest class of fault becomes routine casework.

The Sentries — Matching the Instrument to the Fault's Shape

Sentries differ in what they can see, so the first decision is what shape the fault is expected to take (capturing-transients-and-single-shot-events). Min/max is the cheapest witness. A multimeter's min/max mode records the highest and lowest values seen for as long as it runs — two numbers that prove an extreme visited, at the cost of not knowing when, how often, or how fast — and the mode itself responds over a tenth of a second or more, so events faster than that need a meter's peak-capture mode or the scope's tripwire. The logger adds the axis of time. A data logger or logging meter samples at intervals and stores the series, turning a night into a voltage-versus-time trend where drifts, steps, and daily rhythms become visible — at the cost of missing anything narrower than its sampling. The tripwire holds the fast end. For events too brief for a logger — the millisecond sag, the glitch — the scope's staked trigger in single or normal mode remains the precise witness, catching one node's fast behaviour with a pre-trigger lead-up. The counter measures rate. Where the question is "how often," a counter on the tripwire's output — or the scope's own trigger counter where it exists — turns recurrence into a number that can be compared across conditions. Shape chooses the sentry. An expected drift wants the logger, an expected extreme wants min/max at least, an expected flash wants the tripwire, and an unknown wants layers — because guessing one shape and rigging only for it is how watches miss (why-intermittents-are-the-hardest-faults). Two numbers from min/max, a trend from the logger, the flash from the tripwire, the rate from the counter, and layers when the shape is unknown — and the sentries are chosen. Ask what the fault would look like if it happened, and pick the instruments that would see exactly that.

The Rig — Layering, Workload, and the Environment Channel

A watch is a small installation, and its rigging decides whether day three's data is worth anything (documenting-and-reasoning-about-faults). Layers cover blind spots. The standard rig pairs a coarse, wide sentry with a fine, fast one — a logger trending the rail while a tripwire waits for the moment — so the trend explains the capture and the capture explains the trend. The workload must be real. A device idling politely on a bench is not the device that fails — the watch runs the real workload, the closed case, the true duty cycle, because the earlier sections' lesson stands: the bench hides faults by changing conditions (why-intermittents-are-the-hardest-faults). The environment gets its own channel. A temperature and humidity logger runs beside every serious watch, because the time class is so often the environment class in disguise — and the correlation between a reset log and a humidity curve is a diagnosis (thermal-provocation-forcing-heat-and-cold-dependent-faults). The rig is hardened for duration. Leads are secured and strain-relieved, batteries and storage are checked against the planned length, auto-power-off is confirmed disabled on every meter, the station is labelled do-not-touch, and the armed rig is photographed — because a watch's own decay writes artifacts into its channels, and a meter that sleeps at hour four files a false acquittal. Everything is armed in order and closed in order. A start checklist arms sentries with synchronized clocks and a noted start time; an end checklist saves every record before anything is unclipped — the capture chapter's save-before-re-arm rule, scaled up (capturing-transients-and-single-shot-events). Coarse over fine, the true workload, the environment beside the electronics, hardened for days, and opened and closed by checklist — and the rig is worthy of its wait. Build the watch as an installation, not a bench setup that overstayed.

The Correlation — Clocks, Diaries, and What Else Was Happening

A capture says what happened; the timestamps say when; and when, held against everything else that has a time, is often the diagnosis (documenting-and-reasoning-about-faults). Clocks are synchronized before arming. Every instrument that stamps time — loggers, the scope, the diary phone — is set to one time base at the start, because correlation across drifted clocks manufactures false stories and buries true ones. The diary is the human channel. A notebook at the station — or a card sent home with a field watch — records real-world events with times: the compressor, the rain, the cleaning crew, the welder next door; entries cost seconds and answer questions no instrument can. Alignment turns times into windows. When the fault's timestamp lines up with an environment curve or a diary entry across repeated strikes, the failure window gains a condition it never had — "resets nightly" becomes "resets when the room drops below sixteen degrees" — and the campaign gains a provocation to try (thermal-provocation-forcing-heat-and-cold-dependent-faults). Repetition separates signal from coincidence. One alignment is a story; the same alignment across several strikes is a window — and given enough channels something always lines up once, so single coincidences are logged, not believed (why-intermittents-are-the-hardest-faults). The correlation is confirmed by provocation. A window found by alignment is handed back to the provocation sections — recreate the cold, the damp, the load — because a correlation confirmed by summons is a diagnosis, while an unconfirmed one is a hypothesis with a good story. One time base, a human diary, alignment into windows, repetition against coincidence, and confirmation by summons — and the timestamps earn their keep. Ask what else was happening at every strike, and let the clocks answer.

The Reading — Drift, Events, Artifacts, and Honest Silence

A log is evidence only after it is read critically, and the reading has four disciplines (documenting-and-reasoning-about-faults). Drift and events are different animals. A trend that slides over hours — a rail sagging as temperature climbs — points at margins and aging, while a discrete step or spike points at an event with a cause and a time; the two feed different suspect lists and different provocations. Every channel carries the bench's fingerprints. A probe bumped during the watch, a meter's range change, the shop's own mains events, or a lead that crept loose all write entries that look like findings — which is why the diary logs bench activity too, and why an artifact's timestamp usually matches somebody's presence. Blind spots are read alongside the data. A clean logger trend does not clear narrow events its sampling missed, and a min/max extreme without a time cannot be correlated — each sentry's silence covers only what that sentry could see (capturing-transients-and-single-shot-events). Silence is weighed, not just filed. A watch's quiet is evidence in proportion to its length against the fault's interval — the arithmetic from the campaign's start — so the log's conclusion states both numbers: watched this long, against a fault that strikes this often (why-intermittents-are-the-hardest-faults). Findings survive a second look. Anything about to drive a repair — a captured sag, a correlated window — is re-derived from the raw records once, because a misread channel label or a time-zone slip has sent benches chasing their own logging errors. Drift split from events, artifacts owned, blind spots respected, silence weighed, and findings re-derived — and the log becomes evidence. Read the watch as sceptically as it was rigged carefully, and its answers hold.

The Return — Feeding the Campaign and Going to the Field

The watch is not the end of the campaign; it is the campaign's intelligence arm, and its findings flow back into everything before it (why-intermittents-are-the-hardest-faults). A refined window re-arms the provocations. The correlation's product — cold mornings, damp nights, the welder's hours — is a condition the provocation sections know how to recreate, so the slow discovery becomes a fast summons and the fault finally performs on the bench (thermal-provocation-forcing-heat-and-cold-dependent-faults). A captured event aims the route. The trend and capture say which rail and how it failed, so the diagnostic route from the oscilloscope chapter starts at the guilty node with the failure's shape already known. A silent watch narrows honestly. Quiet channels acquit their nodes for the watched conditions and duration, and the next watch moves — new nodes, new season, new workload — with the log recording what has been covered (documenting-and-reasoning-about-faults). The field watch goes where the fault lives. When the fault refuses the bench entirely, a logger and an event-diary card go home with the device — low-voltage sensing, owner-safe rigging, honest instructions — and the returned pair frequently names a provocation no shop could have guessed. And the verification inherits the watch. A repair against a time-class fault is proven the same way the fault was caught: the same watch, re-run past the fault's old interval, its silence now carrying the arithmetic that makes it proof (capturing-transients-and-single-shot-events). Windows back to provocations, captures to the route, silences to the map, the watch to the field, and the same watch as the proof — and the long game closes the loop. Treat every watch as reconnaissance for the next move, and even the slow faults run out of places to hide.

Common Mistakes

Watching without arithmetic. A quiet afternoon against a weekly fault proves nothing — plan the watch's length against the fault's interval, and state both numbers with the conclusion (why-intermittents-are-the-hardest-faults).
Rigging one sentry for an unknown shape. A logger misses the flash, a tripwire misses the drift — layer coarse and fine until the fault's shape is known (capturing-transients-and-single-shot-events).
Skipping clock synchronization. Correlation across drifted clocks manufactures false stories — one time base across every instrument and the diary, set before arming (documenting-and-reasoning-about-faults).
Letting the device idle. The polite bench idle is not the failing workload — run the real duty cycle, closed case and all, for the whole watch.
Believing a single alignment. Given enough channels, something always lines up once — require repetition across strikes, then confirm the window by provocation.

Troubleshooting Guidance

The long watch runs sentries, rig, correlation, reading, return. If no lever has compressed the window: estimate the fault's interval from the history and plan a watch to beat it (why-intermittents-are-the-hardest-faults). If the expected fault is a drift or a rhythm: lead with the logger; if an extreme, at least min/max; if a flash, the tripwire; if unknown, layer them (capturing-transients-and-single-shot-events). If the watch will run days: harden the rig — secured leads, checked batteries and storage, a labelled station, a photo of the armed setup. If environment is even suspected: run the temperature and humidity logger alongside, because the time class is often the environment class in disguise (thermal-provocation-forcing-heat-and-cold-dependent-faults). If a strike is captured: pull its timestamp against the diary and the environment log, and require the same alignment across repeated strikes before calling it a window (documenting-and-reasoning-about-faults). If a channel shows a finding: check it against the bench's own activity first — bumps, range changes, and mains events write artifacts with matching timestamps. If the watch stays silent: file the duration against the fault's interval, acquit only what the sentries could actually see, and move the next watch accordingly. If the fault will not visit the bench: send a field watch home — logger, diary card, owner-safe rigging — and read the returned pair together. If a window emerges: hand it back to the provocations — recreate the condition, summon the fault, and let the fast campaign finish what the slow one found. The throughline: delegate the patience to instruments, correlate everything that has a clock, read with the blind spots in mind, and feed every finding back into the campaign.

Verification & Testing Methods

Confirm the watch was an instrument problem solved, not a vigil endured:

[ ] I chose sentries by the fault's expected shape — min/max recording for extremes, a logger for trends, the tripwire for flashes, a counter for rate — and layered coarse over fine when the shape was unknown.
[ ] I rigged for duration: real workload running, leads secured and strain-relieved, batteries and storage checked, station labelled, the armed rig photographed, and data logging extended to the environment with a temperature and humidity channel alongside.
[ ] I synchronized every clock to one time base before arming, kept an event diary for the human channel, and practised timestamp correlation — requiring the same alignment across repeated strikes before calling it a window, then confirming the window by provocation.
[ ] I read the log honestly: drift separated from discrete events, bench artifacts checked against the diary before becoming findings, each sentry's blind spots respected, findings re-derived from raw records before driving repairs.
[ ] I weighed silence with arithmetic — watch length against the fault's interval, stated together — fed refined windows back to the provocations, sent a field watch home when the fault stayed there, and proved any repair with the same watch re-run past the old interval.

Then try the practice exercises below — hands-on watch work; scenarios differ from the quiz.

Practice Exercises

Run the cheapest sentry (5 minutes to arm, hands-on). Put a multimeter in min/max on a live rail, note the start time, and return after an hour of real workload to read the two numbers — then write down what they prove, and what they cannot (no times, no count, no shape) (capturing-transients-and-single-shot-events).
Rig a layered watch with one time base (5 minutes to arm, hands-on). Pair a logger's trend with a staked tripwire on the same rail, add the temperature and humidity logger beside them, synchronize every clock and your diary phone, photograph the labelled station, and start an event diary — the full rig, rehearsed small (documenting-and-reasoning-about-faults).
Read a log for drift, event, and artifact (5 minutes, paper). On a real or sample day-long trend, mark one slow drift, one discrete step, and one entry that matches diary-logged bench activity — and state which suspect list each feeds and which one is not a finding at all.
Plan a field watch with honest arithmetic (3 minutes, paper). For a fault that strikes about twice a month at a customer's site, specify the take-home rig — logger channel, sensing point, diary card wording — the watch length whose silence would actually mean something against that interval, and the provocation you would run back at the bench if the returned log correlates with a condition (why-intermittents-are-the-hardest-faults).

These core steps — matching sentries to shapes, the layered rig, one-time-base correlation, the honest reading, and feeding findings back — are tested in the Chapter Quiz at the end of this chapter, where a score of 80% is required to continue.

Key Takeaways

The time-and-environment class is answered by delegated patience: sentries chosen by the fault's expected shape — min/max recording for extremes it cannot timestamp, loggers for trends, the scope's tripwire for flashes, counters for rate — layered so coarse and fine cover each other's blind spots (capturing-transients-and-single-shot-events).
The rig is an installation: data logging runs on the real workload with the case closed, extends to a temperature and humidity channel, and is hardened for duration — secured leads, checked batteries and storage, a labelled and photographed station, armed and closed by checklist (why-intermittents-are-the-hardest-faults).
Timestamp correlation is the watch's second product: one time base across instruments and diary, real-world events logged by hand, alignment demanded across repeated strikes, and every correlated window confirmed by handing it back to the provocations (thermal-provocation-forcing-heat-and-cold-dependent-faults).
The log is read sceptically: drift and discrete events feed different suspect lists, bench activity writes artifacts with matching timestamps, each sentry's silence covers only what it could see, and findings are re-derived from raw records before they drive repairs (documenting-and-reasoning-about-faults).
Silence is evidence with arithmetic attached — watch length against the fault's interval, stated together — and the watch closes the loop: refined windows re-arm the provocations, field watches follow the fault home, and the repair is proven by the same watch re-run past the old interval.

Skills Learned

You can now choose sentries by the fault's expected shape — min/max for extremes, loggers for trends, tripwires for fast events.
You can now rig a layered watch: coarse trend plus fine trigger, the real workload running, environment logged alongside.
You can now correlate timestamps across instruments and a diary of events, turning a capture into a refined window.
You can now read a log honestly — drift versus discrete events versus artifacts, and silence weighed against the fault's interval.
You can now feed the watch's findings back into the campaign — refined windows into provocations, and field watches when the fault stays home.

Glossary Additions

data logging — recording a measurement as a timestamped series over hours or days, using a data logger or logging meter, so that trends, steps, rhythms, and rare events on a watched node become reviewable after the fact. In intermittent diagnosis the logger is the wide, slow layer of a watch: it trades speed for span — anything narrower than its sampling interval is likely to pass unseen, which is why serious rigs layer a scope tripwire beneath the trend — and it extends beyond electrical channels to the environment, with temperature and humidity logged alongside the rails because the correlation between them is often the diagnosis. A logging rig is hardened for its duration — secured leads, checked batteries and storage, a labelled station — and its records are saved by checklist before anything is unclipped.
min/max recording — a multimeter mode that continuously tracks a measurement and retains the highest and lowest values seen while it runs, making it the cheapest long-watch sentry on the bench. Its two numbers prove that an extreme visited a node — a 12-volt input that shows a 9-volt minimum by morning has confessed to a sag — but the mode keeps no timestamps, no count, and no shape: it cannot say when the extreme happened, how often, or how fast. Min/max therefore serves as the first, lowest-effort layer of a watch — deployed in minutes, read at a glance — with its findings handed to a data logger or a staked tripwire to add the time axis and the waveform the two numbers lack.
timestamp correlation — aligning the time of a captured fault against every other record that carries a clock — the data logger's trend, the environment log, and a hand-kept diary of real-world events — to learn not just what happened but what else was happening. The discipline has three rules: every instrument and the diary are synchronized to one time base before the watch is armed, because correlation across drifted clocks manufactures false stories; a single alignment is logged as a hypothesis, not believed, since across enough channels something always lines up once — the same alignment must repeat across strikes; and a correlated window is confirmed by provocation, recreating the aligned condition to summon the fault on demand. Done honestly, correlation turns a timestamp into a refined failure window — "resets nightly" into "resets when the room drops below sixteen degrees."

Suggested Next Sections

Must read next:

From Reproduction to Verified Repair — Section 9.6 closes the campaign: choosing the repair the evidence actually supports, proving it with the recipe or watch that caught the fault, and writing the record that keeps the next device of the family from starting over.

Recommended:

Documenting and Reasoning About Faults — the note-keeping and inference discipline that the watch's diaries, logs, and honest silences are built on.
Thermal Provocation — Forcing Heat- and Cold-Dependent Faults — the provocation that most often receives the watch's findings, since the time class is so frequently the environment class in disguise.