From Reproduction to Verified Repair

Section Overview

The campaign summoned the fault and named its mechanism; this closing section spends that evidence — on the right repair, a designed proof, and a record that outlives the case (why-intermittents-are-the-hardest-faults). The evidence chooses the repair. A confirmed mechanism dictates its own fix — the named joint reflowed, the eroded margin recapped with its numbers, the relaxed contact retensioned or replaced — and evidence that names nothing yet authorises no soldering, which is the discipline that holds off shotgunning (power-and-load-provocation-margins-brownout-and-load-steps). The proof is designed with arithmetic. The verification count is chosen to beat the fault's own reproduction rate, the minimal recipe that summoned the fault is the one that must now stay silent, and the runs sweep the window's whole band — hot and cold, loaded and idle — because a repair can move a window instead of closing it (thermal-provocation-forcing-heat-and-cold-dependent-faults). Failed campaigns end honestly. A no-fault-found is issued only over a documented, explored window; a preventive repair is offered as a labelled wager when reproduction is out of reach; and the return path — a field watch, the next season — stays open in the record (the-long-watch-monitoring-and-logging-over-time). The record is the second product. Window, recipe, captures, margins, repair, and count file under the device family, growing the recipe library that lets the next twin of this fault start ahead. Spend the evidence on its own repair, prove it by the count, end honestly when the campaign fails, and file everything — and the chapter's promise closes end to end.

Why This Matters

Everything before this section finds the fault; this section is why it stays found — and it is where benches that reproduce faults still manage to lose them (why-intermittents-are-the-hardest-faults). This matters because the repair step is where discipline collapses: with the fault finally summoned, the urge to resolder everything nearby is strongest exactly when the evidence has made it least necessary — and a shotgunned board can no longer teach which fix mattered. This matters because unverified repairs of intermittents are coin flips: the fault's own rarity impersonates a cure for days, so a repair without its count ships on luck and returns on schedule (the-long-watch-monitoring-and-logging-over-time). It matters because windows move: a reflowed joint or a recapped rail can shift a thermal or power window rather than close it, and only a band-sweeping verification catches the fault that merely changed address (thermal-provocation-forcing-heat-and-cold-dependent-faults). It matters because honest failure keeps trust: a documented no-fault-found with its explored window, or a preventive repair sold as the wager it is, keeps a customer — while a confident guess that bounces loses them. And it matters because the record compounds: a filed recipe turns the family's next three-week saga into an afternoon, which is how a bench builds the speed that looks like magic from outside (power-and-load-provocation-margins-brownout-and-load-steps). Close the case the way the campaign opened it — on evidence and arithmetic — and the intermittent stays repaired.

Required Prerequisites

Why Intermittents Are the Hardest Faults — Section 9.1 built the window, the recipe, and reproduction as the gate; this section is that gate's second use — the proof — carried through to the filed case.
The Long Watch — Monitoring and Logging Over Time — Section 9.5 taught the watches, the correlation discipline, and honest silence; this section reuses all three as the verification's instruments and bookkeeping.

Recommended Consumables

The campaign's case notes — to carry the window, recipe, captures, and margins into the repair decision (why-intermittents-are-the-hardest-faults)
A case-record template or a dedicated notebook section — to file the case in a form the next technician can actually use
Index cards or a digital note per device family — to grow the recipe library one closed case at a time
Labels for repaired boards — to mark what was replaced and when, so the next opening starts informed
The customer's contact notes — to write the honest ending — verified, preventive, or no-fault-found — in words the owner can weigh

Recommended Practice Hardware

A device whose fault the earlier sections' practice summoned — to carry one case through repair, proof, and filing (why-intermittents-are-the-hardest-faults)
The summoning instruments, still rigged — to re-run the recipe as the verification without rebuilding the setup (the-long-watch-monitoring-and-logging-over-time)
The repair tools the mechanism calls for — iron, contact cleaner, replacement parts — to fix exactly what the evidence named
A thermal or power rig from the provocation sections — to sweep the window's band during verification (thermal-provocation-forcing-heat-and-cold-dependent-faults)
A stack of past case notes, real or sample — to practise writing records that a stranger could pick up
A folder of family recipes — to see how a library turns the next case into a lookup (power-and-load-provocation-margins-brownout-and-load-steps)

Real-World Applications

The endgame is where the campaign's weeks pay out — or leak away. A technician whose freeze spray named one joint reflows that joint alone, re-runs the tap-and-flex recipe twenty times against a fault that used to strike every third run, and files the silent count — the repair proven in an hour because the evidence was already exact (why-intermittents-are-the-hardest-faults). A repairer whose margin sweep showed a 0.5-volt erosion recaps the input stage, re-runs the same sweep, and writes both failing points into the record — the repair verified in volts, and the family's healthy number now on file (power-and-load-provocation-margins-brownout-and-load-steps). Someone whose warm-window fault vanished after a reflow sweeps the band anyway — cold start, warm soak, the transitions — and catches the window merely moved colder, saving the return trip the quick pass would have shipped (thermal-provocation-forcing-heat-and-cold-dependent-faults). A bench that never reproduced a twice-a-year fault writes the honest ending: the explored window, a preventive recap offered as a labelled wager, and a field-watch plan for the next strike — the customer chooses with open eyes (the-long-watch-monitoring-and-logging-over-time). And a shop that files every closed case meets the same model's same complaint a month later and goes from intake to summoned fault in one afternoon — the library doing the diagnosis. The failures this prevents: a shotgunned board that can never teach which fix mattered, a coin-flip repair shipped on three quiet days, a moved window mistaken for a closed one, and a solved case whose lessons evaporated unfiled.

Common Challenges

The evidence feels like permission to do more. With the board finally open and the iron hot, "while I'm in here" grows arms — the difficulty is that every extra touched joint blurs the verification, because a silent recipe can no longer say which change silenced it (why-intermittents-are-the-hardest-faults).
The count is unglamorous. Twenty recipe runs after the fault already "looks fixed" feels like wasted bench time — the difficulty is holding the arithmetic against the optimism, because the fault's rarity is precisely what fakes cures.
Records compete with the next job. The queue is long and the case is closed in every way but paperwork — the difficulty is that the ten-minute record is the only part of the campaign that pays interest, and it is the first part benches skip (the-long-watch-monitoring-and-logging-over-time).

Safety Notes

Risk Level: Medium. This section repeats the campaign's summons as its proof, so every provocation's own discipline returns — plus the repair work between them.

The verification re-runs the provocations — every summons keeps its own rules

Proving a repair means re-running whatever summoned the fault, so each provocation's discipline returns intact: thermal runs gentle, measured, and condensation-aware; mechanical runs insulated, graded, and in millimetres; power runs inside written ratings with the current limit set; and watches rigged for absence with supplies protected (thermal-provocation-forcing-heat-and-cold-dependent-faults). The repair between summons is bench work with its own rules — a discharged board for soldering, verified before power returns — and the first post-repair power-up of a previously faulty device is treated as a first power-up: current-limited, observed, and instrumented (power-and-load-provocation-margins-brownout-and-load-steps). Long verification watches remain unattended powered operation of recently faulty equipment: current-limit or fuse the supply, clear the area, and no overnight power-stage suspects in unoccupied spaces (the-long-watch-monitoring-and-logging-over-time). The takeaway: the proof borrows every tool the campaign used — and every rule that came with it.

Professional Tips Before Starting

Photograph before the iron moves. The pre-repair state — the named joint, the suspect caps — belongs in the record — thirty seconds now completes the case file later (the-long-watch-monitoring-and-logging-over-time).
Fix one thing, then prove it. If the evidence names one mechanism, repair one mechanism — a second change waits until the first is verified, or the verification stops meaning anything (why-intermittents-are-the-hardest-faults).
Write the count before the first run. Decide how many silent runs prove it — from the fault's old rate — before optimism can negotiate it down.

The Endgame — Repair, Proof, Honest Endings, and the Record

Recap and Frame

Five sections built the campaign — window, recipe, three provocations, and the watch; this closer spends what they earned, and the spending has its own discipline (why-intermittents-are-the-hardest-faults). The campaign's product is a named mechanism. A joint that answers to direction, a margin eroded by half a volt, a contact that drops on flex, a window correlated to damp nights — the evidence arrives specific, and specificity is what the endgame protects. The repair must match the name. Evidence this exact dictates its fix, and the endgame's first rule is to repair what was named — no more — because extra changes blur the only proof available (power-and-load-provocation-margins-brownout-and-load-steps). The proof re-uses the summons. The recipe, the sweep, or the watch that caught the fault becomes the verification instrument, run against the repair with a count designed from the fault's old rate (the-long-watch-monitoring-and-logging-over-time). The failures get endings too. Campaigns that never summon end in a documented no-fault-found, a labelled preventive wager, or a field watch — honest closings that keep the return path open. And the record is the second deliverable. Every closed case files its window, recipe, and numbers under the device family, so the library — not the memory — carries the lesson forward. Hold the frame — a named mechanism, a matching repair, a designed proof, honest endings, and a filed record — and the chapter's campaign closes the way it opened: on evidence.

The Repair — Spending Evidence Without Blurring It

The repair step looks like ordinary bench work, but on an intermittent it carries one extra duty: preserving the verification's meaning (why-intermittents-are-the-hardest-faults). The mechanism dictates the fix. A cracked joint confirmed by direction and magnification is reflowed or resoldered; an eroded margin with its failing voltage on file is recapped or its regulator replaced; a relaxed or corroded contact is cleaned, retensioned, or swapped — the fix is the mechanism's mirror, not a menu choice (power-and-load-provocation-margins-brownout-and-load-steps). One change at a time keeps the proof sharp. If the recipe goes silent after one repair, that repair is the cure; after five simultaneous repairs, the silence names nothing — so changes are made and verified singly wherever the fault's stakes allow. Shotgunning is the discipline's named enemy. The resolder-everything reflex — met back in the diagnostic mindset — sometimes works, never teaches, and cannot be verified; the campaign's whole point was to earn the right not to shotgun. Neighbours are repaired with their own justification. The named joint's siblings under the same stress may deserve preventive attention — but as a decision recorded on its own line, made after the named repair is verified, so the proof stays attributable. And the board is left better documented, not just better soldered. The repaired joint, the replaced parts, and the date go on the board's label and in the record, because the next technician opens this device knowing nothing otherwise (the-long-watch-monitoring-and-logging-over-time). The mechanism's mirror, one change at a time, shotgunning refused, neighbours on their own line, and the work documented — and the evidence survives its own repair. Fix exactly what the campaign named, and the proof that follows can actually prove it.

The Proof — Designing the Verification Count

A repair on an intermittent is unproven until arithmetic says otherwise, and the arithmetic is designed before optimism can negotiate (why-intermittents-are-the-hardest-faults). The count comes from the old rate. A fault that struck on roughly every third recipe run sets the bar: the repaired device must stay silent through enough runs — a couple of dozen, not three — that the old rate would almost surely have struck, and the number is written down before the first run. The minimal recipe is the instrument. The shrunken recipe that most reliably summoned the fault — this motion, this temperature, this load step — is the recipe the verification runs, because a proof against the strongest summons is the strongest proof. The band gets swept. Repairs move windows: a reflowed joint can shift a thermal window colder, a recap can move a failing voltage rather than restore it — so the verification sweeps the window's whole band, hot and cold, loaded and idle, the transitions included (thermal-provocation-forcing-heat-and-cold-dependent-faults). Numbers verify where they exist. Where the campaign produced a margin — a failing voltage, a load level — the verification is the same sweep re-run, and the repair is proven by the number moving back where the family's healthy units live (power-and-load-provocation-margins-brownout-and-load-steps). Slow faults verify by watch. A fault caught by the long watch is proven by the same watch re-run past the old interval, its silence now carrying the arithmetic — and for field-only faults, the verification watch goes home with the device too (the-long-watch-monitoring-and-logging-over-time). A count from the rate, the minimal recipe as instrument, the band swept, numbers re-measured, and slow faults out-waited — and the repair is proven, not presumed. Design the proof like the campaign designed the summons, and the silence at the end means something.

The Honest Endings — When the Campaign Does Not Summon

Some campaigns end without a reproduction, and the endgame's quality shows most in how those cases close (why-intermittents-are-the-hardest-faults). No-fault-found is a documented verdict. Issued honestly, it lists the windows explored — the conditions recreated, the watches run with their durations against the fault's interval, the provocations tried — so the verdict reads "not reproduced under these conditions," never "nothing wrong" (the-long-watch-monitoring-and-logging-over-time). The preventive repair is a labelled wager. When evidence is partial — the class known, the mechanism probable, reproduction out of reach — repairing the most probable mechanism can be the right call, offered to the owner as what it is: a repair of the likely suspect, unverifiable by recipe, with the odds stated plainly. Declining is an ending too. A fault too rare to reproduce, too costly to wager on, and too mild to chase may close as documented advice — what to log at home, what symptom would justify reopening — which serves the owner better than a confident guess. The return path is built into the record. A field watch offered for the next strike, the season noted when the window is seasonal, and the case filed as open-with-a-plan turn a defeat into a staged campaign rather than a shrug (thermal-provocation-forcing-heat-and-cold-dependent-faults). And the unreproduced case still feeds the library. The explored windows and quiet watches are data for the family file — the next twin of this complaint starts where this one stopped. A documented verdict, a labelled wager, a principled decline, a built return path, and the defeat filed as data — and even the unsummoned fault ends honestly. Close the failures with the same bookkeeping as the wins, and the bench keeps both its customers and its evidence.

The Record — Filing the Case and Growing the Library

The repaired device leaves; the record stays, and the record is the half of the product that compounds (the-long-watch-monitoring-and-logging-over-time). The case record has a fixed skeleton. Complaint and history, the window as written, the recipe that summoned, the captures and margins, the named mechanism, the repair performed, and the verification count with its arithmetic — one page a stranger could pick up and act on. The recipe library is the record's working half. Per device family, the file keeps the summoning recipes, the healthy and failing margins, and the usual suspects — so the next case of the family opens with a lookup instead of a campaign (power-and-load-provocation-margins-brownout-and-load-steps). Numbers make the library sharp. "Fails below 4.6 volts, healthy units run to 4.2" outlives any prose description, and margin pairs, soak temperatures, and verification counts are the entries that turn the next diagnosis into arithmetic (why-intermittents-are-the-hardest-faults). Silent watches file too. Explored windows, quiet durations, and acquitted nodes are the library's negative space — the next campaign skips what this one already ruled out. The record closes the loop with the owner. The verified count, the labelled wager, or the documented no-fault-found goes to the customer in plain words, because the record's last job is an honest expectation (thermal-provocation-forcing-heat-and-cold-dependent-faults). A fixed skeleton, a working library, numbers over prose, the negatives filed, and the owner told honestly — and the case is closed in both senses. File every campaign, win or lose, and the bench gets faster every month it operates.

The Campaign End to End — One Case, Whole

Assembled, the chapter reads as one continuous method, and a worked case fixes its shape (why-intermittents-are-the-hardest-faults). The complaint frames the window. An espresso machine "sometimes dies mid-shot, worse in winter" — the history's verbs and seasons classify it thermal-plus-power, and the window drafts as cold room, heavy heater load. The provocations summon it. A cold chill power-off, then the heater's own load step under a staked tripwire — and the rail's capture shows the sag bottoming below reset at the boiler's kick, but only cold (thermal-provocation-forcing-heat-and-cold-dependent-faults). The margin names the mechanism. An input sweep finds the machine resetting at a voltage the service data says it should ride through, and the cold-morning pairing points at the input electrolytics — at their worst cold, exactly as the window said (power-and-load-provocation-margins-brownout-and-load-steps). The repair mirrors the name. The input capacitors are replaced — that and nothing else — with the old failing voltage written down and the board labelled. The proof sweeps the band. The same sweep re-runs cold and warm: the failing point has moved back below the family's healthy line, and the heater's load step runs silent through a count that beats the old every-third-shot rate. The record closes it. Window, recipe, captures, both margins, the repair, and the count file under the machine's model — and the winter's next espresso machine with the same complaint is an afternoon's lookup, not a fortnight's campaign (the-long-watch-monitoring-and-logging-over-time). Complaint to window, window to summons, summons to margin, margin to one repair, repair to a banded count, and the whole of it filed — one case, whole. That walk — evidence first, arithmetic last — is what this chapter means by repairing an intermittent.

Common Mistakes

Shotgunning the open board. Resoldering everything near the suspect blurs the proof and teaches nothing — repair what the evidence named, one change at a time (why-intermittents-are-the-hardest-faults).
Negotiating the count downward. Three quiet runs feel convincing against a fault that struck every tenth — write the count from the old rate before the first run, and hold it.
Verifying at one point of the window. A repair can move a window instead of closing it — sweep the band: hot and cold, loaded and idle, the transitions included (thermal-provocation-forcing-heat-and-cold-dependent-faults).
Issuing no-fault-found without its documentation. The bare verdict says "nothing wrong"; the honest one says "not reproduced under these listed conditions" — file the explored windows and watch durations with it (the-long-watch-monitoring-and-logging-over-time).
Closing the case without filing it. The unrecorded campaign evaporates — ten minutes of record and a library card are the only part of the work that pays interest.

Troubleshooting Guidance

The endgame runs repair, proof, honest ending, record. If the campaign named a mechanism: repair its mirror — the joint, the caps, the contact — and nothing else yet (why-intermittents-are-the-hardest-faults). If more than one repair tempts: make and verify them singly where stakes allow, or accept that the proof will name the set, not the fix. If the repair is in: run the minimal recipe as the verification, with a count written from the old rate before the first run. If the window had a band: sweep it — a moved window is this step's classic catch (thermal-provocation-forcing-heat-and-cold-dependent-faults). If the campaign produced a margin: re-run the same sweep and compare failing points against the family's healthy number (power-and-load-provocation-margins-brownout-and-load-steps). If the fault was a long-watch catch: re-run the same watch past the old interval — at the site, if that is where it lived (the-long-watch-monitoring-and-logging-over-time). If the campaign never summoned: close honestly — documented no-fault-found with explored windows, a labelled preventive wager, or advice with a return plan — and file the negatives. If the case is closing: write the record's skeleton and the family's library card, numbers first. The throughline: repair the name, prove it by the count across the band, end the failures honestly, and file everything — the campaign is only over when the record is.

Verification & Testing Methods

Confirm the endgame closed the case rather than just the ticket:

[ ] I repaired what the evidence named — the mechanism's mirror, one change at a time — resisted shotgunning, and put any neighbour work on its own recorded line after the named repair was verified.
[ ] I designed the proof before running it: a verification count written from the fault's old reproduction rate, run on the minimal recipe that summoned the fault, with the count and its arithmetic recorded.
[ ] I swept the window's band during verification — hot and cold, loaded and idle, transitions included — and where the campaign produced a margin, I re-ran the same sweep and compared failing points against the family's healthy number.
[ ] I ended any unsummoned campaign honestly: a no-fault-found documented with its explored windows and watch durations, any preventive repair offered as a labelled wager with the odds stated, and a return path — field watch or season note — built into the record.
[ ] I filed the case — window, recipe, captures, margins, repair, count — and grew the recipe library with the family's numbers and negatives, so the next twin of this fault starts ahead.

Then try the practice exercises below — endgame work; scenarios differ from the quiz.

Practice Exercises

Match evidence to repair (5 minutes, paper). For three confirmed mechanisms — a direction-responding jack joint, a rail failing 0.5 V early with healthy siblings on file, a connector that drops on cable flex — write the one repair each dictates; then take a fourth case with only a probable mechanism and write both endings: the preventive-repair offer as a labelled wager, and the one-line documented no-fault-found you would issue if the owner declines — explored windows listed, return path named (why-intermittents-are-the-hardest-faults).
Design three verification counts (5 minutes, paper). For faults that struck every third recipe run, every tenth, and twice a month on a field watch, write each verification plan: the count or duration, the recipe or watch that runs it, and the band it must sweep (the-long-watch-monitoring-and-logging-over-time).
Run one proof for real (5 minutes, hands-on). On a practice board from the earlier sections' exercises — repaired or standing in as repaired — re-run its minimal summoning recipe to a pre-written count with the observable armed, and record the outcome the way the case file needs it (thermal-provocation-forcing-heat-and-cold-dependent-faults).
Trace one campaign whole, then file it (3 minutes, paper). From any worked case in this chapter's practice — or the espresso case — first trace the campaign end to end, one line per stage: complaint, window, summons, mechanism, repair, count; then write the one-page record and the family library card with its numbers, including one negative (an acquitted node or a quiet watch) (power-and-load-provocation-margins-brownout-and-load-steps).

These core steps — the evidence-matched repair, the designed count, the banded proof, the honest endings, and the filed record — are tested in the Chapter Quiz at the end of this chapter, where a score of 80% is required to continue.

Key Takeaways

The evidence chooses the repair: a confirmed mechanism dictates its mirror — joint reflowed, margin recapped, contact replaced — one change at a time, with shotgunning refused because a blurred board can never say which fix mattered (why-intermittents-are-the-hardest-faults).
The proof is designed before it runs: a verification count written from the fault's old reproduction rate, executed on the minimal recipe that summoned the fault, and swept across the window's whole band — because repairs can move windows instead of closing them (thermal-provocation-forcing-heat-and-cold-dependent-faults).
Where the campaign produced numbers, the proof is arithmetic: the same margin sweep re-run, the failing point compared against the family's healthy line, and long-watch faults out-waited past their old interval — at the site if that is where they lived (power-and-load-provocation-margins-brownout-and-load-steps).
Unsummoned campaigns end honestly: no-fault-found documented with its explored windows, the preventive repair offered as a labelled wager with odds stated, declining as a principled option, and a return path built into the record (the-long-watch-monitoring-and-logging-over-time).
The record is the second product: the case's fixed skeleton filed, and the recipe library grown with the family's recipes, margin pairs, and negatives — so the next twin of this fault starts where this one finished.

Skills Learned

You can now choose the repair the evidence supports — the confirmed mechanism dictating its fix — and resist shotgunning.
You can now design the proof: a verification count that beats the fault's reproduction rate, run on the minimal recipe, across the window's band.
You can now end a failed campaign honestly — a documented no-fault-found, a labelled preventive repair, and an open return path.
You can now write the case record and grow the recipe library that lets the family's next fault start ahead.
You can now run the whole campaign end to end — from complaint to filed proof — as one continuous method.

Glossary Additions

verification count — the number of silent recipe runs (or the watch duration) that proves a repair of an intermittent, designed from the fault's own reproduction rate before the first run: a fault that struck on roughly every third summons needs a couple of dozen quiet runs — not three — before its silence means cure rather than coincidence, and a fault caught by a long watch is proven by the same watch re-run past its old interval. The count is executed on the minimal recipe that most reliably summoned the fault, swept across the window's whole band — hot and cold, loaded and idle — because repairs can move windows instead of closing them, and it is written down with its arithmetic before optimism can negotiate it downward. The recorded count is what separates a verified repair from a lucky quiet spell.
preventive repair — a repair performed on a probable mechanism without a reproduced fault to verify against, legitimate only as a labelled wager: the class is known, the mechanism is likely — aged input electrolytics on a brownout-flavoured complaint, the stressed sibling joints beside a cracked one — but reproduction is out of reach, so the fix cannot be proven by recipe. The label is what keeps it honest: the owner is told it is a repair of the likely suspect with the odds stated plainly, the record files it as unverified, and a return path — a field watch, a symptom to log — is built in, so a wager that loses becomes the next campaign's head start instead of a broken promise.
recipe library — the per-device-family file a bench grows from its closed intermittent cases: the summoning recipes that worked, the soak temperatures and margin pairs (failing voltage against the family's healthy number), the usual mechanisms, the verification counts that proved repairs, and the negatives — explored windows, quiet watches, acquitted nodes. Numbers make the library sharp and prose ages poorly, so entries lead with the figures; the negatives are kept because they spare the next campaign what this one already ruled out. The library is the compounding half of the endgame's two products: the repaired device leaves, the record stays, and the family's next fault opens with a lookup instead of a fortnight.

Suggested Next Sections

Must read next:

Thinking in Trees — Structured Fault Isolation — Chapter 10 closes the volume by turning everything since Chapter 1 into reusable structure: troubleshooting trees that encode a family's diagnostic route into branch-by-branch decisions, built from exactly the records and recipe libraries this section taught you to file.

Recommended:

Thermal Provocation — Forcing Heat- and Cold-Dependent Faults — the summons the verification most often re-runs, band sweep and all.
Power and Load Provocation — Margins, Brownout, and Load Steps — the campaign's numeric provocation, whose margin pairs are the recipe library's sharpest entries.