Cable verdict folded into one console cell (length green, fault name red, xover amber, fail; diag rule lines and counters-reset line gone), reset fires at the press not diag completion, and the measure's blip is suppressed at every counter source instead of counted-hidden-reverted: worker error adds, loss-window write-offs (quiet slides state without charging; leaving the quiet era presumes in-window holes delivered so the gap's tail is never charged late), NIC poller tracks raw without accumulating, socket drops discarded, modules un-primed by the measure itself; display hold deleted so pre-existing errors no longer blink out during a measure

counters.go

@@ -105,13 +105,14 @@ func sumFields(base string, fields []string) uint64 {
 const nicInterval = sampleInterval
 
 type nicPoller struct {
-	txBase string
-	rx     *statReader
-	total  *atomic.Uint64
-	raw    uint64
+	txBase    string
+	rx        *statReader
+	total     *atomic.Uint64
+	measuring *atomic.Bool
+	raw       uint64
 }
 
-func newNICPoller(fd int, txName, rxName string, total *atomic.Uint64) (*nicPoller, error) {
+func newNICPoller(fd int, txName, rxName string, total *atomic.Uint64, measuring *atomic.Bool) (*nicPoller, error) {
 	drv, err := ifDriver(rxName)
 	if err != nil {
 		return nil, err
@@ -125,9 +126,10 @@ func newNICPoller(fd int, txName, rxName string, total *atomic.Uint64) (*nicPoll
 		return nil, err
 	}
 	return &nicPoller{
-		txBase: "/sys/class/net/" + txName,
-		rx:     rx,
-		total:  total,
+		txBase:    "/sys/class/net/" + txName,
+		rx:        rx,
+		total:     total,
+		measuring: measuring,
 	}, nil
 }
 
@@ -142,10 +144,12 @@ func (p *nicPoller) read() uint64 {
 // Nic counters run from boot and restart from zero whenever the driver resets
 // its statistics, so only their forward motion is accumulated. Taking raw
 // differences instead charges a boot's worth of errors to the first sample and
-// turns a reset into a near-2^64 underflow.
+// turns a reset into a near-2^64 underflow. The hardware cannot pause its
+// counters for a cable measure, so its blip is suppressed here at the read:
+// the baseline tracks the raw values without accumulating until it is over.
 func (p *nicPoller) poll() {
 	raw := p.read()
-	if raw > p.raw {
+	if raw > p.raw && !p.measuring.Load() {
 		p.total.Add(raw - p.raw)
 	}
 	p.raw = raw

docs/state.md

@@ -75,5 +75,5 @@ In `~/work/` alongside the phydiag artifacts, ready to fold into the repo's `ker
 ## Standing goals
 
 - **Length is a product goal again** — the FS (ECD length) rides alongside the Wiitek (SNR) in the test pair, which the tree implements. The Fibergaga (Aquantia 1E.C884 ±1 m + 1E.C800 verdicts, documented) remains the alternate length path if the pair mix ever changes.
-- **BCM ECD works** (recipe recovered from the OpenBCM SDK, validated on the FS — modules/fs/): per-pair lengths meter-accurate against a known ~45 m cable. cabletest runs it at bringup and on every reset, re-baselining counters after the relink so the blip is never charged (`phy.go`). Remaining work is characterizing the link blip the run causes (length stays a between-measurements operation until then — open-questions).
+- **BCM ECD works** (recipe recovered from the OpenBCM SDK, validated on the FS — modules/fs/): per-pair lengths meter-accurate against a known ~45 m cable. cabletest runs it at bringup and on every reset; while a measure runs, every failure counter is suppressed at its source (worker error adds, loss-window write-offs, NIC-poller accumulation, socket drops, module latch priming), so the blip is never counted anywhere rather than counted and reverted (`phy.go` `cableDiag.measuring`). Remaining work is characterizing the link blip the run causes (length stays a between-measurements operation until then — open-questions).
 - **Pre-FEC verification** on the Aquantia — counters documented; needs the graded-noise correlation run (design: modules/fibergaga/).

loss.go

@@ -12,12 +12,13 @@ const (
 )
 
 type lossWindow struct {
-	mu     sync.Mutex
-	base   uint64
-	inited bool
-	bits   []uint64
-	lost   atomic.Uint64
-	late   atomic.Uint64
+	mu       sync.Mutex
+	base     uint64
+	inited   bool
+	quietEra bool
+	bits     []uint64
+	lost     atomic.Uint64
+	late     atomic.Uint64
 
 	// The sending half of this stream. Both halves are in this process and the
 	// receiving socket ignores its own outgoing frames, so this window is fed by
@@ -47,11 +48,24 @@ func newLossWindows(tx []*txStats) []lossWindow {
 // the run. Nothing that arrives later is evidence enough to undo that, which is
 // why the sender's own frontier is the guard rather than a limit on how far a
 // single step may move.
-func (w *lossWindow) observe(seq uint64) bool {
+// quiet slides the window without charging: gaps written off during a cable
+// measure are its own link blip, suppressed at the source while the state
+// machine stays in sync with the wire.
+func (w *lossWindow) observe(seq uint64, quiet bool) bool {
 	if seq >= w.sent.Load() {
 		return false
 	}
 	w.mu.Lock()
+	if w.quietEra && !quiet {
+		// Leaving a measure: the window was full when it began, so every hole
+		// still inside went missing while failures were suppressed. Marking
+		// them delivered keeps the write-off complete however the eviction
+		// timing falls, instead of charging the gap's in-window tail later.
+		for i := range w.bits {
+			w.bits[i] = ^uint64(0)
+		}
+	}
+	w.quietEra = quiet
 	if !w.inited {
 		// Start half a window below the first sequence seen, so anything the
 		// sender put on the wire before it lands inside the window rather than
@@ -63,11 +77,13 @@ func (w *lossWindow) observe(seq uint64) bool {
 	}
 	if seq < w.base {
 		w.mu.Unlock()
-		w.late.Add(1)
+		if !quiet {
+			w.late.Add(1)
+		}
 		return true
 	}
 	if seq >= w.base+lossSlots {
-		w.evict(seq - lossSlots + 1)
+		w.evict(seq-lossSlots+1, quiet)
 	}
 	idx := seq & (lossSlots - 1)
 	w.bits[idx>>6] |= 1 << (idx & 63)
@@ -75,7 +91,7 @@ func (w *lossWindow) observe(seq uint64) bool {
 	return true
 }
 
-func (w *lossWindow) evict(newBase uint64) {
+func (w *lossWindow) evict(newBase uint64, quiet bool) {
 	span := newBase - w.base
 	if span >= lossSlots {
 		var missing uint64
@@ -83,7 +99,9 @@ func (w *lossWindow) evict(newBase uint64) {
 			missing += uint64(64 - bits.OnesCount64(w.bits[i]))
 			w.bits[i] = 0
 		}
-		w.lost.Add(missing + span - lossSlots)
+		if !quiet {
+			w.lost.Add(missing + span - lossSlots)
+		}
 		w.base = newBase
 		return
 	}
@@ -97,6 +115,8 @@ func (w *lossWindow) evict(newBase uint64) {
 		}
 		w.bits[word] &^= bit
 	}
-	w.lost.Add(missing)
+	if !quiet {
+		w.lost.Add(missing)
+	}
 	w.base = newBase
 }

loss_test.go

@@ -14,7 +14,7 @@ func newWindow(sent uint64) *lossWindow {
 func TestLossWindowContiguousLosesNothing(t *testing.T) {
 	w := newWindow(lossSlots + 1000)
 	for seq := uint64(0); seq < lossSlots+1000; seq++ {
-		w.observe(seq)
+		w.observe(seq, false)
 	}
 	if got := w.lost.Load(); got != 0 {
 		t.Errorf("lost = %d, want 0", got)
@@ -30,7 +30,7 @@ func TestLossWindowCountsGapOnceEvicted(t *testing.T) {
 		if seq == 100 {
 			continue
 		}
-		w.observe(seq)
+		w.observe(seq, false)
 	}
 	if got := w.lost.Load(); got != 1 {
 		t.Errorf("lost = %d, want 1", got)
@@ -42,13 +42,13 @@ func TestLossWindowCountsGapOnceEvicted(t *testing.T) {
 func TestLossWindowOutOfOrderIsNotLoss(t *testing.T) {
 	w := newWindow(lossSlots + 1000)
 	for seq := uint64(99); ; seq-- {
-		w.observe(seq)
+		w.observe(seq, false)
 		if seq == 0 {
 			break
 		}
 	}
 	for seq := uint64(100); seq < lossSlots+1000; seq++ {
-		w.observe(seq)
+		w.observe(seq, false)
 	}
 	if got := w.lost.Load(); got != 0 {
 		t.Errorf("lost = %d, want 0", got)
@@ -59,8 +59,8 @@ func TestLossWindowOutOfOrderIsNotLoss(t *testing.T) {
 // the window held plus the sequences that never landed in it at all.
 func TestLossWindowJumpBeyondWindow(t *testing.T) {
 	w := newWindow(200001)
-	w.observe(0)
-	w.observe(200000)
+	w.observe(0, false)
+	w.observe(200000, false)
 
 	// Everything below the new base except the one sequence that was seen.
 	want := uint64(200000 - lossSlots + 1 - 1)
@@ -74,8 +74,8 @@ func TestLossWindowJumpBeyondWindow(t *testing.T) {
 
 func TestLossWindowBelowBaseIsLate(t *testing.T) {
 	w := newWindow(100001)
-	w.observe(100000)
-	w.observe(1000)
+	w.observe(100000, false)
+	w.observe(1000, false)
 	if got := w.late.Load(); got != 1 {
 		t.Errorf("late = %d, want 1", got)
 	}
@@ -88,7 +88,7 @@ func TestLossWindowBelowBaseIsLate(t *testing.T) {
 // another worker is still holding land inside rather than arriving late.
 func TestLossWindowStartsHalfAWindowBack(t *testing.T) {
 	w := newWindow(100001)
-	w.observe(100000)
+	w.observe(100000, false)
 	if w.base != 100000-lossSlots/2 {
 		t.Errorf("base = %d, want %d", w.base, 100000-lossSlots/2)
 	}
@@ -103,11 +103,11 @@ func TestLossWindowRefusesUnsentSeq(t *testing.T) {
 	const first = 100000
 	w := newWindow(first + 2000)
 	for seq := uint64(first); seq < first+1000; seq++ {
-		w.observe(seq)
+		w.observe(seq, false)
 	}
 	base := w.base
 
-	if w.observe(1 << 62) {
+	if w.observe(1<<62, false) {
 		t.Error("a sequence number far past the sender's frontier was accepted")
 	}
 	if got := w.lost.Load(); got != 0 {
@@ -121,7 +121,7 @@ func TestLossWindowRefusesUnsentSeq(t *testing.T) {
 	// Still tracking the real traffic, rather than reporting every frame late
 	// against a base that ran away.
 	for seq := uint64(first + 1000); seq < first+2000; seq++ {
-		w.observe(seq)
+		w.observe(seq, false)
 	}
 	if got := w.late.Load(); got != 0 {
 		t.Errorf("late = %d, want 0", got)
@@ -131,14 +131,42 @@ func TestLossWindowRefusesUnsentSeq(t *testing.T) {
 	}
 }
 
+// A gap written off while quiet — a cable measure's own link blip — charges
+// nothing, however the eviction timing falls: even the gap's tail still
+// inside the window when the measure ends is presumed delivered. Counting
+// resumes seamlessly and later real gaps are still caught.
+func TestLossWindowQuietSlidesWithoutCharging(t *testing.T) {
+	w := newWindow(4 * lossSlots)
+	for seq := uint64(0); seq < 100; seq++ {
+		w.observe(seq, false)
+	}
+	// The blip: a huge jump arriving during the measure.
+	w.observe(2*lossSlots, true)
+	if got := w.lost.Load(); got != 0 {
+		t.Errorf("lost = %d after a quiet write-off, want 0", got)
+	}
+
+	// The measure ends immediately — the worst case, with the gap's tail
+	// still in-window — and a single real gap follows.
+	for seq := uint64(2*lossSlots + 1); seq < 4*lossSlots-100; seq++ {
+		if seq == 2*lossSlots+500 {
+			continue
+		}
+		w.observe(seq, false)
+	}
+	if got := w.lost.Load(); got != 1 {
+		t.Errorf("lost = %d, want exactly the one real post-measure gap", got)
+	}
+}
+
 // The sender's own frontier is the bound, so the sequence one past it is
 // refused while the one below it is not.
 func TestLossWindowBoundIsExclusive(t *testing.T) {
 	w := newWindow(500)
-	if !w.observe(499) {
+	if !w.observe(499, false) {
 		t.Error("the last sequence the sender put on the wire was refused")
 	}
-	if w.observe(500) {
+	if w.observe(500, false) {
 		t.Error("a sequence the sender had not reached was accepted")
 	}
 }

main.go

@@ -39,6 +39,10 @@ type direction struct {
 	rxFDs   []int
 	statFD  int
 
+	// While a cable measure runs, every failure counter in this direction is
+	// suppressed at its source rather than counted, hidden and reverted.
+	measuring *atomic.Bool
+
 	// Guards everything the sampler touches. The counters are read on their own
 	// clock and drawn on another, and the two must not read them at once:
 	// sampleDrops consumes what it reads, so a second caller would see a gap.
@@ -250,22 +254,26 @@ func (d *direction) reset() {
 }
 
 // Returns the new start time, so the uptime shown alongside the totals counts
-// from the reset rather than from launch.
-func resetAll(dirs []*direction, mods []*phyModule, stats *streamTable) time.Time {
+// from the reset rather than from launch; the elapsed clock restarting is the
+// visible mark of the re-baseline.
+func resetAll(dirs []*direction, mods []*phyModule) time.Time {
 	for _, d := range dirs {
 		d.reset()
 	}
 	for _, m := range mods {
 		m.reset()
 	}
-	stats.sinceHeader = 0
-	fmt.Println(stats.rule("counters reset"))
 	return time.Now()
 }
 
+// The socket statistic is read-and-clear, so it is always consumed; a drop
+// during a cable measure is the blip's and is discarded at this source.
 func (d *direction) sampleDrops() {
 	for _, fd := range d.rxFDs {
-		d.drops += packetDrops(fd)
+		n := packetDrops(fd)
+		if !d.measuring.Load() {
+			d.drops += n
+		}
 	}
 }
 
@@ -286,7 +294,7 @@ var intervalCols = []colSpec{
 	{group: "OVERALL", title: "elapsed", width: 9, right: true},
 	{group: "OVERALL", title: "packets", width: 9, right: true},
 	{group: "OVERALL", title: "bytes", width: 9, right: true},
-	{group: "OVERALL", title: "metres", width: 6, right: true},
+	{group: "OVERALL", title: "cable", width: 6, right: true},
 	{group: "OVERALL", title: "corrected", width: 9, right: true},
 	{group: "OVERALL", title: "lost", width: 9, right: true},
 	{group: "OVERALL", title: "corrupt", width: 9, right: true},
@@ -331,17 +339,11 @@ func (d *direction) counters(now counterSet) view {
 	}
 }
 
-// Errors seen while a measure is in flight are the diag's own link blip and
-// are re-based away at its completion; until then they are held off the
-// display rather than shown as the cable's.
-func measureView(diag *cableDiag, modules []*phyModule, v view) (view, phyDisplay) {
+// Nothing to hold or revert here: failures during a measure were never
+// counted, so the view is always the counters as they stand.
+func phyView(diag *cableDiag, modules []*phyModule) phyDisplay {
 	info, measuring := diag.snapshot()
-	phy := phyDisplayFrom(info, measuring, modules[0].view(), modules[1].view())
-	if measuring {
-		v.window, v.since = errs{}, errs{}
-		phy.corrected, phy.recent = 0, 0
-	}
-	return v, phy
+	return phyDisplayFrom(info, measuring, modules[0].view(), modules[1].view())
 }
 
 func totalView(views []view) view {
@@ -417,7 +419,7 @@ func (d *direction) primeCounters() {
 	d.reset()
 }
 
-func buildDirection(label string, tx, rx endpoint) (*direction, error) {
+func buildDirection(label string, tx, rx endpoint, measuring *atomic.Bool) (*direction, error) {
 	// Built before the windows, since each window judges sequence numbers against
 	// the frontier its own sender publishes.
 	txs := make([]*txStats, numStreams)
@@ -425,8 +427,9 @@ func buildDirection(label string, tx, rx endpoint) (*direction, error) {
 		txs[i] = &txStats{}
 	}
 	d := &direction{
-		txStats: txs,
-		streams: newLossWindows(txs),
+		txStats:   txs,
+		streams:   newLossWindows(txs),
+		measuring: measuring,
 	}
 	// Held open for the life of the run: the stats ioctl is issued five times a
 	// second and reopening a socket for each one is pure overhead.
@@ -436,7 +439,7 @@ func buildDirection(label string, tx, rx endpoint) (*direction, error) {
 	}
 	d.statFD = statFD
 
-	d.poller, err = newNICPoller(statFD, tx.name, rx.name, &d.nic)
+	d.poller, err = newNICPoller(statFD, tx.name, rx.name, &d.nic, measuring)
 	if err != nil {
 		return nil, fmt.Errorf("%s: %w", label, err)
 	}
@@ -466,12 +469,13 @@ func buildDirection(label string, tx, rx endpoint) (*direction, error) {
 func (d *direction) start(wg *sync.WaitGroup, done *atomic.Bool, rxReady *sync.WaitGroup, startTx <-chan struct{}) {
 	for i, fd := range d.txFDs {
 		w := &txWorker{
-			fd:      fd,
-			stream:  uint16(i),
-			spec:    d.specs[i],
-			batch:   batchSize,
-			stats:   d.txStats[i],
-			startTx: startTx,
+			fd:        fd,
+			stream:    uint16(i),
+			spec:      d.specs[i],
+			batch:     batchSize,
+			stats:     d.txStats[i],
+			measuring: d.measuring,
+			startTx:   startTx,
 		}
 		wg.Add(1)
 		go func() {
@@ -482,13 +486,14 @@ func (d *direction) start(wg *sync.WaitGroup, done *atomic.Bool, rxReady *sync.W
 	}
 	for i, fd := range d.rxFDs {
 		w := &rxWorker{
-			fd:     fd,
-			batch:  batchSize,
-			stream: uint16(i),
-			spec:   d.specs[i],
-			stats:  d.rxStats[i],
-			loss:   &d.streams[i],
-			ready:  rxReady,
+			fd:        fd,
+			batch:     batchSize,
+			stream:    uint16(i),
+			spec:      d.specs[i],
+			stats:     d.rxStats[i],
+			loss:      &d.streams[i],
+			measuring: d.measuring,
+			ready:     rxReady,
 		}
 		wg.Add(1)
 		go func() {
@@ -657,7 +662,7 @@ func run(aName, bName string) (err error) {
 
 	var dirs []*direction
 	for _, p := range [][2]endpoint{{a, b}, {b, a}} {
-		d, err := buildDirection(p[0].name+"->"+p[1].name, p[0], p[1])
+		d, err := buildDirection(p[0].name+"->"+p[1].name, p[0], p[1], &diag.measuring)
 		if err != nil {
 			return err
 		}
@@ -768,10 +773,18 @@ func run(aName, bName string) (err error) {
 	}
 
 	start := time.Now()
+	// A reset resets at the press, then re-measures; the measure's failures
+	// are suppressed at their sources while it runs, so there is nothing to
+	// hide or revert afterwards.
+	kickMeasure := func() {
+		if diag.kick(&done) {
+			start = resetAll(dirs, modules)
+		}
+	}
 	close(startTx)
 	// The first measure rides the same async path as a reset, so startup never
-	// waits on it; counters re-baseline when its link blip is over.
-	diag.kick(&done)
+	// waits on it.
+	kickMeasure()
 	tick := time.NewTicker(reportInterval)
 	defer tick.Stop()
 
@@ -784,32 +797,21 @@ func run(aName, bName string) (err error) {
 			return fmt.Errorf("%v", p)
 		case <-sig:
 			return nil
-		// A reset re-measures the cable first; the counters re-baseline at diag
-		// completion, so its link blip is never charged to the fresh run.
+		// The verdict lives in the cable cell, not its own line.
 		case <-space:
-			if diag.kick(&done) {
-				fmt.Println(stats.rule("measuring cable"))
-			}
-		case err := <-diag.completed:
-			if err != nil {
-				fmt.Println(stats.rule("cable diag failed: " + err.Error()))
-			} else {
-				info, _ := diag.snapshot()
-				fmt.Println(stats.rule("cable diag: " + cableLine(info)))
-			}
-			start = resetAll(dirs, modules, stats)
+			kickMeasure()
 		case <-disp.fb.flips:
 			now := time.Now()
 			px, py, down := touch.get()
 			x, y := disp.fb.fromPanel(px, py)
 			if disp.holdReset(x, y, down, now) {
-				diag.kick(&done)
+				kickMeasure()
 			}
 			disp.showVersion = down && disp.versionSpot.contains(x, y)
 			for i, d := range dirs {
 				views[i] = d.displayView()
 			}
-			v, phy := measureView(diag, modules, totalView(views))
+			v, phy := totalView(views), phyView(diag, modules)
 			if err := disp.render(v, now.Sub(start), phy,
 				noise.view()); err != nil {
 				return err
@@ -819,7 +821,7 @@ func run(aName, bName string) (err error) {
 			for i, d := range dirs {
 				rows[i] = d.displayView()
 			}
-			v, phy := measureView(diag, modules, totalView(rows))
+			v, phy := totalView(rows), phyView(diag, modules)
 			for _, m := range modules {
 				for _, n := range m.takeNotes() {
 					fmt.Println(stats.rule(n))

phy.go

@@ -769,6 +769,15 @@ func (m *phyModule) reset() {
 	m.mu.Unlock()
 }
 
+// Latches drained after a measure belong to its blip: un-priming makes the
+// next poll an origin only, without discarding the pre-measure totals.
+func (m *phyModule) forgive() {
+	m.mu.Lock()
+	m.primed = false
+	m.recentDelta = 0
+	m.mu.Unlock()
+}
+
 type phyModView struct {
 	fresh   bool
 	link    bool
@@ -802,6 +811,7 @@ type cableInfo struct {
 	ecd      ecdResult
 	maps     [2]byte
 	haveMaps [2]bool
+	failed   bool
 }
 
 func (c cableInfo) metresString() string {
@@ -845,8 +855,6 @@ type phyDisplay struct {
 	metresClass int
 }
 
-func pairLetter(i int) string { return string(rune('A' + i)) }
-
 // Each end resolves MDI on its own, so a swap at either known end counts; an
 // end with no readable map abstains.
 func pairSwapped(i int, c cableInfo) bool {
@@ -858,32 +866,35 @@ func pairSwapped(i int, c cableInfo) bool {
 	return false
 }
 
+// The whole cable verdict in one cell: the length when healthy, the first
+// faulted pair's verdict when not, "xover" for a pair swap, "fail" for a diag
+// that never delivered.
 func cableSummary(cable cableInfo, measuring bool) (string, int) {
 	if measuring {
-		return "...", clsNone
+		return "-", clsNone
 	}
-	anyData, anyFault, anySwap := false, false, false
+	if cable.failed {
+		return "fail", clsBad
+	}
+	anyData, anySwap := false, false
 	for i, v := range cable.ecd.verdicts {
 		if v != 0 {
 			anyData = true
 		}
 		if v != 0 && v != pairOK {
-			anyFault = true
+			return pairVerdicts[v], clsBad
 		}
 		if pairSwapped(i, cable) {
 			anySwap = true
 		}
 	}
-	s := cable.metresString()
 	switch {
 	case !anyData:
 		return "-", clsNone
-	case anyFault:
-		return s, clsBad
 	case anySwap:
-		return s, clsWarn
+		return "xover", clsWarn
 	}
-	return s, clsGood
+	return cable.metresString(), clsGood
 }
 
 // The margin is the worst pair across the ends that measure SNR (the Wiitek's
@@ -977,6 +988,7 @@ func measureCable(mods []*phyModule, waitRelink bool, done *atomic.Bool) (cableI
 	}
 	defer func() {
 		for _, m := range mods {
+			m.forgive()
 			m.busy.Store(false)
 		}
 	}()
@@ -1010,83 +1022,45 @@ func measureCable(mods []*phyModule, waitRelink bool, done *atomic.Bool) (cableI
 }
 
 type cableDiag struct {
-	mods      []*phyModule
-	completed chan error
+	mods []*phyModule
 
-	mu      sync.Mutex
-	info    cableInfo
-	running bool
+	// While set, every failure counter is suppressed at its source: the
+	// measure's own link blip is never counted anywhere, rather than counted,
+	// hidden and reverted.
+	measuring atomic.Bool
+
+	mu   sync.Mutex
+	info cableInfo
 }
 
 func newCableDiag(mods []*phyModule, info cableInfo) *cableDiag {
-	return &cableDiag{mods: mods, completed: make(chan error, 1), info: info}
+	return &cableDiag{mods: mods, info: info}
 }
 
 func (c *cableDiag) snapshot() (cableInfo, bool) {
 	c.mu.Lock()
 	defer c.mu.Unlock()
-	return c.info, c.running
+	return c.info, c.measuring.Load()
 }
 
 func (c *cableDiag) kick(done *atomic.Bool) bool {
-	c.mu.Lock()
-	if c.running {
-		c.mu.Unlock()
+	if !c.measuring.CompareAndSwap(false, true) {
 		return false
 	}
-	c.running = true
-	c.mu.Unlock()
-
 	go func() {
 		defer holdPanic()
 		info, _, err := measureCable(c.mods, true, done)
 		if err != nil {
-			info = cableInfo{}
+			info = cableInfo{failed: true}
 		}
 		c.mu.Lock()
 		c.info = info
-		c.running = false
 		c.mu.Unlock()
-		select {
-		case c.completed <- err:
-		default:
-		}
+		c.measuring.Store(false)
 	}()
 	return true
 }
 
-func mapString(m byte, have bool) string {
-	if !have {
-		return "unread"
-	}
-	if m == pairIdentityMap {
-		return "straight"
-	}
-	out := make([]string, 4)
-	for i := range out {
-		out[i] = pairLetter(int(m>>(2*i)) & 3)
-	}
-	return "swapped to " + strings.Join(out, "")
-}
-
-func verdictString(r ecdResult) string {
-	bad := []string{}
-	for i, v := range r.verdicts {
-		if v != pairOK {
-			s, ok := pairVerdicts[v]
-			if !ok {
-				s = fmt.Sprintf("%d", v)
-			}
-			bad = append(bad, fmt.Sprintf("%s %s at %dm", pairLetter(i), s, r.metres[i]))
-		}
-	}
-	if len(bad) > 0 {
-		return strings.Join(bad, ", ")
-	}
-	return fmt.Sprintf("all pairs ok, %d/%d/%d/%d m",
-		r.metres[0], r.metres[1], r.metres[2], r.metres[3])
-}
-
 func openModules(names [2]string) ([]*phyModule, [2]string, error) {
 	mods := make([]*phyModule, 0, 2)
 	var idents [2]string
@@ -1218,7 +1192,3 @@ func moduleChecks(mods []*phyModule, names [2]string) []checkResult {
 	return out
 }
 
-func cableLine(c cableInfo) string {
-	return fmt.Sprintf("%s; map %s / %s",
-		verdictString(c.ecd), mapString(c.maps[0], c.haveMaps[0]), mapString(c.maps[1], c.haveMaps[1]))
-}

phy_test.go

@@ -19,8 +19,8 @@ func TestPhyDisplayWorstMarginAndFault(t *testing.T) {
 	if !d.haveSNR || d.worstMargin != 0.5 {
 		t.Errorf("worstMargin = %v (have %v), want 0.5", d.worstMargin, d.haveSNR)
 	}
-	if d.metres != "44" || d.metresClass != clsBad {
-		t.Errorf("metres = %q class %d, want the healthy mean 44 painted bad", d.metres, d.metresClass)
+	if d.metres != "OPEN" || d.metresClass != clsBad {
+		t.Errorf("cable cell = %q class %d, want the fault named bad", d.metres, d.metresClass)
 	}
 }
 
@@ -51,10 +51,13 @@ func TestCableSummary(t *testing.T) {
 		class     int
 	}{
 		{"clean", cableInfo{ecd: healthy, maps: [2]byte{pairIdentityMap, pairIdentityMap}, haveMaps: [2]bool{true, true}}, false, "49", clsGood},
-		{"far-end swap", cableInfo{ecd: healthy, maps: [2]byte{pairIdentityMap, 0xE1}, haveMaps: [2]bool{true, true}}, false, "49", clsWarn},
+		{"far-end swap", cableInfo{ecd: healthy, maps: [2]byte{pairIdentityMap, 0xE1}, haveMaps: [2]bool{true, true}}, false, "xover", clsWarn},
 		{"one-end map only", cableInfo{ecd: healthy, maps: [2]byte{pairIdentityMap, 0}, haveMaps: [2]bool{true, false}}, false, "49", clsGood},
+		{"fault", cableInfo{ecd: ecdResult{verdicts: [4]int{pairOK, pairShort, pairXtalk, pairOK}, metres: [4]int{50, 12, 30, 51}}}, false, "SHORT", clsBad},
+		{"fault beats swap", cableInfo{ecd: ecdResult{verdicts: [4]int{pairOpen, pairOK, pairOK, pairOK}}, maps: [2]byte{pairIdentityMap, 0xE1}, haveMaps: [2]bool{true, true}}, false, "OPEN", clsBad},
+		{"diag failed", cableInfo{failed: true}, false, "fail", clsBad},
 		{"no diag yet", cableInfo{}, false, "-", clsNone},
-		{"measuring", cableInfo{ecd: healthy, maps: [2]byte{pairIdentityMap, pairIdentityMap}, haveMaps: [2]bool{true, true}}, true, "...", clsNone},
+		{"measuring", cableInfo{ecd: healthy, maps: [2]byte{pairIdentityMap, pairIdentityMap}, haveMaps: [2]bool{true, true}}, true, "-", clsNone},
 	} {
 		s, cl := cableSummary(c.cable, c.measuring)
 		if s != c.want || cl != c.class {
@@ -104,13 +107,3 @@ func TestSNRClass(t *testing.T) {
 	}
 }
 
-func TestVerdictString(t *testing.T) {
-	ok := ecdResult{verdicts: [4]int{1, 1, 1, 1}, metres: [4]int{45, 45, 41, 46}}
-	if got := verdictString(ok); got != "all pairs ok, 45/45/41/46 m" {
-		t.Errorf("healthy = %q", got)
-	}
-	bad := ecdResult{verdicts: [4]int{1, 2, 4, 1}, metres: [4]int{45, 12, 30, 46}}
-	if got := verdictString(bad); got != "B OPEN at 12m, C XTALK at 30m" {
-		t.Errorf("faulted = %q", got)
-	}
-}

rx.go

@@ -110,13 +110,14 @@ func fillBack(frames, bytes []uint64) {
 }
 
 type rxWorker struct {
-	fd     int
-	batch  int
-	stream uint16
-	spec   *frameSpec
-	stats  *rxStats
-	loss   *lossWindow
-	ready  *sync.WaitGroup
+	fd        int
+	batch     int
+	stream    uint16
+	spec      *frameSpec
+	stats     *rxStats
+	loss      *lossWindow
+	measuring *atomic.Bool
+	ready     *sync.WaitGroup
 }
 
 func (w *rxWorker) run(done *atomic.Bool) {
@@ -132,9 +133,12 @@ func (w *rxWorker) run(done *atomic.Bool) {
 	w.ready.Done()
 
 	for !done.Load() {
+		// One flag load covers the batch: failures seen while a cable measure
+		// runs are its own link blip and are never counted.
+		quiet := w.measuring.Load()
 		n, err := recvmmsg(w.fd, hdrs, unix.MSG_WAITFORONE)
 		if n <= 0 {
-			if err != nil && err != unix.EAGAIN && err != unix.EINTR {
+			if err != nil && err != unix.EAGAIN && err != unix.EINTR && !quiet {
 				w.stats.rxErrs.Add(1)
 			}
 			continue
@@ -145,6 +149,9 @@ func (w *rxWorker) run(done *atomic.Bool) {
 			buf := bufs[i][:int(hdrs[i].len)]
 			p, st := parseHeader(buf)
 			if st != hdrOK {
+				if quiet {
+					continue
+				}
 				if st == hdrForeign {
 					w.stats.badMagic.Add(1)
 				} else {
@@ -158,18 +165,22 @@ func (w *rxWorker) run(done *atomic.Bool) {
 			// The ethertype this socket is bound to already says which stream the
 			// frame belongs to, so a header naming another one is damaged, as is a
 			// sequence number the sender never reached.
-			if p.stream != w.stream || !w.loss.observe(p.seq) {
-				w.stats.badHdr.Add(1)
+			if p.stream != w.stream || !w.loss.observe(p.seq, quiet) {
+				if !quiet {
+					w.stats.badHdr.Add(1)
+				}
 				continue
 			}
 
 			want, ok := w.spec.expectedCRC(p.patIdx, p.payLen)
 			if !ok {
-				w.stats.badLen.Add(1)
+				if !quiet {
+					w.stats.badLen.Add(1)
+				}
 				continue
 			}
 			pay := buf[minFrame : minFrame+p.payLen]
-			if crc32.Checksum(pay, crcTable) != want {
+			if crc32.Checksum(pay, crcTable) != want && !quiet {
 				w.stats.crcErr.Add(1)
 			}
 		}

tx.go

@@ -17,12 +17,13 @@ type txStats struct {
 }
 
 type txWorker struct {
-	fd      int
-	stream  uint16
-	spec    *frameSpec
-	batch   int
-	stats   *txStats
-	startTx <-chan struct{}
+	fd        int
+	stream    uint16
+	spec      *frameSpec
+	batch     int
+	stats     *txStats
+	measuring *atomic.Bool
+	startTx   <-chan struct{}
 }
 
 func (w *txWorker) run(done *atomic.Bool) {
@@ -63,7 +64,8 @@ func (w *txWorker) run(done *atomic.Bool) {
 		}
 		// Taking fewer of the vector than offered is the ring's room, not a frame
 		// lost: the rest go on the next pass.
-		if n < 0 && err != unix.EINTR && err != unix.EAGAIN && err != unix.ENOBUFS {
+		if n < 0 && err != unix.EINTR && err != unix.EAGAIN && err != unix.ENOBUFS &&
+			!w.measuring.Load() {
 			w.stats.errs.Add(1)
 		}
 	}

ui.go

@@ -386,10 +386,12 @@ func (d *display) errCounts(x, w, y int, e errs) int {
 
 func metresStat(phy phyDisplay) statCell {
 	col := uiFg
+	unit := "m"
 	if phy.metresClass != clsGood {
 		col = classColor(phy.metresClass)
+		unit = "cable"
 	}
-	return statCell{phy.metres, "m", col}
+	return statCell{phy.metres, unit, col}
 }
 
 func (d *display) panel(p rect, bad bool) (int, int) {