feat: autotune PSU capacity from the same full-load run as fan ceilings

Extract the fan peak-tracking into observedPeakStore (observe max under
load, hold >= minHold to reject spikes, round, persist JSON) and add a
second instance for PSU draw (psu-observation.json, keyed by PSU
ordinal). Fed from samplePSUPower like fans are from sampleFanSpeeds, so
any full-load run refines it — the Fan Ceiling Check (which also samples
PSU power at a slow cadence off its loop and writes psu_<i>_peak_w), a
burn, thermal cycling, and the 5s web metrics collector.

/topo PSU cards now scale the load fill by wattage_w when the BMC
reports it, else by the observed peak draw — marked "~N% load". This
MSI stand's BMC gives only instantaneous input power, so the observed
peak is the only capacity figure available.

Fan behaviour is unchanged (tests exercise updateFanObservation /
estimateFanDutyCyclePctFromObservation / ResolveFanMaxRPM through the
new store).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019VHG21rgTUiR1G3qFHTVmN

audit/internal/platform/sat_fan_stress_test.go

@@ -8,15 +8,20 @@ import (
 	"time"
 )
 
+// resetPeakStore points a store at a fresh temp file and clears its cache for
+// the duration of the test.
+func resetPeakStore(t *testing.T, s *observedPeakStore) {
+	t.Helper()
+	old := *s
+	s.path = filepath.Join(t.TempDir(), "peaks.json")
+	s.loaded = false
+	s.peaks = nil
+	s.candidates = nil
+	t.Cleanup(func() { *s = old })
+}
+
 func TestResolveFanMaxRPM(t *testing.T) {
-	oldPath := fanObservationStatePath
-	oldInit := fanObservationInit
-	fanObservationStatePath = filepath.Join(t.TempDir(), "fan-observation.json")
-	fanObservationInit = false
-	t.Cleanup(func() {
-		fanObservationStatePath = oldPath
-		fanObservationInit = oldInit
-	})
+	resetPeakStore(t, fanPeaks)
 
 	// No persisted file yet: unknown fans fall back to their own current RPM.
 	got := ResolveFanMaxRPM(map[string]float64{"A": 4000, "B": 9000})
@@ -24,7 +29,7 @@ func TestResolveFanMaxRPM(t *testing.T) {
 		t.Fatalf("no-persist fallback: got %v", got)
 	}
 
-	if err := os.WriteFile(fanObservationStatePath, []byte(`{"max_rpm":{"A":17000}}`), 0644); err != nil {
+	if err := os.WriteFile(fanPeaks.path, []byte(`{"max_rpm":{"A":17000}}`), 0644); err != nil {
 		t.Fatal(err)
 	}
 	got = ResolveFanMaxRPM(map[string]float64{"A": 4000, "B": 9000, "C": 5000})
@@ -34,6 +39,49 @@ func TestResolveFanMaxRPM(t *testing.T) {
 	}
 }
 
+func TestObservedPeakStore(t *testing.T) {
+	s := &observedPeakStore{
+		path:    filepath.Join(t.TempDir(), "peaks.json"),
+		jsonKey: "max_w",
+		roundUp: 50,
+		minHold: time.Second,
+	}
+	t0 := time.Unix(0, 0)
+
+	// A single spike does not stick.
+	s.observe(map[string]float64{"0": 900}, t0)
+	if len(s.snapshot()) != 0 {
+		t.Fatalf("transient spike should not persist: %v", s.snapshot())
+	}
+	// Held past minHold → sticks, rounded up to the next 50.
+	s.observe(map[string]float64{"0": 920}, t0.Add(1200*time.Millisecond))
+	if got := s.snapshot()["0"]; got != 950 {
+		t.Fatalf("held peak: got %v want 950", got)
+	}
+	// A lower reading never lowers the peak.
+	s.observe(map[string]float64{"0": 400}, t0.Add(5*time.Second))
+	if got := s.snapshot()["0"]; got != 950 {
+		t.Fatalf("peak must not drop: got %v", got)
+	}
+	// Fresh store reloads from disk.
+	s2 := &observedPeakStore{path: s.path, jsonKey: "max_w"}
+	if got := s2.snapshot()["0"]; got != 950 {
+		t.Fatalf("reload from disk: got %v want 950", got)
+	}
+}
+
+func TestUpdatePSUObservationKeyedByOrdinal(t *testing.T) {
+	resetPeakStore(t, psuPeaks)
+	now := time.Unix(0, 0)
+	psus := []PSUReading{{Name: "PSU1", PowerW: 1200}, {Name: "PSU2", PowerW: 1400}}
+	updatePSUObservation(psus, now)
+	updatePSUObservation(psus, now.Add(1200*time.Millisecond))
+	got := ObservedPSUMaxW()
+	if got["0"] != 1200 || got["1"] != 1400 {
+		t.Fatalf("keyed-by-ordinal peaks: got %v", got)
+	}
+}
+
 func TestApplyFanCheckDefaults(t *testing.T) {
 	var o FanCheckOptions
 	applyFanCheckDefaults(&o)
@@ -109,22 +157,7 @@ func TestParseFanDutyCyclePctSensorsJSON(t *testing.T) {
 }
 
 func TestEstimateFanDutyCyclePctFromObservation(t *testing.T) {
-	t.Parallel()
-
-	oldPath := fanObservationStatePath
-	oldState := fanObservation
-	oldInit := fanObservationInit
-	oldCandidates := fanPeakCandidates
-	fanObservationStatePath = filepath.Join(t.TempDir(), "fan-observation.json")
-	fanObservation = fanObservationState{}
-	fanObservationInit = false
-	fanPeakCandidates = make(map[string]fanPeakCandidate)
-	t.Cleanup(func() {
-		fanObservationStatePath = oldPath
-		fanObservation = oldState
-		fanObservationInit = oldInit
-		fanPeakCandidates = oldCandidates
-	})
+	resetPeakStore(t, fanPeaks)
 
 	start := time.Unix(100, 0)
 	updateFanObservation([]FanReading{{Name: "FAN1", RPM: 5000}}, start)
@@ -143,9 +176,9 @@ func TestEstimateFanDutyCyclePctFromObservation(t *testing.T) {
 		t.Fatalf("got=%v want ~43.3", got)
 	}
 
-	fanObservation = fanObservationState{}
-	fanObservationInit = false
-	fanPeakCandidates = make(map[string]fanPeakCandidate)
+	fanPeaks.loaded = false
+	fanPeaks.peaks = nil
+	fanPeaks.candidates = nil
 	got, ok = estimateFanDutyCyclePctFromObservation([]FanReading{{Name: "FAN1", RPM: 2600}})
 	if !ok {
 		t.Fatalf("expected persisted observed max to be reloaded from disk")