signalk-pypilot-newui 2.4.1 → 2.5.0

package/CHANGELOG.md

@@ -1,5 +1,123 @@
 # Changelog
 
+## 2.5.0 — 2026-09-09 — Rev154..Rev174
+
+> ## ⚠️ Version con funciones aun sin probar en el mar
+>
+> Esta version trae funciones NUEVAS que **no hemos podido validar
+> navegando** todavia (el dia de sacarla no habia viento).
+> Instalala si te apetece ayudarnos a probarlas — y si algo va raro,
+> **avisa**: https://github.com/Aitonos/signalk-pypilot-newui/issues
+>
+> Tu feedback nos vale ORO para tunear los limites antes de dejar
+> estas funciones a punto para todos.
+>
+> ### ⛵ Que tenemos que probar en el mar (checklist rapida)
+>
+> 1. **Contador de virada** (grande sobre el barco): al pulsar TACK
+>    con el AP enganchado, cuenta atras rojo (5-4-3-2-1) y luego pasa
+>    a naranja "VIRANDO" mientras el pypilot vira. Tocar el circulo
+>    cancela.
+> 2. **Resumen despues de virar** (panel que sale 2 min): compara la
+>    velocidad y el angulo antes y despues del bordo. Verde = buena.
+>    Amarillo = normal. Rojo = perdida.
+> 3. **Historial de bordos** (deja pulsado el barco -> "Historial"):
+>    guarda cada bordo con retencion de velocidad y tiempo de
+>    recuperacion. Filtro por dias, resalta el mejor.
+> 4. **Marca de rafagas** (activable en "Cambiar esquinas" ->
+>    Overlays): cuando entra una rafaga fuerte, deja una flecha
+>    amarilla fantasma con la velocidad del viento durante 10 segundos.
+> 5. **Cerebro auto** (opcional, en la ficha Smart Pilot): si lo
+>    activas, cambia el perfil del pypilot segun el viento (suave /
+>    normal / duro) y reacciona a rafagas segun la estrategia elegida
+>    (avisar / congelar rumbo / subir D / cambiar a perfil duro).
+> 6. **Corta automatica** (opcional): si el timon toca los topes
+>    demasiado tiempo, suelta el AP para que retomes tu.
+>
+> **Como nos ayudas**: usalas, y si algo se comporta raro o no te
+> gusta, mandanos: que hiciste, que esperabas, que paso. Con eso
+> tunearemos los limites antes de sacar la 2.5.0 estable.
+
+### English (Rev154..Rev174, pending sea-trial confirmation)
+
+**Added — supervisor layer ("brain over pypilot")**
+
+- **Auto-profile by wind** — the plugin can switch the active pypilot
+  profile between light / medium / heavy based on rolling TWS. Bands
+  configurable in the visor's Smart Pilot card. Disabled by default.
+  Rev164.
+- **Gust strategy** — configurable response when a remarkable AWS jump
+  is detected: `off`, `warn`, `freeze-target`, `boost-D`, `temp-heavy`.
+  Replaces the Rev165 warn-only detector. Restore-first logic returns
+  the pilot to the prior state when the gust window closes. Rev167.
+- **Auto-disengage on lost authority** — if the servo saturates against
+  its rudder stops for a sustained window, the AP releases so a manned
+  helm regains control. Rev166.
+- **Config surface moved to the visor** — the Smart Pilot card in the
+  plugin's own UI now owns every supervisor toggle. Nothing to touch in
+  SK Admin. Rev167.
+
+**Added — rose UI**
+
+- **Tack countdown overlay over the boat sprite** — huge red digit
+  during the pre-tack delay, then flips to orange `VIRANDO` + elapsed
+  seconds during execution. Click cancels the maneuver. Fresh delay is
+  fetched every arm so an edit in Calibration takes effect without a
+  page reload. Rev171/172.
+- **Post-tack stats HUD** — for 2 minutes after a completed tack the
+  boat sprite glows and a small panel shows TWS / |AWA| / SOG averaged
+  30 s before vs. 30 s after (measured t+30..t+60 so the boat has time
+  to settle). Verdict is colour-coded by SOG retention. Rev171.
+- **Persistent tack log (long-press on the boat)** — every completed
+  tack is stored locally with pre/post averages, SOG retention %, and
+  time to recover 95 % of pre-SOG. The action sheet on the boat sprite
+  now offers "Cambiar esquinas" (the old corner selector) or
+  "Historial de bordos". The log modal filters by day range and
+  highlights the best tack by retention and the fastest to recover.
+  Rev172.
+- **Gust ghost arrow (opt-in in the corner-selector modal)** — a
+  remarkable AWS jump (≥ 6 kn over the 30-s baseline or ≥ 1.55×) or a
+  sharp AWA rotation (≥ 30° in 5 s) leaves a translucent A-shaped
+  arrow frozen at the gust angle, with the AWS reading printed
+  radially outside it, fading to zero over 10 s. Cooldown 30 s and
+  single-ghost policy so it never chases the live arrow. Card lives
+  alongside the A / T / COG / Current overlay cards. Rev172-174.
+- **Mode selector reads full phrases** — `APARENTE` / `REAL` shown on
+  the collapsed button, `VIENTO APARENTE` / `VIENTO REAL` on the open
+  dropdown (swapped on pointerdown / blur). Rev172.
+- **T over A on overlap** — when both wind pieces cover the same angle
+  the teal T now draws on top of the amber A, so TWA stays readable at
+  a glance. Rev172.
+- **Rose corner sub-line resized (Carlos)** — the wind speed under the
+  angle in the top corners is now a first-class figure in portrait and
+  a compact one in landscape (media-query split). Bottom AWS/TWS chip
+  also trimmed so it stops covering half the boat on a portrait phone.
+  Rev171-173.
+- **Tack countdown click-to-cancel** — tapping the huge red circle at
+  any moment (delay or VIRANDO) sends `ap.tack.state=none`. Guard added
+  so TACK does nothing if the AP is not engaged. Rev171.
+
+**Changed — internal**
+
+- Shared `SERVO_ON_MIN_A` constant across doctor / kpis / servo-health.
+  Rev154.
+- Alarm rules extended with `servo-motor-temp`, `servo-load-high`
+  (deviation-ratio > 1.65), and `cruise-drift`. Rev155.
+- `doctor.applyAll()` applies only the top-priority suggestion and
+  returns `mustRunFreshSession` + `remaining` so the caller can drive a
+  next-iteration loop. Rev160.
+- Session recorder JSONL now enforces a 200 MB FIFO cap. Rev158.
+- Servo-health snapshot persists to disk between restarts. Rev158.
+- Hardware-ceiling finding in `doctor` (duty>0.75 & rms>5°) short-
+  circuits further advice with a "you have out-tuned the actuator"
+  message. Rev163.
+
+**Removed**
+
+- All "helm is manned" confirmation modals on engage / tack / Doctor
+  apply. Confirming every action was noise; the physical presence of
+  the helm is a given while at sea. Rev170.
+
 ## 2.4.1 — 2026-09-08 — Rev146..Rev153
 
 Patch release. Fixes + UI polish on top of 2.4.0.

package/README.md

@@ -1,5 +1,15 @@
 # signalk-pypilot-newui
 
+> ### ⚠️ 2.5.0 trae funciones que aun no hemos probado en el mar
+>
+> El dia de sacarla no habia viento suficiente para validar el
+> contador de virada, el resumen post-virada, el historial de bordos,
+> la marca de rafagas ni el "cerebro auto" que gestiona perfiles y
+> rafagas por ti. **Nos vale ORO tu feedback:** instalala, usala y si
+> algo va raro cuentanoslo en
+> https://github.com/Aitonos/signalk-pypilot-newui/issues
+> Con eso tuneamos los limites antes de dar todo por bueno.
+
 **PyPilot New-UI + SK Paths** — a modern, touch-first control panel for the
 open-source [pypilot](https://github.com/pypilot/pypilot) autopilot, plus every
 pypilot value exposed as a first-class Signal K path so tools like

package/dist/alarms.js

@@ -50,6 +50,30 @@ exports.DEFAULT_RULES = [
             return `AP heading error ${(r * 180 / Math.PI).toFixed(1)}° sustained`;
         },
     },
+    {
+        // Rev160 (Carlos): cruise-drift catches the slow bleed the RMS
+        // rule cannot see - a boat that consistently sits 6-8° off the
+        // target because of an unbalanced sail plan or current. Uses
+        // meanErrorRad instead of RMS. Threshold 5° | 120 s so trimming
+        // manoeuvres (short excursions) don't trip it.
+        id: "cruise-drift",
+        label: "Sustained heading drift",
+        severity: "warn",
+        defaultEnabled: true,
+        sustainSec: 120,
+        description: "Signed mean heading error > 5° for over 2 min - possible sail imbalance, current, or misaligned target.",
+        evaluate: (c) => {
+            if (!c.kpis || !c.sample || !c.sample.engaged)
+                return false;
+            const mean = c.kpis.window1m.meanErrorRad;
+            return typeof mean === "number" && Math.abs(mean) > (5 * Math.PI / 180);
+        },
+        message: (c) => {
+            const m = (c.kpis?.window1m.meanErrorRad ?? 0) * 180 / Math.PI;
+            const side = m > 0 ? "starboard" : "port";
+            return `Boat drifting ${Math.abs(m).toFixed(1)}° to ${side} of target`;
+        },
+    },
     {
         id: "unable-to-steer",
         label: "Unable to steer",
@@ -99,6 +123,47 @@ exports.DEFAULT_RULES = [
         },
         message: (c) => `Servo temp ${(c.sample?.servoTemp ?? 0).toFixed(0)} °C`,
     },
+    {
+        // Rev156 (Carlos): coil temperature complements servo-temp-high,
+        // which watches the controller. In practice the coil heats first
+        // under prolonged full-power operation. Threshold 70 C conservative
+        // (most pypilot motors alarm at 80-85).
+        id: "servo-motor-temp",
+        label: "Servo motor temperature high",
+        severity: "warn",
+        defaultEnabled: true,
+        sustainSec: 15,
+        description: "Motor coil temperature above 70 C.",
+        evaluate: (c) => {
+            if (!c.sample || typeof c.sample.servoMotorTemp !== "number")
+                return false;
+            return c.sample.servoMotorTemp > 70;
+        },
+        message: (c) => `Servo motor temp ${(c.sample?.servoMotorTemp ?? 0).toFixed(0)} °C`,
+    },
+    {
+        // Rev156 (Carlos): ServoHealth already grades the drive as
+        // learning/good/elevated/high on the deviation ratio (recent avg
+        // vs learned baseline). Surface "high" as a notification so KIP /
+        // WilhelmSK show the same red banner the visor card does. Only
+        // fires once the baseline is actually learned (deviation != null).
+        id: "servo-load-high",
+        label: "Autopilot load high",
+        severity: "warn",
+        defaultEnabled: true,
+        sustainSec: 60,
+        description: "Servo current is running >65% above the learned baseline. Check for extra drag, hard rudder, clutch slip, or heavy weather.",
+        evaluate: (c) => {
+            const sh = c.servoHealth;
+            if (!sh || sh.status !== "high" || sh.deviationRatio == null)
+                return false;
+            return sh.deviationRatio > 1.65;
+        },
+        message: (c) => {
+            const r = c.servoHealth?.deviationRatio ?? 0;
+            return `AP load ${(r * 100).toFixed(0)}% of baseline (drag / heavy weather?)`;
+        },
+    },
     {
         id: "low-voltage",
         label: "Low battery voltage",

package/dist/constants.js

@@ -0,0 +1,17 @@
+"use strict";
+// Rev154 (Carlos): shared numeric constants that used to be
+// duplicated as literals across doctor.ts / kpis.ts / servo-health.ts.
+// Keeping a single source of truth so a policy change (e.g. move to
+// 0.5 A for a bigger servo) does not miss one of the callers.
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.SERVO_ON_MIN_A = void 0;
+/**
+ * Servo current above which we consider the drive to be "on".
+ * Below this the servo is idle (or noise from the ADC) and the
+ * historian tick should NOT count toward runtime / duty / energy.
+ * Same threshold used by:
+ *  - Doctor authority analysis (servoOn ratio)
+ *  - KPIs servo runtime + duty cycle + amp-hour integration
+ *  - ServoHealth baseline learner (only counts on-load samples)
+ */
+exports.SERVO_ON_MIN_A = 0.3;

package/dist/doctor.js

@@ -16,6 +16,7 @@
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.DoctorEngine = void 0;
 const kpis_1 = require("./kpis");
+const constants_1 = require("./constants");
 const MIN_DURATION_SEC = 30;
 const DEFAULT_DURATION_SEC = 180;
 class DoctorEngine {
@@ -157,7 +158,7 @@ class DoctorEngine {
         const meanDeg = mean * 180 / Math.PI;
         const rmsDeg = rms * 180 / Math.PI;
         // Servo duty (fraction of engaged samples with servo drawing)
-        const servoOnCount = engaged.filter((s) => typeof s.servoCurrent === "number" && s.servoCurrent > 0.3).length;
+        const servoOnCount = engaged.filter((s) => typeof s.servoCurrent === "number" && s.servoCurrent > constants_1.SERVO_ON_MIN_A).length;
         const servoDuty = engaged.length > 0 ? servoOnCount / engaged.length : 0;
         // Zero-crossing count -> rough oscillation period estimate
         let crossings = 0;
@@ -180,6 +181,29 @@ class DoctorEngine {
             });
             return this.buildResult(session, now, samples, engaged, findings, suggestions);
         }
+        // Rev163 (Carlos, per Sean D'Epagnier forum #6): hardware ceiling
+        // finding. Sean's core message was that the best gain tuning
+        // cannot beat a saturated servo - a stronger motor is the fix.
+        // We flag "hardware ceiling" when:
+        //   - servo duty > 0.75 (drive is on 3/4 of the time)  AND
+        //   - RMS heading error still > 5° (i.e. tracking is bad)
+        // AND emit it as a finding WITHOUT any P/I/D suggestion so the
+        // skipper doesn't chase gains that cannot help. The message
+        // explicitly names hardware, not tuning, as the fix.
+        if (servoDuty > 0.75 && rmsDeg > 5) {
+            findings.push({
+                category: "hardware-ceiling",
+                severity: "critical",
+                message: `Servo running at ${(servoDuty * 100).toFixed(0)}% duty with ${rmsDeg.toFixed(1)}° RMS error. Tuning cannot close a saturated actuator - consider a more powerful drive or reducing steering friction.`,
+                metric: `duty=${(servoDuty * 100).toFixed(0)}% rms=${rmsDeg.toFixed(1)}°`,
+                messageKey: "doctor.finding.hardwareCeiling",
+                messageArgs: { duty: (servoDuty * 100).toFixed(0), rms: rmsDeg.toFixed(1) },
+            });
+            // Intentional early return: the follow-on rules would suggest
+            // gain changes that cannot help while the drive is saturated.
+            // Present just the honest verdict.
+            return this.buildResult(session, now, samples, engaged, findings, suggestions);
+        }
         // ---- Rule 1: Persistent bias -> increase I ----
         if (Math.abs(meanDeg) > 3) {
             findings.push({
@@ -393,21 +417,41 @@ class DoctorEngine {
         s.dismissedTs = Date.now();
         return { ok: true, message: "Dismissed." };
     }
+    // Rev162 (Carlos, per Sean D'Epagnier): apply ONLY ONE gain at a
+    // time. The old applyAll iterated blindly through P + I + D in the
+    // same tick, which contradicts Sean's forum advice ("tweak the
+    // values one at a time") and made it impossible to isolate which
+    // change helped. Now applyAll picks the top-priority suggestion,
+    // applies it, and marks the rest as "waiting for a fresh Doctor
+    // session before the next change". The visor uses `mustRunFreshSession`
+    // in the response to render the banner.
     applyAll() {
         const applied = [];
         const failed = [];
         if (!this.result)
             return { ok: false, applied, failed };
-        for (const s of this.result.suggestions) {
-            if (s.applied || s.dismissed)
-                continue;
-            const r = this.applySuggestion(s.id);
-            if (r.ok)
-                applied.push(s.id);
-            else
-                failed.push({ id: s.id, message: r.message });
+        // Priority order: bias > authority > oscillation > noise. Same as
+        // pypilot forum guidance - fix the biggest-error direction first.
+        const priority = { bias: 0, authority: 1, oscillation: 2, noise: 3 };
+        const pending = this.result.suggestions
+            .filter((s) => !s.applied && !s.dismissed)
+            .sort((a, b) => (priority[a.category] ?? 99) - (priority[b.category] ?? 99));
+        if (pending.length === 0) {
+            return { ok: true, applied, failed, mustRunFreshSession: false, remaining: 0 };
         }
-        return { ok: failed.length === 0, applied, failed };
+        const first = pending[0];
+        const r = this.applySuggestion(first.id);
+        if (r.ok)
+            applied.push(first.id);
+        else
+            failed.push({ id: first.id, message: r.message });
+        return {
+            ok: failed.length === 0,
+            applied,
+            failed,
+            mustRunFreshSession: pending.length > 1,
+            remaining: pending.length - 1,
+        };
     }
     reset() {
         if (this.timer) {