Les requetes s'executaient, les grandeurs etaient fausses

Trois defauts trouves en LISANT les tableaux, qu'aucune des 19 requetes ne
signalait puisqu'elles s'executaient toutes sans erreur.

1. `historise-evcs.mjs` horodate chaque mesure avec son propre last_updated —
   le vrai instant de la mesure, et c'est le bon choix. Mais deux grandeurs ne
   partagent alors presque jamais le meme ts : 2656 lignes pour 2656 instants
   distincts. Mes GROUP BY ts rendaient une ligne par grandeur avec une seule
   colonne remplie, donc la courbe "Excedent" — la grandeur meme du §36 —
   etait VIDE sur les 2650 lignes.

2. Le service n'ecrit que les changements, donc un regroupement a la minute
   laisse des trous : la difference n'existait que 73 minutes sur 434. Report
   de la derniere valeur connue, ce que fait aussi le controleur en temps
   reel. 430/435 apres.

3. L'energie horaire par sum(valeur)/60 suppose 60 echantillons par heure ; il
   y en a eu 26 a 47. Integration ponderee par la duree, plafonnee a 300 s
   pour qu'un arret du service n'extrapole pas une valeur perimee. ~9 %
   d'ecart. Le "25 a 58 %" que j'avais d'abord annonce etait faux : il
   comparait la somme a la moyenne, qui ignore la duree.

Plus deux corrections d'affichage : la colonne `jour` s'affichait
"2026-09-04 02:00:00" (date nue rendue comme instant UTC), et le decompte des
chargeurs MPPT est de six entites publiees, pas sept.

Lecon : verifier-dashboards.mjs prouve qu'une requete s'EXECUTE, pas qu'elle
rend la grandeur voulue — une colonne integralement nulle passe sans un mot.
Ce sont les courbes regardees a l'ecran qui ont attrape les trois.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KKzkkVTG6KXnGXFFC7X9XJ

deploy/historisation/grafana/dashboards/nuit.json

@@ -356,7 +356,7 @@
           "format": "time_series",
           "rawQuery": true,
           "editorMode": "code",
-          "rawSql": "SELECT ts AS time,\n       max(valeur) FILTER (WHERE cle = 'sensor.victron_grid_power_l1')    AS \"Reseau\",\n       max(valeur) FILTER (WHERE cle = 'sensor.victron_battery_power')    AS \"Batterie\",\n       max(valeur) FILTER (WHERE cle = 'sensor.victron_consumption_power_l1') AS \"Maison\",\n       max(valeur) FILTER (WHERE cle = 'sensor.victron_evcs_ac_power')    AS \"Voiture\"\nFROM mesure\nWHERE $__timeFilter(ts)\nGROUP BY ts\nORDER BY ts"
+          "rawSql": "WITH par_minute AS (\n  SELECT date_trunc('minute', ts) AS t,\n         avg(valeur) FILTER (WHERE cle = 'sensor.victron_grid_power_l1') AS Reseau,\n         avg(valeur) FILTER (WHERE cle = 'sensor.victron_battery_power') AS Batterie,\n         avg(valeur) FILTER (WHERE cle = 'sensor.victron_consumption_power_l1') AS Maison,\n         avg(valeur) FILTER (WHERE cle = 'sensor.victron_evcs_ac_power') AS Voiture\n  FROM mesure WHERE $__timeFilter(ts) GROUP BY 1\n),\npaliers AS (\n  SELECT t, Reseau, Batterie, Maison, Voiture,\n         count(Reseau) OVER (ORDER BY t) AS g_Reseau,\n         count(Batterie) OVER (ORDER BY t) AS g_Batterie,\n         count(Maison) OVER (ORDER BY t) AS g_Maison,\n         count(Voiture) OVER (ORDER BY t) AS g_Voiture\n  FROM par_minute\n),\nreport AS (\n  SELECT t,\n         first_value(Reseau) OVER (PARTITION BY g_Reseau ORDER BY t) AS Reseau,\n         first_value(Batterie) OVER (PARTITION BY g_Batterie ORDER BY t) AS Batterie,\n         first_value(Maison) OVER (PARTITION BY g_Maison ORDER BY t) AS Maison,\n         first_value(Voiture) OVER (PARTITION BY g_Voiture ORDER BY t) AS Voiture\n  FROM paliers\n)\nSELECT t AS time,\n         Reseau AS \"Reseau\",\n         Batterie AS \"Batterie\",\n         Maison AS \"Maison\",\n         Voiture AS \"Voiture\"\nFROM report ORDER BY t"
         }
       ],
       "gridPos": {

deploy/historisation/grafana/dashboards/prevision.json

@@ -74,7 +74,7 @@
           "format": "table",
           "rawQuery": true,
           "editorMode": "code",
-          "rawSql": "WITH du_matin AS (\n  SELECT DISTINCT ON (jour) jour, publie_a, horaire\n  FROM prevision\n  WHERE publie_a::date = jour\n    AND jour BETWEEN $__timeFrom()::date AND $__timeTo()::date\n  ORDER BY jour, publie_a ASC\n),\nprevu AS (\n  SELECT (jour + (h - 1) * interval '1 hour') AS heure,\n         horaire[h] AS kwh\n  FROM du_matin, generate_subscripts(horaire, 1) AS h\n),\nproduit AS (\n  -- Un echantillon toutes les 60 s : chaque point vaut 1/60 h.\n  SELECT date_trunc('hour', ts) AS heure,\n         sum(valeur) / 60.0 / 1000.0 AS kwh\n  FROM mesure\n  WHERE cle = 'sensor.victron_pv_power_system' AND $__timeFilter(ts)\n  GROUP BY 1\n)\nSELECT coalesce(prevu.heure, produit.heure) AS time,\n       round(prevu.kwh::numeric, 2)         AS \"Prevu le matin meme\",\n       round(produit.kwh::numeric, 2)       AS \"Reellement produit\"\nFROM prevu FULL OUTER JOIN produit USING (heure)\nWHERE coalesce(prevu.heure, produit.heure) BETWEEN $__timeFrom() AND $__timeTo()\nORDER BY 1"
+          "rawSql": "WITH du_matin AS (\n  SELECT DISTINCT ON (jour) jour, publie_a, horaire\n  FROM prevision\n  WHERE publie_a::date = jour\n    AND jour BETWEEN $__timeFrom()::date AND $__timeTo()::date\n  ORDER BY jour, publie_a ASC\n),\nprevu AS (\n  SELECT (jour + (h - 1) * interval '1 hour') AS heure,\n         horaire[h] AS kwh\n  FROM du_matin, generate_subscripts(horaire, 1) AS h\n),\n-- INTEGRATION PONDEREE PAR LA DUREE, et non somme/60.\n-- Le service n'ecrit une mesure que lorsqu'elle change : compter les\n-- echantillons revient a supposer qu'il y en a 60 par heure alors qu'il y\n-- en a eu 26 a 47 le 04/09 : l'energie etait sous-estimee d'environ 9 %.\n-- Chaque point vaut donc son intervalle jusqu'au suivant, plafonne a 300 s\n-- pour qu'un arret du service n'extrapole pas une valeur perimee.\npas AS (\n  SELECT ts, valeur,\n         LEAST(EXTRACT(epoch FROM (lead(ts) OVER (ORDER BY ts) - ts)), 300) AS dt\n  FROM mesure\n  WHERE cle = 'sensor.victron_pv_power_system' AND $__timeFilter(ts)\n),\nproduit AS (\n  SELECT date_trunc('hour', ts) AS heure,\n         sum(valeur * coalesce(dt, 0)) / 3600.0 / 1000.0 AS kwh\n  FROM pas GROUP BY 1\n)\nSELECT to_char(coalesce(prevu.heure, produit.heure), 'DD/MM HH24') || ' h' AS \"Heure\",\n       round(prevu.kwh::numeric, 2)   AS \"Prevu le matin meme\",\n       round(produit.kwh::numeric, 2) AS \"Reellement produit\"\nFROM prevu FULL OUTER JOIN produit USING (heure)\nWHERE coalesce(prevu.heure, produit.heure) BETWEEN $__timeFrom() AND $__timeTo()\nORDER BY coalesce(prevu.heure, produit.heure)"
         }
       ],
       "gridPos": {
@@ -205,7 +205,7 @@
           "format": "table",
           "rawQuery": true,
           "editorMode": "code",
-          "rawSql": "SELECT publie_a AS \"Publiee le\", jour AS \"Pour le jour\",\n       round(total_kwh::numeric, 1) AS \"Total annonce (kWh)\",\n       round((horaire[8]  + horaire[9]  + horaire[10] + horaire[11])::numeric, 2) AS \"07-11 h\",\n       round((horaire[12] + horaire[13] + horaire[14] + horaire[15])::numeric, 2) AS \"11-15 h\",\n       round((horaire[16] + horaire[17] + horaire[18] + horaire[19])::numeric, 2) AS \"15-19 h\"\nFROM prevision\nWHERE $__timeFilter(publie_a)\nORDER BY publie_a DESC, jour"
+          "rawSql": "SELECT publie_a AS \"Publiee le\",\n       to_char(jour, 'DD/MM') AS \"Pour le jour\",\n       round(total_kwh::numeric, 1) AS \"Total annonce (kWh)\",\n       round((horaire[8]  + horaire[9]  + horaire[10] + horaire[11])::numeric, 2) AS \"07-11 h\",\n       round((horaire[12] + horaire[13] + horaire[14] + horaire[15])::numeric, 2) AS \"11-15 h\",\n       round((horaire[16] + horaire[17] + horaire[18] + horaire[19])::numeric, 2) AS \"15-19 h\"\nFROM prevision\nWHERE $__timeFilter(publie_a)\nORDER BY publie_a DESC, jour"
         }
       ],
       "gridPos": {

deploy/historisation/grafana/dashboards/solaire.json

@@ -74,7 +74,7 @@
           "format": "time_series",
           "rawQuery": true,
           "editorMode": "code",
-          "rawSql": "SELECT ts AS time,\n       max(valeur) FILTER (WHERE cle = 'sensor.victron_pv_power_system')     AS \"Production\",\n       max(valeur) FILTER (WHERE cle = 'sensor.victron_consumption_power_l1') AS \"Maison\",\n       max(valeur) FILTER (WHERE cle = 'sensor.victron_evcs_ac_power')       AS \"Voiture\",\n       max(valeur) FILTER (WHERE cle = 'sensor.victron_pv_power_system')\n     - max(valeur) FILTER (WHERE cle = 'sensor.victron_consumption_power_l1') AS \"Excedent\"\nFROM mesure\nWHERE $__timeFilter(ts)\nGROUP BY ts\nORDER BY ts"
+          "rawSql": "WITH par_minute AS (\n  SELECT date_trunc('minute', ts) AS t,\n         avg(valeur) FILTER (WHERE cle = 'sensor.victron_pv_power_system') AS Production,\n         avg(valeur) FILTER (WHERE cle = 'sensor.victron_consumption_power_l1') AS Maison,\n         avg(valeur) FILTER (WHERE cle = 'sensor.victron_evcs_ac_power') AS Voiture\n  FROM mesure WHERE $__timeFilter(ts) GROUP BY 1\n),\npaliers AS (\n  SELECT t, Production, Maison, Voiture,\n         count(Production) OVER (ORDER BY t) AS g_Production,\n         count(Maison) OVER (ORDER BY t) AS g_Maison,\n         count(Voiture) OVER (ORDER BY t) AS g_Voiture\n  FROM par_minute\n),\nreport AS (\n  SELECT t,\n         first_value(Production) OVER (PARTITION BY g_Production ORDER BY t) AS Production,\n         first_value(Maison) OVER (PARTITION BY g_Maison ORDER BY t) AS Maison,\n         first_value(Voiture) OVER (PARTITION BY g_Voiture ORDER BY t) AS Voiture\n  FROM paliers\n)\nSELECT t AS time,\n         Production AS \"Production\",\n         Maison AS \"Maison\",\n         Voiture AS \"Voiture\",\n         Production - Maison AS \"Excedent\"\nFROM report ORDER BY t"
         }
       ],
       "gridPos": {
@@ -270,7 +270,7 @@
     {
       "type": "state-timeline",
       "title": "Ce que les chargeurs declarent eux-memes",
-      "description": "mppt_active = ce chargeur donne tout ce qu'il peut. voltage_current_limited ou off = il se retient, faute de preneur. Le 03/09 ces sept capteurs ont bascule 24 minutes avant le drapeau du bilan — c'est le signal le plus direct dont on dispose.",
+      "description": "mppt_active = ce chargeur donne tout ce qu'il peut. voltage_current_limited ou off = il se retient, faute de preneur. Le 03/09 ces capteurs ont bascule 24 minutes avant le drapeau du bilan — c'est le signal le plus direct dont on dispose.",
       "datasource": {
         "type": "influxdb",
         "uid": "evcs-influx"
@@ -357,7 +357,7 @@
           "format": "time_series",
           "rawQuery": true,
           "editorMode": "code",
-          "rawSql": "SELECT ts AS time,\n       max(valeur) FILTER (WHERE cle = 'sensor.victron_battery_soc') AS \"SOC\"\nFROM mesure WHERE $__timeFilter(ts) GROUP BY ts ORDER BY ts"
+          "rawSql": "SELECT date_trunc('minute', ts) AS time, avg(valeur) AS \"SOC\"\nFROM mesure WHERE cle = 'sensor.victron_battery_soc'\n  AND $__timeFilter(ts) GROUP BY 1 ORDER BY 1"
         },
         {
           "refId": "B",
@@ -368,7 +368,7 @@
           "format": "time_series",
           "rawQuery": true,
           "editorMode": "code",
-          "rawSql": "SELECT ts AS time,\n       max(valeur) FILTER (WHERE cle = 'sensor.victron_battery_current') AS \"Courant\"\nFROM mesure WHERE $__timeFilter(ts) GROUP BY ts ORDER BY ts"
+          "rawSql": "SELECT date_trunc('minute', ts) AS time, avg(valeur) AS \"Courant\"\nFROM mesure WHERE cle = 'sensor.victron_battery_current'\n  AND $__timeFilter(ts) GROUP BY 1 ORDER BY 1"
         }
       ],
       "gridPos": {