SPB Git forge
spboucher.ai ↗

spb/qc26

Public
10commits 1branches 0releases
2.8 MBsize
maindefault branch
17 days agolast push
Python 51.6% TypeScript 46.7% CSS 1.7%

Modèle 1.1.0 : recensement 2021 (StatCan ADA → 127 circonscriptions), swing par segments langue/région, calibration nationale, sondages de référence 2022

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Simon-Pierre Boucher committed 18 days ago (Sep 6, 2026) parent 4f94192

15 changed files +794 −23

added engine/data/derived/riding_demographics_2021.SOURCE.txt +3 −0
@@ -0,0 +1,3 @@
1 +Statistique Canada, Recensement de la population 2021, Profil du recensement (98-401-X2021012, ADA) + limites des ADA 2021 ; agrégation surfacique QC26 vers la carte électorale 2026
2 +computed_at=2026-09-06T05:33:24.708846+00:00
3 +ADA=1175
added engine/data/derived/riding_demographics_2021.csv +128 −0
@@ -0,0 +1,128 @@
1 +riding_code,population_2021,density,median_age,pct_65_plus,pct_french_mt,pct_english_mt,pct_other_mt,pct_plop_french,pct_immigrants,pct_visible_minority,median_household_income,pct_owner,pct_bachelor_plus,unemployment_rate,ada_count
2 +117,59877,65.2,50.1,26.9,92.97,4.84,1.06,94.27,1.85,1.39,61100.7,74.39,11.56,6.5,20
3 +119,69357,1174.5,42.6,22.74,84.32,7.49,5.79,88.64,7.17,7.83,62714.8,52.26,22.16,5.7,16
4 +121,66763,1913.5,42.2,24.17,83.35,3.37,10.55,91.63,11.76,13.03,55492.0,35.82,37.79,7.8,18
5 +127,54813,373.9,54.3,29.89,83.05,12.62,2.01,85.04,5.08,1.79,72900.0,73.63,27.5,6.7,18
6 +131,56240,50.3,43.8,19.71,94.34,3.06,1.35,95.99,2.41,1.52,73336.9,75.63,13.24,5.3,28
7 +137,80331,491.5,41.5,18.12,91.74,3.75,3.01,94.55,4.29,3.88,83355.4,74.05,26.58,4.4,25
8 +139,61070,918.9,41.9,18.8,94.26,1.18,3.45,97.48,3.75,4.57,69336.2,62.35,15.27,5.4,21
9 +141,62826,630.0,47.1,24.97,95.27,1.23,2.37,97.72,3.28,3.42,66088.8,62.12,16.36,5.3,31
10 +147,77182,880.6,47.3,25.77,96.0,0.75,2.51,98.24,2.55,3.19,64761.0,65.22,15.86,4.4,23
11 +151,49305,35.7,47.5,24.71,97.24,0.81,1.3,98.62,1.63,1.18,66585.7,75.14,16.2,4.8,20
12 +157,64821,120.1,49.4,26.33,79.17,15.53,2.34,81.27,4.75,1.88,72814.2,68.66,24.27,5.4,15
13 +159,72314,1061.7,47.6,25.15,91.86,2.18,4.35,95.99,5.16,5.27,66039.2,55.52,16.31,5.5,14
14 +161,64739,487.4,42.6,19.23,93.77,2.65,2.24,95.9,3.47,2.53,78527.2,71.8,14.95,5.1,23
15 +167,76657,1650.5,44.8,21.65,91.93,2.65,3.37,95.36,4.73,4.44,80892.9,58.5,19.49,6.0,18
16 +171,55823,29.5,44.9,20.73,80.34,14.46,2.37,82.49,3.63,1.86,74679.0,77.75,11.41,5.7,17
17 +179,68337,744.5,45.5,22.94,91.7,3.93,2.15,94.02,3.23,3.19,69240.8,60.41,13.41,6.2,12
18 +181,76066,242.3,42.7,16.96,65.12,23.14,7.53,67.74,9.27,7.81,100223.6,84.04,25.33,5.8,13
19 +187,76138,1893.2,42.2,17.06,49.8,24.73,18.87,55.39,20.26,22.17,92829.4,76.61,30.89,7.0,12
20 +191,71424,1517.6,42.0,18.26,62.01,19.85,13.44,69.63,17.72,19.53,84979.4,74.24,21.77,7.3,14
21 +197,57004,1855.7,41.4,16.13,82.04,4.65,10.2,88.43,10.87,11.35,93281.3,74.47,21.93,5.5,11
22 +199,64791,1234.3,41.9,16.31,75.02,6.42,14.69,82.16,15.14,15.97,105402.6,75.79,40.16,5.6,17
23 +201,80349,2807.4,43.8,21.29,39.2,11.45,42.22,53.06,40.72,50.45,87852.6,71.35,46.89,9.5,15
24 +207,70831,1181.2,41.0,15.1,87.61,3.78,6.11,92.26,8.22,8.2,103824.0,78.97,31.2,4.6,20
25 +211,74747,1986.9,42.2,18.23,70.9,6.11,18.53,82.62,21.24,24.8,90027.7,76.5,30.35,6.5,18
26 +217,73974,3719.1,43.7,21.86,60.84,12.89,20.7,72.01,23.68,25.89,74918.9,52.78,38.3,8.2,18
27 +219,64727,4757.8,40.5,18.43,74.4,3.52,17.59,87.32,21.36,27.13,60149.4,33.78,30.26,9.1,15
28 +221,72574,3682.3,43.1,20.54,76.96,2.94,16.09,89.08,19.77,23.17,76799.9,50.81,36.43,7.6,16
29 +227,66775,831.7,47.4,24.22,85.66,4.57,7.06,90.71,9.75,8.15,116165.7,84.73,52.41,5.1,15
30 +231,69700,599.9,42.2,16.21,93.14,1.59,3.54,96.69,5.23,5.31,102135.3,79.39,27.07,4.7,14
31 +237,71701,912.8,44.1,20.85,92.57,2.74,2.86,95.43,4.89,3.45,96153.9,77.59,29.1,5.0,19
32 +239,71549,1229.9,45.8,24.14,92.16,0.89,5.58,97.32,6.43,8.05,67928.8,53.48,17.1,6.0,20
33 +241,65121,778.3,49.0,26.25,96.15,1.07,1.71,98.19,2.82,2.74,69471.0,67.74,13.16,7.1,21
34 +247,69406,8526.7,40.5,17.64,56.01,15.86,22.42,64.21,24.42,24.27,73646.3,37.35,52.55,9.5,13
35 +251,80452,5953.6,41.8,20.2,31.72,25.35,35.81,42.71,35.54,45.14,64183.9,39.19,32.38,11.1,14
36 +257,64922,3675.4,43.0,19.23,41.98,28.86,22.5,49.29,24.84,30.54,70287.8,47.29,37.03,9.2,18
37 +259,62063,1684.8,47.7,24.65,21.3,47.68,24.51,24.17,25.52,22.5,110007.3,75.52,55.6,8.2,16
38 +261,80169,2359.1,44.1,18.1,24.78,31.09,35.33,31.67,35.66,38.83,100494.9,76.09,44.79,8.9,17
39 +267,75174,3409.1,42.7,19.73,18.8,33.3,38.54,26.95,41.27,45.94,92134.2,71.64,42.45,10.1,16
40 +271,94189,4642.8,39.4,16.13,28.61,14.63,47.31,45.71,48.31,56.59,79582.8,49.86,48.46,11.0,25
41 +277,84880,9328.6,40.9,20.77,19.94,32.17,38.31,27.56,45.72,44.55,74444.0,35.11,53.57,11.6,24
42 +279,72786,8575.2,40.1,17.38,23.6,36.48,31.85,30.81,34.87,38.1,70685.7,37.12,57.55,10.6,18
43 +281,86296,7756.5,36.4,13.18,48.96,20.09,23.97,58.26,26.23,31.84,65631.0,34.69,50.82,10.1,20
44 +287,62926,10583.9,35.6,13.92,60.88,12.16,21.15,70.64,26.08,27.61,57828.8,27.56,58.96,11.0,18
45 +291,87978,12918.4,36.5,16.83,26.29,27.57,37.64,32.44,32.56,43.38,69579.2,29.91,72.49,12.6,21
46 +297,104357,8131.6,36.8,16.56,37.89,16.09,37.32,50.29,38.06,41.15,81779.1,34.41,62.73,11.2,31
47 +299,82425,7453.0,40.9,18.94,32.08,8.09,50.43,54.58,51.27,53.54,64715.6,38.21,43.22,12.2,19
48 +301,65691,6755.3,41.9,20.15,63.87,3.69,26.7,83.79,28.28,32.48,63619.9,36.03,45.89,8.7,15
49 +307,74423,14226.6,36.1,13.14,47.26,8.06,37.49,60.18,31.93,40.12,58569.0,27.09,47.73,10.8,19
50 +311,61347,10884.1,36.4,11.44,75.11,6.57,13.88,84.77,18.3,15.28,62888.9,30.43,57.43,8.0,17
51 +313,67818,12928.7,34.3,10.54,63.14,13.84,17.62,71.13,22.41,15.99,66470.4,31.21,65.07,10.1,17
52 +317,58580,9224.1,36.1,11.97,76.37,4.67,14.69,86.58,19.02,22.07,58551.3,28.53,44.33,9.2,14
53 +319,77301,9292.6,41.9,19.39,69.38,3.61,22.11,85.88,24.21,25.18,59665.8,32.03,45.06,8.7,17
54 +321,69941,8917.2,37.5,14.32,40.44,4.12,47.01,73.39,44.12,59.6,55238.7,28.47,27.65,13.0,19
55 +327,81233,8389.8,40.6,18.61,46.03,4.72,41.54,78.66,39.76,58.23,57553.5,28.96,17.31,11.6,17
56 +331,79197,6764.5,42.4,20.82,29.88,8.27,52.61,62.66,46.94,47.1,66517.3,34.92,29.57,10.9,18
57 +337,74680,6241.9,41.3,18.37,71.68,3.2,20.42,87.53,22.71,28.17,64559.2,38.65,31.26,9.8,15
58 +339,57720,3426.1,44.0,19.25,39.03,11.82,41.62,61.89,34.87,38.32,83951.0,66.87,22.01,8.1,15
59 +341,54216,3515.4,44.8,19.19,79.32,3.12,13.78,91.43,16.46,22.16,70618.5,54.73,18.8,7.9,16
60 +347,77402,4325.3,42.4,21.21,57.27,5.06,31.19,76.66,33.11,37.42,69575.3,44.09,32.39,9.4,14
61 +351,73429,4629.2,45.8,24.47,32.57,9.98,48.08,50.0,45.02,39.06,68166.4,52.34,28.87,12.7,13
62 +357,81641,2545.4,41.9,14.83,46.59,10.61,35.2,60.53,31.73,31.61,101326.0,79.15,33.9,8.6,13
63 +359,78797,2585.4,42.2,15.92,59.98,6.18,27.53,75.71,27.82,30.11,99449.0,76.67,35.16,7.4,12
64 +569,57409,2473.8,44.0,17.72,58.49,8.28,27.11,75.25,23.56,26.49,96527.3,77.84,30.75,6.8,12
65 +571,65396,2004.2,43.3,18.08,59.33,7.1,27.55,77.1,26.1,30.3,97522.7,77.44,31.14,6.9,14
66 +577,67073,1910.2,44.1,20.75,77.92,5.82,12.6,84.52,10.51,9.51,86600.5,60.73,27.74,7.2,15
67 +581,63017,2480.1,44.3,20.17,81.06,6.37,9.08,87.54,10.75,11.13,83249.7,69.91,22.06,6.2,15
68 +587,72233,836.7,38.9,13.07,88.14,3.49,5.88,92.77,7.36,7.61,93699.4,76.95,20.25,5.2,16
69 +589,47868,70.2,52.5,27.05,81.3,13.35,2.53,83.42,4.46,1.53,64921.0,73.34,14.92,8.9,16
70 +591,55062,685.1,39.5,14.32,92.68,2.16,3.27,95.85,5.11,4.96,88038.1,71.87,17.24,5.6,12
71 +597,51876,1695.3,48.0,26.59,92.29,1.51,4.27,96.53,5.46,6.09,57604.0,43.68,12.96,8.1,10
72 +601,63562,867.3,38.8,12.89,88.51,2.13,6.61,95.19,8.83,13.09,88363.9,72.9,13.17,6.3,21
73 +607,79090,1932.2,42.1,15.1,81.73,4.46,10.29,89.07,12.02,12.39,111062.0,79.51,32.42,5.9,17
74 +609,66645,1590.4,42.5,17.47,84.51,2.62,9.6,93.26,11.64,15.95,95655.6,72.33,23.06,5.9,15
75 +611,70409,1146.3,41.3,16.4,86.98,2.54,7.76,93.9,9.47,12.04,97504.2,77.85,22.23,5.2,16
76 +617,65430,2784.0,45.7,21.74,85.81,1.68,9.62,95.16,12.13,16.89,93212.2,74.66,22.57,6.1,14
77 +621,72908,1138.2,41.4,16.21,92.16,1.23,4.84,97.31,6.96,9.68,88060.2,75.41,16.47,5.7,23
78 +627,59251,40.6,51.6,26.38,93.22,1.17,4.58,97.47,2.0,1.44,62206.0,74.36,11.41,7.9,19
79 +629,71156,563.7,47.7,26.46,95.23,0.96,2.61,98.17,3.46,3.64,65642.4,55.8,16.83,6.8,15
80 +631,58649,125.6,39.9,15.69,94.97,1.6,2.0,97.36,2.94,2.7,74790.8,75.25,7.96,6.6,18
81 +637,61482,224.0,46.3,20.51,91.85,3.78,2.52,94.27,4.39,1.86,84292.0,79.02,20.53,7.1,14
82 +641,67015,64.4,53.8,27.66,90.32,5.11,2.59,92.69,4.98,1.93,63954.6,72.88,19.23,9.4,16
83 +649,61835,17.1,54.2,29.34,93.45,3.85,1.43,95.01,2.62,1.29,60029.1,72.33,14.85,10.5,14
84 +651,79806,2179.5,40.1,17.21,65.29,12.27,16.21,74.62,21.28,26.77,77872.2,46.22,45.9,10.7,14
85 +657,85981,1173.0,40.8,15.46,52.82,28.96,12.07,58.67,15.59,18.91,93247.2,71.78,37.32,8.5,13
86 +661,76752,143.2,45.4,18.09,75.16,17.36,3.89,77.79,5.37,4.4,101914.6,84.65,28.34,7.3,26
87 +667,78741,2277.8,43.8,19.04,79.2,6.75,9.59,86.42,12.13,14.56,79381.2,62.49,23.18,10.0,15
88 +669,69997,620.9,44.3,19.24,88.97,6.18,2.29,91.67,3.76,4.42,79539.8,72.91,14.92,8.2,20
89 +671,48909,13.8,44.2,20.92,90.97,6.11,1.45,92.7,1.91,2.08,78438.1,71.01,19.17,5.3,12
90 +681,45353,19.9,45.3,22.03,97.41,0.82,1.02,98.67,0.94,1.13,72304.9,72.6,12.88,6.1,14
91 +689,52770,219.5,43.1,19.87,93.03,2.47,2.76,96.06,2.01,3.17,75197.3,58.06,16.34,5.5,13
92 +691,64320,1923.1,47.6,28.39,91.64,1.22,5.36,96.82,6.0,9.07,57689.0,43.2,28.45,7.9,14
93 +693,63232,144.6,48.3,24.49,96.66,0.96,1.6,98.33,2.2,2.25,71852.9,74.83,19.74,6.0,17
94 +697,76888,378.6,52.9,29.55,92.42,1.12,4.98,97.94,1.72,1.6,57139.8,61.24,13.65,7.5,22
95 +701,68480,962.8,49.1,25.96,96.53,1.04,1.54,98.36,2.68,2.71,65751.2,65.96,17.73,6.1,20
96 +707,69150,2661.6,40.7,24.76,82.07,2.21,12.77,92.63,14.67,18.44,79992.6,43.21,60.05,8.1,15
97 +709,61599,1647.8,48.1,24.87,91.01,1.54,5.64,96.28,7.45,7.58,99679.2,75.51,50.22,5.6,14
98 +727,66308,77.8,45.1,21.87,97.05,1.14,0.9,98.26,1.63,1.38,78346.8,79.2,17.2,5.2,15
99 +729,74046,1158.9,40.6,17.12,93.42,1.89,3.11,96.91,4.86,5.15,91563.9,68.71,26.5,5.1,21
100 +749,68146,2233.8,47.2,25.49,90.95,1.02,6.3,96.77,8.1,9.71,74462.8,50.04,32.81,7.2,22
101 +751,63140,7250.7,45.6,25.85,88.29,2.36,7.14,94.51,10.14,9.77,56719.2,30.02,51.11,9.4,15
102 +757,58658,4322.3,43.3,23.45,89.54,1.24,7.06,96.32,10.84,12.56,50073.8,25.04,29.21,9.1,16
103 +761,75350,1668.9,44.0,19.16,95.14,0.79,2.86,98.3,5.08,5.23,86404.1,69.31,24.39,5.7,17
104 +767,67951,2478.3,46.5,25.53,92.25,1.03,5.09,97.48,8.05,8.37,73128.5,57.6,27.81,7.0,15
105 +769,85039,992.9,41.4,16.02,95.07,1.33,2.34,97.77,4.25,3.57,94543.0,76.62,26.76,5.8,22
106 +771,65739,67.0,50.4,26.6,97.46,0.88,0.98,98.74,2.24,1.48,78844.5,76.91,21.24,7.2,15
107 +777,64329,125.0,45.2,22.43,97.01,0.85,1.56,98.46,1.47,2.27,65958.5,70.34,13.47,5.5,17
108 +781,57258,53.5,42.4,20.66,97.35,0.72,1.33,98.68,1.5,1.87,77869.8,77.56,16.32,4.3,18
109 +787,73978,58.0,47.7,26.01,96.82,0.97,1.32,98.34,1.57,2.09,66333.0,74.66,12.98,5.4,21
110 +789,77526,770.7,42.0,17.24,96.3,1.03,1.71,98.28,3.05,3.12,99297.2,76.89,33.18,4.6,17
111 +791,56340,1585.1,48.8,27.05,94.93,1.0,2.82,97.96,4.36,5.01,71176.2,54.2,29.71,6.6,12
112 +797,63024,103.5,46.3,23.36,97.12,0.76,1.46,98.64,2.17,2.67,74650.3,74.56,16.94,5.3,19
113 +799,61225,30.6,52.7,29.43,98.33,0.53,0.74,99.09,1.16,1.23,61381.0,73.56,13.76,6.7,14
114 +801,63599,234.8,50.7,28.18,98.2,0.56,0.72,99.08,1.24,1.45,62381.9,69.96,15.8,6.9,12
115 +803,56729,981.6,47.9,25.68,97.39,0.69,1.22,98.93,2.15,2.22,69830.1,64.64,27.74,7.9,19
116 +807,56851,44.8,52.3,28.27,98.48,0.62,0.34,99.15,0.81,0.83,56902.8,71.72,12.79,8.5,18
117 +809,35672,10.9,54.0,29.62,91.52,6.51,0.75,92.63,1.87,1.37,62587.2,72.98,14.3,11.5,9
118 +811,12251,67.5,54.4,28.37,93.98,4.98,0.28,94.54,0.76,0.48,75500.0,75.56,16.99,12.2,1
119 +813,47903,404.4,43.6,18.97,78.37,7.95,11.23,88.32,1.7,1.8,87018.5,65.74,14.07,7.6,18
120 +817,40192,249.6,49.8,24.17,92.87,0.54,5.92,98.63,1.07,1.17,70259.7,71.78,13.92,7.9,10
121 +819,50216,77.9,45.7,22.21,98.11,0.88,0.42,98.75,0.77,0.74,79436.0,76.72,15.73,5.9,26
122 +821,53423,1169.6,47.4,26.73,96.63,0.76,1.84,98.59,1.93,3.14,66411.0,56.57,27.05,5.9,10
123 +823,53208,1063.5,46.4,24.4,98.01,0.74,0.66,99.01,1.11,1.51,70752.6,62.37,19.66,5.6,11
124 +827,50728,214.5,47.4,24.38,98.71,0.41,0.54,99.48,0.76,1.12,70821.8,72.74,14.77,5.4,18
125 +829,47203,29.0,50.3,26.69,98.49,0.35,0.72,99.54,0.89,1.04,64957.0,72.87,14.21,7.0,12
126 +831,45781,14.8,30.7,9.12,29.95,6.65,59.49,31.14,1.21,1.81,98836.5,35.62,9.62,7.6,33
127 +833,62053,5390.7,44.4,22.51,54.55,4.28,34.54,80.4,36.0,37.74,67848.6,41.03,34.94,9.4,19
128 +837,40585,14.8,53.8,30.31,87.45,10.24,0.85,88.58,1.14,0.63,61573.9,72.29,15.0,11.2,10
modified engine/qc26/api/public.py +23 −2
@@ -19,10 +19,20 @@ from ..connectors.parties.documents import TOPIC_LABELS, TOPICS
19 19 from ..db import (AuditLog, CampaignEvent, Candidate, CompassQuestion, DocumentVersion, ElectionEvent, ForecastHouseEffect,
20 20 ForecastPartyResult, ForecastRidingResult, ForecastRidingSummary, ForecastRun, HistoricalResult, IngestionJob,
21 21 LiveRidingState, LiveSnapshot, NowcastRun, Party, PartyDocument, PartyPosition, Poll, PollWeight, Pollster,
22 − PolicyProposal, Region, Riding, RidingBaseline, SessionLocal, Source, get_setting, db_latency_ms)
22 + PolicyProposal, Region, Riding, RidingBaseline, RidingDemographics, SessionLocal, Source, get_setting, db_latency_ms)
23 +
24 +
25 +def _demo_payload(d: RidingDemographics | None) -> dict | None:
26 + if not d:
27 + return None
28 + return {"population2021": d.population_2021, "density": d.density, "medianAge": d.median_age, "pct65Plus": d.pct_65_plus,
29 + "pctFrenchMt": d.pct_french_mt, "pctEnglishMt": d.pct_english_mt, "pctOtherMt": d.pct_other_mt, "pctPlopFrench": d.pct_plop_french,
30 + "pctImmigrants": d.pct_immigrants, "pctVisibleMinority": d.pct_visible_minority, "medianHouseholdIncome": d.median_household_income,
31 + "pctOwner": d.pct_owner, "pctBachelorPlus": d.pct_bachelor_plus, "unemploymentRate": d.unemployment_rate, "adaCount": d.ada_count,
32 + "source": d.source, "computedAt": d.computed_at.isoformat() if d.computed_at else None}
23 33 from ..live.feed import broadcaster
24 34 from ..parties import OTHERS_COLOR, OTHERS_LABEL, PARTY_IDS, registry_payload
25 −from ..regions import POLL_REGION_LABELS
35 +from ..regions import POLL_REGION_LABELS, poll_region_for
26 36 from ..compass import positions as compass_positions
27 37 from ..compass.questions import AXES
28 38
@@ -208,12 +218,15 @@ def ridings(s: Session = Depends(db)):
208 218 pwin[fr.riding_code][fr.party_id] = round(fr.p_win, 3)
209 219 live = {st.riding_code: st for st in s.scalars(select(LiveRidingState)).all()}
210 220 regions = {r.id: r for r in s.scalars(select(Region)).all()}
221 + demo = {d.riding_code: d for d in s.scalars(select(RidingDemographics)).all()}
211 222 out = []
212 223 for r in s.scalars(select(Riding).order_by(Riding.sort_key)).all():
213 224 x = summ.get(r.code)
214 225 st = live.get(r.code)
226 + dm = demo.get(r.code)
215 227 out.append({"code": r.code, "name": r.name, "slug": r.slug, "region": r.region_id, "regionName": regions[r.region_id].name if r.region_id in regions else None,
216 228 "electors": r.electors_2026, "isNew": r.is_new_2026, "changed": r.boundary_changed_2026,
229 + "pctFrench": dm.pct_french_mt if dm else None, "pctImmigrants": dm.pct_immigrants if dm else None, "medianIncome": dm.median_household_income if dm else None,
217 230 "incumbent": {"party": r.incumbent_party_id, "name": r.incumbent_name, "running": r.incumbent_running},
218 231 "centroid": [r.centroid_lon, r.centroid_lat],
219 232 "forecast": {"favorite": x.favorite_party_id, "p": round(x.favorite_p, 3), "runnerUp": x.runner_up_party_id, "margin": round(x.margin_mean, 1),
@@ -249,6 +262,8 @@ def riding_detail(slug: str, s: Session = Depends(db)):
249 262 "elected": h.elected, "turnout": h.turnout})
250 263 d["history"] = dict(hist)
251 264 d["historyNote"] = "Résultats officiels sur les limites de l'époque (carte 2017) ; la ligne « 2022 transposé » applique les limites 2026." if hist else "Circonscription nouvelle ou renommée : pas d'historique sous ce nom."
265 + d["demographics"] = _demo_payload(s.get(RidingDemographics, r.code))
266 + d["pollRegion"] = poll_region_for(r.code)
252 267 if run:
253 268 x = s.scalar(select(ForecastRidingSummary).where(ForecastRidingSummary.run_id == run.id, ForecastRidingSummary.riding_code == r.code))
254 269 res = s.scalars(select(ForecastRidingResult).where(ForecastRidingResult.run_id == run.id, ForecastRidingResult.riding_code == r.code).order_by(desc(ForecastRidingResult.p_win))).all()
@@ -673,6 +688,12 @@ def status(s: Session = Depends(db)):
673 688
674 689
675 690 CHANGELOG = [
691 + {"version": "1.1.0", "date": "2026-09-06", "title": "Recensement 2021 et swing par segments",
692 + "notes": ["Recensement de la population 2021 (Statistique Canada, profil des aires de diffusion agrégées + limites) agrégé par intersection surfacique aux 127 circonscriptions : langue maternelle, première langue officielle parlée, immigration, minorités visibles, âge, revenu, tenure, scolarité, chômage.",
693 + "Excès de swing par langue : ventilations francophones / non-francophones des sondages (90 jours, pondérées récence × √n) comparées à une référence 2022 par régression écologique des résultats sur la part francophone ; appliqué à chaque circonscription selon sa composition (κ = 0,6).",
694 + "Excès de swing par région : ventilations RMR de Montréal / RMR de Québec / reste du Québec comparées aux résultats 2022 agrégés par le même regroupement (κ = 0,5) ; regroupement affiné circonscription par circonscription.",
695 + "Calibration : la moyenne des parts par circonscription pondérée par les électeurs inscrits est ramenée au vote national simulé à chaque tirage.",
696 + "« Pourquoi ? » : facteurs linguistique et régional affichés avec les valeurs réellement utilisées."]},
676 697 {"version": "1.0.0", "date": "2026-09-06", "title": "Lancement de QC26",
677 698 "notes": ["Agrégation des sondages depuis novembre 2022 : pondération récence × taille × qualité, effets maison (leave-one-pollster-out, rétrécis, bornés à ±3 pt).",
678 699 "État latent du vote par filtre de Kalman (marche aléatoire, espace logit), lissage RTS, dérive de campagne jusqu'au scrutin.",
modified engine/qc26/config.py +1 −1
@@ -52,7 +52,7 @@ class Settings(BaseSettings):
52 52 # ils dérivent de la liste officielle des circonscriptions (127 en 2026 → 64).
53 53
54 54 # --- Modèle ---
55 − model_version: str = "1.0.0"
55 + model_version: str = "1.1.0"
56 56 n_simulations: int = 20000
57 57 random_seed: int | None = None
58 58 recency_halflife_days: float = 14.0 # hors campagne
modified engine/qc26/connectors/polls/registry.py +27 −0
@@ -219,6 +219,33 @@ def verify_bootstrap_documents(s: Session, limit: int = 20, only_ids: list[str]
219 219 return {"verified": done, "failed": failed, "skipped": skipped, "remaining": max(0, len(polls) - done - failed - skipped)}
220 220
221 221
222 +# Sondages de référence de fin de campagne 2022 (ventilations linguistiques) : servent de base
223 +# 2022 par segment au modèle (model/segments.py). Documents originaux de la firme.
224 +REFERENCE_POLLS_2022 = [
225 + ("leger", "https://leger360.com/wp-content/uploads/2024/02/20221002-Rapport-fin-de-campagne-Election-Quebec-2022.pdf"),
226 + ("leger", "https://leger360.com/wp-content/uploads/2024/02/20220927-Rapport-Intentions-de-vote-post-debat-Radio-Canada-Election-Quebec-2022.pdf"),
227 +]
228 +
229 +
230 +def ensure_reference_polls(s: Session) -> int:
231 + """Ingère (une seule fois) les sondages de référence 2022 s'ils sont absents de la base."""
232 + known = set(s.scalars(select(PollDocument.url)).all())
233 + n = 0
234 + for pid, url in REFERENCE_POLLS_2022:
235 + if url in known:
236 + continue
237 + try:
238 + poll = connector_for(pid).process(s, DocumentCandidate(url=url, pollster_id=pid))
239 + s.commit()
240 + if poll:
241 + n += 1
242 + log.info("sondage de référence 2022 ingéré : %s", poll.id)
243 + except Exception as e: # noqa: BLE001
244 + log.warning("sondage de référence %s : %s", url, e)
245 + s.rollback()
246 + return n
247 +
248 +
222 249 def watch_new_polls(s: Session) -> dict:
223 250 """Cycle du Poll Watcher : découverte chez chaque firme, ingestion des documents inconnus."""
224 251 known_urls = {d.url for d in s.scalars(select(PollDocument.url)).all()} if False else set(s.scalars(select(PollDocument.url)).all())
modified engine/qc26/db.py +25 −0
@@ -93,6 +93,31 @@ class RidingBaseline(Base):
93 93 source: Mapped[str] = mapped_column(String(240))
94 94
95 95
96 +class RidingDemographics(Base):
97 + """Recensement 2021 (Statistique Canada) agrégé aux circonscriptions 2026 par intersection
98 + surfacique des aires de diffusion agrégées (ADA). Parts en % de la population concernée."""
99 + __tablename__ = "riding_demographics"
100 + riding_code: Mapped[int] = mapped_column(ForeignKey("ridings.code"), primary_key=True)
101 + population_2021: Mapped[int] = mapped_column(Integer)
102 + density: Mapped[float | None] = mapped_column(Float)
103 + median_age: Mapped[float | None] = mapped_column(Float)
104 + pct_65_plus: Mapped[float | None] = mapped_column(Float)
105 + pct_french_mt: Mapped[float | None] = mapped_column(Float) # langue maternelle française (réponses uniques)
106 + pct_english_mt: Mapped[float | None] = mapped_column(Float)
107 + pct_other_mt: Mapped[float | None] = mapped_column(Float)
108 + pct_plop_french: Mapped[float | None] = mapped_column(Float) # première langue officielle parlée : français
109 + pct_immigrants: Mapped[float | None] = mapped_column(Float)
110 + pct_visible_minority: Mapped[float | None] = mapped_column(Float)
111 + median_household_income: Mapped[float | None] = mapped_column(Float)
112 + pct_owner: Mapped[float | None] = mapped_column(Float)
113 + pct_bachelor_plus: Mapped[float | None] = mapped_column(Float) # 25-64 ans
114 + unemployment_rate: Mapped[float | None] = mapped_column(Float)
115 + ada_count: Mapped[int] = mapped_column(Integer, default=0)
116 + coverage: Mapped[float | None] = mapped_column(Float) # part de la population de la circ. couverte par des ADA
117 + source: Mapped[str] = mapped_column(String(300))
118 + computed_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), default=utcnow)
119 +
120 +
96 121 class HistoricalResult(Base):
97 122 """Résultats officiels par circonscription (cartes antérieures)."""
98 123 __tablename__ = "historical_results"
added engine/qc26/ingest/census.py +222 −0
@@ -0,0 +1,222 @@
1 +"""Recensement de la population 2021 (Statistique Canada) → circonscriptions 2026.
2 +
3 +Produit : Profil du recensement 2021, aires de diffusion agrégées (98-401-X2021012, CSV)
4 +Géométrie : fichier des limites des ADA 2021 (lada000b21a, Lambert Statistique Canada)
5 +
6 +Méthode : chaque ADA du Québec (1 175) est intersectée avec les 127 circonscriptions 2026 ;
7 +les effectifs (population, langue, immigration, tenure…) sont répartis au prorata de la
8 +surface intersectée, les indicateurs (âge médian, revenu médian, taux de chômage) sont
9 +moyennés pondérés par la population allouée. Le résultat est écrit dans
10 +`data/derived/riding_demographics_2021.csv` (versionné) puis chargé en base.
11 +"""
12 +from __future__ import annotations
13 +
14 +import csv
15 +import io
16 +import json
17 +import logging
18 +import zipfile
19 +from collections import defaultdict
20 +from pathlib import Path
21 +
22 +from sqlalchemy import select
23 +from sqlalchemy.orm import Session
24 +
25 +from ..config import settings
26 +from ..db import Riding, RidingDemographics, Source, utcnow
27 +
28 +log = logging.getLogger("qc26.census")
29 +DERIVED = settings.data_dir / "derived"
30 +OUT_CSV = DERIVED / "riding_demographics_2021.csv"
31 +SOURCE = ("Statistique Canada, Recensement de la population 2021, Profil du recensement (98-401-X2021012, ADA) "
32 + "+ limites des ADA 2021 ; agrégation surfacique QC26 vers la carte électorale 2026")
33 +
34 +# identifiants de caractéristiques (fichier français)
35 +IDS = {
36 + "pop": 1, "density": 6, "age65": 24, "median_age": 40,
37 + "mt_total": 393, "mt_en": 396, "mt_fr": 397, "mt_other": 398,
38 + "plop_total": 388, "plop_fr": 390,
39 + "income_median": 243,
40 + "tenure_total": 1414, "owner": 1415,
41 + "immig_total": 1527, "immigrants": 1529,
42 + "vismin_total": 1683, "vismin": 1684,
43 + "educ_total": 2014, "bachelor": 2024, "unemployment": 2230,
44 +}
45 +BACHELOR_LABEL = "Baccalauréat ou grade supérieur" # id 2024 (25-64 ans)
46 +
47 +
48 +def _num(x: str) -> float | None:
49 + x = (x or "").strip()
50 + if not x or x in ("...", "..", "x", "F"):
51 + return None
52 + try:
53 + return float(x.replace(",", "."))
54 + except ValueError:
55 + return None
56 +
57 +
58 +def read_ada_profile(csv_path: Path) -> dict[str, dict[str, float | None]]:
59 + """Lit (en flux) le CSV national et ne garde que les ADA du Québec (DGUID 2021S051624…)."""
60 + out: dict[str, dict[str, float | None]] = {}
61 + wanted = set(IDS.values())
62 + bachelor_id: int | None = None
63 + with open(csv_path, encoding="latin-1", newline="") as fh:
64 + r = csv.reader(fh)
65 + next(r)
66 + cur = None
67 + for row in r:
68 + dg = row[1]
69 + if not dg.startswith("2021S051624"):
70 + if out and cur and not dg.startswith("2021S0516"):
71 + break
72 + continue
73 + cid = int(row[8])
74 + if cid not in wanted:
75 + continue
76 + if cid == IDS["bachelor"] and row[9].strip() != BACHELOR_LABEL:
77 + log.warning("libellé inattendu pour l'id %d : %r", cid, row[9])
78 + bachelor_id = IDS["bachelor"]
79 + if dg != cur:
80 + cur = dg
81 + out[dg] = {}
82 + key = next(k for k, v in IDS.items() if v == cid)
83 + out[dg][key] = _num(row[11]) # C1_CHIFFRE_TOTAL
84 + log.info("ADA Québec lues : %d (id baccalauréat = %s)", len(out), bachelor_id)
85 + return out
86 +
87 +
88 +def read_ada_geometries(zip_path: Path) -> dict[str, object]:
89 + import pyproj
90 + import shapefile
91 + from shapely.geometry import shape
92 + from shapely.ops import transform
93 + z = zipfile.ZipFile(zip_path)
94 + base = next(n[:-4] for n in z.namelist() if n.endswith(".shp"))
95 + rd = shapefile.Reader(shp=io.BytesIO(z.read(base + ".shp")), dbf=io.BytesIO(z.read(base + ".dbf")), shx=io.BytesIO(z.read(base + ".shx")))
96 + prj = z.read(base + ".prj").decode()
97 + tr = pyproj.Transformer.from_crs(pyproj.CRS.from_wkt(prj), "EPSG:4326", always_xy=True).transform
98 + fields = [f[0] for f in rd.fields[1:]]
99 + i_dg = fields.index("IDUGD")
100 + geoms = {}
101 + for sr in rd.iterShapeRecords():
102 + dg = sr.record[i_dg]
103 + if not str(dg).startswith("2021S051624"):
104 + continue
105 + g = shape(sr.shape.__geo_interface__)
106 + if g.is_empty:
107 + continue
108 + geoms[dg] = transform(tr, g).buffer(0)
109 + log.info("géométries ADA Québec : %d", len(geoms))
110 + return geoms
111 +
112 +
113 +def aggregate(profile: dict, geoms: dict, ridings_geo: dict) -> dict[int, dict]:
114 + """Répartition surfacique ADA → circonscriptions 2026."""
115 + from shapely.strtree import STRtree
116 + codes = list(ridings_geo)
117 + polys = [ridings_geo[c] for c in codes]
118 + tree = STRtree(polys)
119 + acc: dict[int, dict[str, float]] = defaultdict(lambda: defaultdict(float))
120 + counts: dict[int, set] = defaultdict(set)
121 + count_keys = ["pop", "age65", "mt_total", "mt_en", "mt_fr", "mt_other", "plop_total", "plop_fr", "tenure_total", "owner",
122 + "immig_total", "immigrants", "vismin_total", "vismin", "educ_total", "bachelor"]
123 + rate_keys = ["median_age", "income_median", "unemployment", "density"]
124 + for dg, g in geoms.items():
125 + p = profile.get(dg)
126 + if not p or not p.get("pop"):
127 + continue
128 + area = g.area
129 + if area <= 0:
130 + continue
131 + for idx in tree.query(g, predicate="intersects"):
132 + inter = g.intersection(polys[idx])
133 + if inter.is_empty:
134 + continue
135 + w = inter.area / area
136 + if w < 1e-4:
137 + continue
138 + code = codes[idx]
139 + counts[code].add(dg)
140 + pop_w = (p["pop"] or 0) * w
141 + for k in count_keys:
142 + if p.get(k) is not None:
143 + acc[code][k] += p[k] * w
144 + for k in rate_keys:
145 + if p.get(k) is not None:
146 + acc[code][f"{k}_wsum"] += p[k] * pop_w
147 + acc[code][f"{k}_w"] += pop_w
148 + out: dict[int, dict] = {}
149 + for code, a in acc.items():
150 + pop = a["pop"]
151 + def pct(n, d):
152 + return round(100 * a[n] / a[d], 2) if a.get(d) else None
153 + def wavg(k):
154 + return round(a[f"{k}_wsum"] / a[f"{k}_w"], 1) if a.get(f"{k}_w") else None
155 + out[code] = {
156 + "population_2021": int(round(pop)), "density": wavg("density"), "median_age": wavg("median_age"),
157 + "pct_65_plus": pct("age65", "pop"), "pct_french_mt": pct("mt_fr", "mt_total"), "pct_english_mt": pct("mt_en", "mt_total"),
158 + "pct_other_mt": pct("mt_other", "mt_total"), "pct_plop_french": pct("plop_fr", "plop_total"),
159 + "pct_immigrants": pct("immigrants", "immig_total"), "pct_visible_minority": pct("vismin", "vismin_total"),
160 + "median_household_income": wavg("income_median"), "pct_owner": pct("owner", "tenure_total"),
161 + "pct_bachelor_plus": pct("bachelor", "educ_total"), "unemployment_rate": wavg("unemployment"),
162 + "ada_count": len(counts[code]),
163 + }
164 + return out
165 +
166 +
167 +def build(csv_path: Path, geo_zip: Path) -> Path:
168 + """Calcule le fichier dérivé (à exécuter là où les fichiers StatCan sont disponibles)."""
169 + from shapely.geometry import shape
170 + gj = json.loads((settings.official_dir / "circonscriptions_electorales_sans_eau_2026.json").read_text())
171 + ridings_geo = {int(f["properties"]["CO_CEP"]): shape(f["geometry"]).buffer(0) for f in gj["features"]}
172 + profile = read_ada_profile(csv_path)
173 + geoms = read_ada_geometries(geo_zip)
174 + agg = aggregate(profile, geoms, ridings_geo)
175 + DERIVED.mkdir(parents=True, exist_ok=True)
176 + cols = ["riding_code", "population_2021", "density", "median_age", "pct_65_plus", "pct_french_mt", "pct_english_mt", "pct_other_mt",
177 + "pct_plop_french", "pct_immigrants", "pct_visible_minority", "median_household_income", "pct_owner", "pct_bachelor_plus",
178 + "unemployment_rate", "ada_count"]
179 + with open(OUT_CSV, "w", newline="", encoding="utf-8") as fh:
180 + w = csv.writer(fh)
181 + w.writerow(cols)
182 + for code in sorted(agg):
183 + w.writerow([code] + [agg[code].get(c) for c in cols[1:]])
184 + (DERIVED / "riding_demographics_2021.SOURCE.txt").write_text(SOURCE + f"\ncomputed_at={utcnow().isoformat()}\nADA={len(geoms)}\n")
185 + log.info("fichier dérivé écrit : %s (%d circonscriptions)", OUT_CSV, len(agg))
186 + return OUT_CSV
187 +
188 +
189 +def load(s: Session, path: Path = OUT_CSV) -> int:
190 + """Charge le fichier dérivé en base (idempotent)."""
191 + if not path.exists():
192 + log.warning("fichier démographique absent : %s", path)
193 + return 0
194 + ridings = {r.code for r in s.scalars(select(Riding)).all()}
195 + total_pop = 0
196 + n = 0
197 + with open(path, encoding="utf-8") as fh:
198 + for row in csv.DictReader(fh):
199 + code = int(row["riding_code"])
200 + if code not in ridings:
201 + continue
202 + d = s.get(RidingDemographics, code) or RidingDemographics(riding_code=code)
203 + for k, v in row.items():
204 + if k == "riding_code":
205 + continue
206 + val = None if v in ("", "None") else (int(float(v)) if k in ("population_2021", "ada_count") else float(v))
207 + setattr(d, k, val)
208 + d.source = SOURCE
209 + d.coverage = 1.0
210 + s.add(d)
211 + total_pop += d.population_2021
212 + n += 1
213 + src = s.get(Source, "statcan-census-2021")
214 + if not src:
215 + src = Source(id="statcan-census-2021", name="Statistique Canada — Recensement 2021 (profil ADA + limites)", tier=1, kind="official",
216 + url="https://www12.statcan.gc.ca/census-recensement/2021/dp-pd/prof/details/download-telecharger.cfm?Lang=F")
217 + s.add(src)
218 + src.last_checked_at = src.last_ok_at = utcnow()
219 + src.status = "operational"
220 + src.detail = f"{n} circonscriptions, population totale {total_pop:,}".replace(",", " ")
221 + log.info("démographie chargée : %d circonscriptions, population %d", n, total_pop)
222 + return n
modified engine/qc26/main.py +11 −0
@@ -33,6 +33,17 @@ def bootstrap() -> None:
33 33 log.info("base vide : ingestion des données officielles")
34 34 official.run_all(s, refresh=True)
35 35 polls_bootstrap.load(s)
36 + from .db import RidingDemographics
37 + from .ingest import census
38 + if not s.scalar(select(RidingDemographics).limit(1)):
39 + census.load(s) # fichier dérivé versionné (data/derived)
40 + try:
41 + from .connectors.polls.registry import ensure_reference_polls
42 + if ensure_reference_polls(s):
43 + from .model.run import run_forecast
44 + run_forecast(s, trigger="reference_polls_2022")
45 + except Exception as e: # noqa: BLE001
46 + log.warning("sondages de référence 2022 : %s", e)
36 47 if not s.scalar(select(ForecastRun).where(ForecastRun.status == "ok").limit(1)):
37 48 from .model.run import run_forecast
38 49 try:
modified engine/qc26/model/ridings.py +37 −14
@@ -1,16 +1,17 @@
1 −"""Modèle de circonscription : priors 2022 transposés, swing proportionnel régional,
2 −effets candidats, simulation Monte Carlo à erreurs corrélées.
1 +"""Modèle de circonscription (v1.1) : priors 2022 transposés, swing proportionnel national,
2 +excès de swing par segments (langue × composition du recensement, région), effets candidats,
3 +calibration nationale et simulation Monte Carlo à erreurs corrélées.
3 4
4 5 Pour chaque tirage national v (vote %), la part du parti p dans la circonscription r :
5 − share_r,p ∝ baseline_r,p × (v_p / national2022_p)^β (swing proportionnel, β=1)
6 − × exp(ε_région,p) × exp(ε_circ,p) × exp(effets candidats)
7 −puis normalisation. Les erreurs régionales et locales sont tirées par simulation
8 −(indépendantes entre régions/circonscriptions, corrélées via le tirage national).
6 + logit-ratio_r,p = log(baseline_r,p) − log(national2022_p) + adj_segments_r,p + effets_candidats_r,p
7 + share_r,p ∝ exp(log v_p + logit-ratio_r,p + ε_région,p + ε_circ,p)
8 +puis normalisation, puis calibration : la moyenne des parts pondérée par les électeurs
9 +inscrits est ramenée exactement au vote national simulé (une itération multiplicative).
9 10 Les partis absents d'une circonscription en 2022 reçoivent un plancher (0,3 %).
10 11 """
11 12 from __future__ import annotations
12 13
13 −from dataclasses import dataclass
14 +from dataclasses import dataclass, field
14 15
15 16 import numpy as np
16 17
@@ -27,11 +28,14 @@ class RidingInput:
27 28 incumbent_party: str | None
28 29 incumbent_running: bool | None
29 30 candidates: set[str] # partis présentant un·e candidat·e (DGEQ)
31 + electors: int = 50000
32 + pct_french: float | None = None # langue maternelle française (recensement 2021)
33 + poll_region: str = "reste"
30 34
31 35
32 36 def simulate_ridings(ridings: list[RidingInput], national_draws: np.ndarray, national_2022: dict[str, float],
33 37 rng: np.random.Generator, regional_sd: float | None = None, riding_sd: float | None = None,
34 − candidate_effects: bool = True) -> tuple[np.ndarray, np.ndarray]:
38 + candidate_effects: bool = True, seg_adj: np.ndarray | None = None, calibrate: bool = True) -> tuple[np.ndarray, np.ndarray]:
35 39 """national_draws : (S, P) en % ; retourne (winners (S, R) indices de parti, shares (S, R, P) en %)."""
36 40 regional_sd = settings.regional_error_sd if regional_sd is None else regional_sd
37 41 riding_sd = settings.riding_error_sd if riding_sd is None else riding_sd
@@ -40,9 +44,6 @@ def simulate_ridings(ridings: list[RidingInput], national_draws: np.ndarray, nat
40 44 regions = sorted({r.region_id for r in ridings})
41 45 reg_idx = {g: i for i, g in enumerate(regions)}
42 46 nat22 = np.array([max(national_2022.get(p, 0.5), 0.5) for p in PARTIES_ALL]) / 100.0
43 − # La règle « pas de candidature → part résiduelle » ne s'applique qu'aux partis dont la
44 − # liste est (quasi) complète : avant la clôture des candidatures, une absence n'est pas
45 − # une information.
46 47 counts = {p: sum(1 for r in ridings if p in r.candidates) for p in PARTY_IDS}
47 48 complete = {p for p, n in counts.items() if n >= 0.9 * R}
48 49 base = np.zeros((R, P))
@@ -54,6 +55,8 @@ def simulate_ridings(ridings: list[RidingInput], national_draws: np.ndarray, nat
54 55 base[i, j] = 0.001 # pas de candidat·e officiel·le : part résiduelle
55 56 if candidate_effects and r.incumbent_party and r.incumbent_party in PARTIES_ALL and r.incumbent_running is False:
56 57 adj[i, PARTIES_ALL.index(r.incumbent_party)] -= settings.incumbent_retirement_penalty
58 + if seg_adj is not None:
59 + adj = adj + seg_adj
57 60 log_ratio = np.log(base) - np.log(nat22)[None, :] + adj # (R, P)
58 61 reg_eps = rng.normal(0, regional_sd, size=(S, len(regions), P))
59 62 reg_eps[:, :, -1] *= 0.5
@@ -64,6 +67,14 @@ def simulate_ridings(ridings: list[RidingInput], national_draws: np.ndarray, nat
64 67 logits = nat[:, None, :] + log_ratio[None, :, :] + reg_eps[:, reg_of, :] + rid_eps
65 68 ex = np.exp(logits - logits.max(axis=2, keepdims=True))
66 69 shares = ex / ex.sum(axis=2, keepdims=True)
70 + if calibrate:
71 + w = np.array([max(r.electors or 1, 1) for r in ridings], dtype=float)
72 + w = w / w.sum()
73 + for _ in range(2):
74 + implied = np.einsum("r,srp->sp", w, shares) # (S, P)
75 + ratio = np.clip((national_draws / 100.0) / np.clip(implied, 1e-6, None), 0.5, 2.0)
76 + shares = shares * ratio[:, None, :]
77 + shares = shares / shares.sum(axis=2, keepdims=True)
67 78 # « Autres » agrège des candidatures hétérogènes : jamais déclaré gagnant (v1)
68 79 winners = shares[:, :, : len(PARTY_IDS)].argmax(axis=2)
69 80 return winners, shares * 100.0
@@ -80,7 +91,8 @@ def rating(p_win: float) -> str:
80 91
81 92
82 93 def explain_factors(r: RidingInput, national_now: dict[str, float], national_2022: dict[str, float],
83 − favorite: str, share_now: dict[str, float]) -> list[dict]:
94 + favorite: str, share_now: dict[str, float], seg_adj_fav: float | None = None,
95 + seg_detail: dict | None = None) -> list[dict]:
84 96 """Facteurs réellement utilisés par le modèle pour la circonscription (« Pourquoi ? »)."""
85 97 out = []
86 98 b = r.baseline.get(favorite, 0.0)
@@ -90,14 +102,25 @@ def explain_factors(r: RidingInput, national_now: dict[str, float], national_202
90 102 out.append({"factor": "tendance_provinciale", "label": "Tendance provinciale (sondages agrégés)",
91 103 "detail": f"{favorite.upper()} : {national_now.get(favorite, 0):.1f} % aujourd'hui vs {national_2022.get(favorite, 0):.1f} % en 2022 ({swing:+.1f} pt)",
92 104 "direction": "+" if swing > 0 else "-"})
93 − out.append({"factor": "swing_proportionnel", "label": "Swing proportionnel appliqué au territoire",
105 + if r.pct_french is not None and seg_detail and seg_detail.get("language"):
106 + lang = seg_detail["language"]
107 + out.append({"factor": "composition_linguistique", "label": "Composition linguistique (Recensement 2021) × ventilations des sondages",
108 + "detail": (f"{r.pct_french:.0f} % de langue maternelle française. {favorite.upper()} chez les francophones : {lang['fr_now']:.0f} % "
109 + f"(vs {lang['fr_2022']:.0f} % en 2022) ; non-francophones : {lang['nonfr_now']:.0f} % (vs {lang['nonfr_2022']:.0f} %)."),
110 + "direction": "+" if (seg_adj_fav or 0) > 0.02 else "-" if (seg_adj_fav or 0) < -0.02 else "±"})
111 + if seg_detail and seg_detail.get("region"):
112 + reg = seg_detail["region"]
113 + out.append({"factor": "region_sondages", "label": f"Ventilation régionale des sondages ({reg['label']})",
114 + "detail": f"{favorite.upper()} : {reg['now']:.0f} % dans la région aujourd'hui vs {reg['2022']:.0f} % en 2022, contre {swing:+.1f} pt au national.",
115 + "direction": "±"})
116 + out.append({"factor": "swing_proportionnel", "label": "Swing proportionnel appliqué au territoire, calibré au national",
94 117 "detail": f"part projetée {share_now.get(favorite, 0):.1f} % (moyenne des simulations)", "direction": "±"})
95 118 if r.incumbent_party:
96 119 if r.incumbent_running is False:
97 120 out.append({"factor": "sortant", "label": "Député·e sortant·e ne se représente pas",
98 121 "detail": f"pénalité de ~1,3 pt appliquée à {r.incumbent_party.upper()}", "direction": "-" if r.incumbent_party == favorite else "+"})
99 122 elif r.incumbent_running:
100 − out.append({"factor": "sortant", "label": "Député·e sortant·e candidat·e", "detail": f"{r.incumbent_party.upper()} (aucun bonus appliqué ; effet neutre dans le modèle v1)", "direction": "±"})
123 + out.append({"factor": "sortant", "label": "Député·e sortant·e candidat·e", "detail": f"{r.incumbent_party.upper()} (aucun bonus appliqué ; effet neutre dans le modèle)", "direction": "±"})
101 124 missing = [p for p in PARTY_IDS if r.candidates and p not in r.candidates]
102 125 if missing:
103 126 out.append({"factor": "candidatures", "label": "Candidatures non encore enregistrées (DGEQ)",
modified engine/qc26/model/run.py +22 −4
@@ -13,10 +13,12 @@ from sqlalchemy.orm import Session
13 13
14 14 from ..config import settings
15 15 from ..db import (Candidate, ForecastHouseEffect, ForecastPartyResult, ForecastRidingResult, ForecastRidingSummary,
16 − ForecastRun, Party, Poll, PollWeight, Riding, RidingBaseline, get_setting, set_setting, utcnow)
16 + ForecastRun, Party, Poll, PollWeight, Riding, RidingBaseline, RidingDemographics, get_setting, set_setting, utcnow)
17 17 from ..parties import PARTY_IDS
18 +from ..regions import POLL_REGION_LABELS, poll_region_for
18 19 from .aggregate import PARTIES_ALL, Obs, compute_weights, election_day_distribution, house_effects, kalman_latent
19 20 from .ridings import RidingInput, explain_factors, rating, simulate_ridings
21 +from .segments import build_segment_model, riding_adjustments
20 22
21 23 log = logging.getLogger("qc26.model")
22 24
@@ -51,8 +53,11 @@ def load_ridings(s: Session) -> tuple[list[RidingInput], dict[str, float]]:
51 53 for code, pid in s.execute(select(Candidate.riding_code, Candidate.party_id)).all():
52 54 if pid:
53 55 cands[code].add(pid)
56 + demo = {d.riding_code: d for d in s.scalars(select(RidingDemographics)).all()}
54 57 out = [RidingInput(code=r.code, region_id=r.region_id or "inconnue", baseline=baselines.get(r.code, {}),
55 − incumbent_party=r.incumbent_party_id, incumbent_running=r.incumbent_running, candidates=cands.get(r.code, set()))
58 + incumbent_party=r.incumbent_party_id, incumbent_running=r.incumbent_running, candidates=cands.get(r.code, set()),
59 + electors=r.electors_2026 or 50000, pct_french=(demo[r.code].pct_french_mt if r.code in demo else None),
60 + poll_region=poll_region_for(r.code))
56 61 for r in rows if baselines.get(r.code)]
57 62 national_2022 = {p.id: (p.vote_2022 or 0.0) for p in s.scalars(select(Party)).all()}
58 63 national_2022["aut"] = max(0.5, 100.0 - sum(national_2022.values()))
@@ -86,7 +91,10 @@ def run_forecast(s: Session, trigger: str = "manual", cutoff: date | None = None
86 91 national = _draw_national(mu, cov, n_sims, rng) # (S, P) %
87 92
88 93 ridings, national_2022 = load_ridings(s)
89 − winners, shares = simulate_ridings(ridings, national, national_2022, rng)
94 + national_now_mean = {p: float(national[:, j].mean()) for j, p in enumerate(PARTIES_ALL)}
95 + seg_model, f_r = build_segment_model(s, cutoff, national_now_mean, national_2022)
96 + seg_adj = riding_adjustments(seg_model, [r.code for r in ridings], f_r)
97 + winners, shares = simulate_ridings(ridings, national, national_2022, rng, seg_adj=seg_adj)
90 98 S, R, P = shares.shape
91 99 seats_total = int(get_setting(s, "seats_total", R))
92 100 majority = int(get_setting(s, "majority_threshold", seats_total // 2 + 1))
@@ -139,10 +147,18 @@ def run_forecast(s: Session, trigger: str = "manual", cutoff: date | None = None
139 147 fav_id = PARTIES_ALL[fav]
140 148 margin = share_mean[fav_id] - share_mean[PARTIES_ALL[second]]
141 149 swing = share_mean[fav_id] - r.baseline.get(fav_id, 0.0)
150 + seg_detail = {}
151 + if "fr" in seg_model.current and "nonfr" in seg_model.current and fav_id in seg_model.current["fr"]:
152 + seg_detail["language"] = {"fr_now": seg_model.current["fr"][fav_id], "fr_2022": seg_model.baseline.get("fr", {}).get(fav_id, 0),
153 + "nonfr_now": seg_model.current["nonfr"][fav_id], "nonfr_2022": seg_model.baseline.get("nonfr", {}).get(fav_id, 0)}
154 + if r.poll_region in seg_model.current and fav_id in seg_model.current[r.poll_region]:
155 + seg_detail["region"] = {"label": POLL_REGION_LABELS.get(r.poll_region, r.poll_region), "now": seg_model.current[r.poll_region][fav_id],
156 + "2022": seg_model.baseline.get(r.poll_region, {}).get(fav_id, 0)}
142 157 summ = ForecastRidingSummary(run_id=run.id, riding_code=r.code, favorite_party_id=fav_id, favorite_p=float(pw[fav]),
143 158 runner_up_party_id=PARTIES_ALL[second], margin_mean=float(margin),
144 159 volatility=float(1 - pw[fav]), rating=rating(float(pw[fav])), swing_from_2022=float(swing),
145 − factors=explain_factors(r, national_now, national_2022, fav_id, share_mean))
160 + factors=explain_factors(r, national_now, national_2022, fav_id, share_mean,
161 + seg_adj_fav=float(seg_adj[i, fav]), seg_detail=seg_detail))
146 162 s.add(summ)
147 163 riding_rows.append((r.code, fav_id, float(pw[fav]), PARTIES_ALL[second], float(margin)))
148 164 battlegrounds.append((float(1 - pw[fav]), r.code))
@@ -185,6 +201,8 @@ def run_forecast(s: Session, trigger: str = "manual", cutoff: date | None = None
185 201 "leader": top[0][0], "poll_count": len(obs), "latest_poll": max(o.end for o in obs).isoformat(),
186 202 "national_now": national_now, "cutoff": cutoff.isoformat(), "days_to_election": (settings.election_date - cutoff).days,
187 203 "battlegrounds": [c for _, c in sorted(battlegrounds, reverse=True)[:15]],
204 + "segments": seg_model.payload(),
205 + "demographics_coverage": sum(1 for r in ridings if r.pct_french is not None),
188 206 "ridings": {str(c): {"fav": f, "p": round(p, 3), "second": sec, "margin": round(m, 1)} for c, f, p, sec, m in riding_rows},
189 207 }
190 208 diff = None
added engine/qc26/model/segments.py +235 −0
@@ -0,0 +1,235 @@
1 +"""Swing par segments (langue, région) — modèle 1.1.
2 +
3 +Idée : le swing n'est pas uniforme. Les firmes publient des ventilations par langue
4 +(francophones / non-francophones) et par région (RMR de Montréal / RMR de Québec / reste).
5 +On en tire, pour chaque parti, l'écart entre le swing du segment et le swing national ;
6 +cet « excès de swing » est appliqué à chaque circonscription selon sa composition :
7 +- langue : part de langue maternelle française (Recensement 2021, agrégé aux circonscriptions) ;
8 +- région : regroupement « sondeurs » de la circonscription.
9 +
10 +Références 2022 par segment :
11 +- région : résultats 2022 transposés, agrégés par regroupement (exact) ;
12 +- langue : ventilations francophones / non-francophones des sondages de fin de campagne 2022
13 + (terrain dans les 14 jours précédant le 3 octobre 2022, documents originaux archivés),
14 + recalées sur le résultat officiel (chaque parti est multiplié par résultat/sondage national,
15 + puis renormalisé). Sans sondage de référence, la dimension linguistique est désactivée.
16 + La régression écologique (résultats 2022 ~ part francophone) est publiée à titre de
17 + diagnostic mais n'est PAS utilisée comme référence : l'extrapolation à f = 0 est instable.
18 +Les excès de swing sont rétrécis (κ) car les sous-échantillons sont bruités.
19 +"""
20 +from __future__ import annotations
21 +
22 +import math
23 +from collections import defaultdict
24 +from dataclasses import dataclass, field
25 +from datetime import date
26 +
27 +import numpy as np
28 +from sqlalchemy import select
29 +from sqlalchemy.orm import Session
30 +
31 +from ..config import settings
32 +from ..db import Poll, PollSubsample, RidingBaseline, RidingDemographics, Riding
33 +from ..parties import PARTY_IDS
34 +from ..regions import poll_region_for
35 +from .aggregate import PARTIES_ALL
36 +
37 +KAPPA_LANGUAGE = 0.6
38 +KAPPA_REGION = 0.5
39 +WINDOW_DAYS = 90
40 +HALFLIFE_DAYS = 21.0
41 +MIN_POLLS = 2
42 +
43 +LANGUAGE_MAP = {"francophones": "fr", "francophone": "fr", "non-francophones": "nonfr", "non francophones": "nonfr", "nonfrancophones": "nonfr",
44 + "anglophones": "nonfr", "allophones": "nonfr"}
45 +REGION_MAP = {"montréal rmr": "montreal_rmr", "montreal rmr": "montreal_rmr", "rmr de montréal": "montreal_rmr", "grand montréal": "montreal_rmr",
46 + "québec rmr": "quebec_rmr", "quebec rmr": "quebec_rmr", "rmr de québec": "quebec_rmr", "région de québec": "quebec_rmr",
47 + "reste du québec": "reste", "ailleurs au québec": "reste", "autres régions": "reste", "régions": "reste"}
48 +
49 +
50 +def canonical(dimension: str, segment: str) -> str | None:
51 + s = segment.strip().lower()
52 + if dimension == "language":
53 + return LANGUAGE_MAP.get(s)
54 + if dimension == "region":
55 + return REGION_MAP.get(s)
56 + return None
57 +
58 +
59 +def _logit(p: float) -> float:
60 + p = min(max(p, 0.003), 0.997)
61 + return math.log(p / (1 - p))
62 +
63 +
64 +@dataclass
65 +class SegmentModel:
66 + current: dict[str, dict[str, float]] = field(default_factory=dict) # seg → party → %
67 + baseline: dict[str, dict[str, float]] = field(default_factory=dict) # seg → party → % (2022)
68 + n_polls: dict[str, int] = field(default_factory=dict)
69 + excess: dict[str, dict[str, float]] = field(default_factory=dict) # seg → party → excès de swing (logit)
70 + national_now: dict[str, float] = field(default_factory=dict)
71 + national_2022: dict[str, float] = field(default_factory=dict)
72 + language_regression: dict[str, dict[str, float]] = field(default_factory=dict) # party → {a, b, r2} (diagnostic)
73 + reference_polls_2022: list[str] = field(default_factory=list)
74 +
75 + def payload(self) -> dict:
76 + return {"current": self.current, "baseline2022": self.baseline, "nPolls": self.n_polls,
77 + "excessSwingLogit": {k: {p: round(v, 3) for p, v in d.items()} for k, d in self.excess.items()},
78 + "languageRegressionDiagnostic": self.language_regression, "referencePolls2022": self.reference_polls_2022,
79 + "languageEnabled": "fr" in self.excess and "nonfr" in self.excess,
80 + "kappa": {"language": KAPPA_LANGUAGE, "region": KAPPA_REGION}, "window_days": WINDOW_DAYS}
81 +
82 +
83 +def current_segment_support(s: Session, cutoff: date) -> tuple[dict[str, dict[str, float]], dict[str, int]]:
84 + since = date.fromordinal(cutoff.toordinal() - WINDOW_DAYS)
85 + rows = s.execute(select(PollSubsample, Poll).join(Poll, Poll.id == PollSubsample.poll_id)
86 + .where(Poll.excluded == False, Poll.validated == True, Poll.fieldwork_end >= since, Poll.fieldwork_end <= cutoff, # noqa: E712
87 + PollSubsample.dimension.in_(["language", "region"]))).all()
88 + acc: dict[str, dict[str, float]] = defaultdict(lambda: defaultdict(float))
89 + wsum: dict[str, float] = defaultdict(float)
90 + polls: dict[str, set] = defaultdict(set)
91 + per_poll_seg: dict[tuple[str, str], dict[str, float]] = defaultdict(dict)
92 + meta: dict[tuple[str, str], tuple[float, int | None]] = {}
93 + for ss, poll in rows:
94 + key = canonical(ss.dimension, ss.segment)
95 + if not key or ss.party_id not in PARTY_IDS:
96 + continue
97 + per_poll_seg[(poll.id, key)][ss.party_id] = ss.value
98 + age = (cutoff - poll.fieldwork_end).days
99 + meta[(poll.id, key)] = (0.5 ** (age / HALFLIFE_DAYS) * math.sqrt(min(ss.n or 300, 1500) / 500.0), ss.n)
100 + for (pid, key), vals in per_poll_seg.items():
101 + if len(vals) < 4:
102 + continue
103 + tot = sum(vals.values())
104 + if not (80 <= tot <= 105):
105 + continue
106 + w = meta[(pid, key)][0]
107 + for party, v in vals.items():
108 + acc[key][party] += w * v
109 + wsum[key] += w
110 + polls[key].add(pid)
111 + out: dict[str, dict[str, float]] = {}
112 + for key, d in acc.items():
113 + if len(polls[key]) < MIN_POLLS:
114 + continue
115 + vals = {p: d[p] / wsum[key] for p in PARTY_IDS if p in d}
116 + rest = max(0.0, 100 - sum(vals.values()))
117 + vals["aut"] = rest
118 + out[key] = {p: round(v, 2) for p, v in vals.items()}
119 + return out, {k: len(v) for k, v in polls.items()}
120 +
121 +
122 +ELECTION_2022 = date(2022, 10, 3)
123 +
124 +
125 +def language_baseline_2022(s: Session, national_2022: dict[str, float]) -> tuple[dict[str, dict[str, float]], list[str]]:
126 + """Soutien 2022 par langue : sondages de fin de campagne 2022 (≤ 14 j avant le vote), recalés sur le résultat."""
127 + since = date.fromordinal(ELECTION_2022.toordinal() - 14)
128 + rows = s.execute(select(PollSubsample, Poll).join(Poll, Poll.id == PollSubsample.poll_id)
129 + .where(Poll.excluded == False, Poll.validated == True, Poll.fieldwork_end >= since, Poll.fieldwork_end < ELECTION_2022, # noqa: E712
130 + PollSubsample.dimension == "language")).all()
131 + per: dict[tuple[str, str], dict[str, float]] = defaultdict(dict)
132 + nat: dict[str, dict[str, float]] = {}
133 + used: set[str] = set()
134 + for ss, poll in rows:
135 + key = canonical(ss.dimension, ss.segment)
136 + if not key or ss.party_id not in PARTY_IDS:
137 + continue
138 + per[(poll.id, key)][ss.party_id] = ss.value
139 + nat[poll.id] = {r.party_id: r.value for r in poll.results}
140 + used.add(poll.id)
141 + if not per:
142 + return {}, []
143 + acc: dict[str, dict[str, list[float]]] = defaultdict(lambda: defaultdict(list))
144 + for (pid, key), vals in per.items():
145 + if len(vals) < 4:
146 + continue
147 + for p, v in vals.items():
148 + # recalage : le sondage sous/sur-estimait le parti au national → même correction dans le segment
149 + pn = nat[pid].get(p)
150 + corr = (national_2022.get(p, pn) / pn) if pn and pn > 0 else 1.0
151 + acc[key][p].append(v * corr)
152 + out: dict[str, dict[str, float]] = {}
153 + for key, d in acc.items():
154 + vals = {p: sum(v) / len(v) for p, v in d.items()}
155 + vals["aut"] = max(0.5, 100 - sum(vals.values()))
156 + tot = sum(vals.values())
157 + out[key] = {p: round(100 * v / tot, 2) for p, v in vals.items()}
158 + return out, sorted(used)
159 +
160 +
161 +def baseline_segments(s: Session, national_2022: dict[str, float] | None = None) -> tuple[dict[str, dict[str, float]], dict[str, dict[str, float]], dict[int, float], list[str]]:
162 + """Références 2022 : régions (agrégation exacte), langue (sondages de fin de campagne recalés) ; régression écologique = diagnostic."""
163 + ridings = {r.code: r for r in s.scalars(select(Riding)).all()}
164 + demo = {d.riding_code: d for d in s.scalars(select(RidingDemographics)).all()}
165 + votes: dict[int, dict[str, int]] = defaultdict(dict)
166 + for b in s.scalars(select(RidingBaseline)).all():
167 + votes[b.riding_code][b.party_id] = b.votes
168 + # régions
169 + reg: dict[str, dict[str, int]] = defaultdict(lambda: defaultdict(int))
170 + for code, d in votes.items():
171 + pr = poll_region_for(code)
172 + for p, v in d.items():
173 + reg[pr][p] += v
174 + region_base = {}
175 + for pr, d in reg.items():
176 + tot = sum(d.values()) or 1
177 + region_base[pr] = {p: round(100 * d.get(p, 0) / tot, 2) for p in PARTIES_ALL}
178 + # langue : régression pondérée share = a + b·f
179 + f_r: dict[int, float] = {}
180 + for code in votes:
181 + dm = demo.get(code)
182 + if dm and dm.pct_french_mt is not None:
183 + f_r[code] = dm.pct_french_mt / 100.0
184 + regression: dict[str, dict[str, float]] = {}
185 + if len(f_r) >= 50:
186 + codes = list(f_r)
187 + X = np.array([f_r[c] for c in codes])
188 + W = np.array([sum(votes[c].values()) for c in codes], dtype=float)
189 + for p in PARTIES_ALL:
190 + y = np.array([100 * votes[c].get(p, 0) / max(1, sum(votes[c].values())) for c in codes])
191 + A = np.vstack([np.ones_like(X), X]).T * np.sqrt(W)[:, None]
192 + coef, *_ = np.linalg.lstsq(A, y * np.sqrt(W), rcond=None)
193 + a, b = float(coef[0]), float(coef[1])
194 + pred = a + b * X
195 + ss_res = float((W * (y - pred) ** 2).sum())
196 + ss_tot = float((W * (y - np.average(y, weights=W)) ** 2).sum()) or 1.0
197 + regression[p] = {"a": round(a, 2), "b": round(b, 2), "r2": round(1 - ss_res / ss_tot, 3)}
198 + lang_base, ref_polls = language_baseline_2022(s, national_2022 or {})
199 + base = {**region_base, **lang_base}
200 + return base, regression, f_r, ref_polls
201 +
202 +
203 +def build_segment_model(s: Session, cutoff: date, national_now: dict[str, float], national_2022: dict[str, float]) -> tuple[SegmentModel, dict[int, float]]:
204 + cur, n_polls = current_segment_support(s, cutoff)
205 + base, regression, f_r, ref_polls = baseline_segments(s, national_2022)
206 + m = SegmentModel(current=cur, baseline=base, n_polls=n_polls, national_now=national_now, national_2022=national_2022, language_regression=regression)
207 + m.reference_polls_2022 = ref_polls
208 + nat_delta = {p: _logit(national_now.get(p, 1) / 100) - _logit(national_2022.get(p, 1) / 100) for p in PARTIES_ALL}
209 + for seg, curvals in cur.items():
210 + if seg not in base:
211 + continue
212 + m.excess[seg] = {}
213 + for p in PARTY_IDS:
214 + if p in curvals and p in base[seg]:
215 + d = _logit(curvals[p] / 100) - _logit(base[seg][p] / 100)
216 + m.excess[seg][p] = d - nat_delta[p]
217 + return m, f_r
218 +
219 +
220 +def riding_adjustments(m: SegmentModel, ridings_codes: list[int], f_r: dict[int, float]) -> np.ndarray:
221 + """Matrice (R, P) d'ajustements logit par circonscription."""
222 + R, P = len(ridings_codes), len(PARTIES_ALL)
223 + adj = np.zeros((R, P))
224 + has_lang = "fr" in m.excess and "nonfr" in m.excess
225 + for i, code in enumerate(ridings_codes):
226 + for j, p in enumerate(PARTY_IDS):
227 + a = 0.0
228 + if has_lang and code in f_r:
229 + f = f_r[code]
230 + a += KAPPA_LANGUAGE * (f * m.excess["fr"].get(p, 0.0) + (1 - f) * m.excess["nonfr"].get(p, 0.0))
231 + pr = poll_region_for(code)
232 + if pr in m.excess:
233 + a += KAPPA_REGION * m.excess[pr].get(p, 0.0)
234 + adj[i, j] = a
235 + return adj
modified engine/qc26/regions.py +20 −0
@@ -88,3 +88,23 @@ RIDING_REGION: dict[int, str] = {
88 88 }
89 89
90 90 assert len(RIDING_REGION) == 127, len(RIDING_REGION)
91 +
92 +# Regroupement « sondeurs » affiné par circonscription : les firmes ventilent Montréal RMR /
93 +# Québec RMR / reste du Québec. Les régions administratives débordent des RMR ; on corrige ici.
94 +_POLL_REGION_BY_REGION = {rid: pr for rid, _, pr in REGIONS}
95 +_POLL_REGION_OVERRIDES: dict[int, str] = {
96 + # Laurentides hors RMR de Montréal
97 + 589: "reste", 641: "reste", 649: "reste", 637: "reste",
98 + # Lanaudière hors RMR
99 + 627: "reste", 629: "reste", 631: "reste",
100 + # Montérégie hors RMR (Saint-Hyacinthe, Granby, Brome-Missisquoi, Iberville, Saint-Jean, Huntingdon, Richelieu, Beauharnois)
101 + 239: "reste", 159: "reste", 157: "reste", 161: "reste", 167: "reste", 171: "reste", 241: "reste", 179: "reste",
102 + # Capitale-Nationale hors RMR de Québec (Charlevoix, Portneuf)
103 + 771: "reste", 727: "reste",
104 + # Chaudière-Appalaches hors RMR (Beauces, Bellechasse, Côte-du-Sud, Lotbinière)
105 + 781: "reste", 777: "reste", 797: "reste", 799: "reste", 787: "reste",
106 +}
107 +
108 +
109 +def poll_region_for(code: int) -> str:
110 + return _POLL_REGION_OVERRIDES.get(code) or _POLL_REGION_BY_REGION.get(RIDING_REGION.get(code, ""), "reste")
modified engine/qc26/scheduler.py +6 −0
@@ -78,9 +78,15 @@ def job_daily_forecast():
78 78
79 79 @_guarded("documents")
80 80 def job_documents():
81 + from sqlalchemy import select
81 82 from .compass.positions import compute_positions
82 83 from .connectors.parties import documents
83 84 with session_scope() as s:
85 + # un travail documentaire lancé hors planificateur (admin, one-off) est peut-être en cours : ne pas doubler les appels LLM
86 + recent = s.scalar(select(IngestionJob).where(IngestionJob.kind.like("party-docs-%"), IngestionJob.status == "running",
87 + IngestionJob.started_at > utcnow() - timedelta(hours=3)))
88 + if recent:
89 + return {"skipped": f"travail {recent.kind} #{recent.id} en cours"}
84 90 res = documents.sync_all(s, analyze=True)
85 91 pos = compute_positions(s, only_missing=True)
86 92 return {"docs": res, "positions": pos}
modified web/src/app/circonscription/[slug]/page.tsx +28 −0
@@ -22,6 +22,8 @@ interface Detail {
22 22 forecast?: { runId: number; completedAt: string | null; modelVersion: string; favorite: string; p: number; runnerUp: string | null; margin: number; rating: string; volatility: number; swing: number; factors: { factor: string; label: string; detail: string; direction: string }[]; parties: { party: string; voteMean: number; vote80: [number, number]; pWin: number }[] };
23 23 live: { status: string; updatedAt: string | null; bureaux: [number, number]; votesValid: number; turnout: number | null; leader: string | null; marginVotes: number | null; marginPct: number | null; candidates: { name: string; party_id: string; votes: number; share: number }[] | null; final: boolean } | null;
24 24 nearby?: { code: number; name: string; slug: string }[];
25 + demographics?: { population2021: number; density: number | null; medianAge: number | null; pct65Plus: number | null; pctFrenchMt: number | null; pctEnglishMt: number | null; pctOtherMt: number | null; pctPlopFrench: number | null; pctImmigrants: number | null; pctVisibleMinority: number | null; medianHouseholdIncome: number | null; pctOwner: number | null; pctBachelorPlus: number | null; unemploymentRate: number | null; adaCount: number; source: string } | null;
26 + pollRegion?: string;
25 27 }
26 28
27 29 export async function generateMetadata({ params }: { params: Promise<{ slug: string }> }): Promise<Metadata> {
@@ -103,6 +105,32 @@ export default async function Page({ params }: { params: Promise<{ slug: string
103 105 </div>
104 106 </section>
105 107
108 + {d.demographics && (
109 + <section>
110 + <SectionHeader eyebrow="Démographie" title="Recensement 2021" description="Statistique Canada, agrégé aux limites 2026 par intersection des aires de diffusion agrégées. La part francophone alimente le swing par segment du modèle." action={<SourceBadge source={{ name: "Statistique Canada — Recensement 2021 (profil ADA + limites)", url: "https://www12.statcan.gc.ca/census-recensement/2021/dp-pd/prof/details/download-telecharger.cfm?Lang=F", detail: `${d.demographics.adaCount} aires de diffusion agrégées intersectées`, tier: "Officiel (StatCan) · agrégation QC26" }} />} />
111 + <div className="grid grid-cols-2 md:grid-cols-4 lg:grid-cols-7 gap-3">
112 + {([
113 + ["Population 2021", int(d.demographics.population2021), ""],
114 + ["Langue maternelle française", d.demographics.pctFrenchMt, " %"],
115 + ["Langue maternelle anglaise", d.demographics.pctEnglishMt, " %"],
116 + ["Autres langues maternelles", d.demographics.pctOtherMt, " %"],
117 + ["Immigrant·es", d.demographics.pctImmigrants, " %"],
118 + ["Minorités visibles", d.demographics.pctVisibleMinority, " %"],
119 + ["65 ans et plus", d.demographics.pct65Plus, " %"],
120 + ["Âge médian", d.demographics.medianAge, " ans"],
121 + ["Revenu médian des ménages", d.demographics.medianHouseholdIncome ? int(d.demographics.medianHouseholdIncome) : null, " $"],
122 + ["Propriétaires", d.demographics.pctOwner, " %"],
123 + ["Baccalauréat ou plus (25-64)", d.demographics.pctBachelorPlus, " %"],
124 + ["Taux de chômage", d.demographics.unemploymentRate, " %"],
125 + ["Densité", d.demographics.density ? int(d.demographics.density) : null, " hab./km²"],
126 + ["Regroupement sondeurs", d.pollRegion === "montreal_rmr" ? "Grand Montréal" : d.pollRegion === "quebec_rmr" ? "Région de Québec" : "Reste du Québec", ""],
127 + ] as [string, string | number | null, string][]).map(([label, value, suffix]) => (
128 + <div key={label} className="card p-3"><div className="text-[11px] text-ink-3 leading-tight">{label}</div><div className="num font-bold text-[18px] mt-1">{value === null || value === undefined ? "—" : typeof value === "number" ? value.toLocaleString("fr-CA", { maximumFractionDigits: 1 }) : value}<span className="text-[12px] font-medium text-ink-2">{value === null ? "" : suffix}</span></div></div>
129 + ))}
130 + </div>
131 + </section>
132 + )}
133 +
106 134 {years.length > 0 && (
107 135 <section>
108 136 <SectionHeader eyebrow="Historique" title="Résultats officiels" description={d.historyNote} />
modified web/src/app/methodologie/page.tsx +6 −2
@@ -30,7 +30,11 @@ export default async function Page() {
30 30
31 31 <Section id="temporel" title="Modèle temporel"><p>L&apos;état latent de l&apos;opinion est estimé par un <b>filtre de Kalman</b> (marche aléatoire quotidienne, écart-type 0,012 dans l&apos;espace logit) avec lissage de Rauch–Tung–Striebel. La variance d&apos;observation de chaque sondage vient de sa taille effective (n divisé par un design effect selon le mode : web 1,30, IVR 1,40, téléphone 1,20) et d&apos;une erreur non-échantillonnale de 1,5 point (« total survey error »). La projection au jour du vote ajoute une dérive proportionnelle aux jours restants (inflation 1,5) et une <b>erreur d&apos;industrie</b> corrélée entre partis (σ ≈ 2,2 pt), calibrée sur l&apos;écart entre sondages finaux et résultats en 2018 et 2022 : les firmes se trompent ensemble, et le modèle le sait.</p></Section>
32 32
33 − <Section id="regional" title="Projection régionale"><p>Version 1 : le swing est <b>proportionnel</b> et uniforme à l&apos;échelle du Québec, puis perturbé par une erreur régionale (σ 0,12 logit) indépendante entre 17 régions (classification QC26 alignée sur les régions administratives). Les ventilations régionales publiées par les firmes sont archivées (visibles dans le Poll Inspector) mais n&apos;entrent pas encore dans le modèle ; c&apos;est la prochaine étape du changelog.</p></Section>
33 + <Section id="regional" title="Projection régionale et swing par segments"><p>Le swing de base est <b>proportionnel</b> à l&apos;échelle du Québec. Depuis la version 1.1, il est corrigé par un <b>excès de swing par segment</b> tiré des ventilations publiées par les firmes dans les 90 derniers jours (pondérées récence × √n) :</p>
34 + <ul className="list-disc pl-5 space-y-1"><li><b>Langue</b> : soutien chez les francophones et les non-francophones aujourd&apos;hui, comparé à la référence 2022 (sondages Léger de fin de campagne 2022, documents archivés, recalés sur le résultat officiel). L&apos;écart entre le swing du segment et le swing national est appliqué à chaque circonscription selon sa part de langue maternelle française au <b>Recensement 2021</b> (Statistique Canada, aires de diffusion agrégées intersectées avec les limites 2026), avec un rétrécissement κ = 0,6.</li>
35 + <li><b>Région</b> : RMR de Montréal, RMR de Québec, reste du Québec, comparés aux résultats 2022 agrégés selon le même regroupement (affiné circonscription par circonscription), κ = 0,5.</li>
36 + <li><b>Calibration</b> : à chaque tirage, la moyenne des parts pondérée par les électeurs inscrits est ramenée au vote national simulé.</li></ul>
37 + <p>Une erreur régionale (σ 0,12 logit) indépendante entre 17 régions s&apos;ajoute. La régression écologique des résultats 2022 sur la part francophone est publiée dans l&apos;API à titre de diagnostic mais n&apos;est pas utilisée : son extrapolation aux circonscriptions sans francophones est instable.</p></Section>
34 38
35 39 <Section id="circonscriptions" title="Modèle de circonscription"><p>Le point de départ est le <b>résultat 2022 transposé sur la carte 2026</b> : les résultats officiels par section de vote (Élections Québec) sont affectés à la circonscription 2026 contenant le centroïde de la section, puis les totaux officiels de chaque circonscription 2022 sont répartis selon cette distribution (hypothèse : les votes par anticipation suivent les votes du jour). Cette transposition reproduit exactement les chiffres publiés lors de la refonte de la carte (CAQ 93, PLQ 20, QS 11, PQ 3). Pour chaque simulation, la part du parti p dans la circonscription r vaut baseline<sub>r,p</sub> × (vote national simulé<sub>p</sub> / vote national 2022<sub>p</sub>), multipliée par les erreurs régionale et locale (σ 0,20 logit) et par un effet candidat : −0,06 logit (≈ 1,3 pt) quand le·la député·e sortant·e ne se représente pas. Un parti sans candidature n&apos;est pénalisé que lorsque sa liste est quasi complète (≥ 90 % des circonscriptions), pour ne pas interpréter un retard d&apos;enregistrement.</p></Section>
36 40
@@ -42,7 +46,7 @@ export default async function Page() {
42 46
43 47 <Section id="boussole" title="Boussole"><p>48 énoncés sur 6 axes. Les positions des partis sont dérivées exclusivement des propositions extraites de leurs documents officiels (page + citation vérifiée dans le texte) : un modèle de langage situe le parti sur l&apos;échelle −2…+2 <i>uniquement</i> si les propositions le permettent, sinon la position est « introuvable » et la question ne compte pas pour ce parti. La concordance utilisateur–parti est 1 − |u − p|/4, pondérée par l&apos;importance. Les réponses restent dans le navigateur.</p></Section>
44 48
45 − <Section id="sources" title="Sources"><ul className="list-disc pl-5 space-y-1"><li><b>Élections Québec</b> (données ouvertes officielles) : liste et géométrie des 127 circonscriptions 2026, électeurs inscrits, candidatures, partis autorisés et chef·fes, résultats 2018 et 2022 par circonscription et par section de vote, résultats du soir du vote.</li><li><b>Firmes de sondage</b> : documents originaux (Léger, Pallas Data, Synopsis, Angus Reid, Mainstreet, SEGMA, Abacus, Research Co., etc.).</li><li><b>Sites officiels des partis</b> : plateformes, programmes, engagements, cadres financiers, communiqués — versionnés.</li><li>Un index tiers (Wikipédia) sert uniquement à amorcer la liste des sondages et à repérer les documents originaux ; il est remplacé dès que le document est archivé.</li></ul></Section>
49 + <Section id="sources" title="Sources"><ul className="list-disc pl-5 space-y-1"><li><b>Élections Québec</b> (données ouvertes officielles) : liste et géométrie des 127 circonscriptions 2026, électeurs inscrits, candidatures, partis autorisés et chef·fes, résultats 2018 et 2022 par circonscription et par section de vote, résultats du soir du vote.</li><li><b>Firmes de sondage</b> : documents originaux (Léger, Pallas Data, Synopsis, Angus Reid, Mainstreet, SEGMA, Abacus, Research Co., etc.).</li><li><b>Sites officiels des partis</b> : plateformes, programmes, engagements, cadres financiers, communiqués — versionnés.</li><li><b>Statistique Canada</b> : Recensement de la population 2021, profil des aires de diffusion agrégées et fichier des limites, agrégés aux 127 circonscriptions (langue, immigration, âge, revenu, tenure, scolarité, chômage).</li><li>Un index tiers (Wikipédia) sert uniquement à amorcer la liste des sondages et à repérer les documents originaux ; il est remplacé dès que le document est archivé.</li></ul></Section>
46 50
47 51 <Section id="limites" title="Limites"><ul className="list-disc pl-5 space-y-1"><li>Le swing est uniforme ; les dynamiques régionales spécifiques (Québec, Montréal, régions) ne sont captées que par un bruit régional, pas encore par les ventilations des sondages.</li><li>Les effets candidats se limitent au retrait du sortant ; notoriété locale et vedettes ne sont pas modélisées.</li><li>L&apos;erreur d&apos;industrie est calibrée sur deux élections seulement.</li><li>Le nombre de sondages par firme est faible pour plusieurs maisons : leurs effets maison ne sont pas estimés (aucune correction appliquée).</li><li>L&apos;extraction documentaire dépend d&apos;un modèle de langage ; les propositions à faible confiance sont mises en file de révision et non publiées.</li></ul><div className="mt-4"><Disclaimer /></div></Section>
48 52 </article>
49 53