diff --git a/apps/web/cypress/e2e/admin-monitoring.cy.ts b/apps/web/cypress/e2e/admin-monitoring.cy.ts
index fd6ba78..733e184 100644
--- a/apps/web/cypress/e2e/admin-monitoring.cy.ts
+++ b/apps/web/cypress/e2e/admin-monitoring.cy.ts
@@ -53,21 +53,33 @@ describe("Admin monitoring", () => {
   });
 
   // The page route `/admin/monitoring` and the API path `GET /admin/monitoring`
-  // are identical — a plain `**/admin/monitoring` glob would also match the
+  // are identical — a plain `**/admin/monitoring` glob also matches the
   // `cy.visit()` document request (and whether that's same-origin or not
-  // depends on `VITE_API_URL`, which isn't set in CI). `resourceType: "fetch"`
-  // pins the intercept to the `AdminApiClient` XHR only.
-  const monitoringApi = {
-    method: "GET",
-    url: "**/admin/monitoring",
-    resourceType: "fetch",
-  } as const;
+  // depends on `VITE_API_URL`, which isn't set in CI).
+  //
+  // This used to be disambiguated with a declarative `resourceType: "fetch"`
+  // route matcher, but that stopped working (Cypress 13.17.0 on this
+  // machine's Chrome 152): the *declarative* `resourceType` matcher no
+  // longer catches this request at all — confirmed by logging every request
+  // Cypress hands to a `{resourceType: "fetch"}` intercept, which only ever
+  // saw `/admin/auth/me`, never `/admin/monitoring` — even though the very
+  // same request's own `req.resourceType`, read *inside* a handler matched
+  // by url/method alone, reliably reads back `"fetch"`. So the request *is*
+  // correctly typed; only the pre-filtering by that type is broken. Reading
+  // `req.resourceType` inside the handler and branching manually (as below)
+  // isn't affected by that bug and reproduced correctly in every check.
+  function interceptMonitoringApi(response: { statusCode: number; body: unknown }) {
+    return cy.intercept("GET", "**/admin/monitoring", (req) => {
+      if (req.resourceType === "document") {
+        req.continue();
+        return;
+      }
+      req.reply(response);
+    });
+  }
 
   it("renders one card per service with its status and details", () => {
-    cy.intercept(monitoringApi, {
-      statusCode: 200,
-      body: monitoringFixture(),
-    }).as("getMonitoring");
+    interceptMonitoringApi({ statusCode: 200, body: monitoringFixture() }).as("getMonitoring");
     cy.visit("/admin/monitoring");
     cy.wait("@getMonitoring");
 
@@ -85,11 +97,17 @@ describe("Admin monitoring", () => {
   });
 
   it("shows an error state when the request fails", () => {
-    cy.intercept(monitoringApi, {
-      statusCode: 500,
-      body: { code: 5000, message: "x" },
-    });
+    // Aliased and waited on (unlike the other tests' plain intercepts) so
+    // this genuinely proves the 500 came from the mock — without a live API
+    // server, an unmatched request fails at the network level too, which
+    // `MonitoringPage`'s `.catch()` can't tell apart from a real 500 (see
+    // the comment above `interceptMonitoringApi`, this is exactly the class
+    // of bug that let this test pass for the wrong reason before).
+    interceptMonitoringApi({ statusCode: 500, body: { code: 5000, message: "x" } }).as(
+      "getMonitoringError",
+    );
     cy.visit("/admin/monitoring");
+    cy.wait("@getMonitoringError");
     cy.contains("Impossible de charger").should("be.visible");
   });
 });
diff --git a/apps/web/cypress/e2e/recipes.cy.ts b/apps/web/cypress/e2e/recipes.cy.ts
index a3190df..d3a4958 100644
--- a/apps/web/cypress/e2e/recipes.cy.ts
+++ b/apps/web/cypress/e2e/recipes.cy.ts
@@ -69,6 +69,64 @@ const omeletteDetail = {
   ],
 };
 
+/**
+ * A 3-step recipe whose middle step already has an existing technique match
+ * carrying attached ingredient/utensil metadata — for the step-numbering
+ * regression test below, which needs a correction popover with several
+ * `<li>` rows of its own open at once (see that test's own comment).
+ */
+const tarteDetail = {
+  id: 5,
+  name: "Tarte",
+  description: null,
+  picture: null,
+  portions: 4,
+  authorId: 1,
+  visibility: "PERSONAL",
+  allergens: [],
+  diets: [],
+  isFavorite: false,
+  ingredients: [],
+  steps: [
+    { id: 1, description: "Préchauffer le four.", picture: null, order: 1, techSteps: [] },
+    {
+      id: 2,
+      description: "Faire revenir les oignons avec une poêle.",
+      picture: null,
+      order: 2,
+      techSteps: [
+        {
+          techStep: { id: 2, key: "panFry" },
+          start: 0,
+          end: 13,
+          source: "auto",
+          ingredients: [
+            {
+              ingredient: {
+                id: 10,
+                key: "onion",
+                icon: "VEGETABLE",
+                category: "freshProduce",
+                subcategory: "vegetables",
+                allergens: [],
+                diets: [],
+              },
+              quantity: 2,
+              unit: { id: 2, key: "piece", type: "COUNT", toBaseFactor: 1 },
+              start: 18,
+              end: 25,
+              source: "auto",
+            },
+          ],
+          utensils: [{ utensil: { id: 20, key: "pan" }, start: 31, end: 40, source: "auto" }],
+          temperatures: [],
+        },
+      ],
+    },
+    { id: 3, description: "Servir chaud.", picture: null, order: 3, techSteps: [] },
+  ],
+};
+
 function interceptAuth() {
   cy.intercept("GET", "**/auth/me", { statusCode: 200, body: authenticatedProfile });
   cy.intercept("GET", "**/profile/disliked-ingredients", { statusCode: 200, body: [] });
@@ -187,6 +245,54 @@ describe("Recipe catalog", () => {
     });
   });
 
+  // Regression: `.recipe-detail-panel__steps`'s CSS step counter
+  // (recipes.scss) used a bare descendant `li` selector, matching *any*
+  // `<li>` nested inside a step — including ones several levels down
+  // inside `TechStepCorrectionPopover` (each attached ingredient/utensil/
+  // temperature row, and the technique picker's own `<li>` list). Opening
+  // a correction with attached metadata inflated the counter, throwing off
+  // every *later* step's own displayed number — exactly the "les étapes
+  // affichées sont numérotées n'importe comment" bug report this covers.
+  it("keeps every step's own number correct while a correction popover with attached metadata is open", () => {
+    cy.viewport(1600, 900);
+    cy.intercept("GET", /\/recipes\?/, { statusCode: 200, body: [] });
+    cy.intercept("GET", "**/recipes/5", { statusCode: 200, body: tarteDetail });
+    cy.intercept("GET", "**/reference/tech-steps", {
+      statusCode: 200,
+      body: [{ id: 2, key: "panFry" }],
+    });
+
+    cy.visit("/recettes/5");
+    cy.get(".recipe-detail-panel__steps > li").should("have.length", 3);
+
+    // Open the middle step's correction — pre-seeded with one ingredient
+    // and one utensil, so its popover renders two `.tech-step-correction-popover__row`
+    // `<li>`s of its own (plus the technique picker's own `<li>` list once
+    // "Changer" is clicked) sitting *inside* the second step's own `<li>`.
+    cy.contains(".step-tech-step", "Faire revenir").click();
+    cy.get(".tech-step-correction-popover__row").should("have.length", 2);
+    cy.contains(".tech-step-correction-popover__technique-change", "Changer").click();
+    cy.get(".catalog-search-picker__list li").should("have.length.greaterThan", 0);
+
+    // None of the popover's own nested `<li>`s join the step counter chain
+    // — `getComputedStyle(...).content` isn't reliable for reading a
+    // *resolved* counter() value across browsers (it can return the raw
+    // `counter(step) ". "` declaration instead), so this checks the actual
+    // `counter-increment` computed property instead: a real, directly
+    // testable signal for exactly what `> li` (recipes.scss) does and
+    // doesn't scope to.
+    cy.get(".tech-step-correction-popover__row").each(($row) => {
+      expect(window.getComputedStyle($row[0]).counterIncrement).to.equal("none");
+    });
+    cy.get(".catalog-search-picker__list li").each(($item) => {
+      expect(window.getComputedStyle($item[0]).counterIncrement).to.equal("none");
+    });
+    // The step `<li>`s themselves are still correctly in the chain.
+    cy.get(".recipe-detail-panel__steps > li").each(($step) => {
+      expect(window.getComputedStyle($step[0]).counterIncrement).to.equal("step 1");
+    });
+  });
+
   it("shows a not-found message for a selected id the API rejects", () => {
     cy.intercept("GET", /\/recipes\?/, { statusCode: 200, body: [] });
     cy.intercept("GET", "**/recipes/999", {
diff --git a/apps/web/cypress/support/step_definitions/common.steps.ts b/apps/web/cypress/support/step_definitions/common.steps.ts
index c74aac9..272a2ef 100644
--- a/apps/web/cypress/support/step_definitions/common.steps.ts
+++ b/apps/web/cypress/support/step_definitions/common.steps.ts
@@ -180,7 +180,12 @@ Then("the ingredient {string} should be visible in the picker", (name: string) =
 });
 
 Then("the recipe should include the ingredient {string}", (name: string) => {
-  cy.contains(".ingredient-row__name", name).should("be.visible");
+  // Same reasoning as "I should see the highlighted technique" above: on the
+  // recipe-edit form the ingredients list can sit below the fold of
+  // `.app-content`'s own scroll (Nom/Description/Photo/Portions/Visibilité/
+  // Régimes all come first), and a bare `.should("be.visible")` doesn't
+  // auto-scroll — only interaction commands like `.click()`/`.check()` do.
+  cy.contains(".ingredient-row__name", name).scrollIntoView().should("be.visible");
 });
 
 Then("there should be {int} ingredient rows", (count: number) => {
diff --git a/apps/web/src/features/recipes/recipes.scss b/apps/web/src/features/recipes/recipes.scss
index de13290..8f6b260 100644
--- a/apps/web/src/features/recipes/recipes.scss
+++ b/apps/web/src/features/recipes/recipes.scss
@@ -630,7 +630,17 @@
     gap: var(--space-md);
     counter-reset: step;
 
-    li {
+    // `> li`, not a bare descendant `li` — this used to match *any* `<li>`
+    // nested anywhere inside a step, not just the step's own direct one.
+    // Harmless while `StepDescription`'s editable mode only ever added
+    // plain text/buttons, but `TechStepCorrectionPopover` renders its own
+    // `<li>`s a few levels down (each attached ingredient/utensil/
+    // temperature row, `.tech-step-correction-popover__row`; the technique
+    // picker's `.catalog-search-picker__list`, up to ~74 entries) —
+    // every one of those was incrementing this *same* counter, throwing
+    // every step's own number off as soon as any correction popover with
+    // attached metadata (or its technique picker) was open.
+    > li {
       counter-increment: step;
 
       &::before {
@@ -639,14 +649,20 @@
         color: var(--color-primary);
       }
 
-      img {
+      // Same descendant-selector fix as `> li` above, for the same reason:
+      // both the step's own picture and its `StepDescription`-rendered `<p>`
+      // are direct children of this `<li>` — `TechStepCorrectionPopover`'s
+      // own `<p>`s (the selection line, hints, `.field-error`) sit deeper
+      // inside it and must not inherit "inline, no margin" meant only for
+      // the step's own paragraph.
+      > img {
         display: block;
         max-width: 100%;
         border-radius: var(--radius-base);
         margin-bottom: var(--space-xs);
       }
 
-      p {
+      > p {
         display: inline;
         margin: 0;
       }