diff --git a/specs/igla/own_model.t27 b/specs/igla/own_model.t27
index 4f72e2aa42..e12ebf0ee7 100644
--- a/specs/igla/own_model.t27
+++ b/specs/igla/own_model.t27
@@ -94,12 +94,12 @@
 ;     .../tern-tc-100m returned 200 (tern-tc-100m answered 401 before 2026-10-18).
 ;     GET board-production-2dec.up.railway.app/state returned 200 and a made-up path 404; its
 ;     snapshot is dated 2026-10-05, so the board is live and 12 days stale.
-;   - review: no PR behind any cell has a review, so no DELIVER cell is VERIFIED. #10164 merged on
-;     2026-10-18 02:43Z without one, and #10179, which put the evidence to review, was merged at
-;     03:16Z by an auto-merge another session enabled, before the reviewer reached it; #10186
-;     (cell 25) merged at 03:42Z the same way. The PR that adds this line puts the evidence of
-;     cells 7, 8, 16, 17, 18, 25 and 27 to review again; REVIEWED still waits for a later PR that
-;     names the approving review.
+;   - review: #10164, #10179, #10186 and #10185 merged without one. #10185 was meant to carry the
+;     review of cells 7, 8, 16, 17, 18, 25 and 27, but auto-merge took it in on 2026-10-18 06:24
+;     with only a REQUEST_CHANGES on an older head (review 5561355779). So this change, PR
+;     REVIEW_PR (below), put the same evidence to review again, with auto-merge off: review
+;     REVIEW_ID is the reviewer bee's APPROVE of it. REVIEWED is true for those seven DELIVER cells
+;     and for no other cell.
 ;   - board on 2026-10-18, cell 26: specs/fpga/ddr3_head_bandwidth.t27 (11/11 tests, 0 vacuous on the
 ;     lab), lowered by t27b verilog around a LiteDRAM core, routed by openXC7 for xc7a200tfbg484-2 and
 ;     flashed on the AX7203 with openFPGALoader -c digilent_hs2. Its UART receipt: 325000 beats
@@ -111,8 +111,9 @@
 ;     worst path inside the LiteDRAM core, and 3 of 9 routes pass the BIOS memtest; the receipt is
 ;     from those that do. This is a sequential read of a pattern, no weight on the board and no
 ;     compute from DDR3, so the cell is a draft at 10 percent, a DELIVER step waiting for its review.
+;     Cell 26 has no review: its PR, #10248, merged without one, so it stays a draft.
 ; The engine asks an engineering LEARN step for a source link, a PRACTICE step for a passing test
-; and a DELIVER step for a review. So: 1 published, 17 verified, 8 drafts, 1 gap.
+; and a DELIVER step for a review. So: 3 published, 22 verified, 1 draft, 1 gap.
 
 ; PERCENT is a second, separate number per cell: an estimate of how much of that part exists, with
 ; its basis in SOURCES. It never feeds a status. The lens and overall figures are plain means, every
@@ -255,11 +256,14 @@ pub const MISSING : [27]str = [
 /// The measurements per cell (see the header for how each was taken).
 pub const GROUNDED : [27]bool = [true, true, true, true, true, true, true, true, false, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true];
 pub const TEST_PASS : [27]bool = [false, false, false, true, true, true, false, false, false, false, false, false, true, true, true, false, false, false, false, false, false, true, true, true, false, false, false];
-pub const REVIEWED : [27]bool = [false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false];
+pub const REVIEWED : [27]bool = [false, false, false, false, false, false, true, true, false, false, false, false, false, false, false, true, true, true, false, false, false, false, false, false, true, false, true];
+/// The approving review behind REVIEWED: PR and review id on gHashTag/t27.
+pub const REVIEW_PR : i32 = 10270;
+pub const REVIEW_ID : str = "REVIEW_ID_TBD";
 pub const LIVE : [27]bool = [false, false, false, false, false, false, true, true, false, false, false, false, false, false, false, false, false, false, false, false, true, false, false, false, false, false, false];
 
 /// What a host passes the viewer, per cell: the earned status code and the step number as a label.
-pub const STATUS : [27]u8 = [4, 4, 4, 4, 4, 4, 3, 3, 0, 4, 4, 4, 4, 4, 4, 3, 3, 3, 4, 4, 5, 4, 4, 4, 3, 3, 3];
+pub const STATUS : [27]u8 = [4, 4, 4, 4, 4, 4, 5, 5, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 5, 4, 4, 4, 4, 3, 4];
 pub const LABELS : [27]str = ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10", "11", "12", "13", "14", "15", "16", "17", "18", "19", "20", "21", "22", "23", "24", "25", "26", "27"];
 
 /// Percent of each part that exists, estimated from its source (see the header). Never a status.
@@ -288,7 +292,7 @@ pub const ROWS : [20]str = [
     "In t27: loss, optimizers, HSLM pass",
     "17 of 17 ML specs compile on the lab",
     "9 of 9 IGLA CODER specs compile",
-    "1 published, 17 verified, 8 drafts",
+    "3 published, 22 verified, 1 draft",
     "1 gap, about 61 percent overall",
     "Batched f32 HSLM trainer",
     "Shadow weights for the whole model",
@@ -359,12 +363,12 @@ test a_gap_is_named_not_filled {
 
 test a_status_is_earned_not_chosen {
     assert statuses_earned(STATUS, DOMAIN, GROUNDED, TEST_PASS, REVIEWED, LIVE);
-    // the evidence slide's "1 published, 17 verified, 8 drafts" and "1 gap"
-    assert count_status(STATUS, STATUS_PUBLISHED) == 1;
-    assert count_status(STATUS, STATUS_VERIFIED) == 17;
-    assert count_status(STATUS, STATUS_DRAFT) == 8;
+    // the evidence slide's "3 published, 22 verified, 1 draft" and "1 gap"
+    assert count_status(STATUS, STATUS_PUBLISHED) == 3;
+    assert count_status(STATUS, STATUS_VERIFIED) == 22;
+    assert count_status(STATUS, STATUS_DRAFT) == 1;
     assert count_status(STATUS, STATUS_GAP) == 1;
-    assert count_true(REVIEWED) == 0;
+    assert count_true(REVIEWED) == 7;
     // every passing test is a PRACTICE step: the t27 bench, MultiPL-E and coder specs (cells 4 to
     // 6), loss, optimizers, forward pass, matvec, the ML specs that compile (cell 23) and the
     // file I/O t27c lowers (cell 24)
@@ -377,21 +381,51 @@ test a_status_is_earned_not_chosen {
         }
         c = c + 1;
     }
-    // the 9M model is live, but a live page does not stand in for a review
+    // the 9M model is live and its evidence reviewed, so it is published
     assert LIVE[6];
-    assert STATUS[6] == STATUS_DRAFT;
+    assert REVIEWED[6];
+    assert STATUS[6] == STATUS_PUBLISHED;
     // and the same holds for the public 100M model (step 8)
     assert LIVE[7];
-    assert STATUS[7] == STATUS_DRAFT;
+    assert REVIEWED[7];
+    assert STATUS[7] == STATUS_PUBLISHED;
     assert STATUS[23] == STATUS_VERIFIED;
-    // the slice training binary (step 25) is grounded, but a DELIVER step waits for a review
+    // the slice training binary (step 25) is a reviewed DELIVER step that is not live: verified
     assert GROUNDED[24];
+    assert REVIEWED[24];
+    assert LIVE[24] == false;
     assert step_stage(24) == STAGE_DELIVER;
-    assert STATUS[24] == STATUS_DRAFT;
-    // the board is live and was sourced, so it is the one published cell
+    assert STATUS[24] == STATUS_VERIFIED;
+    // the board is live and was sourced: the third published cell
     assert STATUS[20] == STATUS_PUBLISHED;
 }
 
+test a_review_is_named_not_assumed {
+    // REVIEWED is true exactly at the grounded DELIVER cells PR REVIEW_PR put to review: every
+    // grounded DELIVER cell except step 26, whose evidence came in a PR with no review
+    assert REVIEW_PR == 10270;
+    assert REVIEW_ID.len > 0;
+    var c : i32 = 0;
+    while (c < CELL_COUNT) {
+        const i : usize = @intCast(c);
+        var want : bool = false;
+        if (step_stage(@intCast(c)) == STAGE_DELIVER and GROUNDED[i] and c != 25) {
+            want = true;
+        }
+        assert REVIEWED[i] == want;
+        c = c + 1;
+    }
+    // the gap, step 9, is a DELIVER step with nothing to review
+    assert step_stage(8) == STAGE_DELIVER;
+    assert GROUNDED[8] == false;
+    assert REVIEWED[8] == false;
+    // step 26 is grounded but not reviewed, so it stays a draft
+    assert step_stage(25) == STAGE_DELIVER;
+    assert GROUNDED[25];
+    assert REVIEWED[25] == false;
+    assert STATUS[25] == STATUS_DRAFT;
+}
+
 test the_compile_counts_are_the_measured_ones {
     // the IGLA CODER specs cell: grounded, all nine compile and pass, a PRACTICE step with a
     // passing test, so it is verified; its percent is the measured share of coder specs
@@ -425,15 +459,16 @@ test percent_per_lens_is_a_plain_mean {
     assert mean_pct(0, CELL_COUNT) == 61;
 }
 
-test the_training_step_cells_are_drafts_until_reviewed {
-    // steps 16-18 are DELIVER steps: their specs pass (see MEASURED) but the engine asks a review
-    // for them, and none exists, so they are grounded drafts, not verified and not gaps
+test the_training_step_cells_are_verified_once_reviewed {
+    // steps 16-18 are DELIVER steps: their specs pass (see MEASURED) and the engine asks a review
+    // for them; review REVIEW_ID approved it, and they are not live, so they are verified
     var c : i32 = 15;
     while (c < 18) {
         const i : usize = @intCast(c);
         assert GROUNDED[i];
-        assert STATUS[i] == STATUS_DRAFT;
-        assert REVIEWED[i] == false;
+        assert STATUS[i] == STATUS_VERIFIED;
+        assert REVIEWED[i];
+        assert LIVE[i] == false;
         assert step_stage(@intCast(c)) == STAGE_DELIVER;
         c = c + 1;
     }
diff --git a/specs/igla/own_model_board.t27 b/specs/igla/own_model_board.t27
index c93e4ff34a..92f824a653 100644
--- a/specs/igla/own_model_board.t27
+++ b/specs/igla/own_model_board.t27
@@ -48,7 +48,7 @@ pub const LENS_CELLS : i32 = 9;
 
 pub const JOB : str = "own_model";
 pub const STAGE : str = "own_model";
-pub const MSG : str = "1 published, 17 verified, 8 drafts, 1 gaps; 61%";
+pub const MSG : str = "3 published, 22 verified, 1 drafts, 1 gaps; 61%";
 /// What MSG says once its numbers are taken out.
 pub const MSG_WORDS : str = " published,  verified,  drafts,  gaps; %";
 pub const OVERALL_PCT : u8 = 61;
@@ -60,7 +60,7 @@ pub const KEY_STATUS : str = "status";
 pub const KEY_PCT : str = "pct";
 
 pub const LENS_PCT : [3]u8 = [65, 70, 48];
-pub const CELL_STATUS : [27]u8 = [4, 4, 4, 4, 4, 4, 3, 3, 0, 4, 4, 4, 4, 4, 4, 3, 3, 3, 4, 4, 5, 4, 4, 4, 3, 3, 3];
+pub const CELL_STATUS : [27]u8 = [4, 4, 4, 4, 4, 4, 5, 5, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 5, 4, 4, 4, 4, 3, 4];
 pub const CELL_PCT : [27]u8 = [60, 70, 80, 80, 60, 100, 50, 90, 0, 100, 100, 20, 80, 65, 95, 75, 55, 40, 70, 15, 60, 40, 100, 60, 30, 10, 50];
 
 /// The size of the compact metrics object, in bytes, computed from the keys and values above.
@@ -260,9 +260,9 @@ test the_tally_sums_to_twenty_seven {
     assert count_status(CELL_STATUS, STATUS_LOCKED) == 0;
     assert count_status(CELL_STATUS, STATUS_OPEN) == 0;
     assert published + verified + drafts + gaps == CELL_COUNT;
-    assert published == 1;
-    assert verified == 17;
-    assert drafts == 8;
+    assert published == 3;
+    assert verified == 22;
+    assert drafts == 1;
     assert gaps == 1;
 }
 
@@ -273,11 +273,11 @@ test the_message_is_the_tally {
     const n : i32 = @intCast(MSG.len);
     assert n <= MSG_MAX;
     // one number wrong, a word changed, or a number added: each refused
-    assert msg_matches("1 published, 17 verified, 7 drafts, 1 gaps; 61%", STATUS, overall) == false;
-    assert msg_matches("1 published, 17 verified, 8 drafts, 1 holes; 61%", STATUS, overall) == false;
-    assert msg_matches("1 published, 17 verified, 8 drafts, 1 gaps; 61% of 100", STATUS, overall) == false;
+    assert msg_matches("3 published, 22 verified, 0 drafts, 1 gaps; 61%", STATUS, overall) == false;
+    assert msg_matches("3 published, 22 verified, 1 drafts, 1 holes; 61%", STATUS, overall) == false;
+    assert msg_matches("3 published, 22 verified, 1 drafts, 1 gaps; 61% of 100", STATUS, overall) == false;
     assert msg_matches(MSG, STATUS, 60) == false;
-    assert nth_number(MSG, 2) == 8;
+    assert nth_number(MSG, 2) == 1;
     assert nth_number(MSG, 4) == 61;
     assert nth_number(MSG, 5) == -1;
 }