diff --git a/.github/workflows/scripts/check-review-execution.sh b/.github/workflows/scripts/check-review-execution.sh
index d74c588a..f1c1889f 100755
--- a/.github/workflows/scripts/check-review-execution.sh
+++ b/.github/workflows/scripts/check-review-execution.sh
@@ -627,11 +627,75 @@ review_text_file="$(mktemp)"
 # the reader cannot see. A single-verdict transcript (the common case, and
 # the gha#173 wrap-up-after-verdict shape) selects exactly that one block,
 # unchanged.
+# gha#805: the gha#710 widening has a failure mode of its own. A reviewer that
+# REDRAFTS its final message -- re-running an instrument between drafts, each
+# draft a complete review with its own `### Verdict` heading and
+# structured-review-data block -- produced three verdict-bearing blocks, and
+# the span rule concatenated all three (measured on Morrison-Lab/ai-config#2966,
+# run 33594599768: three `## Review` headings, three verdicts, two
+# Stopping-Point lines in one comment). The two shapes are told apart by the
+# HEADING form: a complete draft carries `### Verdict` (or any `#` heading
+# naming it), while the gha#710 follow-up tail writes a line-start `Verdict:`
+# line (verdict-split-across-blocks.json, block 2), which $vidx matches and
+# this does not. So when more than one block carries a verdict HEADING, the
+# span starts at the LAST such block: the earlier drafts were superseded by
+# their author, and the tail after the last draft is still kept. One heading
+# (the gha#710 shape, and the common case) leaves the span rule exactly as it
+# was.
+#
+# A heading is AUTHORED only outside a fenced code block and outside a
+# blockquote. This corpus quotes the literal heading constantly -- CLAUDE.md,
+# this file, every fixture, and a review OF this file -- so a later block that
+# merely shows the heading shape in a fence, or blockquotes the previous
+# verdict for context, must not read as a fresh draft: with the bare regex it
+# did, and the span then started at the quotation and dropped the entire real
+# review (gha#808 review). That is the failure gha#710 exists to prevent, and
+# it is worse than the concatenation this fixes. A fence closes only on a run
+# of the SAME character at least as long as the opener (CommonMark, and what
+# strip-non-invoking-markup.sh implements); a first draft closed on any
+# fence line, so a backtick fence containing a tilde line "closed" early and
+# its heading counted (gha#808 review round 1). An unclosed fence runs to the
+# end of its block, again per CommonMark: a redraft whose own code sample is
+# never closed hides its own heading from this test, exactly as GitHub would
+# hide it from the reader, and the block then falls back to the gha#710 span
+# rule. That is malformed input rendered faithfully, not a case to special-
+# case, since ignoring unclosed fences would re-admit the quoted-heading drop.
+# The third literal construct, an INDENTED code block, needs no state at all:
+# a heading may be indented by at most three spaces (CommonMark), so a line
+# at four columns or a tab is never a heading, whether it is code or a lazy
+# paragraph continuation. The heading test says {0,3} rather than [ \t]* for
+# exactly that reason (gha#808 review round 3, which reproduced the drop with
+# indentation instead of a fence).
 jq -r '
+  def authored_heading:
+    ( split("\n")
+      | reduce .[] as $l ({fence: "", flen: 0, out: []};
+          # Spaces only in the indentation allowance: a tab is four columns in
+          # CommonMark, so a tab-led fence line is indented code, not a fence
+          # (Copilot on gha#808).
+          ( [ $l | capture("^ {0,3}(?<run>`{3,}|~{3,})(?<rest>.*)$") ] | first ) as $f
+          | if $f != null and .fence == ""
+              then .fence = ($f.run[0:1]) | .flen = ($f.run | length)
+            # A closer carries nothing but whitespace after its run; a run
+            # followed by text is fence content (CommonMark; the info-string
+            # form is legal only on an opener). gha#808 review round 2.
+            elif $f != null and ($f.run[0:1]) == .fence and ($f.run | length) >= .flen
+                 and ($f.rest | test("^[ \\t]*$"))
+              then .fence = "" | .flen = 0
+            elif .fence != "" or ($l | test("^[ \\t]*>")) then .
+            else .out += [$l] end)
+      | .out | join("\n") )
+    # One to six hashes, then at least one space or tab: seven hashes, or
+    # hashes run into the word, are paragraph text in CommonMark (Copilot on
+    # gha#808). The awk invariant in run-fixture-tests.sh mirrors both limits.
+    | test("(?im)^ {0,3}#{1,6}[ \\t]+verdict\\b");
   . as $blocks
   | [ range(0; $blocks | length)
       | select($blocks[.] | test("(?im)^[\\s>*_#-]*verdict\\b")) ] as $vidx
-  | if ($vidx | length) > 1
+  | [ $vidx[] | select($blocks[.] | authored_heading) ] as $hidx
+  | if ($hidx | length) > 1
+      then $blocks[($hidx | last):(($vidx | last) + 1)] | join("\n\n")
+    elif ($vidx | length) > 1
       then $blocks[($vidx | first):(($vidx | last) + 1)] | join("\n\n")
     elif ($vidx | length) == 1
       then $blocks[$vidx | first]
diff --git a/.github/workflows/scripts/tests/fixtures/verdict-redraft-after-tab-fence.json b/.github/workflows/scripts/tests/fixtures/verdict-redraft-after-tab-fence.json
new file mode 100644
index 00000000..1bb4a095
--- /dev/null
+++ b/.github/workflows/scripts/tests/fixtures/verdict-redraft-after-tab-fence.json
@@ -0,0 +1,38 @@
+[
+  {
+    "type": "system",
+    "subtype": "init"
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "## Review\n\nThe iota-pass first draft of the analysis.\n\n### Verdict\n\n**Ready for merge.**\n"
+        }
+      ]
+    }
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "## Review\n\nThe iota-pass second draft, whose tab-led backtick line is indented code and opens no fence:\n\n\t```\n\n### Verdict\n\n**Ready for merge.**\n"
+        }
+      ]
+    }
+  },
+  {
+    "type": "result",
+    "subtype": "success",
+    "is_error": false,
+    "num_turns": 9,
+    "duration_ms": 120000,
+    "total_cost_usd": 1.46,
+    "permission_denials_count": 0,
+    "permission_denials": []
+  }
+]
diff --git a/.github/workflows/scripts/tests/fixtures/verdict-redrafted-thrice.json b/.github/workflows/scripts/tests/fixtures/verdict-redrafted-thrice.json
new file mode 100644
index 00000000..e26a62fa
--- /dev/null
+++ b/.github/workflows/scripts/tests/fixtures/verdict-redrafted-thrice.json
@@ -0,0 +1,71 @@
+[
+  {
+    "type": "system",
+    "subtype": "init"
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "## Review\n\nThe change is sound. First draft names the alpha-pass fixture table.\n\n### Verdict\n\n**Ready for merge.**\n\n<!-- review-data:\n{\"schema\": \"1.1\", \"verdict\": \"ready\", \"draft\": 1}\n-->\n"
+        }
+      ]
+    }
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "Confirmed via the authoritative instrument: exit 1, five in-progress bullets, no failures. My content verdict stands unchanged."
+        }
+      ]
+    }
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "## Review\n\nThe change is sound. Second draft names the beta-pass fixture table.\n\n### Verdict\n\n**Ready for merge.**\n\n<!-- review-data:\n{\"schema\": \"1.1\", \"verdict\": \"ready\", \"draft\": 2}\n-->\n"
+        }
+      ]
+    }
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "## Review\n\nThe change is sound. Third draft names the gamma-pass fixture table.\n\n### Verdict\n\n**Ready for merge.**\n\n<!-- review-data:\n{\"schema\": \"1.1\", \"verdict\": \"ready\", \"draft\": 3}\n-->\n"
+        }
+      ]
+    }
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "Instrument re-run after the third draft: still exit 0, the delta-pass tail is retained.\n\nVerdict: Ready for merge."
+        }
+      ]
+    }
+  },
+  {
+    "type": "result",
+    "subtype": "success",
+    "is_error": false,
+    "num_turns": 9,
+    "duration_ms": 120000,
+    "total_cost_usd": 2.34,
+    "permission_denials_count": 0,
+    "permission_denials": []
+  }
+]
diff --git a/.github/workflows/scripts/tests/fixtures/verdict-then-indented-heading.json b/.github/workflows/scripts/tests/fixtures/verdict-then-indented-heading.json
new file mode 100644
index 00000000..585a0752
--- /dev/null
+++ b/.github/workflows/scripts/tests/fixtures/verdict-then-indented-heading.json
@@ -0,0 +1,38 @@
+[
+  {
+    "type": "system",
+    "subtype": "init"
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "## Review\n\nThe theta-pass analysis of the diff: sound, and the fixture discriminates.\n\n### Verdict\n\n**Ready for merge.**\n"
+        }
+      ]
+    }
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "For reference, the heading shape as an indented code sample:\n\n    ### Verdict\n\nand a paragraph continuation that is not code:\nsome prose\n    ### Verdict continued on an indented line\n\nNeither is a new verdict."
+        }
+      ]
+    }
+  },
+  {
+    "type": "result",
+    "subtype": "success",
+    "is_error": false,
+    "num_turns": 9,
+    "duration_ms": 120000,
+    "total_cost_usd": 1.45,
+    "permission_denials_count": 0,
+    "permission_denials": []
+  }
+]
diff --git a/.github/workflows/scripts/tests/fixtures/verdict-then-mismatched-fence.json b/.github/workflows/scripts/tests/fixtures/verdict-then-mismatched-fence.json
new file mode 100644
index 00000000..2e3199e1
--- /dev/null
+++ b/.github/workflows/scripts/tests/fixtures/verdict-then-mismatched-fence.json
@@ -0,0 +1,38 @@
+[
+  {
+    "type": "system",
+    "subtype": "init"
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "## Review\n\nThe zeta-pass analysis of the diff: sound, and the fixture discriminates.\n\n### Verdict\n\n**Ready for merge.**\n"
+        }
+      ]
+    }
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "The stripper's own test table shows the shape:\n\n```\n~~~\n### Verdict\n~~~\n```\n\nThat is documentation, not a new verdict."
+        }
+      ]
+    }
+  },
+  {
+    "type": "result",
+    "subtype": "success",
+    "is_error": false,
+    "num_turns": 9,
+    "duration_ms": 120000,
+    "total_cost_usd": 1.43,
+    "permission_denials_count": 0,
+    "permission_denials": []
+  }
+]
diff --git a/.github/workflows/scripts/tests/fixtures/verdict-then-quoted-heading.json b/.github/workflows/scripts/tests/fixtures/verdict-then-quoted-heading.json
new file mode 100644
index 00000000..1b251e42
--- /dev/null
+++ b/.github/workflows/scripts/tests/fixtures/verdict-then-quoted-heading.json
@@ -0,0 +1,38 @@
+[
+  {
+    "type": "system",
+    "subtype": "init"
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "## Review\n\nThe epsilon-pass analysis of the diff: the guard change is sound and the fixture discriminates.\n\n### Verdict\n\n**Ready for merge.**\n"
+        }
+      ]
+    }
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "For the record, the heading shape the extractor keys on looks like this:\n\n```\n### Verdict\n```\n\nand the previous round said:\n\n> ### Verdict\n> Needs more work.\n\n### Verdicts of prior rounds\n\n####### Verdict\n\n###Verdict\n\nNeither of those is a new verdict; the one above stands."
+        }
+      ]
+    }
+  },
+  {
+    "type": "result",
+    "subtype": "success",
+    "is_error": false,
+    "num_turns": 9,
+    "duration_ms": 120000,
+    "total_cost_usd": 1.42,
+    "permission_denials_count": 0,
+    "permission_denials": []
+  }
+]
diff --git a/.github/workflows/scripts/tests/fixtures/verdict-then-tab-inside-fence.json b/.github/workflows/scripts/tests/fixtures/verdict-then-tab-inside-fence.json
new file mode 100644
index 00000000..3f3df483
--- /dev/null
+++ b/.github/workflows/scripts/tests/fixtures/verdict-then-tab-inside-fence.json
@@ -0,0 +1,38 @@
+[
+  {
+    "type": "system",
+    "subtype": "init"
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "## Review\n\nThe kappa-pass analysis of the diff: sound, and the fixture discriminates.\n\n### Verdict\n\n**Ready for merge.**\n"
+        }
+      ]
+    }
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "The shape, with a tab-led backtick line inside a real fence (fence content, not a closer):\n\n```\n\t```\n### Verdict\n```\n\nThat is documentation, not a new verdict."
+        }
+      ]
+    }
+  },
+  {
+    "type": "result",
+    "subtype": "success",
+    "is_error": false,
+    "num_turns": 9,
+    "duration_ms": 120000,
+    "total_cost_usd": 1.47,
+    "permission_denials_count": 0,
+    "permission_denials": []
+  }
+]
diff --git a/.github/workflows/scripts/tests/fixtures/verdict-then-trailing-text-closer.json b/.github/workflows/scripts/tests/fixtures/verdict-then-trailing-text-closer.json
new file mode 100644
index 00000000..0b2c93cd
--- /dev/null
+++ b/.github/workflows/scripts/tests/fixtures/verdict-then-trailing-text-closer.json
@@ -0,0 +1,38 @@
+[
+  {
+    "type": "system",
+    "subtype": "init"
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "## Review\n\nThe eta-pass analysis of the diff: sound, and the fixture discriminates.\n\n### Verdict\n\n**Ready for merge.**\n"
+        }
+      ]
+    }
+  },
+  {
+    "type": "assistant",
+    "message": {
+      "content": [
+        {
+          "type": "text",
+          "text": "The shape, with a line that looks like a closer but is not:\n\n```\nfoo\n```not-a-closer\n### Verdict\n```\n\nThat is documentation, not a new verdict."
+        }
+      ]
+    }
+  },
+  {
+    "type": "result",
+    "subtype": "success",
+    "is_error": false,
+    "num_turns": 9,
+    "duration_ms": 120000,
+    "total_cost_usd": 1.44,
+    "permission_denials_count": 0,
+    "permission_denials": []
+  }
+]
diff --git a/.github/workflows/scripts/tests/run-fixture-tests.sh b/.github/workflows/scripts/tests/run-fixture-tests.sh
index db3aa67b..a680c0cd 100755
--- a/.github/workflows/scripts/tests/run-fixture-tests.sh
+++ b/.github/workflows/scripts/tests/run-fixture-tests.sh
@@ -102,6 +102,30 @@ declare -A expected=(
   # must_contain needle is block A's analysis, which the pre-#710 last-block
   # extraction dropped.
   [verdict-split-across-blocks.json]=pass
+  # gha#805: three complete drafts, each with a `### Verdict` heading. Only
+  # the last is posted; the gha#710 span rule alone concatenates all three.
+  [verdict-redrafted-thrice.json]=pass
+  # gha#808 review: a later block that only QUOTES a verdict heading (in a
+  # fence, and in a blockquote) is not a draft; the real review before it
+  # must be posted, not dropped.
+  [verdict-then-quoted-heading.json]=pass
+  # gha#808 review round 1: a backtick fence containing a tilde line does not
+  # close on the tilde; the heading after it is still fenced.
+  [verdict-then-mismatched-fence.json]=pass
+  # gha#808 review round 2: a same-character run followed by text is fence
+  # content, not a closer.
+  [verdict-then-trailing-text-closer.json]=pass
+  # gha#808 review round 3: a heading quoted as INDENTED code is not a draft.
+  [verdict-then-indented-heading.json]=pass
+  # Copilot on gha#808: a tab-led backtick line is indented code, not a
+  # fence, so the real heading after it is authored and the last draft wins.
+  [verdict-redraft-after-tab-fence.json]=pass
+  # Copilot on gha#808: a tab-led backtick line INSIDE a real fence is
+  # content. A tab-admitting extractor reads it as the closer, counts the
+  # fenced heading as authored, and drops the real review (the must_contain
+  # needle); a tab-admitting awk counts two headings in the correct span.
+  # One fixture, both halves.
+  [verdict-then-tab-inside-fence.json]=pass
   [verdict-not-last-block.json]=pass
   [verdict-via-inline-comment-tool.json]=pass
   [verdict-via-gh-comment-heredoc.json]=pass
@@ -142,6 +166,16 @@ declare -A must_contain=(
   # pre-#710 tail-only extraction (which drops it along with block A) and a
   # hypothetical first+last-only join (which keeps A but drops it).
   [verdict-split-across-blocks.json]='middle-pass rerun of the suite'
+  # Two needles in one: the third draft, and the tail AFTER it (a line-start
+  # Verdict: line, no heading), which must be kept -- narrowing the span end
+  # to the last heading block drops the tail.
+  [verdict-redrafted-thrice.json]='gamma-pass fixture table'
+  [verdict-then-quoted-heading.json]='epsilon-pass analysis'
+  [verdict-then-mismatched-fence.json]='zeta-pass analysis'
+  [verdict-then-trailing-text-closer.json]='eta-pass analysis'
+  [verdict-then-indented-heading.json]='theta-pass analysis'
+  [verdict-redraft-after-tab-fence.json]='iota-pass second draft'
+  [verdict-then-tab-inside-fence.json]='kappa-pass analysis'
   # gha#391: confirms review_text_file carries the actual posted verdict, not
   # just an empty/fallback string from the is_error early-fail path.
   [is-error-success-with-verdict.json]='Ready for merge'
@@ -172,7 +206,17 @@ declare -A must_contain=(
   # posted verdict rather than an empty fallback from the error path.
   [quota-exhausted-midrun-with-verdict.json]='Ready for merge'
 )
+# gha#808 review: a second must-contain needle where one fixture pins two
+# claims. Checked exactly like must_contain.
+declare -A must_also_contain=(
+  [verdict-redrafted-thrice.json]='delta-pass tail is retained'
+)
+
 declare -A must_not_contain=(
+  # gha#805: the superseded first draft must not be posted. Its needle is
+  # what the pre-#805 span rule (first verdict block through last) keeps.
+  [verdict-redrafted-thrice.json]='alpha-pass fixture table'
+  [verdict-redraft-after-tab-fence.json]='iota-pass first draft'
   [verdict-not-last-block.json]="I've posted my findings"
   [verdict-via-inline-comment-tool.json]="Posted the inline finding and a summary comment ending in"
   [verdict-via-gh-comment-heredoc.json]='gh pr comment'
@@ -263,6 +307,13 @@ declare -A expected_cost=(
   [genuine-finished-review.json]=0.42
   [spawn-denials-only-retryable.json]=4.21
   [verdict-split-across-blocks.json]=1.11
+  [verdict-redrafted-thrice.json]=2.34
+  [verdict-then-quoted-heading.json]=1.42
+  [verdict-then-mismatched-fence.json]=1.43
+  [verdict-then-trailing-text-closer.json]=1.44
+  [verdict-then-indented-heading.json]=1.45
+  [verdict-redraft-after-tab-fence.json]=1.46
+  [verdict-then-tab-inside-fence.json]=1.47
   [spawn-denials-plus-starved-calls.json]=3.9
   [stub-background-agents-executed.json]=4.19
   [stub-background-agents-omitted-param.json]=4.18
@@ -421,9 +472,52 @@ assert_pass() {
   if [[ -n "${must_contain[$fixture]:-}" ]] && ! grep -qF "${must_contain[$fixture]}" "$posted_file"; then
     return 1
   fi
+  if [[ -n "${must_also_contain[$fixture]:-}" ]] && ! grep -qF "${must_also_contain[$fixture]}" "$posted_file"; then
+    return 1
+  fi
   if [[ -n "${must_not_contain[$fixture]:-}" ]] && grep -qF "${must_not_contain[$fixture]}" "$posted_file"; then
     return 1
   fi
+  # gha#805, as an invariant over every posted review rather than one
+  # fixture: a comment carries at most ONE authored verdict heading. A second
+  # means two complete drafts were concatenated, whichever fixture produced
+  # them. The gha#710 tail writes `Verdict:` without a heading, so it does
+  # not count; nor does a heading inside a fenced block or a blockquote,
+  # excluded here the same way the extractor excludes them (gha#808 review:
+  # the first draft of this counted them, so a correct single-draft review
+  # that quoted one example heading would have failed). No interval
+  # expression in the awk, per this repo's mawk rule. This is a shape check
+  # on our own extraction, not a verdict parse.
+  local headings
+  headings="$(awk '
+    # A fence closes only on the same character, at least as long as the
+    # opener (CommonMark), mirroring the jq. No interval expressions.
+    # Spaces only: a tab is four columns, so a tab-led fence line is
+    # indented code (Copilot on gha#808). No interval expressions.
+    match($0, /^ ? ? ?(```+|~~~+)/) {
+      run = substr($0, RSTART, RLENGTH); sub(/^ +/, "", run)
+      ch = substr(run, 1, 1); len = length(run)
+      rest = substr($0, RSTART + RLENGTH)
+      if (fence == "") { fence = ch; flen = len; next }
+      if (ch == fence && len >= flen && rest ~ /^[ \t]*$/) { fence = ""; flen = 0; next }
+    }
+    fence != "" { next }
+    /^[ \t]*>/ { next }
+    # At most three leading spaces: a four-column or tab indent is indented
+    # code or a lazy continuation, never a heading (gha#808 review round 3).
+    # The trailing class is the word boundary the jq spells verdict\b, so
+    # a "Verdicts" heading counts in neither; at least one space after the
+    # hashes, and at most six of them, measured by run length rather than an
+    # interval expression (Copilot on gha#808).
+    tolower($0) ~ /^ ? ? ?#+[ \t]+verdict([^a-z0-9_]|$)/ {
+      hashes = $0; sub(/^ */, "", hashes); sub(/[^#].*$/, "", hashes)
+      if (length(hashes) <= 6) n++
+    }
+    END { print n + 0 }' "$posted_file")"
+  if [[ "$headings" -gt 1 ]]; then
+    echo "::error::$fixture: posted review carries $headings verdict headings (gha#805)"
+    return 1
+  fi
   return 0
 }
 
diff --git a/CLAUDE.md b/CLAUDE.md
index 0e0700b7..51fbc7a3 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -2001,6 +2001,78 @@ pins that ordering, and it passes with or without the fix by design -- the
 fixture that actually fails when the fix is removed is
 `quota-exhausted-midrun.json`, confirmed by mutation rather than assumed.
 
+**A reviewer that redrafts its final message defeats the gha#710 span rule,
+and the heading form is what tells the two shapes apart.**
+gha#710 widened the posted text to the whole span from the first
+verdict-bearing block to the last, because a review split across blocks was
+being posted as its tail alone.
+gha#805 is that rule's own failure: a reviewer re-ran an instrument between
+drafts and produced three complete reviews in a row, each with its own
+`### Verdict` heading and structured-review-data block, and the span rule
+concatenated all three (Morrison-Lab/ai-config#2966, run 33594599768).
+A complete draft carries a verdict *heading*; the gha#710 follow-up tail
+writes a line-start `Verdict:` line, never a heading (a mid-sentence "my
+verdict stands unchanged" matches neither regex, since both anchor at line
+start).
+So when more than one block carries an authored heading, the span starts at
+the last such block and still runs to the last verdict-bearing block, which
+drops the superseded drafts and keeps the tail; one heading leaves gha#710's
+behaviour untouched.
+**Authored means outside a fence, outside a blockquote, and outside an
+indented code block.**
+This corpus quotes the literal heading constantly, and a review of this very
+script does too, so a later block that only shows the heading shape in a
+code fence or blockquotes the previous verdict must not read as a new
+draft; with the bare regex it did, and the span then started at the
+quotation and dropped the entire real review, which is the failure gha#710
+exists to prevent (gha#808 review).
+Both the jq detector and the bash invariant skip fenced and blockquoted
+lines.
+A fence closes only on the same character, at least as long as the opener,
+with nothing but whitespace after it (CommonMark, as
+`strip-non-invoking-markup.sh` does);
+a first draft closed on any fence line, so a backtick fence holding a tilde
+line leaked its heading, and a second closed on a run followed by text.
+Both accept at most three leading spaces before the `#` or before a fence
+run (spaces only: a tab is four columns, so a tab-led backtick line is
+indented code rather than a fence), so a line at four columns or a tab is
+never a heading, whether indented code or a lazy paragraph continuation;
+round 3 of the same review reproduced the drop with indentation, and no
+state machine is needed for it.
+The awk that counts headings carries no interval expression, per this file's
+mawk rule, spells the jq's word boundary as a trailing class, so a plural
+"Verdicts" heading counts in neither, and mirrors the jq's one-to-six hash
+limit by run length and its space-after-hashes requirement, so
+`####### Verdict` and `###Verdict` count in neither.
+An unclosed fence runs to the end of its block, so a redraft whose own code
+sample is never closed hides its own heading and falls back to the gha#710
+span rule; that is malformed input rendered as GitHub renders it, and
+ignoring unclosed fences would re-admit the quoted-heading drop.
+`verdict-redrafted-thrice.json` pins the rule with a must-contain on the
+third draft, a second must-contain on the tail after it, and a
+must-not-contain on the first draft;
+`verdict-then-quoted-heading.json` pins that a quoted heading is not a
+draft, `verdict-then-mismatched-fence.json` that a tilde line does not
+close a backtick fence, and `verdict-then-trailing-text-closer.json` that a
+run followed by text does not close one either, and
+`verdict-then-indented-heading.json` that an indented heading is not a
+draft, `verdict-redraft-after-tab-fence.json` that a tab-led backtick line
+opens no fence, and `verdict-then-tab-inside-fence.json` that one inside a
+real fence closes none (there a tab-admitting extractor un-fences the quoted
+heading and drops the real review, the same silent-drop class as the fence
+and blockquote cases, and a tab-admitting awk counts two headings in the
+correct span, so the one fixture pins both halves);
+and `assert_pass` holds every posted review to at most one authored heading,
+so a future concatenation fails on whichever fixture produces it.
+Nine mutations turn a named case red: disabling the multi-heading branch,
+narrowing the span end to the last heading block, dropping the
+fence-and-blockquote exclusion, closing a fence on any delimiter, closing
+on a run followed by text, widening the heading indent back to any
+whitespace, admitting a tab into the fence indentation, lifting the
+six-hash limit, and dropping the awk's trailing word-boundary class.
+That count is a shape check on our own extraction, not a verdict parse: it
+never reads which verdict was stated.
+
 **`permission_denials_count` can be absent from the real execution file even
 though `claude-code-action` prints it to the job log, because the log line is
 a display value the action computes, not a field it always writes to disk.**
diff --git a/changelog.d/post-last-complete-review-draft.fixed.md b/changelog.d/post-last-complete-review-draft.fixed.md
new file mode 100644
index 00000000..765dbfe2
--- /dev/null
+++ b/changelog.d/post-last-complete-review-draft.fixed.md
@@ -0,0 +1,9 @@
+- **`claude-code-review` posts only the reviewer's last complete draft**
+  (#805).
+  A reviewer that redrafted its final message produced three complete
+  reviews, each with its own `### Verdict` heading, and the gha#710 span
+  rule concatenated all three into one comment.
+  When more than one block carries an authored verdict heading (outside a
+  code fence or blockquote, and indented at most three spaces), the posted
+  text now starts at the last such block;
+  a single heading, including gha#710's split-review shape, is unchanged.