diff --git a/scripts/report-test-coverage.sh b/scripts/report-test-coverage.sh
new file mode 100755
index 0000000..90d5150
--- /dev/null
+++ b/scripts/report-test-coverage.sh
@@ -0,0 +1,141 @@
+#!/usr/bin/env sh
+# Per-workspace-crate line and branch coverage report.
+#
+# This is deliberately a report over every host-testable workspace package,
+# not a misleading aggregate. STM32F7 platform/board images are target-only:
+# their `panic-halt` handler cannot link into the host test harness, so they
+# are reported as N/A and must be covered by target checks plus bench
+# evidence instead.
+#
+# Prerequisites (kept out of the firmware dependency graph):
+#   cargo install cargo-llvm-cov --locked
+#   rustup toolchain install nightly --component llvm-tools-preview
+#   python3 (standard library only, for strict source-level accounting)
+set -eu
+
+if ! cargo llvm-cov --version >/dev/null 2>&1; then
+    echo "cargo-llvm-cov is required; install it with: cargo install cargo-llvm-cov --locked" >&2
+    exit 127
+fi
+
+if ! rustup toolchain list | grep -q '^nightly'; then
+    echo "nightly with llvm-tools-preview is required for branch coverage" >&2
+    exit 127
+fi
+
+if ! python3 --version >/dev/null 2>&1; then
+    echo "python3 is required for strict source-level coverage accounting" >&2
+    exit 127
+fi
+
+python3 scripts/test-summarize-test-coverage.py
+
+# Keep instrumentation products outside a possibly shared target directory.
+: "${CARGO_TARGET_DIR:=/tmp/kaptan-test-coverage}"
+export CARGO_TARGET_DIR
+mkdir -p "$CARGO_TARGET_DIR"
+
+# `cargo llvm-cov` merges every instrumented object below `CARGO_TARGET_DIR`.
+# A shared target directory therefore makes a later strict-core report include
+# an unexecuted copy of a shared dependency built while testing an earlier
+# package. Keep reports together but isolate each strict-core invocation. The
+# non-core packages retain the shared directory: several intentionally
+# target-only crates have no host-instrumented object to export by themselves.
+coverage_root=$CARGO_TARGET_DIR
+
+core_incomplete=0
+
+report_package() {
+    package=$1
+    source_path=$2
+    shift 2
+    case "$package" in
+        kaptan-math|kaptan-navigation|kaptan-control|kaptan-safety)
+            package_target_dir="$coverage_root/$package"
+            lcov_report="$coverage_root/$package.lcov"
+            json_report="$coverage_root/$package.json"
+            report_format=strict-core
+            ;;
+        *)
+            package_target_dir=$coverage_root
+            lcov_report="$coverage_root/$package.lcov"
+            report_format=lcov
+            ;;
+    esac
+
+    case "$report_format" in
+        strict-core)
+            # Keep the strict core isolated, then export both representations
+            # from the same test run. LCOV `DA` defines source lines; JSON
+            # coordinates define branch identities across generic instances.
+            RUSTC_BOOTSTRAP=1 CARGO_TARGET_DIR="$package_target_dir" cargo llvm-cov \
+                --branch \
+                --no-report \
+                -p "$package" \
+                "$@"
+            RUSTC_BOOTSTRAP=1 CARGO_TARGET_DIR="$package_target_dir" cargo llvm-cov report \
+                --lcov \
+                --output-path "$lcov_report"
+            RUSTC_BOOTSTRAP=1 CARGO_TARGET_DIR="$package_target_dir" cargo llvm-cov report \
+                --json \
+                --output-path "$json_report"
+            totals=$(python3 scripts/summarize-test-coverage.py \
+                "$lcov_report" "$json_report" "$source_path")
+            ;;
+        lcov)
+            RUSTC_BOOTSTRAP=1 CARGO_TARGET_DIR="$package_target_dir" cargo llvm-cov \
+                --branch \
+                --lcov \
+                --output-path "$lcov_report" \
+                -p "$package" \
+                "$@"
+            totals=$(awk -v source_path="$source_path" '
+                /^SF:/ { selected = index($0, source_path) != 0 }
+                selected && /^LF:/ { lines_found += substr($0, 4) }
+                selected && /^LH:/ { lines_hit += substr($0, 4) }
+                selected && /^BRF:/ { branches_found += substr($0, 5) }
+                selected && /^BRH:/ { branches_hit += substr($0, 5) }
+                END {
+                    printf "%d %d %d %d", lines_hit, lines_found, branches_hit, branches_found
+                }
+            ' "$lcov_report")
+            ;;
+    esac
+    set -- $totals
+    printf '%-26s lines %s/%s, branches %s/%s\n' "$package" "$1" "$2" "$3" "$4"
+
+    case "$package" in
+        kaptan-math|kaptan-navigation|kaptan-control|kaptan-safety)
+            if [ "$1" -ne "$2" ] || [ "$3" -ne "$4" ]; then
+                core_incomplete=1
+            fi
+            ;;
+    esac
+}
+
+echo "Host test coverage (reports: $CARGO_TARGET_DIR; strict core unions LCOV DA lines and JSON branches)"
+report_package kaptan-actuator /libs/kaptan-actuator/
+report_package kaptan-board /libs/kaptan-board/
+report_package kaptan-control /libs/kaptan-control/
+report_package kaptan-data-channel /libs/kaptan-data-channel/
+report_package kaptan-dma /libs/kaptan-dma/
+report_package kaptan-drivers /libs/kaptan-drivers/
+report_package kaptan-i2c /libs/kaptan-i2c/
+report_package kaptan-math /libs/kaptan-math/
+report_package kaptan-mavlink /libs/kaptan-mavlink/ --features dialect-kaptan
+report_package kaptan-navigation /libs/kaptan-navigation/ --features sitl
+report_package kaptan-safety /libs/kaptan-safety/
+report_package kaptan-sensor /libs/kaptan-sensor/
+report_package kaptan-spi /libs/kaptan-spi/
+report_package kaptan-time /libs/kaptan-time/
+report_package kaptan-tx /libs/kaptan-tx/
+report_package kaptan-sitl /tools/sitl/
+
+echo "platform-stm32f7          target-only/N/A (run cargo check -p platform-stm32f7 --lib)"
+echo "kaptan-geprc-f722-v2      target-only/N/A (run its thumbv7em-none-eabihf check)"
+echo "kaptan-kakute-f7-mini-v3  target-only/N/A (run its thumbv7em-none-eabihf check)"
+
+if [ "$core_incomplete" -ne 0 ]; then
+    echo "Strict core coverage is incomplete; reports above identify each crate." >&2
+    exit 1
+fi
diff --git a/scripts/summarize-test-coverage.py b/scripts/summarize-test-coverage.py
new file mode 100644
index 0000000..929edb8
--- /dev/null
+++ b/scripts/summarize-test-coverage.py
@@ -0,0 +1,197 @@
+#!/usr/bin/env python3
+"""Summarize strict-core LCOV lines and LLVM JSON branches by source location.
+
+`llvm-cov` emits one coverage record for every generic monomorphization.  Its
+per-file totals add those records, which can report an uncovered duplicate even
+when the corresponding source line or branch has been exercised by another
+monomorphization.  This helper instead applies the gate to source locations:
+
+* a source line is identified by its LCOV `DA` file and line number;
+* a branch outcome is identified by its full source span; and
+* a location is covered when any matching monomorphization executed it.
+
+JSON segments are deliberately not used as a line denominator: they mark LLVM
+instrumentation-region boundaries, not every source line represented by an
+LCOV `DA` record.  Replacing `DA` with segments would silently lower the strict
+line denominator.
+
+Macro-expansion records are deliberately rejected.  Their source span denotes
+the invocation, not necessarily the expanded predicate, so folding them into
+the source-level key would make the gate unsound.  Extend this helper with an
+explicit expansion identity before adding macro-expanded strict-core coverage.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+from collections.abc import Iterable
+from pathlib import Path
+import sys
+
+
+class CoverageFormatError(ValueError):
+    """Coverage records cannot be safely summarized by source location."""
+
+
+def selected_files(report: dict[object, object], source_path: str) -> Iterable[dict[object, object]]:
+    """Yield selected file records after rejecting ambiguous macro expansion."""
+    exports = report.get("data")
+    if not isinstance(exports, list):
+        raise CoverageFormatError("LLVM JSON has no data array")
+
+    for export in exports:
+        if not isinstance(export, dict):
+            raise CoverageFormatError("LLVM JSON data array contains a non-object")
+        files = export.get("files")
+        if not isinstance(files, list):
+            raise CoverageFormatError("LLVM JSON export has no files array")
+        for file_record in files:
+            if not isinstance(file_record, dict):
+                raise CoverageFormatError("LLVM JSON files array contains a non-object")
+            filename = file_record.get("filename")
+            if not isinstance(filename, str) or source_path not in filename:
+                continue
+            expansions = file_record.get("expansions", [])
+            if not isinstance(expansions, list):
+                raise CoverageFormatError(f"{filename}: expansions is not an array")
+            if expansions:
+                raise CoverageFormatError(
+                    f"{filename}: macro-expansion coverage is present; "
+                    "refusing to merge it by invocation source span"
+                )
+            yield file_record
+
+
+def lcov_line_totals(report: str, source_path: str) -> tuple[int, int, set[str]]:
+    """Return the union of selected LCOV `DA` lines and their source files."""
+    lines: dict[tuple[str, int], bool] = {}
+    files: set[str] = set()
+    filename: str | None = None
+
+    for record in report.splitlines():
+        if record.startswith("SF:"):
+            filename = record[3:]
+            if source_path in filename:
+                files.add(filename)
+            continue
+        if record == "end_of_record":
+            filename = None
+            continue
+        if filename is None or source_path not in filename or not record.startswith("DA:"):
+            continue
+        fields = record[3:].split(",", 2)
+        if len(fields) < 2:
+            raise CoverageFormatError(f"{filename}: malformed LCOV DA record")
+        try:
+            line = int(fields[0])
+            count = int(fields[1])
+        except ValueError as error:
+            raise CoverageFormatError(f"{filename}: non-integer LCOV DA record") from error
+        if line <= 0 or count < 0:
+            raise CoverageFormatError(f"{filename}: invalid LCOV DA location or count")
+        key = (filename, line)
+        lines[key] = lines.get(key, False) or count > 0
+
+    if not files:
+        raise CoverageFormatError(f"no LCOV source file contains {source_path!r}")
+    if not lines:
+        raise CoverageFormatError(f"no LCOV DA records found under {source_path!r}")
+    return sum(lines.values()), len(lines), files
+
+
+def branch_totals(report: dict[object, object], source_path: str) -> tuple[int, int, set[str]]:
+    """Return the union of selected JSON branch outcomes and source files."""
+    branches: dict[tuple[str, int, int, int, int], list[bool]] = {}
+    files: set[str] = set()
+
+    for file_record in selected_files(report, source_path):
+        filename = file_record["filename"]
+        assert isinstance(filename, str)
+        files.add(filename)
+
+        raw_branches = file_record.get("branches")
+        if not isinstance(raw_branches, list):
+            raise CoverageFormatError(f"{filename}: branches is not an array")
+        for branch in raw_branches:
+            if (
+                not isinstance(branch, list)
+                or len(branch) != 9
+                or not all(type(value) is int for value in branch)
+            ):
+                raise CoverageFormatError(f"{filename}: invalid branch record")
+            (
+                start_line,
+                start_column,
+                end_line,
+                end_column,
+                true_count,
+                false_count,
+                file_id,
+                expanded_file_id,
+                _kind,
+            ) = branch
+            if (
+                start_line <= 0
+                or start_column <= 0
+                or end_line < start_line
+                or end_column <= 0
+                or (end_line == start_line and end_column < start_column)
+                or true_count < 0
+                or false_count < 0
+                or file_id < 0
+                or expanded_file_id < 0
+            ):
+                raise CoverageFormatError(f"{filename}: invalid branch location or count")
+            if file_id != expanded_file_id:
+                raise CoverageFormatError(
+                    f"{filename}:{start_line}:{start_column}: macro-expanded branch is present; "
+                    "refusing to merge it by invocation source span"
+                )
+            key = (filename, start_line, start_column, end_line, end_column)
+            outcomes = branches.setdefault(key, [False, False])
+            outcomes[0] = outcomes[0] or true_count > 0
+            outcomes[1] = outcomes[1] or false_count > 0
+
+    if not files:
+        raise CoverageFormatError(f"no LLVM JSON source file contains {source_path!r}")
+    branch_hit = sum(sum(outcomes) for outcomes in branches.values())
+    return branch_hit, 2 * len(branches), files
+
+
+def coverage_totals(
+    lcov_report: str, json_report: dict[object, object], source_path: str
+) -> tuple[int, int, int, int]:
+    """Return strict source-level line and branch coverage for one core crate."""
+    lines_hit, lines_found, lcov_files = lcov_line_totals(lcov_report, source_path)
+    branch_hit, branches_found, json_files = branch_totals(json_report, source_path)
+    if lcov_files != json_files:
+        raise CoverageFormatError(
+            "LCOV/JSON source-file mismatch: "
+            f"LCOV={sorted(lcov_files)!r}, JSON={sorted(json_files)!r}"
+        )
+    return lines_hit, lines_found, branch_hit, branches_found
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("lcov_report", type=Path, help="LCOV report from cargo llvm-cov")
+    parser.add_argument("json_report", type=Path, help="LLVM JSON report from cargo llvm-cov")
+    parser.add_argument("source_path", help="workspace-relative path fragment to include")
+    arguments = parser.parse_args()
+
+    try:
+        lcov_report = arguments.lcov_report.read_text(encoding="utf-8")
+        with arguments.json_report.open(encoding="utf-8") as report_file:
+            json_report = json.load(report_file)
+        if not isinstance(json_report, dict):
+            raise CoverageFormatError("LLVM JSON root is not an object")
+        print(*coverage_totals(lcov_report, json_report, arguments.source_path))
+    except (CoverageFormatError, OSError, json.JSONDecodeError) as error:
+        print(f"coverage summary error: {error}", file=sys.stderr)
+        return 2
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/scripts/test-summarize-test-coverage.py b/scripts/test-summarize-test-coverage.py
new file mode 100644
index 0000000..8a7206d
--- /dev/null
+++ b/scripts/test-summarize-test-coverage.py
@@ -0,0 +1,77 @@
+#!/usr/bin/env python3
+"""Synthetic regression tests for strict source-level coverage accounting."""
+
+from __future__ import annotations
+
+import importlib.util
+import unittest
+
+
+def load_summary_module():
+    spec = importlib.util.spec_from_file_location(
+        "summarize_test_coverage", "scripts/summarize-test-coverage.py"
+    )
+    assert spec is not None and spec.loader is not None
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    return module
+
+
+SUMMARY = load_summary_module()
+SOURCE = "/libs/kaptan-navigation/"
+FILENAME = "/repo/libs/kaptan-navigation/src/eskf.rs"
+
+
+def json_report(*, expansions=None, expanded_file_id=0):
+    return {
+        "data": [
+            {
+                "files": [
+                    {
+                        "filename": FILENAME,
+                        "expansions": [] if expansions is None else expansions,
+                        # Segments intentionally differ from DA lines: strict
+                        # line coverage must not lower its denominator to them.
+                        "segments": [[10, 1, 3, True, True, False]],
+                        "branches": [
+                            [10, 5, 10, 10, 0, 3, 0, expanded_file_id, 4],
+                            [10, 5, 10, 10, 2, 0, 0, expanded_file_id, 4],
+                            [11, 5, 11, 10, 0, 0, 0, expanded_file_id, 4],
+                        ],
+                    }
+                ]
+            }
+        ]
+    }
+
+
+LCOV = """SF:/repo/libs/kaptan-navigation/src/eskf.rs
+DA:10,0
+DA:11,0
+end_of_record
+SF:/repo/libs/kaptan-navigation/src/eskf.rs
+DA:10,3
+DA:12,1
+end_of_record
+"""
+
+
+class CoverageSummaryTests(unittest.TestCase):
+    def test_unions_da_lines_and_full_span_branch_outcomes(self):
+        self.assertEqual(
+            SUMMARY.coverage_totals(LCOV, json_report(), SOURCE),
+            # DA lines: 10 is covered by the second generic instance; 11 is
+            # uncovered; 12 is covered. Branch outcomes: the two duplicate
+            # source spans cover both outcomes, while line 11 covers neither.
+            (2, 3, 2, 4),
+        )
+
+    def test_rejects_macro_expansion_records(self):
+        with self.assertRaises(SUMMARY.CoverageFormatError):
+            SUMMARY.coverage_totals(LCOV, json_report(expansions=[{}]), SOURCE)
+        with self.assertRaises(SUMMARY.CoverageFormatError):
+            SUMMARY.coverage_totals(LCOV, json_report(expanded_file_id=1), SOURCE)
+
+
+if __name__ == "__main__":
+    unittest.main()