feat(lib): add experimental _EXP_STRING backed by std::string

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

1 files changed, 1 insertions, 1 deletions

diff --git a/tests/README.md b/tests/README.md
index 3b9cfc7..ab05c48 100644
--- a/tests/README.md
+++ b/tests/README.md
@@ -11,7 +11,7 @@ perl tests/summarize.pl build/test-results
 
 Use `ctest --test-dir build -L '^ii$' --output-on-failure` for one suite, or `-R '^i/arrays$'` for one case. CTest returns failure if any case fails; it continues through the other cases. There are no expected-failure exemptions. Rejection of valid CIE syntax must not be treated as a successful negative test.
 
-The suites are **i**: independently traced fixed outputs, **ii**: C++ differential tests with Perl inputs, and **iii**: rejected programs with required causes and source lines. Fixed expectations are **pending human review**; each case's notes explain the trace and distinguish CIE rules from project extensions. REAL calculations are observed through BOOLEAN/INTEGER output; REAL text and REAL-to-string formatting remain outside this suite.
+The suites are **i**: independently traced fixed outputs, **ii**: C++ differential tests with Perl inputs, and **iii**: rejected programs with required causes and source lines. Fixed expectations are **pending human review**; each case's notes explain the trace and distinguish CIE rules from project extensions. REAL calculations are observed through BOOLEAN/INTEGER output; The experimental string-library fixture additionally checks its specified C++17 numeric text formatting.
 
 The default is 10,000 random records per differential group, plus fixed boundary records. Both generated native programs and the configured `-r` path are checked; NONE builds only run native programs. Each source is compiled once per execution path and processes its records in one batch. Tests normalize only CRLF to LF; spaces, blank lines, case and final newlines remain significant. Generated text files use the same comparison rule.