Task: Keep the forward rate agreement's direction

A trade the document states as Short is stored as Short and exported as Short.

• The import stores the direction the document states, not a constant.
• The export writes the stored direction back into the document.
• A test proves a Short forward rate agreement survives both directions, with the document's own value as its expectation.

1. Write the failing test first: an XML portfolio stating <LongShort>Short</LongShort> through forward_fra and back through reverse_fra, expecting Short in the row and in the exported document. The corpus states Long, so the document is the suite's only Short witness and the codec's parse is part of the proof.
2. Read the direction on import with to_string, and write it back with the mapper's own parse_code, which round-trips the generated to_string so the spelling has one source of truth.
3. Check the money path while the mapper is open, since the same function converts a notional and a strike. Add nothing: mapper_fra_holds_the_digits_of_its_strike_and_notional already asserts the lexical form and the serialised <Strike>0.005</Strike>.

Every number this task claims about a tree, with the command that produced it and the commit it was measured at. A number without both is not a measurement, because a clean-up record describes a tree and trunk moves under a branch: a consumer count, a gate's component count and a suite's size all change with no code changing, and the reader of a bare number cannot tell which tree it came from. Where the raw output matters, put it in an evidence/ file beside the task and name that file in the Command column.

• The review graded this finding Critical because the defect is silent: the store agrees with itself and the document disagrees with both. The fix is small because the column, its check and the schema member were already right; only the mapper ignored them.
• The direction parse uses parse_code, the mapper's existing helper, rather than the ternary a neighbouring product uses. parse_code scans the generated to_string across long_short_count entries, so the spellings have one source. The count itself is hand-maintained, as it is for every other enum in this mapper, so a third spelling would fall back silently until someone bumped it. Recorded because the first version of this note claimed the helper removed the second table, and it does not.
• Money: this unit changes no money conversion, and the FRA's seven sites were already exact. mapper_fra_holds_the_digits_of_its_strike_and_notional proves it by asserting the lexical form and the serialised <Strike>0.005</Strike>

Manual QA scenarios (scaffolded via compass add test_scenario) that verify this task. Link new ones here as they're created; the scenario doc itself links back via its "Verifies task" field.

A short forward rate agreement is now stored, exported and read back as Short. The import reads fra.LongShort through the generated to_string, and the export writes the stored code back through parse_code outside the leg block, so a row whose leg children were never written still states its direction. Index and Strike still come from those leg rows, so such a row exports its direction and neither of those; that is the shared-leg design and the leg path belongs to the unit that sets the leg's own direction.

The case is a document, not a hand-set binding: mapper_fra_reads_a_short_document_and_writes_it_back parses an XML portfolio stating <LongShort>Short</LongShort> and asserts both the stored code and the exported <LongShort>Short</LongShort>. The codec's parse of the spelling is therefore part of what it proves, and the suite carries its only Short witness.

Red then green, executed: with the mapper reverted to the merge base the case fails at xml_ir_mapper_roundtrip_tests.cpp:236; with the fix it passes. The same run also proves the second half, because the export assertion at :244 would fail on its own if the reverse wrote the constant back. The CTest artefact holds only the last run, so the red half is reproduced by reverting the mapper rather than by reading a file.

Money: the unit changes no conversion, and the FRA's four import and three export sites were already exact. The family already proves it: mapper_fra_holds_the_digits_of_its_strike_and_notional asserts the lexical form and the serialised document text, so this unit's money check is a negative result, recorded because a fix that touches a money-carrying mapper should say so.

Rows imported before this fix hold Long for a Short trade and cannot be recovered from the row: the defect wrote a valid value, so nothing in the store distinguishes them. Re-import the affected trades from their documents.

The gate ran in both fleet states: build, db, ctest, ore-xml-roundtrip and the fourteen static checks with the fleet down, and step-consumers with it up.