# Explicit placeholder result inputs — 2026-10-04

The placeholder updater resolves its capability catalogue from the owning
harness, using the existing source-layout helper. Development retains its
default results-file selection. A standalone archive requires explicitly
named results TSVs, including for a dry run; it cannot discover an enclosing
checkout's campaign data.

The entire original row-update loop is byte- and AST-identical to baseline
`82bff0850831c7eba6afc34497898d5929de7211`. Existing rows, unknown columns,
capability status labels, dry-run behaviour and idempotence retain their
original implementation. Missing supported cells receive `todo`, never a
successful measurement label. No canonical wire, clock, engine, Stage 2,
root payload or retained result changes.

## Actual archive checks

Source pin `d6c7ce0d65d36c29db29b7a6fee23b8f5c55f19c` supplies the whole
**746-file flat manifest archive**, materialized inside an unrelated Git
repository with a foreign results sentinel. Eight actual updater CLI
operations pass their expected outcomes:

- Missing explicit files refuse, with and without `--dry-run`.
- Explicit dry-run leaves a results file unchanged.
- The write fills the complete capability surface while preserving the
  original row's exact bytes and its future column.
- A repeated write is idempotent.
- Non-matrix input is skipped without changing its bytes.
- A multi-file dry-run preserves both inputs.
- Missing owning layout support refuses without borrowing parent support.

These are lightweight local text-publication checks, without a compiler,
SDK build, benchmark or host timing claim.

## Original failure and separate completion

After all eight operations, the outer checker tries to rename the deliberately
hidden helper in the wrong direction. Its `FileNotFoundError`, failed receipt
and original checker remain retained. This failure is not relabelled as a
successful original run.

A separate read-only completion restores the helper only after checking its
original hash. All 746 source bytes and modes match the pinned archive; there
are no generated source outputs and the foreign results sentinel is unchanged.
The development default dry-run succeeds and preserves all ten committed
results TSVs. Passing archive operations are not repeated. The owned
materialization is then removed.

Evidence is under `evidence/placeholder-inputs-20261004/`; the whole source
tarball and original sources remain in
`~/archive/libhft/placeholder-inputs-20261004/`, outside website downloads.
Readiness still selects 743 sources, with zero scrub, private/unresolved
reference or dead-link findings. Its 68 textual dependency escapes remain a
worklist.

This pass adds no physical Stage 5, timing, report or standalone release
admission. Outstanding SDK consumers, licence review, root installations,
comparison-fixture decisions and coordinated publication remain open.