# Indicative loopback TSV input discovery

The existing loopback log converter now takes explicit `FLAVOUR=RUNDIR`,
`--capabilities`, `--output` and optional comment-only `--header` inputs.
Capabilities default to the owning harness file; output follows
`LIBHFT_BUILD_ROOT/load-reports/indicative-results.tsv`, or the harness's own
`target/load-reports`. Import no longer parses unrelated arguments or writes
files. Invalid input/log values refuse before replacing prior output.

The original `agg_rows` and `cell` functions and full capability/status loop
remain AST identical. Median central percentiles, maximum tails across client
processes, capability order/markers, failed `smoke-fail` and missing `todo`
statuses are unchanged. A private original-source replay redirects only CAP/OUT
paths to synthetic fixtures and supplies the preserved historical header:
the old and new complete TSV bytes match exactly.

Original scripts/header and source/SHA mapping are retained under
`internal/bench/load/indicative-inputs-originals-20261003/`, pinned to private
main `7299b10314af`. Three live wrapper comments drop the private machine name;
wrapper runtime commands/flags, historical raw data and installed jobs are
unchanged. Live wrappers still need their development SDK/artifact layout.
No standalone wrapper producer qualification is claimed.

Source `48e6a5f30f0e` is preserved on
`s/codex-fullbench-indicative-tsv-source-20261003`. Its actual complete
**745-file** archive passes help, inert import, invalid-input refusal, actual
owning default-output generation and **seven tests plus six subchecks, zero
skips**. Synthetic log tests cover space-containing paths, median/max
aggregation, missing/bypass/unsupported cells, failed clients and refusal
before replacing old output. Local checks also run the private exact replay:
**eight tests plus six subchecks, zero skips**. All archive source bytes remain
identical; the disposable materialization is removed after verification.

Original sealed evidence lives at
`~/archive/libhft/indicative-tsv-20261003-48e6a5f30f0e`, with
[repository evidence](evidence/indicative-tsv-20261003/SHA256SUMS).
Readiness selects **742 source files**, with **75 scrub findings in 14 files**,
87 textual source-root escapes and zero dead documentation links. Aggregation
establishes no runtime placement, physical link, rate/stability or report
admission. Standalone release and full campaign gates remain open.

Safe integration of the microbench owner's primitive JMH parking changes
leaves 740/745 qualified files identical. Only that lane's guide, launcher,
utility class, guide link and unused pom declarations differ; all converter,
capability, test and support bytes are unchanged. Eight checks and six
subchecks pass again without skips. Current integrated selection is 739 source
files, with the same 75 scrub findings/14 files, 85 textual escapes and zero
dead links. The original archive generation and evidence remain unchanged.