# Explicit handbook inputs and retained report fixtures — 2026-10-04

## Scope and preserved implementation

The standalone handbook generator requires `--data-dir` and `--host`.
It reads the declared template, result/provenance TSVs, progress JSON and
optional smoke evidence. Placement profiles have a separate optional input.
Development retains its original campaign input/primary-host defaults; the
development sweep now passes its publication input/output directories
explicitly. Normal outputs default to the session build root, while
`--artifact-body` writes only stdout.

Missing required data, template, host, profile directory or explicitly named
smoke file refuses before writing. An invalid status preserves a previous
page. The index keeps separate schedule versions, actual observation counts,
failed/saturated labels, evidence timestamps and withdrawn observations.

Nine of thirteen original generator function ASTs remain unchanged. The
other four change source bindings and CLI/output setup, preserving the
rendering/calculation rules. No engine, canonical wire, measurement clock,
Stage 2, root payload or retained benchmark data changes.

The mixed handbook fixture retains its synthetic checks publicly and moves
eleven actual campaign/template checks into private fixtures. All **28
original handbook tests and 214 assertion calls** remain. Of 35 original
test/helper functions, 32 remain AST-identical; the other three only bind
temporary data directories. The original private-template DOM fixture also
moves intact: **27 of 28 function ASTs and all 116 assertion calls** remain;
one capability lookup follows its owning public load directory. The status
fixture retains all eight original assertions and supplies complete owning
sources and explicit command arguments.

## Original failures retained

The first capsule preparation at source
`ab7dd38e73193a20cb786516baedbb67ec158897` assumes the optional smoke console
is committed. It is absent at that pin; preparation stops before any host
job. Its original preparation script, partial archive and failed receipt
remain preserved. The second preparation correctly includes optional smoke
only when tracked, without changing generator sources.

HP's first high-priority shared archive run at that same source executes
61 public tests. Three status-rendering subchecks fail because the copied
fixture omits the now-required explicit primary host. Every other public
case passes; the private suite is not reached. The original source archive,
all stdout/stderr, lease log and failed receipt remain unchanged.

The fixture correction adds that one host argument. It changes no assertion,
generator source or report policy. Original failures are never relabelled
as successful runs.

## Complete archive qualification

Corrected source `5aa4a2ed0aef087d45dd5769cc654f0951c27f17` supplies the
whole **746-file flat harness archive**. An unrelated enclosing Git
repository has a foreign result sentinel. A separate development-layout
materialization uses that same complete harness and **71 explicit private
inputs**, including the original baseline generator and retained evidence.
Neither materialization borrows another checkout's source, identity or data.

Under a high-priority HP shared CPU lease:

- All **61 public tests** pass with **zero skips**, including eight actual
  standalone CLI tests, the original synthetic handbook/status cases and
  all 34 OPEN publication/controller fixtures.
- All **27 private tests** pass with **zero skips**: the eleven original
  retained campaign/template checks and sixteen original inert Node DOM
  checks. No test starts a benchmark, root launcher or network job.
- The actual original baseline generator and corrected CLI emit
  **byte-identical retained handbook bodies**, SHA-256
  `a14245f0a8aaed4cf36ede5fe27de97b9ef8dec96eaedddcf87ce935af0102cb`.
- All 746 harness files in both layouts and all 71 private inputs retain
  their bytes and modes. Both source trees have no generated output; the
  foreign result sentinel is unchanged and owned materializations are removed.

The original baseline is pinned to
`6df8a622322fdb6ad71acf8af7ec7e7ebee0867b`. Exact-body comparison includes
its current/retained schedule index, provenance, profile rules, warnings and
smoke availability; no measurement or historical wording is invented.

## Retention and admission

Evidence lives under `evidence/handbook-inputs-20261004/`. Original sources,
whole source/private-input tarballs and preparation/qualification scripts
remain under `~/archive/libhft/handbook-inputs-20261004/`, outside website
downloads. Failed materializations are removed only after original source
bytes/modes and retained copies are verified.

Readiness selects 743 sources, with zero scrub findings, private/unresolved
references or dead documentation links. Its 68 textual dependency escapes
remain a worklist. This closes a publication consumer and preserves its
private regression evidence; it supplies no quiet timing, physical Stage 5,
report admission, SDK attestation or coordinated standalone release.

## Concurrent integration and cleanup

The normal rebase preserves other sessions' changes. At integrated pin
`90034120cac14fb1e1c762fc278b43a1a0a85f63`, all 746 qualified harness
Git blobs/modes and all 70 current private Git inputs match the qualified
source. The seventy-first input remains the explicitly pinned original
baseline, rather than a different current generator.

A separate high-priority HP shared cleanup verifies all original failed
materialization source/data bytes and modes, both complete file sets and its
parent sentinel, then removes only that owned materialization. The original
failed capsule, source archives and stdout/stderr remain retained.