Foundations Baseline — 2026-05-31

SHA-pinned "before" snapshot for the foundations programme. Every count names the exact command that produced it so a clean checkout at the pinned SHA can reproduce it (FR-001, FR-009, SC-001). This is the measurement floor every later stage proves "no regression" / "parity" against.

Pinned context

git rev-parse HEAD          # 34faf1ed61ec0ec2a8a2a81168517cb5ccf499d1
date -u +%Y-%m-%dT%H:%M:%SZ # 2026-05-31T13:16:44Z

Working-tree note (spec Edge Cases — recorded, not failed): the snapshot was taken while feature 039 itself was in progress. Every corpus below is measured from committed content via git ls-files, so the in-flight, not-yet-committed 039 artifacts (the new build/** projects, ADRs, fixtures, baseline/outcome docs) are excluded — these numbers are the genuine "before" state. The two new build-tooling projects did not exist as committed files at the pinned commit.

Script size — orchestration vs validation (FR-001)

wc -l build.fsx                         # 4688
grep -cE 'StartTarget "' build.fsx      # 45   (orchestration: target dispatch cases)
grep -cE 'let .*[Vv]alidate' build.fsx  # 22   (validation: Validate* functions)

build.fsx is the monolith the programme decomposes: a thin orchestration spine (the 45 target cases routed through the MEL interpret loop) wrapping a large bulk of inline validation logic (the 22 Validate* functions plus helpers). The orchestration/validation split is reported structurally (marker counts); exact per-line attribution is deferred to the Stage 4 port, which moves the validation bulk into the compiled governance library. The Python evidence engine (below) is the first validation slice already extracted out of build.fsx.

Governance Markdown volume (FR-001)

git ls-files '.claude/skills/*.md' | wc -l                                   # 19
git ls-files '.agents/skills/*.md' | wc -l                                   # 19
git ls-files '.claude/skills/*.md' '.agents/skills/*.md' | xargs cat | wc -l # 5854
wc -l .specify/memory/constitution.md                                        # 336
git ls-files '.specify/templates/*.md' '.specify/presets/*/templates/*.md' | xargs cat | wc -l  # 1508
git ls-files 'specs/**/*.md' | wc -l                                         # 773
git ls-files 'specs/**/*.md' | xargs cat | wc -l                            # 58880

The .claude/.agents skill pair is a byte-identical mirror (synchronized peers, 19 ↔ 19); the 5854-line combined count is the governance-skill prose the single-source-generation stage (Stage 6) targets. The 58880-line specs/** volume is dominated by per-feature ceremony artifacts across 40 features.

F# / Bash / Python LOC mix (FR-001)

git ls-files '*.fs' '*.fsi' '*.fsx' | xargs cat | wc -l   # 44398  (191 files)
git ls-files '*.sh'                  | xargs cat | wc -l   #  3744  (17 files)
git ls-files '*.py'                  | xargs cat | wc -l   #  1460  (2 files)

The two Python files are the evidence engine (.specify/extensions/evidence/scripts/python/compute-task-graph.py + audit-status-scan.py) — the exact code Stage 4 ports to compiled F# and proves parity against the golden fixtures below. The F#/Bash/Python balance is the migration dial the programme moves (Python+Bash governance tooling → compiled F#).

Per-feature ceremony-time estimate (FR-001 — explicit estimate, exempt from the measurement-command rule)

There are 40 feature directories under specs/ at the pinned SHA:

find specs -maxdepth 1 -mindepth 1 -type d | wc -l   # 40

Estimate: ~12–14 h of ceremony per feature under the current single-tier process. This figure is carried forward from the foundations implementation plan (docs/reports/2026-05-31-1049-foundations-implementation-plan.md) and is an author estimate, not an instrumented timing (no timing harness exists yet, so per FR-001 it is exempt from the measurement-command rule and labelled an estimate). Derivation inputs: authoring spec + plan + tasks + tasks.deps.yml, running the serialized FAKE gate sequence, capturing readiness evidence, and resolving evidence-audit findings — repeated per feature. The two-tier Route (Stage 1) is expected to cut this for framework-author-loop features; this baseline is the "before" the cut is measured against.

Golden-fixture manifest (FR-002 / SC-001) — Stage 4 parity oracle

Fixtures committed under tests/Governance.Tests/fixtures/evidence-golden/<feature>/ — each with task-graph.json, task-graph.md, and audit-counts.txt. Every fixture's source_commit equals this baseline's git_commit (34faf1ed61ec0ec2a8a2a81168517cb5ccf499d1). Reproducibility is verified byte-for-byte (SHA-1) by re-running the existing engine (FR-003, SC-002); the capture/verify procedure is in the fixtures README.md.

Substitution (recorded per spec Edge Cases / FR-003): the plan named 017-synthetic-error-evidence as the third source, but 017 does not produce a stable evidence output at the pinned SHA — its graph compute fails (exit 3, verdict: error) because its skilled tasks have no committed readiness/skill-loading-evidence.md, so the audit halts before a count block. Rather than commit an unstable fixture, 017 is replaced by 036-archive-readiness-api-docs, the merged feature that passes graph compute deterministically and carries an accepted [SEH] task (accepted-seh=1, T005), preserving the synthetic-propagation coverage 017 was chosen for. See the fixtures README for the coverage-honesty note (no stable source exercises [S*] auto-synthetic / unaccepted counts; that is a documented follow-up).

This fixture set is designated the Stage 4 parity oracle: the ported F# engine must reproduce these files byte-for-byte before the Python is deleted.

Programme meta-process link (FR-008 / SC-007)

The authoritative programme meta-process — the default process tier for foundations features and the named dogfood feature set — is recorded in specs/039-foundations-baseline-spike/plan.md §Programme Meta-Process. That section is the single discoverable record for SC-007; this baseline links to it rather than duplicating it.

Spike outcome link

The D2 build-library spike outcome — D2 confirmed (clean compiled front-end drives a target whose body lives in the governance library; no FSharp.Compiler.Service) — is recorded in 2026-05-31-spike-d2-outcome.md.