git_commit

4276bd061d95d47c61deb141a3b4bb65ccebe4e0

branch

047-foundations-programme-closeout

Toolchain

dotnet 10.0.300 · FAKE library Fake.Core.Target 6.1.4 (compiled, no FSX runner)

1

build.fsx size

4,688 lines

0 (deleted in full, no shim)

git ls-files build.fsx \| xargs -r wc -l → (empty / 0 files)

met — target was "deleted or ≤ 200-line shim" (feature 045).

3

Evidence-path languages

F# + Bash + Python

F# only

git ls-files '.specify/**/*.py' \| wc -l → 0; git ls-files '*.py' \| wc -l → 0; git ls-files '**/run-audit.sh' \| wc -l → 0

met — the evidence path is in-process compiled F# (feature 043).

5

Governance Markdown (rules)

plan claimed ~23,000 lines / 21:1; corrected to ≈ 6,882 (feature 046)

6,876 lines (.agents/skills 4,059 + .specify/**/*.md 2,817); rules enforced by code

find .agents/skills -name '*.md' \| xargs wc -l \| tail -1 → 4059; find .specify -name '*.md' \| xargs wc -l \| tail -1 → 2817 (sum 6,876)

rationale — the plan's ~23,000-line / 21:1 figure was an over-estimate; feature 046 established the corrected rule/guidance baseline at ≈ 6,882, after-trim 6,876 (−6). The literal "low hundreds" target is not reached and is not the real story: the remaining prose doubles as pinned author guidance (its tokens are required by the 041–044 generation-currency term-checks), and the rules themselves are now machine-enforced — Routing.fs, Guidance.fs, GeneratedProductContract.fs, and the in-process EvidenceGraph/EvidenceAudit — so no rule depends on "read the prose and comply." Feature 055 (FR-008): the "low hundreds" figure is now retired as the live target — feature 055 decoupled author-guidance prose from the generation-currency anchors (the literal-substring table no longer freezes prose), so the corpus can shrink; tracking is against the corrected ≈6,882 baseline with the actual large-scale reduction a bounded follow-up.

7

Framework-author process

full consumer ceremony (~12–14 h/feature)

*inner-loop* (Dev + surface check) is the Route default for framework-internal changes; full pipeline reserved for consumers + dogfood

mechanism: Routing.fs innerLoopGates = [ Targets.Dev ] (line 225) + internalInnerLoopApplicable (line 97), proven by tests/Governance.Tests/RoutingTests.fs

met (mechanism) — the hour figure is an author estimate (no timing harness; same Stage-0 exemption) and lives in Section B; the mechanism (light default tier) is realized and tested. A live Route on this feature's diff escalates (it touches governance docs) — that escalation is itself row 8's evidence.

9

Framework-owned config

YAML, stringly-typed, runtime-parsed

compiled F# values/predicates in the library; build-time checked; no FCS

grep -n 'dogfoodFeatureIds' build/Governance/Routing.fs → 235:; git ls-files '.specify/**/*.py' \| wc -l → 0; ADR 0005 (D6)

met — routing/tier/feature policy is typed F#; no YAML governance config parsed at runtime, no FSharp.Compiler.*.

11

Runtime architecture (Scene → SkiaViewer → Elmish)

sound

unchanged

git diff --stat -- 'src/**' → (empty)

met — product runtime / .fsi untouched (FR-010, SC-006); see readiness/runtime-untouched.md.

Per-feature ceremony time

~12–14 h/feature (Stage-0 author estimate)

inner-loop framework changes: hours, not a full-pipeline day (author estimate)

estimate — no timing harness exists (the same Stage-0 exemption). The realized mechanism is the inner-loop light tier now being the Route default for framework-internal changes (Section A row 7); the hour delta is not instrumented and is not asserted as a reproducible number.

Warm-build time

FSX-runner compile tax per invocation (dotnet fake recompiled build.fsx)

lower — compiled build/Build.fsproj exe; no per-run FSX recompile, no FCS (D2 spike confirmed)

estimate / measured-where-possible — wall-clock build time is machine-dependent and not pinned to a SHA; the structural change (no per-invocation FSX compile, no FSharp.Compiler.* in the transitive graph) is the recorded basis.