Plan · CHG-2026-014913 · Agentic SDLC

change_id: CHG-2026-014913
risk_class: R2
autonomy_tier: A2
controls: ['TRC-01', 'CHG-04']
data_classification: internal
originator: olaf@freundcloud.com
agent_identity: svc-agent-platform
model_route: gateway/tier-frontier
supersedes: None

Approach

The playbook is not a standalone document. Seven files quote it, one script publishes it, one map contracts against it, and one recording says its plane count out loud. Every edit was checked against those before a word was written, which is why this change adds no plane and touches no policy.

Files

Agentic-SDLC-Playbook.md
docs/playbook-map.md
site/** — the page builder, the mermaid include, the layout, the stylesheet and the version strings the pages carry
README.md

Sequence

Verify every citation at its primary source. → verify: Regulation (EU) 2026/1744 confirmed at EUR-Lex; the doubt was wrong and the document was right.
Grep the couplings: five planes, Playbook §, v1.0. → verify: seven quoting files and one cast identified before any edit.
Front matter — who reads what, plain terms, the contractor analogy, version bump. → verify: renders; build_pages.py lede matches.
§2, §3 — regulatory currency, UK, METR and DORA, sourced statistics. → verify: no bare statistic remains.
§5 — identity into 5.2, MCP registry risk into 5.3, AI-BOM into 5.5, LiteLLM into 5.1. → verify: still five planes, everywhere.
§6 — spec-driven development named; diagrams. → verify: ASCII in fences, as §6.1 already does; the site has no mermaid plugin.
§8.4, §8.5 — standards references, OWASP ASI mapping, named incidents. → verify: all ten ASI ids addressed.
§9, §10.2 — maintainability metric, perception gap, literacy and workforce.
Appendix D rebuilt; Appendix E glossary added.
docs/playbook-map.md — new rows only where this repository actually executes something; everything else into Deliberate departures. → verify: no row without an artifact.
Re-run everything. → verify: gates green, substitution still 12/12.

Risks

Risk	Likelihood	Mitigation
A new §8.4 row claims a control with no gate	high — it is the failure mode this repo exists to refuse	Art. 4 and maintainability go to Deliberate departures, not to the map’s control rows
Adding a sixth plane falsifies the screencast	high if attempted	Not attempted; identity folds into §5.2, and §5’s own independence test says it should
Touching `policy/` escalates this to R3	certain if touched	`policy/` is out of scope; no new machine-checkable control exists here
Appendix C drifts from `substitution_test.py`	medium	Appendix C unchanged; the script defines the twelve
A cited statistic cannot be verified at origin	medium	Every citation resolved before use; secondary-source figures dropped rather than dressed up
The site publishes a stale version string	high	`build_pages.py` lede updated in the same change

Rejected

A §5.6 identity plane. §5 requires each plane be independently sourceable, replaceable and auditable; identity is cross-cutting. It would also require editing a recording.
Editing site/assets/casts/control-layer.cast. It is a recording of a real session. The screencast page argues at length that a reconstruction is the artefact that reads as evidence while being a performance. Editing the cast to match a document would be exactly that.
New Substitution Test checks. Twelve is what substitution_test.py implements.
Mirroring Article 4 into policy/controls.yaml. Literacy is not machine-checkable here, and a declared control with no evidence is worse than an undeclared one.
Adding duplication to the map’s “metrics this repository actually emits”. It does not emit it. It goes in the playbook’s Tier 2 as a recommendation and nowhere else.

Tests

make build test lint gates
make substitution — expect 12/12 unchanged
python site/build_pages.py and a render check
grep -rn "five planes" — count unchanged

Rollback

Revert. The change is documentation and site generation; no gate, policy, workflow or service behaviour depends on it.