ccm-migration

Deterministic mechanism for migrating AEM Forms (XDP) toward Quadient Inspire, built for the customer proof of concept. It implements the design in deliverable 7-devin-mechanism-rulebook-cim-validators-in-git.md: the conversion layer is code and data under test; the agent layer (.agents/skills, playbooks, knowledge) only tells Devin how to work here.

Status: proof-of-concept scaffold over three partner test forms. Gates G0-G4 run offline. G5 (Designer import) and G6 (composed PDF comparison) require Quadient access and return external until a person supplies the evidence. See docs/11-validation-and-gap-report.md for what is done, what is not, and every open question.

Pipeline

XDP ──xdp_to_cim──▶ CIM (JSON, schema-validated) ──rulebook──▶ build plan (JSON)
                       │                                           │
                       │ G1 shape, G2 completeness, G3 readiness   │ Groovy ▶ migration-stack ▶ WFD  (G5, external)
                       │                                           │
   reference XML ◀── G0 pairing / G4 coverage ───────────────────  composed PDF (G6, external)

gate	question	runs offline
G0	Is this the artifact pair we think it is (hash, revision drift, static-text coverage >= 95%)?	yes
G1	Does the CIM satisfy the schema and reference only ids that exist?	yes
G2	Did extraction account for every source node (field, draw, subform, page)?	yes
G3	Is the form migration-ready: bindings resolved, no draft rules, no manual items outstanding?	yes
G4	Does the reference Quadient XML cover the source fields, sections and static text?	yes
G5	Is the target XML structurally valid and does Designer import it cleanly?	structural only
G6	Does the composed PDF match the reference (structure + visual)?	inspection only

How a form gets converted

One form, one Devin session, one PR. The commands below are what the session runs; they are not the process. The process is .agents/skills/convert-xdp-form/SKILL.md, launched from playbooks/convert-form.md, and it ends with a PR that carries the form's decision record (src/decisions/<form_id>.yaml, every blocker/major finding routed with a rationale and the session URL), the dictionary and rule proposals it produced, the Designer handoff package, and the pinned metrics. G3 counts unrouted findings (findings_unrouted), so a pipeline run without the session's routing is visibly incomplete. Humans own the acceptance step: decisions (decided_by), rules (reviewed_by), Designer import (G5) and composed output (G6). Skills for each stage:

stage	skill	playbook
size and order a batch	assess-form-corpus	playbooks/assess-corpus.md
convert one form	convert-xdp-form	playbooks/convert-form.md
hand the plan to Designer, record the result	prepare-designer-handoff	(inside convert-form)
turn import errors into checks and rules	triage-designer-errors	playbooks/triage-import-errors.md
compare AEM and Quadient PDFs, mutate payload	validate-pdf-output	playbooks/validate-pdf-output.md
move a rule from draft to verified	review-rulebook-change	(reviewer PR)

Commands the session runs

pip install -e ".[dev]"
cp /path/to/artifacts/* src/fixtures/raw/         # not committed; see src/fixtures/raw/README.md
ruff check . && pytest -q
python -m tools.pipeline -o out/              # readiness.json/.md, decisions routed vs routable per form
python -m tools.corpus /path/to/xdps -o out/  # batch tiering and suggested order
cat out/readiness.md

Current readiness for the supplied forms (pinned in tests/test_golden.py):

form	G0	G1	G2	G3	G4	G5	G6
17-0574	fail (static text 77%)	pass	pass	fail (draft rules, unresolved bindings)	fail (fields 50%)	external	external
18-1026	pass	pass	pass	fail (draft rules, unresolved bindings)	fail (fields 63%)	external	external
18-1721	pass (source only)	pass	pass	fail	skipped	skipped	skipped

Those fail verdicts are the point: no rule has a named verifier yet and no target grammar has been supplied, so no form can honestly be called ready. The pipeline exits 0 unless an integrity gate (G1/G2) fails; --strict makes readiness failures fatal.

Layout

src/cim/                    JSON Schema for the Canonical Intermediate Model
src/converters/xdp_to_cim/  XDP -> CIM extractor (no script execution, size-capped, master pages included)
src/converters/xpr_to_cim/  xPression -> CIM: raises until an export format is supplied
src/converters/cim_to_inspire/  CIM -> build plan (JSON) + draft Groovy for quadient/migration-stack
src/rulebook/               YAML rules by family with class/status/origin/fixtures + linter (README.md: rule shape)
src/gates/                  G0..G6 + target XML inventory + Designer import-error lint catalogue
src/dictionary/             canonical business elements + per-form crosswalk (review-gated)
src/fixtures/               pairs.yaml (hashes only), raw/ (gitignored), import_logs/, composition_logs/
src/decisions/              per-form decision records written by the converting session, accepted by a human
src/tools/                  pipeline.py (all forms), convert.py (one form), corpus.py (batch), report.py (rulebook)
references/             target grammar evidence the rulebook needs; currently a wanted-list only
tests/                  behaviour tests + golden metrics for the three proof forms
.agents/skills/         assess-form-corpus, convert-xdp-form, prepare-designer-handoff,
                        triage-designer-errors, validate-pdf-output, review-rulebook-change
playbooks/              one-page launch prompts, one per session type
knowledge/              account facts that cannot be code (stubs to fill)
docs/                   validation and gap report

Relationship to quadient/migration-stack

The build plan is the hand-off. src/converters/cim_to_inspire/groovy/CimBuildPlanImport.groovy reads a plan and calls DocumentObjectBuilder, VariableBuilder, TextStyleBuilder, ParagraphStyleBuilder and PageOptions. It has not been executed: Gradle/Maven Central and Inspire Designer were not reachable where it was written. Treat it as a hypothesis to run with a Quadient practitioner (see the gap report, section on G5).