Task: B5: Implement supported/target set model for physical space binding

Introduce the supported/target set model for physical-space generation, following the MASD variability convention used by Dogen. Physical space is a "set of sets": technical-space → facet → archetype. Two named sets govern generation for each run:

• Supported set (S_e) — what entity e can generate. Derived from ores.* properties in its :PROPERTIES: drawer (specificity-resolved) intersected with the model-types filter. Entities with no ores.* properties: S_e = all archetypes matching the model-type (backward compatible with current behaviour).
• Target set (T) — what to generate this run. Default: T = S_e. The CLI --address argument sets T = the set of all archetypes matching that address (hierarchically expanded from the facet catalogue).

What actually generates for entity e = T ∩ S_e. Warning condition: T ∩ S_e = ∅ → emit warning "<entity>: nothing to generate for address <addr>="; continue with other entities. Error condition: --address value is not a valid address in the facet catalogue → error.

# Generate everything the entity supports (T = S_e)
./projects/ores.codegen/codegen.sh generate --component refdata

# Generate only the sql technical space (T = all sql archetypes)
./projects/ores.codegen/codegen.sh generate --component refdata --address sql

# Generate only the qt facet (T = cpp.qt archetypes)
./projects/ores.codegen/codegen.sh generate --component refdata --address cpp.qt

Field	Value
State	DONE
Parent story	Resolve codegen model unification blockers
Now	Nothing — shipped via PR #1361 (merged 2026-06-28).
Waiting on	Nothing. (B1 landed first via PR #1355, 2026-06-27.)
Next	Follow-ups tracked as B9/B10/B11 (rename templates, retire facet_catalogue, migrate –profile callers).
Last touched	2026-06-29

Design decisions taken (2026-06-27):

1. Doc-per-node graph — one org document per node at every level: a root (ores.org), one per technical space (ores.cpp.org), one per facet (ores.cpp.qt.org), one per archetype (ores.cpp.qt.controller.org). Each doc links down to its children (a table of id: links) and up to its parent. The monolithic facet_catalogue.org (all facets as headings in one file) is wrong and gets decomposed into these per-node docs, then retired. Likewise the duplicated literate facet=/=facet_group docs collapse into this single graph.
2. Scaffold every node via compass — never hand-write these org files. doc_generate.py grows the doc types needed: technical_space (lists its facets) and archetype (the leaf); facet already exists (it lists its archetypes); the root ores.org. Any doc-template gap is fixed in compass, not worked around.

Codegen traverses this doc graph: root → technical space → facet → archetype, every edge an org-roam id: link, with each node linking back up to its parent.

Core invariant — node name = address = drawer key, normalised with the ores root prefix everywhere (node names, --address values, and :PROPERTIES: drawer keys are the identical string):

Level	Node name / address / drawer key
root (general TS index)	ores
technical space	ores.cpp / ores.sql
facet	ores.cpp.qt / ores.sql.schema
archetype	ores.cpp.qt.controller / ores.sql.schema.create

The ores root node IS the general technical-spaces index page; it links to each technical space. :ores.cpp.qt.enabled: false disables a facet and :ores.cpp.qt.controller.enabled: false disables one archetype, by pure name match — no separate address→node lookup table. --address ores.cpp.qt uses the same string.

Archetype naming follows the same standard convention (matching Dogen's masd.cpp.types.class_header): one archetype per output artefact, so a header and its implementation are two distinct archetypes, each with a clean descriptive leaf — never the raw filename, never an _hpp=/=_cpp suffix. E.g. cpp_domain_type_class.hpp.mustache → archetype ores.cpp.domain.class_header; ..._entity.cpp.mustache → ores.cpp.domain.entity_implementation. The template filename is an attribute of the archetype node, not its identity. A standard filename→leaf mapping is derived once in Stage 4 and applied across all ~106 archetypes.

Node set (each its own document, name = address):

• Technical spaces: ores.cpp, ores.sql, ores.doc (ores.cmake=/ =ores.assets fold in later). Each lists its facets.
  • ores.doc is the documentation technical space — the doc-generation system is just its slice of the same graph. Its archetypes are the doc_*.org.mustache templates (doc_facet, doc_knowledge, doc_technical_space, doc_archetype, the agile/task/story scaffolds, …), grouped into doc facets. Self-hosting: the templates that scaffold technical-space / facet / archetype nodes are themselves archetypes in ores.doc. The literate doc.org facet + doc_group.org are its seed.
• Facets (TS-qualified names): each lists its archetypes.
  • ores.sql → ores.sql.schema (schema create/drop/trigger — what the monolith mislabels ** sql :facet:), ores.sql.populate, ores.sql.service, ores.sql.non-temporal.
  • ores.cpp → ores.cpp.domain, ores.cpp.generator, ores.cpp.repository, ores.cpp.service, ores.cpp.protocol, ores.cpp.nats-eventing, ores.cpp.nats-handler, ores.cpp.qt, ores.cpp.non-temporal-domain, ores.cpp.non-temporal-repository, ores.cpp.enum, ores.cpp.field-group.
• Archetypes: one per output artefact (the existing templates/*.mustache.org literate docs), retyped archetype and addressed ores.{ts}.{facet}.{leaf}.

NB: the old B5 spec's TS→facet map (sql → {sql, non-temporal-sql}) was imprecise — sql is the technical space, its create facet is ores.sql.schema.

Stage 3c decisions (2026-06-27):

• Scope: the ~50 entity-facet templates (15 facets). Accurate counts — domain 7, qt 12, repository 6, non-temporal-repository 6, sql.schema 7, generator 2, service 2, non-temporal-domain 3, and singletons.
• Shared archetypes: json_io.hpp/.cpp are duplicated per facet — ores.cpp.domain.json_io_* AND ores.cpp.non-temporal-domain.json_io_*, each owning its own node tangling the same source.
• plantuml: its own technical space ores.plantuml (not under ores.doc); plantuml_er → an archetype under an ores.plantuml facet.
• Component-scaffolding archetypes (cmake_*, cpp_component_*, cpp_service_app_*, sql_service_*, shell_*, the * Component archetypes) are deferred to a follow-up task — component-scope, driven by other commands, not needed for B5's per-entity --address model.
• Mustache filenames match the address: ores.cpp.domain.class_header → ores_cpp_domain_class_header.mustache (dots→underscores). Renamed in Stage 5 (atomically with the :tangle targets, core.py output-path map, and generator lookup) to avoid a broken intermediate.

Stages (commit each):

1. ✓ Link the 106 archetype cells in facet_catalogue.org to their archetype docs; parser strips the link to the bare template name. (Interim — the monolith is decomposed away in later stages.)
2. Extend compass: add technical_space + archetype doc types to doc_generate.py (templates + registration + compass add help).
3. Scaffold the node docs via compass add: root ores.org, the technical spaces (ores.cpp, ores.sql), the facets, retype the archetype docs.

4. Wire the graph by automation, never by hand. Up-links are frontmatter parent keywords set once at scaffold time (#+facet_group: on a facet, new #+facet: on an archetype; technical spaces are discovered by type). Down-link child tables are generated: extend the existing scripts/regenerate_facet_inventories.py into a graph-wide inventory generator that rewrites, from the doc frontmatter:

   • ores.org * Technical spaces ← all #+type: technical_space docs
   • each TS * Facets ← #+type: facet with matching #+facet_group:
   • each facet * Archetypes ← #+type: archetype with matching #+facet:

   Keep the --check CI gate. Migrate facet/archetype content out of the monolith + duplicated literate facet=/=facet_group docs; retire them.

5. Rewire codegen to traverse the doc graph (root → TS → facet → archetype) instead of reading facet_catalogue.org; implement codegen/physical_space.py (compute_supported_set / compute_target_set / resolve_generation_set). (The .mustache rename to address form is deferred to task B9 — it is cosmetic, touches the legacy get_schema_template_mappings path, and is safer as a self-contained change. The graph uses the :tangle indirection, so filenames need not change for the rewire to work.) 5f — functional retirement done: nothing reads facet_catalogue.org for generation any more (load_profiles=/=resolve_profile_templates have no callers — dead code). The physical deletion of the monolith + 18 legacy literate facet/group docs + dead code + repointing the hundreds of inbound id: links (56 to the monolith alone) is deferred to task B10 — a delicate graph-wide link migration best done on its own.
6. Remove --profile; add --address — –address added, –profile kept as a deprecated shim (5e). Updating all compass commands + skills off --profile is the remaining part of stage 6.
7. tests/test_physical_space.py covering every supported/target combination.

Merge gate (decided 2026-06-28): do NOT merge this branch until all of B8, B9, B10, B11 are complete and the in-flight commissions are off --profile. Merging B5 alone would leave main in a half-migrated state (stale facet_catalogue.org monolith + duplicated legacy docs + dead load_profiles code + mixed --profile=/–address= callers). The whole physical-space migration lands as one coherent change.

Mitigation for the long-held branch: rebase on origin/main periodically (the commissions are actively changing codegen) to bound rebase drift.

Replaces the originally planned B5 (#+sql_only: true) and B6 (#+has_qt: false) tasks. See Applied MASD and ORE Studio Variability Model for the full design.

PR	Title
#1374	[agile] Reconcile sprint-21 task bookkeeping drift
#1361	[codegen] B5: physical-space address graph (merged 2026-06-28)

Round 1 (PR #1361, @claude, 2026-06-28):

#	Comment summary	File	Decision	Notes
P1	facets_under("ores") dead code — –address ores returned 0 on live graph	physical_space.py	Accepted	Fixed c8bf447c5: root check moved first
P2	No live-graph test for facets_under("ores")	test_physical_space.py	Accepted	Fixed c8bf447c5: added regression test (16 tests)
m1	Silent skip of archetypes with empty template/output or bad pattern	generate.py	Accepted	Fixed b32c04df1: log.debug on both skip paths
m2	–profile deprecation silent at runtime	generate.py	Accepted	Fixed b32c04df1: runtime warning naming the –address equivalent
m3	Keyword regex inconsistent with inventory script	physical_space.py	Accepted	Fixed b32c04df1: unified to \w[\w-]*
m4	Facets without model_types implicitly universal	facet docs	Deferred	Empty facets own no archetypes; B8 adds model_types when populating

Round 2 (PR #1374 — bookkeeping reconciliation, @claude, 2026-06-29):

#	Comment summary	File	Decision	Notes
1	"Future" date 2026-06-30 in verify_codegen	task_verify_codegen.org / story.org	Declined	Accurate — that task was reconciled after midnight (journal 2026-06-30 00:00); B5/qt-upgrade fixes were 2026-06-29
2	Story-row description still ABANDONED-language after DONE	commission_country/story.org	Accepted	Fixed 69c48269d: reworded to "goal met via unified-model migration tasks; see Result"
3	#+environment merry_newton→prime_origin not noted	b5 / qt-upgrade scaffold tasks	Acknowledged	Auto-stamped by compass task done from the running environment; expected noise

Shipped via PR #1361 (merged 2026-06-28, merge commit 65eaed5d3), with B1 (PR #1355) landing first as required. The original merge gate ("hold until B8–B11") was lifted and B5 landed on its own; the remaining migration work is tracked as separate backlog tasks (B9 template rename, B10 facet_catalogue retirement, B11 –profile caller migration).

What landed on main:

• codegen/physical_space.py — the isolated supported/target set engine: compute_supported_set (resolves ores.* drawer properties with specificity ordering), compute_target_set (expands a --address to its facet set), resolve_generation_set (= supported ∩ target, warns on empty).
• The per-node doc graph: ores.org root → technical-space docs (ores.cpp.org, ores.sql.org) → facet docs → archetype docs (85 archetype-typed + 11 technical_space-typed docs), each linked via org-roam id: up/down edges; codegen traverses this graph instead of the monolith. Node name = address = drawer key, ores-prefixed throughout.
• CLI: --address added (optional filter, defaults to the entity's supported set); --profile retained as a deprecated runtime-warning shim (full removal deferred to B11).
• tests/test_physical_space.py — covers backward-compat (no properties), TS-level disable, facet-level disable, combined override, normal filter, empty-intersection warn, and unknown-address error (16 tests).

Review round 1 (6 comments) addressed in c8bf447c5 / b32c04df1; one item (m4) deferred to B8. Acceptance criteria for supported/target resolution, intersection, CLI, facet catalogue, and isolation/testing all met.

Bookkeeping note: this task's state was corrected to DONE on 2026-06-29 — the PR merged 2026-06-28 but the sprint health review updated only B1, leaving B5 showing STARTED. No code change; bookkeeping reconciliation only.