docs: align work-record authority canon
All checks were successful
CI Smoke / host-smoke (push) Successful in 0s
CI Smoke / container-smoke (push) Successful in 1s

This commit is contained in:
codex 2026-08-31 21:17:38 +02:00
parent 6475a2e1ca
commit d3c6f13d7a
5 changed files with 202 additions and 210 deletions

View file

@ -4,255 +4,244 @@ type: architecture-decision-record
title: "Workplans and Work Items Are Repository Artefacts"
status: accepted
owner: the-custodian
revision: "accepted-1"
last_reviewed: "2026-02-28"
revision: "accepted-2"
last_reviewed: "2026-08-31"
review_interval: 6m
decided_by: Bernd Worsch
date: "2026-02-28"
tags: ["architecture", "state-hub", "workplans", "source-of-truth", "rebuild-principle"]
tags: ["architecture", "state-hub", "workplans", "source-of-truth", "rebuild-principle", "forgejo", "projection"]
---
# ADR-001: Workplans and Work Items Are Repository Artefacts
## Status
Accepted.
Accepted 2026-02-28.
Amended 2026-08-31 to distinguish file-backed work records from hub-native
records, identify the Forge default branch as the central projection baseline,
and align identity, lifecycle, reconciliation, and closure with ADR-007,
ADR-010, ADR-011, ADR-012, and the work-record standards. The central decision
is unchanged.
## Context
During early State Hub development (v0.1v0.4), workstreams and tasks were
created directly in the PostgreSQL database via MCP bootstrap tools
(`create_workstream`, `create_task`). This made the database the **origin** of
work items — not a cache or index. The pattern was convenient for rapid
bootstrapping but is architecturally wrong for a system built on the values of
auditability, reversibility, and local-first sovereignty.
During early State Hub development, workstreams and tasks were created directly
in PostgreSQL through bootstrap APIs. This made a database the origin of durable
work rather than a projection of repository-owned artefacts. The pattern was
convenient, but it made work difficult to audit, review, recover, and carry
between generations of tooling.
The trigger for formalising this decision was the creation of the v0.5
workplan ("Dynamic Domains & Multi-Repo") directly in the state-hub database
without a corresponding file artefact in any repository.
The original decision correctly moved durable workplans and their work items
into repositories. It overstated the consequence, however, by saying that the
entire Hub database and every item that matters for coordination must be
reconstructible from repository files. Later decisions established two
different persistence classes:
- **file-backed records**, whose durable meaning originates in a repository;
- **hub-native records**, such as append-only progress and runtime facts, whose
durable meaning originates in the Hub.
Later decisions also established that a central projection reads the pushed
default branch in Forgejo. An arbitrary workstation checkout is a workspace,
not the shared baseline. Unpushed local work may be represented only as an
explicit preliminary overlay.
## Decision
**Workplans and work items MUST originate as Markdown files in the repository
that owns them.** The Custodian State Hub indexes and caches those artefacts
but is never their origin.
### 1. File-backed work originates in the owning repository
Formally: the state-hub must (theoretically, given sufficient compute and
time) be able to **rebuild its full representation** of repositories, their
workplans, tasks, decisions, and dependencies by reading only the files in the
registered repositories. No information that matters for coordination should
exist solely in the database.
Workplans and their file-backed work items **MUST originate as repository
artefacts in the repository that owns the work**. The Hub indexes and projects
those artefacts but is not their origin.
### Corollaries
This includes the workplan, embedded tasks, declared dependencies, residual
handoffs, and durable governed decision artefacts when those are represented as
files. No authoritative field of a file-backed record may exist solely in a Hub
database.
1. **Repository is authoritative.** A workplan file is the canonical record.
The state-hub database row is a materialized cache of that file.
The owning repository is identified by repository identity, not merely by a
domain or topic. Domains and topics classify work; they do not own its source.
2. **Database is disposable.** Dropping and re-creating the database from
registered repository files must produce an equivalent state. The database
is an operational convenience, not a primary store.
### 2. Authority is explicit per persistence class
3. **MCP bootstrap tools become index/sync tools.** `create_workstream` and
`create_task` are acceptable as convenience wrappers only if they write the
file first and then register the row. Using them to write DB-only records
violates this ADR.
Every record contract must declare whether the record is file-backed or
hub-native. The same record must not be writable as authoritative in both
places.
4. **The rebuild principle implies a sync mechanism.** There must be a defined
path (`make sync-workplans` or equivalent) by which the state-hub reads
workplan files from registered repositories and upserts its database state.
- For a **file-backed record**, the repository artefact is authoritative and
the Hub row is derived state.
- For a **hub-native record**, the Hub is authoritative. Progress events,
messages, token events, run history, and similar runtime facts are not made
fictional repository records merely to satisfy a rebuild slogan.
## Workplan File Convention
The word `decision` is used by more than one subsystem. Architecture decisions
and other governed decision artefacts remain files. A runtime decision event may
be hub-native only where its schema says so explicitly; it does not replace the
governed artefact.
Each workplan lives in a `workplans/` directory in the repository that owns
the work. The owning repository is identified by domain.
This replaces the original blanket rejection of a hybrid architecture. The
architecture is hybrid **by declared record class**, never ambiguous within one
record.
### Location
### 3. Forge is the central projection baseline
```
<repo-root>/workplans/<id>-<slug>.md
```
For the shared central view, the source is the pushed default-branch state held
by Forgejo, as decided by ADR-012. Every projected record must be attributable
to its source repository, path, and commit.
Examples:
- `the-custodian/workplans/CUST-WP-0005-dynamic-domains.md`
- `railiance/workplans/RAIL-WP-0001-three-phoenix.md`
A working copy remains the authoring workspace. Unpushed work may appear in the
Hub only as an attributed preliminary overlay. It must not silently replace or
be presented as the Forge-derived baseline. When its commit reaches Forge, the
baseline absorbs it and the overlay retires.
### Frontmatter Schema
### 4. Mutations of file-backed state are file-first
```yaml
---
id: CUST-WP-0005 # human-readable workplan ID, unique per repo
type: workplan
title: "State Hub v0.5 — Dynamic Domains & Multi-Repo"
domain: custodian # must match a registered domain slug
status: active # active | completed | archived
owner: custodian
topic_slug: custodian # maps to a state-hub Topic slug
created: "2026-02-28"
updated: "2026-02-28"
---
```
Creating or changing a file-backed work record means changing its repository
artefact first. A convenience tool or API is conformant only when it performs a
governed repository mutation and leaves a reviewable file and commit. Writing
only the projected database row is not a durable update.
### Task Items
Direct Hub APIs remain valid for hub-native records. They may also provide
diagnostics or propose repository patches, but a successful database PATCH is
not evidence that a file-backed status changed.
Tasks are embedded in the workplan file as headed sections. Each task
section carries its own YAML block:
### 5. The rebuild guarantee applies to the projection, not the whole database
```markdown
## P1.1 — Create `domains` table + Alembic migration
Given a repository and a specific Forge commit, the system must be able to
reconcile that repository's file-derived projection so that it is equivalent to
the records derived from that commit. Reconciliation must be idempotent,
verifiable, and scoped per repository; a fleet operation is iteration over the
same per-repository operation.
```task
id: CUST-WP-0005-T001
status: todo
priority: high
```
Reconciliation retires file-derived records that no longer derive. It does not
delete hub-native history attached to them. It must refuse and report a
repository whose records exist only in the projection until those records have
an explicit disposition. A Forge rebuild does not reconstruct preliminary
overlays and must disclose their retirement before proceeding.
Task description prose here.
```
Therefore the **file-derived projection is disposable**. The database as a
whole is not disposable when it also contains hub-native facts.
The state-hub parses these embedded task blocks during ingestion and upserts
rows in the `tasks` table. The `id` field is the stable external key; the
state-hub UUID is internal and opaque.
### 6. Identity and lifecycle are delegated contracts
### Decision Items
This ADR does not define a second identity or lifecycle schema.
Decisions are separate files or embedded sections following the same pattern,
using `type: decision` in frontmatter.
- Workplan and task identity follow ADR-007 as amended by ADR-011: global
identity is namespace-aware, and derivable Hub identifiers are governed by
that contract. Existing historical identifiers are grandfathered according
to its migration rules.
- Work-record kinds, locations, lifecycle values, and residual handling follow
`canon/standards/work-record-types_v0.1.md` and
`canon/standards/workplan-terminology-fleet_v0.1.md`.
## Rebuild Principle
New normative text uses `workplan` and `work record`. `workstream` remains only
as a metered compatibility term for legacy database and API surfaces.
The rebuild sequence for a clean state-hub:
### 7. Reconciliation belongs at the repository boundary
1. `make migrate` — create schema
2. `make seed-domains` — insert domain rows (domains.yaml in canon/)
3. For each registered repository: `make sync-workplans REPO=<slug>` — parse
workplan files and upsert workstreams, tasks, decisions
4. `make sync-progress` — replay progress events from episodic memory logs
Repo Manager owns discovery, parsing, identity checks, and reconciliation for
file-backed records. State Hub owns the projection and hub-native records. The
implementation may distribute fetch and parse work, but it must preserve that
authority boundary.
After step 4 the database must be functionally equivalent to the live state.
Legacy `create_workstream`, `create_task`, workstation-driven
`sync-workplans`, and similarly named commands are not normative interfaces.
They are conformant only if their current implementation satisfies decisions
3 and 4; otherwise they are transitional or retired.
## Workplan closure
A workplan cannot be marked `finished` merely by changing its Hub row. Before
closure, the responsible worker must:
1. review the source file and the projected view for unfinished tasks;
2. record completed and cancelled outcomes in the source task blocks using the
canonical task lifecycle;
3. turn every actionable carry-forward item into a live work record with
`origin: residual` and an `origin_ref`, or into another explicitly linked
workplan;
4. set the source workplan to `finished`, update it, and commit the result;
5. publish the commit to Forge for the shared baseline and verify that the
projection reconciles to that commit.
A local reconciliation before push may expose preliminary state, but it does
not complete step 5.
Automated stale-task cleanup must report file/projection drift. It may not
silently cancel a file-backed task only in the database. An automated repair may
change such a task only through the same governed repository mutation and
commit path as any other file-backed change. Cleanup of hub-native records is
governed by their own retention contract.
## Consequences
### Immediate
### Positive
- The v0.5 and v0.3 workplans created DB-first in this session are **legacy
records** that violate this ADR. Remediation: write the corresponding
workplan files, then mark the DB rows as `source: db-legacy` until a sync
mechanism can reconcile them.
- Durable work remains inspectable, reviewable, and recoverable through Git.
- The central view has an exact repository and commit provenance instead of
reflecting whichever workstation synced most recently.
- Reconciliation can be exercised per repository without destroying progress
history or other hub-native evidence.
- Multiple contributors share a published baseline while retaining an honest
representation of preliminary work.
- Service ownership is clearer: repositories own file-backed truth; the Hub
owns runtime facts and projection services.
- The state-hub CLAUDE.md design-boundary note must be updated: the MCP
bootstrap tools are permitted only as write-through tools (file + DB), never
as DB-only tools.
### Negative
### Medium Term
- File-backed changes require a repository mutation and normally a push before
they become shared baseline state.
- Forge availability affects projection freshness, although it does not prevent
local authoring.
- Preliminary, stale, and retired projection states must be visible in APIs and
user interfaces.
- Git merge conflicts become the explicit conflict mechanism for simultaneous
edits to the same authoritative artefact.
- Tools that previously corrected only database state must be changed to emit a
diagnostic or perform a governed repository update.
- A `make sync-workplans` command must be implemented as part of the
managed-repos / contribution-tracking infrastructure (see v0.3 workplan).
## Migration
- The `managed_repos` table is the prerequisite: the state-hub must know which
repositories to scan.
No bulk rename, UUID rewrite, or historical file rewrite is required by this
amendment. Existing identifiers and legacy terminology retain the grandfathering
rules of ADR-007, ADR-011, and the terminology standard.
- Workplan file format must be versioned and parsed by a dedicated loader
(`state-hub/scripts/sync_workplans.py`).
Implementations must audit and retire DB-first creation, closure, and stale-task
cleanup paths. Existing file-backed projection rows without a Forge source
commit are migration state: they must be matched to a repository artefact or
explicitly dispositioned before reset. Forge-derived reconciliation and
preliminary overlays are implemented under ADR-012 rather than duplicated here.
### Long Term
The `accepted-1` publication remains the immutable historical revision. This
document is `accepted-2`.
- When the state-hub grows to cover multiple users or teams, this principle
ensures that no coordination state can be lost by a database failure.
Every repository is its own resilient shard of the coordination graph.
## Alternatives considered
- This is the foundation for the "transgenerational" property: workplans in
git survive database migrations, cloud provider changes, and system
rebuilds.
**Database-first with export.** Rejected. Export can be skipped, making the
reviewable artefact secondary and stale.
## Alternatives Considered
**Files only, with no projection database.** Rejected. Cross-repository queries,
runtime views, and append-only operational facts need indexed services.
**Database-first with export:** Create in DB, export to files on demand.
Rejected: export is easily skipped and files become secondary/stale.
**Treat every Hub record as reconstructible from Git.** Rejected. This either
loses runtime truth during rebuild or creates artificial files whose only
purpose is mirroring a database.
**Files-only, no database:** Parse files on every query.
Rejected: impractical at scale; the database is a necessary cache for
cross-repo aggregation and real-time dashboard queries.
**Hybrid with explicit sync flag:** Mark some records as "db-authoritative"
and others as "file-authoritative."
Rejected: introduces ambiguity about which records matter; violates the
"single source of truth" principle.
## Workplan Closure Protocol
When a workplan is about to be marked `finished`, the responsible agent
MUST perform a closure review before writing the status change. This prevents
the stale-task accumulation that this ADR was designed to make detectable.
### Steps
1. **Query all non-done tasks** in the workplan via
`GET /tasks/?workplan_id=<uuid>` (legacy alias: `workstream_id`; filter for
`todo`, `in_progress`, `blocked`).
2. **Classify each task** into one of three outcomes:
| Outcome | Action |
|---------|--------|
| **Done** — work was completed, DB record just wasn't updated | `PATCH /tasks/{id}/ {"status": "done"}` |
| **Cancelled** — dropped, superseded, or out of scope | `PATCH /tasks/{id}/ {"status": "cancelled", "blocking_reason": "<why>"}` |
| **Carry-forward** — genuinely unfinished, belongs in the next run | Leave open; note in closure review; trigger new workplan |
3. **Append a `## Closure Review` section** to the workplan file:
```markdown
## Closure Review — YYYY-MM-DD
**Outcome:** All tasks completed / N tasks carried forward / N tasks dropped.
### Completed (DB updated)
- TASK-ID — title
### Cancelled (dropped)
| Task | Reason |
|------|--------|
| TASK-ID — title | Superseded by X |
### Carried forward
| Task | Target workplan |
|------|----------------|
| TASK-ID — title | CUST-WP-XXXX |
```
4. **If any tasks are carried forward**: do not mark the workplan
`finished` yet. Create the new workplan file (or amend an existing active
one), then close the current workplan.
5. **Update the workplan frontmatter** `status: finished` and `updated:` date.
6. **Mark the workplan `finished`** in the state hub via MCP or API (`update_workplan_status`).
### Daily Stale-Task Cleanup
As a safety net for cases where the closure review was skipped or incomplete,
a cleanup script cancels any surviving open tasks in completed/archived
workstreams:
```bash
cd ~/the-custodian/state-hub
make cleanup-stale # run immediately
# or add to cron:
# 0 3 * * * cd ~/the-custodian/state-hub && make cleanup-stale
```
The script (`scripts/cleanup_stale_tasks.py`) emits a `cleanup` progress event
recording which tasks were cancelled and in which workstreams. Tasks cancelled
by the cleanup carry a `blocking_reason` noting they should be verified against
the workplan file.
The closure review is the primary mechanism; the cleanup is the fallback. If
the cleanup regularly cancels tasks, it signals that closure reviews are being
skipped — that is the process failure to address, not just the stale tasks.
**Let working copies push authoritative projection rows.** Rejected by ADR-012.
It makes the baseline depend on the last workstation to reconcile and cannot be
audited against a shared commit.
## Related
- Custodian Constitution v0.1 §2 (Powers) — canon changes require review gate
- ADR-000 (forthcoming) — overall Custodian architecture principles
- State Hub v0.3 workplan — `sync_workplans.py` is a Phase 4 deliverable
- `canon/values/foundational_values_v0.1.md` — Local-first, Auditability,
Reversibility
- ADR-003 — materialized and derived state
- ADR-005 — cross-repository work ownership
- ADR-007 — workplan identity and repository worker topology
- ADR-010 — Hub authority and local cache model
- ADR-011 — namespace-aware federation and reconciliation limits
- ADR-012 — Forge projection source and preliminary overlay
- `canon/standards/work-record-types_v0.1.md` — work-record kinds, lifecycle,
residuals, and reconciliation
- `canon/standards/workplan-terminology-fleet_v0.1.md` — canonical terminology
- `canon/values/foundational_values_v0.1.md` — local-first operation,
auditability, and reversibility

View file

@ -4,8 +4,8 @@ type: architecture-decision-record
title: "Materialized Derived State with Fingerprint Invalidation for Repo-Sourced Data"
status: accepted
owner: the-custodian
revision: "accepted-1"
last_reviewed: "2026-03-20"
revision: "accepted-2"
last_reviewed: "2026-08-31"
review_interval: 6m
decided_by: Bernd Worsch
date: "2026-03-20"

View file

@ -4,8 +4,8 @@ type: architecture-decision-record
title: "Workplan Identity Uniqueness, Single Registrar, and Repo Worker Topology"
status: accepted
owner: the-custodian
revision: "accepted-1"
last_reviewed: "2026-08-17"
revision: "accepted-2"
last_reviewed: "2026-08-31"
review_interval: 6m
decided_by: Bernd Worsch
date: "2026-08-16"

View file

@ -4,8 +4,8 @@ type: architecture-decision-record
title: "Hub Authority, Local Cache, and the Two Kinds of Hub Data"
status: proposed
owner: the-custodian
revision: "draft-1"
last_reviewed: "2026-08-17"
revision: "draft-2"
last_reviewed: "2026-08-31"
review_interval: 6m
decided_by: Bernd Worsch
date: "2026-08-17"

View file

@ -3,9 +3,9 @@ id: coulomb-estate-architecture
title: "Coulomb estate architecture"
status: proposed
owner: the-custodian
revision: "draft-2"
revision: "draft-3"
version: "0.1"
last_reviewed: "2026-08-19"
last_reviewed: "2026-08-31"
review_interval: 6m
---
@ -174,26 +174,28 @@ already in §4 and the ADRs in §9.
## 9. Architecture Decisions
Estate ADRs in `the-custodian/canon/architecture/`. Publication ids
still need repo prefixes (WP-0003). Status is the source front-matter.
Estate ADRs in `the-custodian/canon/architecture/`. Status is the source
front-matter.
| Id | Status | Decision |
| --- | --- | --- |
| `CUST-ADR-001` | accepted | Workplans and tasks originate as repo files; the hub is a read model. |
| `CUST-ADR-001` | accepted | File-backed work originates in repositories; the Hub projects it and separately owns declared Hub-native records. |
| `CUST-ADR-002` | accepted | Custodian agent runtime design. |
| `CUST-ADR-003` | accepted | Hub state is derived and fingerprint-invalidated. |
| `CUST-ADR-004` | accepted | Connectivity-first network posture. |
| `CUST-ADR-005` | accepted | Cross-repo work lives in a `prj-*` repo. |
| `CUST-ADR-006` | accepted | Three canons; import, do not redefine. |
| `CUST-ADR-007` | accepted | Globally unique workplan ids; one registrar. |
| `CUST-ADR-007` | accepted | Namespace-aware work-record identity, deterministic Hub ids, and repository worker topology. |
| `adr-008` | superseded | Relocated to NetKingdom Tenancy Posture. |
| `CUST-ADR-010` | proposed | Two kinds of hub data; local cache vs authority. |
| `CUST-ADR-011` | proposed | Federated namespaces and reconciliation limits. |
| `CUST-ADR-012` | accepted | Forge is the projection source; unpushed working-copy state is preliminary. |
Related and published elsewhere: policy-nexus ADR-0001; Tenancy Posture
and IAM Profile v0.3; railiance-master ADR-00010008; activity-core
ACT-ADR-001005; ops-warden ADR-00010005; railiance-platform
RPLAT-ADR-00010003.
and IAM Profile v0.3 plus the current NetKingdom security standards;
NetKingdom ADR-00060008 and 00100015; railiance-master ADR-00010009;
activity-core ACT-ADR-001007; ops-warden ADR-00010005 and 00070010;
railiance-platform RPLAT-ADR-00010003.
Unresolved WP-0003 conflicts (hosts/infra duplicate ADR-003/004,
coulomb-social ADR-0002 partial supersession) are **not** listed as
@ -210,8 +212,9 @@ mechanical checks (`fix-consistency`, `make currency`).
## 11. Risks and Technical Debt
N/A in this revision. Known residual: this workstation cannot mint hub
UUIDs (ADR-007 registrar). Markitect `arc42-v1` is not registered yet.
N/A in this revision. Known residuals are the Forge-derived reset and
preliminary-overlay implementation under ADR-012. Markitect `arc42-v1` is not
registered yet.
---