the-custodian/canon/architecture/adr-007-workplan-identity-and-repo-worker-topology.md
codex d3c6f13d7a
All checks were successful
CI Smoke / host-smoke (push) Successful in 0s
CI Smoke / container-smoke (push) Successful in 1s
docs: align work-record authority canon
2026-08-31 21:17:38 +02:00

324 lines
15 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

---
id: CUST-ADR-007
type: architecture-decision-record
title: "Workplan Identity Uniqueness, Single Registrar, and Repo Worker Topology"
status: accepted
owner: the-custodian
revision: "accepted-2"
last_reviewed: "2026-08-31"
review_interval: 6m
decided_by: Bernd Worsch
date: "2026-08-16"
accepted: "2026-08-17"
tags: ["architecture", "workplans", "identity", "state-hub", "repo-manager", "agents", "registrar"]
---
# ADR-007: Workplan Identity Uniqueness, Single Registrar, and Repo Worker Topology
## Status
Accepted 2026-08-17. Identifier uniqueness, the registrar model, lifecycle
protection, and worker topology are settled.
**Remediation of existing collisions (§ Migration) remains an open ruling.** It
is disruptive, touches six repositories, and no active work depends on it — all
five duplicated identifiers are `finished`.
## Context
On 2026-08-16, `STATE-WP-0080` was found registered **twice, in two different
databases**, with two different workplan UUIDs and two disjoint sets of task
UUIDs:
| Registrar | Workplan UUID | Reachable at `127.0.0.1:8000` |
|---|---|---|
| Workstation hub (`make api`, local postgres) | `03f38314` | yes |
| A second instance, over a second database | `bbfce36a` | **404** |
The second instance is *not* identifiable from the commit author.
`custodian-sync@railiance.local` is a hardcoded default git identity in
`scripts/consistency_check.py:1899` (`GIT_SYNC_USER_NAME` /
`GIT_SYNC_USER_EMAIL`), so every machine running `fix-consistency` commits under
that name. The discriminator is the timezone: sync commits appear under both
`+0000` and `+0200` (the workstation's offset), which is independent evidence of
two machines writing to one repository.
Both write their IDs into the same git-tracked workplan file, so each sync
overwrites the other's IDs and the file flip-flops on every round trip. The same
commit (`ff909e1`, "renormalize lifecycle state [auto]") also promoted the
workplan `proposed``active` without human review.
A fleet scan of 955 hub workplans (525 with parseable `PREFIX-WP-NNNN`
identifiers) found this is not an isolated incident. Two distinct identity
defects are live:
**Prefix reuse across repositories:**
```text
CUST-WP- → state-hub, the-custodian
RAIL-BS-WP- → railiance-bootstrap, railiance-cluster
RAILIANCE-WP- → railiance-apps, railiance-forge, railiance-platform, railiance-telemetry
```
`PRJ-WP-` is a latent fourth: `statehub register` derives it from the `prj-`
flavor marker, so every project repository would collide (see `RMGR-WP-0004`).
**Running-number reuse:**
```text
CUST-WP-0000 the-custodian ×2
CUST-WP-0010 the-custodian ×2
CUST-WP-0045 the-custodian ×2
RAILIANCE-WP-0015 railiance-platform, railiance-apps
RAILIANCE-WP-0016 railiance-platform ×2, railiance-apps
```
`RAILIANCE-WP-0015` is actively cited in operational memory as the cnpg-backup
gate in `railiance-apps`; a second record of the same name in
`railiance-platform` makes that citation ambiguous.
The `C-26` consistency check already flags prefix nonconformance *within* a repo
against a canonical prefix, but nothing enforces uniqueness *across* repos, and
nothing prevents number reuse.
## Decision
**1. A workplan identifier is globally unique.** `PREFIX-WP-NNNN` names exactly
one workplan across the entire fleet, for all time.
> **Amended 2026-08-17 by `ADR-011` decision 2.** Uniqueness and forward-only
> allocation are **namespace-scoped**, not global; global identity is the pair
> `(namespace, identifier)`, written `PREFIX-WP-NNNN@namespace` when foreign.
> Global sequential allocation would require a central coordinator — the exact
> dependency federation must survive. Everything below holds unchanged **within**
> a namespace, which is where all current work sits.
- A workplan prefix is owned by exactly one repository. No two repositories may
use the same prefix.
- A running number is never reused within a prefix, including after a workplan is
cancelled, archived, or deleted. Numbers are allocated forward only.
- Prefixes are derived from the *project or repository identity*, never from a
flavor marker or category (`PRJ-WP-` is invalid by construction).
**2. Hub identifiers stored in repository files must be derivable, not
database-local.** The defect is structural: a database-local key is stored in a
shared artifact, so each database overwrites the other's value on every sync.
It also inverts ADR-001 — a file carrying a hub's private key is the file
holding hub state.
*Target state (C2).* `state_hub_workstream_id` and `state_hub_task_id` become
**deterministic**: UUIDv5 derived from the workplan identifier. Per `ADR-011`
decision 3 the derivation input is the pair `(namespace, identifier)`, not the
identifier alone — deriving from the identifier alone would make two forks
holding unrelated work under the same number compute the same UUID. Every instance computes the same value independently, writeback
becomes idempotent, and any number of hub instances may coexist without
coordination. The field shape is unchanged, so consumers keep working; only the
provenance of the value changes.
This has a hard prerequisite: deterministic derivation from a **non-unique**
identifier would manufacture collisions rather than remove them. Two project
repos sharing `PRJ-WP-` would compute the same UUID for different workplans.
Decision 1 must therefore be enforced before derivation ships.
*Interim state (A).* Until derivation lands, exactly one instance writes hub
identifiers into repository files. Other instances may read, project, and serve,
but must not mint workplan or task UUIDs into git-tracked files.
> **Corrected 2026-08-17, superseded by `ADR-010` decisions 13.** This decision
> originally described workstation hubs as "development read replicas". That was
> wrong on both counts: the workstation instance was not a replica, and it was
> the *larger* of the two, holding 306 more workplans than the primary. The two
> instances were peer databases. `ADR-010` establishes the central hub as
> authoritative and local instances as rebuildable caches, which is what makes
> this interim rule coherent.
The interim is policy, enforced by discipline, and it has a real cost:
registration requires connectivity to the registrar, so disconnected work cannot
register. That cost is accepted only until C2 removes the need for it, at which
point the number of hub instances becomes an availability choice rather than a
correctness constraint.
*Rejected.* Collapsing to a single shared database (workstation hubs pointing at
the production database) would also make identifiers consistent by construction,
but it eliminates offline capability — cutting directly against `STATE-WP-0068`
(offline write buffer and edge relay) — and couples all local work to tunnel
availability.
*Ownership.* Both the interim guard and the derivation belong to `repo-manager`
under decision `747011c6`, which already places file-backed record indexing and
reconciliation there. Building either in `state-hub` would invest in a component
being retired under `STATE-WP-0079`.
Migration scope for C2: **758 workplan files** across the fleet currently carry
these fields.
> **Amended 2026-08-21 — derivation scope for C2.** `ADR-011` decision 7 keys
> derivation on `(namespace, identifier)`, which separates *forks*. It does not
> separate collisions inside one namespace, and the ecosystem's posture is `N1`
> — a single implied namespace. A fleet scan on 2026-08-21 found **20 reused
> identifiers across 48 files**, all within that one namespace, so the
> prerequisite above is not satisfied by `ADR-011` alone.
>
> Ruled: **C2 derives only for live records; archived records keep their minted
> identifiers, frozen.** This is what reconciles § Migration option 2 — under
> which historical files keep colliding identifiers — with the uniqueness
> derivation requires. Of the 20, only **five** collide among live files, so the
> remediation surface is 11 files rather than 48.
>
> Two consequences follow, and both are load-bearing:
>
> 1. **Un-archiving a record with a frozen identifier is a collision hazard.**
> A record returning to live status must be checked against the live
> namespace before it is re-derived, and renumbered if it clashes.
> 2. **Derivation is not retroactive.** Existing live records keep their minted
> UUIDs until they are re-derived deliberately; C2 changes provenance for new
> and re-registered records, not the whole corpus at once.
>
> Rejected: treating a *repository* as the namespace. That would make the
> collisions vanish by construction, but it redefines the term `ADR-011`
> decision 1 fixes as "a fleet instance, a client deployment, an autonomous
> domain", and `ADR-011` decision 9 warns specifically against reading N-plane
> movement into claims it does not support.
**3. Lifecycle status is not automatically promoted.** An automated
normalization pass may report drift; it may not move a workplan from `proposed`
to `active`. `proposed` means awaiting human review, and an automation that
promotes it destroys the meaning of the review gate.
**4. Repository manipulation is performed by a worker agent in that
repository.** This is the default topology.
- A worker acting in repo X owns changes to repo X.
- Multiple independent top-level workers inside a single repository are an
**exception**, requiring an explicit reason, not a routine mode of operation.
- Concurrent independent writers are what turned a two-registrar bug into
repeated git divergence.
**5. Project (`prj-`) repositories may act across their participating
repositories.** When work is governed by a project repo, its tasks may direct
changes across every repository the project names, through the project's work
agent, where that is more efficient than delegating.
This is a deliberate, scoped exception to decision 4: the project repo already
owns cross-repo sequencing and its `SCOPE.md` names its participants, so its
authority is declared rather than ad hoc. It does not license a worker in an
arbitrary repository to reach into others.
## Consequences
**Positive.** Workplan identifiers become citable without qualification.
Hub IDs stop flip-flopping in git. The `proposed` status regains meaning.
Cross-repo authority becomes something a repository declares rather than
something any session assumes.
**Negative.** Existing collisions must be remediated (see below), which is
disruptive. Workstation sessions lose the ability to register workplans directly
and must route through the registrar or a worker in the owning repo. Prefix
allocation needs a fleet-level registry, which is new machinery.
**Enforcement.** Prefix ownership, uniqueness, and forward-only numbering are
repository standards, so they belong to Repo Manager under decision `747011c6`
(`RMGR-WP-0004`), not to a hub. Canon defines the rule; Repo Manager checks it.
## Migration — needs a separate ruling
Three prefixes and five identifiers are already colliding. Remediation options,
in increasing cost:
1. **Freeze and forward-fix.** Accept existing collisions as historical, enforce
uniqueness only for new workplans. Cheapest; leaves `RAILIANCE-WP-0015`
permanently ambiguous.
2. **Renumber the live collisions only.** Fix identifiers that are still cited or
active; leave `finished`/`archived` duplicates alone.
3. **Full renaming.** Give `railiance-apps`, `railiance-forge`,
`railiance-platform`, `railiance-telemetry` distinct prefixes, likewise
`railiance-bootstrap`/`railiance-cluster` and `state-hub`'s legacy `CUST-WP-`
files. Touches six repositories and every inbound reference.
**Ruled 2026-08-17: option 2.** Renumber live collisions; leave
`finished`/`archived` duplicates as historical record.
### The live renumber list is empty
Verified against the fleet scan. All five duplicated identifiers are `finished`:
```text
CUST-WP-0000 the-custodian ×2 finished
CUST-WP-0010 the-custodian ×2 finished
CUST-WP-0045 the-custodian ×2 finished
RAILIANCE-WP-0015 apps, platform finished
RAILIANCE-WP-0016 apps, platform ×2 finished
```
No workplan in a `proposed`, `ready`, `active`, `blocked`, or `backlog` state
shares an identifier with another. **Option 2 therefore requires no renumbering
today.** Historical duplicates stay, including the `RAILIANCE-WP-0015` ambiguity
between `railiance-apps` and `railiance-platform`; citations of it must name the
repository.
### The structural cause is not historical
Option 2 governs *remediation*. It does not exempt anything from decision 1,
which is accepted canon: one prefix, one repository, forward-only numbering.
Three shared prefixes are still in use across seven repositories, and each is a
**single number line being allocated from concurrently**:
```text
RAIL-BS-WP- bootstrap 8, 9
cluster 7, 10, 11, 12, 13, 14
RAILIANCE-WP- platform 5, 8-17 (16 twice — an internal duplicate)
apps 15, 16 <- already collided with platform
forge 2
telemetry 1
CUST-WP- the-custodian 50 plans
state-hub 4 legacy plans (canonical prefix is STATE-WP)
```
`RAIL-BS-` and `RAILIANCE-` are actively growing — `RAIL-BS-WP-0014` (`ready`),
`RAILIANCE-WP-0002` (`ready`), `RAILIANCE-WP-0001` (`proposed`), all created
2026-08-11 or later. The `RAILIANCE-WP-0015`/`0016` collisions were not a
historical accident; they are what concurrent allocation from a shared sequence
produces, and it will recur at the next concurrent allocation.
`CUST-` is dormant on the `state-hub` side — four legacy plans, one in
`backlog` — and needs no split, only a prefix-ownership assertion.
### Prefix assignments
`RAIL-BS-WP-` is **retired** (2026-08-17). Neither repository keeps it:
`railiance-cluster` adopts `RCLUSTER-WP-` for active and future plans;
`railiance-bootstrap` adopts `RBS-WP-` for future plans. Finished and archived
files keep `RAIL-BS-WP-` as historical record, consistent with option 2.
Migrating plans keep their running numbers — the prefix changes, the number does
not. This preserves traceability and cannot violate forward-only allocation,
because neither new prefix has prior history. `railiance-bootstrap` begins at
`RBS-WP-0010`, above its historical maximum, leaving the lower range free should
its finished plans ever be adopted into the new prefix.
`RAILIANCE-WP-` should follow the same pattern — retired rather than awarded to
one repository, since it names a family rather than a repository and so fails
decision 1 for the same reason `PRJ-WP-` does. Assignment of the four successor
prefixes is outstanding.
Execution of each rename belongs to a worker in the owning repository under
decision 4. `RMGR-WP-0004-T09` records assignments and the numbering rule; it
does not perform renames.
**Consequence.** Prefix ownership must be assigned for all three shared prefixes
before the next workplan is created in the affected repositories. This is
forward conformance under decision 1, not migration, and is tracked as
`RMGR-WP-0004-T09`. Renaming the *historical* files is explicitly not required —
that would be option 3, which was rejected.
## References
- Decision `747011c6` — repository standards belong to Repo Manager
- ADR-001 — workplans originate as repo files; hub is a read model
- `RMGR-WP-0004` — repository standards conformance and governed scaffolding
- `STATE-WP-0080` — register scaffolding handoff
- Fleet scan 2026-08-16: 955 hub workplans, 525 parseable identifiers,
3 reused prefixes, 5 reused identifiers