Publish factory efficiency return and point to canonical project
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 2026-09-08 14:54:13 +02:00
parent 389f7e7f1d
commit c1298bd476
8 changed files with 63 additions and 584 deletions

View file

@ -23,20 +23,29 @@ utility; Railiance Fabrics hosted authority still adds prerequisite work.
flags still require source/residual disposition; none was cleared by counting
it as completed. [Correction evidence](2026-09-08-helixforge-factory/human-flag-corrections.json).
The dedicated local project is
[/home/worsch/prj-helixforge-factory](/home/worsch/prj-helixforge-factory/README.md).
The dedicated private project is published at
[prj-helixforge-factory](https://forgejo.coulomb.social/coulomb/prj-helixforge-factory),
with [local source](/home/worsch/prj-helixforge-factory/README.md).
It contains the accepted source-backed decision HFACT-DEC-2026-001, current
workplan, ten exact owner-return records, delivery evidence and an empty
measurement ledger. Hub decision receipt: `21159693-04b7-485e-bbab-1ed8f65aab43`.
The original staged packet is frozen pending project publication.
The original staged packet has been replaced with a pointer, preserving its Git
history. State Hub registration returned repository UUID
`99cc4689-9f73-42a2-9cca-d2e235393014` under the helix-forge topic.
## Remaining gates
Project creation needs the routed Forgejo admin API credential, which currently
has no caller OpenBao login. Organization push-to-create is disabled. The same
login can record both already-integrated PRs as manually merged; the PR metadata
still says open, although the tested source is on main. There is no secret-value
request or additional token provision in this handoff.
The routed OpenBao login completed. Both already-integrated PRs now report
closed/merged with the exact reviewed commit. Forgejo initially refused the
disabled manual-merge style; it was enabled only for each metadata operation and
restored to disabled. Current main ancestry was verified before the operations.
Non-secret [publication receipts](https://forgejo.coulomb.social/coulomb/prj-helixforge-factory/src/branch/main/evidence/pr-merge-receipts.json)
are retained in the project.
The credential proxy reported flex-auth unavailable and its configured
unknown-zone fail-open behavior during this attended publication. HFACT-WP-0001-T03
retains live policy availability and authorization-negative verification before
governed execution acceptance. Login success is not that enforcement proof.
The workplan retains exact native credential/approval/audit receipts through
GLAS-WP-0015, protected runtime placement, natural worker claim/heartbeat/close,
@ -48,3 +57,43 @@ has no consistency failures; historical warnings remain. Reuse-surface has three
archived null-ID failures (REUSE-WP-0017/0018/0019), separate from the successfully
registered new plan. HFACT-WP-0001-T02 retains their disposition; CUST-IN-0017
continues to own the pre-existing Custodian historical consistency issues.
## Efficiency gains already evidenced
| Change | Evidence and practical effect |
| --- | --- |
| Reuse before implementation | vergabe-teilnahme consumed the existing hosted registry with 65 capabilities and a dated, hashed snapshot. Source freshness and completeness checks prevent an unavailable registry from silently authorizing new work. No second registry was built. |
| Repeatable release checks | 208 provider tests and 82 customer application tests passed; application tests and migration checks now run before customer image publication. Two actual release defects were found and fixed: missing source templates and missing runner BuildKit support. |
| Less duplicate CI | Application checks run for PR updates and main, without an additional feature-branch push trigger duplicating the PR run. Main release validation remains explicit. |
| Less review-queue noise | Four completed tasks with explicit published `needs_human: false` were corrected in the Hub. The flagged queue fell from 19 to 15 (21%); the two nonterminal flagged tasks were unchanged. A fresh query after login still returned 15: 12 done, 1 cancelled, 1 todo, 1 wait. The other 13 terminal flags need source/residual review. |
| Clearer coordination | One published project owns factory integration; implementation stays in the two product repositories. Ten dependency-return records name the supplying task, source revision, next action and required evidence. Existing GLAS-WP-0015 coordination is reused. |
| Fewer stale open records | REUSE-WP-0022 is finished, customer source/release tasks are done, and both PRs now show merged. Customer deployment remains an explicit waiting task instead of being mixed with completed source work. |
These are observed improvements in discovery, validation and coordination.
Hours saved, monetary savings and autonomous throughput have not been measured.
The original 94-plan/264-task portfolio baseline is dated; no later fleet-wide
count is implied by these changes. The governed-attempt ledger remains empty and
the fourteen-day acceptance window has not started.
## Next execution sequence
1. Prepare and accept the exact G0 tenant/actor/project/profile, repository
grants, operating owners and enforceable time/spend contract. Project
adoption and publication are complete; unattended operating admission is open.
2. Close the existing GLAS-WP-0015 chain: KeyCape service-client custody and
registration, audit sender admission, deployed approval claim/consume and
native Secrets Engine delivery. Obtain current policy allow/deny evidence.
Protected runtime installation can advance alongside that chain.
3. Prove the current Railiance worker's natural claim, immediate heartbeat,
execution and accepted terminal close, including lease-loss refusal and
durable close replay. Retain one integrated delivery item plus two prerequisite
closures as the project's WIP limit.
4. Use that worker for a useful accepted change; admit and deploy the tested
vergabe-teilnahme image through its exact tenant, database, storage, access
and rollback contract (VERGABE-WP-0018-T03). Keep Railiance Fabric in its
existing medium-term placement work until its additional utility is needed.
5. Complete recovery/limit drills, then measure fourteen days: at least five
useful accepted changes across two repositories, at least 80% without
operator repair, median handling at most 30 minutes per accepted change and
routine operation at most 30 minutes per day. Preserve every attempt,
failure, retry, cost and operator intervention.

View file

@ -1,38 +0,0 @@
# Proposed project agent instructions
This packet is staged in the Custodian. Until project adoption, changes here
prepare the proposal only; do not register it as an unrelated Custodian workplan
or claim it is an active project.
After adoption, repo slug is `prj-helixforge-factory`; proposed workplan prefix
`HFACT-WP-` must be checked with the authoritative registrar before first index.
Orient with `GOAL.md`, `SCOPE.md`, genesis, source workplans and `dependency-map.md`.
Read the generated `.custodian-brief.md` when available. Use State Hub HTTP at
`http://127.0.0.1:8000` locally, `http://10.43.68.154:8000` on railiance01;
use an enabled edge relay only with its explicit configuration. Direct requests
carry `X-StateHub-Component: prj-helixforge-factory` and canonical `/workplans/`
and `/tasks/?workplan_id=...` routes.
Check this project's unread inbox and mark reviewed messages read. Do not send
messages to other owners without user authorization. Refer to existing
GLAS-WP-0015 threads and receipts to avoid repeated requests.
Implementation remains in participating repos under their instructions.
Advance only within existing grants and session authorization. Do not infer
permission for secrets, publication, deployment or paid execution from this
proposed programme. Before any credential need, run `warden route find` and
`warden route show`; ops-warden issues SSH certificates, while OpenBao/platform,
KeyCape and the relevant policy/approval owners retain other responsibilities.
Never place secret values in work records or logs.
Update source task statuses as work progresses. Record significant decisions
as source-backed decision records and the supported Hub decision projection.
At session close log a concise `POST /progress/` event and run
`statehub fix-consistency` after workplan edits. If the CLI/API is unavailable,
retain evidence of the failed sync and request operator recovery; do not claim
queued or failed writes are central success.
Before finishing or archiving, create live owner records for actionable
residuals, name their IDs in the completion record, and verify synchronization.
No production implementation belongs in this coordination repository.

View file

@ -1,90 +0,0 @@
---
repo: prj-helixforge-factory
repo_flavor: project
project_status: draft
started: "2026-09-08"
reviewed: "2026-09-08"
---
# Goal
## Outcome
Establish an internal HelixForge software factory that repeatedly turns bounded
Coulomb demands into validated, reusable capabilities, with agent execution and
at least one resulting software service running on Railiance. Humans retain
capability acceptance and the release decisions required by existing policy.
The first factory serves one explicitly admitted internal tenant. A second
repository demonstrates reuse; it does not introduce a second tenant. The
existing Glas proof's `tenant:platform`, `actor: agt`, `project: glas-local-proof`
remain scoped to that proof. The pilot gets its own reviewed binding where
needed; no identity or authorization transfers merely by copying a profile.
## Invariants
1. Source files and their published revisions own work. Existing owners retain
code, deployment and policy authority. The project links their records.
2. An admitted task states capability intent, scope, tests, allowed repository
effects, actor/tenant, runtime/profile pins, timeout, spend envelope and review
disposition before execution. Repo instructions/prompts cannot expand grants.
3. Repository-changing agents execute inside the demonstrated isolation and
repository transaction boundary. The first pilot permits one active mutator.
4. Source tests, deployed operation and accepted usefulness are separate claims.
A local stub, startup probe or historical profile does not prove the current
production route. The localhost proof is not Railiance admission.
5. Credential custody, policy decisions, approval consumption, audit delivery
and publication use existing admitted owner paths. No secret material goes
into Git, messages or progress evidence.
6. Build/test workers and release authority remain separated by their existing
credential and trust boundaries. An accepted commit grants no implicit push,
package publication or deployment authority.
7. Preserve working coordination and recovery services during migrations.
Source deletion, public exposure changes and broader tenant admission retain
their existing independent gates.
8. No new permanent service or generic framework is established in this project
or in the retiring State Hub. Product implementations remain with owners.
## Success gates
Targets below are proposed for T01 acceptance; they are not measured baselines
or spending approval. Any revision must be decided before its observation window.
| Gate | Required evidence |
| --- | --- |
| G0 — A bounded operating contract | Named capability acceptance owner, operational owner, internal tenant, pilot service, two participating repos, exact grants, budget, allowed release lane, test oracle and source-backed owner dependencies. All pilot dependencies have an owner, return evidence and next review/action. |
| G1 — Real governed model execution | Current GLAS-WP-0012 candidate produces its real artifact with protected owner credentials, correct attribution, declared egress, denial and teardown proof. Only the demonstrated profile becomes ready. |
| G2 — Railiance execution | Admitted installed runtime and worker on railiance01 execute through Activity Core with a current versioned profile and repository grant. Natural claim/heartbeat/close, accepted changed paths, clean source baseline, durable evidence/replay and controlled lease-loss refusal are demonstrated. |
| G3 — Useful delivered capability | One bounded real service/library change has intent and contract, relevant OAS structural/semantic checks, independent tests, accepted review, immutable Forgejo artifact identity, authorized Railiance release, service smoke and rollback proof. A second repo consumes or reuses the capability. Documentation-only or smoke-only output is insufficient for this gate. |
| G4 — Controlled operation | Budget/timeout and grant violations stop work; worker crash, Hub/queue outage, terminal-close replay and sandbox cleanup are exercised without duplicate accepted mutations. Backup and recovery evidence covers the actual worker state, source/artifact path and pilot workload. No stale evidence claim is presented as current. |
| G5 — Repeatability and founder load | Over 14 consecutive calendar days, at least five useful accepted changes across two repos, at least one service release, ≥80% of admitted delivery attempts accepted without operator repair, median human handling ≤30 minutes per accepted change, routine operation ≤30 minutes/day. All attempts, retries, rejects, costs and interventions retained. Report setup and attended security/recovery time separately and in total founder load. |
Suggested initial hard limits for T01: one mutating run; 30 minutes wall clock
per run; EUR 5 equivalent maximum model spend per run; EUR 25 equivalent per
day and EUR 150 total pilot model spend. The admitted provider/runtime must be
able to enforce the selected envelope; token counts alone do not establish a
currency cap. If enforcement is unavailable, adopt and prove an enforceable
conservative limit before admitting paid execution. Infrastructure costs are
measured separately; adding capacity requires an owner decision.
G5 counts one admitted capability-change request as one delivery attempt;
retries consume the same attempt's cost and handling time and must not inflate
the denominator. Cancellation/refusal after admission remains an unsuccessful
attempt with a reason. A human's planned review/release approval is expected;
manual repair, re-prompting to rescue output, or bypassing a failed gate prevents
that attempt from counting as accepted without repair. Five clean trivial runs
cannot stand in for five useful changes with predeclared acceptance criteria.
## Project retirement
Archive only when G0G5 have accepted evidence, the permanent owners have
accepted operating responsibility, and every actionable residual has a live
owner record outside the closing narrative. Update HelixForge's durable
capability/operating documentation and owning repos' runbooks; retain links to
merged/published revisions, deployment evidence, decisions and residual IDs in
a completion record. Do not archive or shut down any participating service as
part of retiring this coordination repository.
External tenant onboarding, autonomous production release, additional rein/model
families, full HA and commercial-company success are outside these gates.
`UPC-WP-0003` retains its separate ≥30-day company-load objective.

View file

@ -4,12 +4,13 @@ The user authorized implementation on 2026-09-08. The dedicated Git project now
lives at [/home/worsch/prj-helixforge-factory](/home/worsch/prj-helixforge-factory/README.md).
Its [current workplan](/home/worsch/prj-helixforge-factory/workplans/HFACT-WP-0001-establish-internal-factory.md)
and [delivery evidence](/home/worsch/prj-helixforge-factory/evidence/2026-09-08-delivery.md)
own ongoing coordination. Forgejo repository creation and central registration
await the routed OpenBao caller login.
own ongoing coordination. The private
[canonical Forgejo project](https://forgejo.coulomb.social/coulomb/prj-helixforge-factory)
is published and registered with State Hub under the helix-forge topic.
The other files in this directory are the **frozen original proposal**, retained
until canonical project publication succeeds. Do not update them as a second
programme. HFACT-WP-0001 in the dedicated project is the current source.
The original proposal files were removed after canonical publication; their
history remains in Git. HFACT-WP-0001 in the dedicated project is the current
source. This directory is only a pointer.
Internal capability: **reuse-surface** (REUSE-WP-0022, completed and published).
Customer service/UI product: **vergabe-teilnahme** (VERGABE-WP-0018; source and

View file

@ -1,55 +0,0 @@
# Scope
## Project authority
This proposed project owns factory success gates, sequencing, integration
acceptance, the dependency map and consolidated evidence. The Custodian
coordinates; `helix-forge` owns the durable product and capability acceptance.
Accountable people/agents and allocated capacity are confirmed in T01, not
assigned to other repositories by writing this proposal.
## Participating repositories
| Owner | Responsibility retained |
| --- | --- |
| helix-forge | Human intent, reusable capability contract, architecture fit and value acceptance |
| the-custodian | Governance, source-record conventions, portfolio assessment and project oversight |
| repo-manager / current State Hub projection | Canonical repository/work identity, files, registration and derived coordination views |
| activity-core | Admitted definitions, ops_run queue, worker identity and leases |
| rein-aharness | Claim worker, repository transaction/acceptance, terminal-close delivery |
| glas-harness | Versioned profile selection and real-profile acceptance; existing owner coordination GLAS-WP-0015 |
| sand-boxer | Runtime provisioning, isolated execution, constrained egress, private state and teardown |
| key-cape, flex-auth (access-engine), approval-engine, secrets-engine, audit-core | Identity, decision, approval, native credential delivery and evidence contracts respectively |
| railiance-platform | Admitted credential custody and shared-service dependencies |
| railiance-master and concrete rail/rapp/reef owners | Execution/workload admission, placement, package deployment and recoverability |
| railiance-forge / railiance-enablement | Runner/artifact operations and existing reusable CI/release paths |
| Chosen source and workload repositories | Pilot code, meaningful tests, release, consumer integration and operating evidence |
| kaizen-agentic / coulomb-loop | Existing improvement definitions and demand/feedback context where appropriate |
| prj-unattended-progress-company | Receives relevant factory-load evidence; retains independent company/revenue gates |
Names above are current checkout identities. This project does not rename
`flex-auth`, replace `coordination-engine`, or create a parallel ownership model.
## In scope
- One internal factory lane and its exact prerequisite closures.
- Two-repository capability delivery and reuse, including a Railiance service release.
- Minimal work-record hygiene needed for dispatch and accountability.
- Fresh operating evidence, recovery, cost and founder-load measurement.
## Out of scope
- Production code or credential material in this repository.
- Clearing the entire estate backlog as an entry gate.
- Full HA, wholesale hub retirement, broad renaming or global schema redesign.
- External tenant onboarding, community campaign expansion or revenue operations.
- Any implicit authorization from a proposed workplan or from another owner's
historic session authorization.
## Work-record rule
The project workplan contains acceptance/handoff tasks and references existing
owner work. It does not duplicate child implementation task lists. A newly
discovered implementation gap receives an owning source-backed task or intake
before it is counted as assigned. Dependencies are current records, not just
messages; an acknowledgement is not a completion receipt.

View file

@ -1,44 +0,0 @@
# Proposed factory dependency map
Date: 2026-09-08. This is a source proposal for acceptance sequencing. Actual
owner task statuses remain in their repositories and the derived Hub.
Stage/receipt is not inferred from the existence of a message or package.
| Project gate | Supplying record(s) | Current return / next evidence | Factory effect |
| --- | --- | --- | --- |
| T03 identity admission | KEY-WP-0013-T02; platform custody record to be resolved through the existing Glas request | Two service clients and protected consumer delivery remain unproven. Tenant choice is resolved. Identify the exact admitted custody record; do not substitute a generic database lane. | Blocks approval service acceptance |
| T03 audit admission | AUDIT-WP-0009-T03/T09; APPROVAL-WP-0002-T01 | Evidence kind is implemented; sender scope/secret policy and credential admission need owner returns. T09 is already high priority/progress. | Blocks approval startup |
| T03 approval | APPROVAL-WP-0002-T01/T03/T05 | Published image is pinned. Latest recorded production review found no deployed service. Need identity, audit, rollout, claim/consume and restart/restore receipts. | Blocks native credential action |
| T03 credentials | SECRETS-WP-0009-T03; SECRETS-WP-0007-T04 and SECRETS-WP-0008-T02/T06 as applicable | September 7 decision-path proof resolved digest normalization; remaining external dependency is the approval claim endpoint. Need real scoped activation/delivery evidence. | Blocks paid real-model proof |
| T04 runtime/profile | SAND-WP-0015-T04; GLAS-WP-0012-T02T06 | Pinned startup candidate and synthetic transport exist. Need protected placement, native credentials, real model, exact binding, teardown and negative evidence. | Blocks current local profile readiness |
| T05 worker | REINAH-WP-0003-T05/T06; ACTIVITY-WP-0032-T05, ACTIVITY-WP-0035-T08, ACTIVITY-WP-0036-T04 | Queue and repository boundaries exist; current source heartbeat correction needs deployed natural trace and controlled late-close evidence. | Blocks governed production lane |
| T05 placement | railiance-master admission contracts; concrete runtime/reef owner record selected in T01 | Existing user-service worker on railiance01 is the starting point. Require explicit ownership and admitted profile/credential/recovery contract for this host. | Blocks claim that the local proof operates on Railiance |
| T06 delivery | Chosen source repo record and consuming repo record selected in T01; existing forge/enablement contracts | No existing end-to-end HelixForge factory acceptance owner was found. Adopt bounded useful-change/release work, with independent tests and actual consumer. | Blocks useful factory result |
| T07 recovery | REINAH-WP-0003-T05; RPF-WP-0038-T04 and actual pilot workload/package owner | Restore demonstrations exist; scheduled caller, inventory/retention and current worker/pilot recovery still need exact coverage. | Bounds operating readiness |
| T08 value/load | HelixForge acceptance; UPC-WP-0003 as receiving context | No current fourteen-day factory delivery/load evidence was established by this assessment. | Blocks repeatability claim |
GLAS-WP-0015 is the existing owner-handoff coordinator for T03/T04. The factory
project consumes its results and adds the software-delivery/operating acceptance
that it does not own. Do not create parallel identity/audit/runtime coordination.
When adopting the map, record canonical task IDs, accountable contact, required
receipt, source revision and next review condition in each owner handoff. Set
structured dependency fields/edges only with the supported source/projection
semantics; do not invent an API convention or overwrite source UUIDs.
Separate lanes retained outside the factory critical path:
- `NK-WP-0033/0034` and `RPF-WP-0027`: known incident/verification closure.
Resolve impact on the proposed privilege boundary before expansion; ordinary
incident priority is not displaced by factory WIP limits.
- `RMASTER-WP-0020-T08`: retained CoulombCore cleanup, with recovery and explicit
destructive-approval gates. It is not a prerequisite to installing the factory.
- `RMASTER-WP-0020-T09`, `RPF-WP-0025`, `RAPP-OPENBAO-WP-0002`, `NK-WP-0032`:
public listener/private login migration. Preserve existing requirements; do not
merge it into native service delivery unless an owner demonstrates dependency.
- `STATE-WP-0079`, `CORE-WP-0010`, relevant `HUB-*` and `RAPPCOREHUB-*` work:
staged hub migration, with receiver and quiet-window gates. Keep existing
coordination operational while the factory consumes stable contracts.
- `CUST-WP-0038`, ThreePhoenix HA, enterprise federation, broad repository
renames, extra rein/model profiles and publication campaigns: separate benefit
decisions; no blanket factory dependency.

View file

@ -1,24 +0,0 @@
# Genesis — 2026-09-08
The operator asked the Custodian to assess the Coulomb, HelixForge and Railiance
ecosystem, especially blocked/open workplan load, and provide a plan for an
agentic software factory.
The pre-proposal Hub snapshot contained 98 open workplan rows. Four explicitly
retired identities reduced the canonical baseline to 94 plans, 27 blocked,
containing 264 open tasks. All 94 matched local source workplan status. The
direct execution cohort contained nine plans and fifteen open tasks; much of
the remaining dependency load concerned identity/custody/audit and Railiance.
The proposal consumes existing GLAS-WP-0015 coordination and implementation
work. Its distinct deliverable is a useful, repeatable capability delivery loop
on Railiance with operating evidence and measurable human load. This avoids
mistaking either source-complete components or the HelixForge publication
programme for a functioning software factory.
Source assessment:
`the-custodian/docs/assessments/2026-09-08-helixforge-factory.md`.
Captured baseline and provenance live beside that report. This packet remains a
draft in the Custodian until HFACT-WP-0001-T01 establishes its dedicated project
source and verifies the prefix with the registrar. No factory work was activated
and no owner was contacted by this assessment.

View file

@ -1,320 +0,0 @@
---
id: HFACT-WP-0001
type: workplan
title: "Establish the internal HelixForge software factory on Railiance"
domain: infotech
repo: prj-helixforge-factory
status: proposed
owner: the-custodian
topic_slug: infotech
created: "2026-09-08"
updated: "2026-09-08"
related:
- GLAS-WP-0012
- GLAS-WP-0015
- SAND-WP-0015
- REINAH-WP-0003
- ACTIVITY-WP-0032
- ACTIVITY-WP-0035
- ACTIVITY-WP-0036
- KEY-WP-0013
- APPROVAL-WP-0002
- SECRETS-WP-0009
- AUDIT-WP-0009
- RPF-WP-0038
- UPC-WP-0003
---
# Establish the internal factory
**Proposal only.** Prepared in the Custodian on 2026-09-08; not yet a registered
project or an execution authorization. Prefix allocation and adoption are T01.
The assessment is complete; the eight tasks below are future factory work.
The [goal](../GOAL.md) defines success. The [dependency map](../dependency-map.md)
identifies existing work to consume. This workplan owns integration acceptance
and handoff; it neither replaces nor copies child implementation plans.
Recommended sequence:
```text
T01 → T02
T01 → T03 ─┐
T01 → T04 ─┴→ T05 → T06 → T08
T01 → T07 ───────────────→ T08
```
T04 runtime preparation can advance before T03 finishes; real model acceptance
cannot. T07 recovery preparation can advance early; its live proof depends on
the deployed T05/T06 configuration. T02 is complete before regular scheduling.
## T01 — Adopt the project and select the first valuable capability
```task
id: HFACT-WP-0001-T01
status: todo
priority: high
assignee: the-custodian
```
Coordinator: the-custodian. Acceptance: the HelixForge product owner and the
operator responsible for Railiance. Estimated effort: 12 engineering days.
Review this packet against current published owner revisions, confirm a unique
prefix, establish the dedicated project repo under ADR-005, publish/register it
through the authorized path, and sync its source records. Retain a pointer from
the Custodian rather than two editable copies of the programme.
Select a **small, useful software service or library capability** with a
repeatable independent test, a known release path and a second consuming repo.
Prefer an existing internal workload and existing CI. Avoid beginning with
identity, credential or production-control code as the agent's pilot target.
HelixForge owns the demand/contract; the actual source and consumer owners
accept the bounded implementation records. Record why this capability matters
and the expected reuse before selecting a convenient task from the queue.
Confirm one internal tenant, exact actor/project/repository/profile tuple,
human acceptance and operating owners, allowed branch/push/release effects,
the proposed G0G5 targets, capacity allocation and an enforceable spend/time
envelope. Assess the outstanding NK-WP-0033/0034 verification incident against
the proposed privilege path; require applicable closure or an explicit scoped
owner disposition before expanding it.
Done when G0 is accepted in a source-backed decision, the project and concrete
pilot work are registered, and a future operator can determine who may run,
review and release it without reconstructing this assessment. Recording a
proposal does not grant execution or spending rights.
## T02 — Make the selected backlog actionable
```task
id: HFACT-WP-0001-T02
status: wait
priority: high
assignee: the-custodian
depends_on: [HFACT-WP-0001-T01]
blocking_reason: "Await project adoption and selection of the pilot dependency chain in T01."
```
Dependency: T01. Coordinator: the-custodian with source/projection owners.
Estimated effort: 1 engineering day, plus genuinely necessary projection fixes.
Use the 94-plan/264-task baseline as the dated starting point. Review the four
retired open identities, seventeen terminal human flags, terminal-parent tasks
and the unmatched local KG-WP-0005 through their existing owning paths. Do not
bulk-delete history or set statuses to improve a dashboard total.
Existing CUST-IN-0017 owns the thirteen Custodian historical consistency
failures and prefix conflict; reuse it rather than duplicate its triage.
For the pilot chain, ensure every waiting task has a supplying task/record,
accountable contact, precise return evidence, actual next action and review
condition; set human-needed only for real human dependencies. Capture published
source revision and freshness. Check projections without making the retiring
State Hub a new permanent owner. Refer off-path cleanup to existing owners;
it does not hold up the factory after its own records are trustworthy.
Adopt a factory WIP limit of one integrated delivery item plus two prerequisite
closures, with a named incident exception. Any wider portfolio reprioritization
remains a recommendation until owners accept it.
Done when the selected chain has complete actionable dependencies, a short
prepared human queue and a repeatable next-pick view; retired/terminal evidence
is excluded from current demand without losing residual ownership.
## T03 — Accept the existing identity, audit, approval and credential chain
```task
id: HFACT-WP-0001-T03
status: wait
priority: high
assignee: the-custodian
depends_on: [HFACT-WP-0001-T01]
blocking_reason: "Await T01 adoption; live returns are owed by KEY-WP-0013-T02, AUDIT-WP-0009-T09, APPROVAL-WP-0002 and SECRETS-WP-0009-T03."
```
Dependency: T01. Existing coordinator: GLAS-WP-0015. Supplying owners:
KEY-WP-0013-T02, AUDIT-WP-0009-T09, APPROVAL-WP-0002-T01/T03/T05,
SECRETS-WP-0009-T03 and exact linked custody/authorization records.
Rough shared effort with T04: 48 engineering days plus owner waiting.
Consume the existing seven-owner handoff packet. Identify the exact custody
record for both service clients and the audit sender; first-provision authority
cannot depend on the unprovisioned service. Keep the accepted tenant spelling
and proved policy/digest behavior pinned. Distinguish service-only admission
from a human callback requirement with owner evidence, not assumption.
Accept the real identity/custody, audit admission, deployed approval claim and
consume, and native credential delivery receipts from the owners. Evidence must
include successful binding and relevant wrong-tenant/scope, denied/reused action
and out-of-scope read rejection, plus revocation/expiry and non-secret audit
references. An image digest or a delivered message alone closes no live gate.
Done when native credential delivery for the exact proof tuple is verified
through the current owner contracts, with no bypass and no outstanding startup
prerequisite. Record later audit freshness/detection obligations separately;
admission success is not a completeness or tamper-evidence claim.
## T04 — Accept the real profile and prepare its Railiance placement
```task
id: HFACT-WP-0001-T04
status: wait
priority: high
assignee: the-custodian
depends_on: [HFACT-WP-0001-T01]
blocking_reason: "Await T01 adoption for preparation; the real-model portion also requires T03 native credential acceptance."
```
Dependency: T01; real model execution also requires T03. Supplying owners:
SAND-WP-0015-T04, GLAS-WP-0012-T02T06 and concrete Railiance admission owners.
Accept protected installation of the already pinned runtime candidate and its
owner configuration. Reuse the existing real-rein acceptance contract: exact
profile/actor/project, provider path, model-created artifact, declared effects,
denial, cleanup and rollback to the retained blocked selection on failure.
Review changed pins as a new compatibility set. Do not re-prove resolved
candidate startup unless the artifact, environment or contract changed.
Then define the production-specific railiance01 placement, runtime owner,
profile/grant, credential path, service restart behavior and recovery inventory.
Start from the evidenced host user-service worker. Satisfy the owning
Railiance rail/rapp/reef admission contract; if the current host service needs
an ownership/admission record, create it in that owner rather than assuming
Kubernetes packaging is the production process. A localhost-only profile needs
explicit target-specific acceptance.
Done when G1 passes, the local result has acknowledged owner receipts, and the
Railiance configuration is concrete and admissible for T05. G2 is not closed
by this task's local proof.
## T05 — Close the current worker and queue loop on Railiance
```task
id: HFACT-WP-0001-T05
status: wait
priority: high
assignee: the-custodian
depends_on: [HFACT-WP-0001-T02, HFACT-WP-0001-T03, HFACT-WP-0001-T04]
blocking_reason: "Await actionable admission records, owner credential chain and accepted profile/placement from T02-T04."
```
Dependencies: T02T04. Supplying work: REINAH-WP-0003-T05/T06,
ACTIVITY-WP-0032-T05, ACTIVITY-WP-0035-T08, ACTIVITY-WP-0036-T04.
Estimated effort: 23 engineering days excluding upstream changes.
Join the admitted task to the existing ActivityDefinition / ops_run intake and
current versioned profile/repository-grant contract. If a small intake adapter
is missing, record it in the proper owner and implement only that bounded gap;
the retired workplan launch endpoint is not a shortcut. Keep the definition
disabled until the owner's live readiness gates are accepted.
Accept the deployed pin and a natural claim → immediate heartbeat → execution
→ accepted terminal close trace. Require a controlled late-close/lease-loss
negative case, correct changed paths and commit ancestry, external metrics,
clean source state and sandbox teardown. Exercise the current durable close
outbox without duplicate mutation. A healthy API poll is not an executed task.
Done when G2 passes under the exact Railiance artifact/configuration and the
Activity Core, rein and Glas owners accept the same run evidence. Return it to
their existing tasks rather than leaving source-complete/live-wait tails open.
## T06 — Deliver and reuse a useful capability through the existing release path
```task
id: HFACT-WP-0001-T06
status: wait
priority: high
assignee: the-custodian
depends_on: [HFACT-WP-0001-T05]
blocking_reason: "Await the current governed Railiance worker proof in T05."
```
Dependency: T05; capability and owning implementation records selected in T01.
Acceptance: HelixForge product owner, source/consumer maintainers and workload
operator. Estimated effort: 35 engineering days.
Drive the selected human demand through capability intent, contract and relevant
OAS structural/semantic validation. The bounded agent change must satisfy an
independent predeclared test oracle and normal maintainer review. Record reuse
search and why an existing capability is extended or a new one is justified.
Use the source repo's current Forgejo workflow, or adopt the existing
railiance-enablement template where needed. Join source commit, CI evidence,
artifact digest/package version and release configuration revision. Apply only
the separately authorized release through the actual workload owner. Validate
the service behavior on Railiance and exercise the agreed rollback. Have the
second repo consume/reuse the capability and retain its acceptance evidence.
Done when G3 passes and the evidence chain links demand → contract → task/run
→ accepted change → CI → immutable artifact → authorized release → operational
result → reuse. Update the durable HelixForge capability/operating entry. A
generated document, empty commit or deterministic smoke alone is insufficient.
## T07 — Establish recoverable, measurable operation
```task
id: HFACT-WP-0001-T07
status: wait
priority: high
assignee: the-custodian
depends_on: [HFACT-WP-0001-T01]
blocking_reason: "Await T01 adoption for design; final live recovery proof requires the deployed T05/T06 configuration."
```
Dependency: T01 for design; T05/T06 for final deployed proof. Supplying owners:
REINAH-WP-0003-T05, the selected workload/package/reef owner and applicable
RPF-WP-0038-T04 recovery/custody coverage. Estimated effort: 23 engineering days.
Consolidate existing receipts for timeout/spend enforcement, worker death,
lease conflict, pending terminal-close replay, credential failure and sandbox
cleanup. Use disposable work and the current deployed configuration. Include
proof that a queue/Hub outage cannot produce duplicate accepted changes and
that restarting restores the right pending evidence. Bind backups and recovery
inventory to the actual host service, forge/artifacts and pilot workload.
Record per attempt queue wait, execution time, accepted/rejected result,
retries, failure class, human handling, model cost, relevant infrastructure cost
and release result. Include refused/admitted failures in denominators. Use
existing evidence and cost owners; start with a small generated scorecard if
automated aggregation is absent. Monitoring must distinguish a stale receipt
from a healthy current worker and alert the responsible operator through an
already authorized channel.
Done when G4 passes, rollback/recovery is executable by the named owner, costs
are bounded, and the measurement process can sustain G5 without reconstructing
individual agent sessions. Required audit claims are either freshly proved or
accurately degraded; unresolved detection work retains live owner records.
## T08 — Accept repeated delivery, transfer ownership and retire the project
```task
id: HFACT-WP-0001-T08
status: wait
priority: high
assignee: the-custodian
depends_on: [HFACT-WP-0001-T06, HFACT-WP-0001-T07]
blocking_reason: "Await useful delivery and operational controls from T06/T07, then the complete fourteen-day observation window."
```
Dependencies: T06 and T07. Reviewers: HelixForge acceptance owner, Railiance
operating owner and the-custodian. Effort: 12 engineering days for review and
handoff plus fourteen calendar days of observation.
Run the accepted fourteen-day G5 window. Admit at least five useful changes
across two repositories and preserve all attempts, costs, failure/repair events
and human time. If a target is missed, classify the limiting defect, create or
update its owner work record, and repeat the affected acceptance window after
correction. Do not redefine the metrics retrospectively to mark the project done.
Publish the operating runbook and capability/contract references in permanent
homes, including correction of HelixForge's stale Inter-Hub/nested State Hub
orientation where relevant. Hand the measured factory evidence to the company
project's UPC-WP-0003 acceptance context without claiming its longer window or
commercial goals complete. Additional tenants, model families, HA or unattended
release receive new benefit/authority decisions, not automatic activation.
Done when G0G5 are accepted, durable owners acknowledge handoff, all actionable
residuals are live records, and the completion record lists owning repos,
merged/published revisions, deployment/rollback evidence and residual IDs.
Only then finish this workplan, synchronize it and archive the project as
read-only provenance. Participant services continue under their owners.