feat: working TOM agent
This commit is contained in:
@@ -0,0 +1,27 @@
|
||||
---
|
||||
status: proposed
|
||||
---
|
||||
|
||||
# The Target operating model is an explicit graph, and the solution is never changed here
|
||||
|
||||
An operating model written as prose cannot be checked: nobody can say whether every part of the solution has someone to run it, or whether anyone is answerable when it fails at night. We decide that the **Target operating model** is one explicit graph, held as data: Parties, Roles, the Responsibilities between Roles and Capabilities, and the Checks each must meet. It is established and approved before anything is written from it, and what is written from it is a view of it.
|
||||
|
||||
This follows the solution architecture agent's ADR-0005, applied to a different subject.
|
||||
|
||||
## What follows
|
||||
|
||||
- **Completeness is a submission rule.** The graph is refused while a Capability has no single accountable Role, or nobody responsible at the hours it is needed, or while something the customer said must be kept true is carried by no Check.
|
||||
- **Views are written by code.** The responsibility matrix, the organisation chart, and staffing and cost are rendered or computed from the graph. No model draws or counts.
|
||||
- **Validation is a verdict per Check,** with scenario walk-throughs in place of a dry run: events generated from the graph and traced through it by code.
|
||||
- **The solution is never changed.** The Items to be operated and the Conditions on them are the customer's description, or the digest of their architecture. Something operations needs that the solution does not provide is a **finding**: it is recorded, a Check that depends on it waits on it, and it goes back to whoever owns the design. Where the spec was digested from a solution architecture, the findings are drafted as a Change request to that agent.
|
||||
|
||||
## Considered Options
|
||||
|
||||
- **Repair gaps in the solution while designing its operations:** rejected. Two agents would then hold different versions of one design, and the architecture's own Checks would no longer describe what is run.
|
||||
- **Refuse to proceed while a gap exists:** rejected. Most operating models can be designed around a known, stated gap, and the customer is better served by a model with its findings listed than by none.
|
||||
- **Hold Capabilities inside the graph:** rejected for now. The capability model is its own Deliverable, so that what must be done is agreed apart from who does it, and a failed Check can reopen one without the other.
|
||||
|
||||
## Consequences
|
||||
|
||||
- A finding is neither a failed Check nor a gap in the model. Nothing is revised for it, and the Execution completes with it listed.
|
||||
- More is decided before detail is written. A fault in the graph costs a Revision of it and of validation.
|
||||
26
docs/adr/0002-an-execution-starts-from-an-operating-spec.md
Normal file
26
docs/adr/0002-an-execution-starts-from-an-operating-spec.md
Normal file
@@ -0,0 +1,26 @@
|
||||
---
|
||||
status: proposed
|
||||
---
|
||||
|
||||
# An Execution starts from an operating spec
|
||||
|
||||
This agent was first conceived as reading the solution architecture agent's files. That would make it usable only after that agent, and tie it to the shape of files that are still changing. We decide that an Execution starts from an **operating spec**: a structured file read against a catalogue in code, complete or partly filled, completed in conversation and approved before anything is derived from it. It is the engine's spec mechanism, with this agent's catalogue.
|
||||
|
||||
The spec carries the solution to be operated inline: its Items, the Conditions on them, and what is known of its build and its risks. It does not point at other files.
|
||||
|
||||
## What follows
|
||||
|
||||
- **Anything supplied is digested into a spec first.** The solution architecture agent's files are one Source, read by a code adapter that keeps every Element and Check with its id. A document is another, drafted by a model with each record quoting where it came from. A sentence is a third.
|
||||
- **Everything downstream reads only the approved spec.** No Specialist opens an architecture file. Completeness is checked against the spec's Items and Conditions.
|
||||
- **A digest error is caught at the scoping Approval,** where the customer sees what was digested, and not later.
|
||||
- **The agent can design an operating model for a solution the architecture agent never saw.**
|
||||
|
||||
## Considered Options
|
||||
|
||||
- **Read the architecture agent's files directly:** rejected by the owner. It is the only way in, and it copies that agent's readers.
|
||||
- **Keep the inventory in separate files the spec points to:** rejected by the owner. It brings back the dependency the spec removes.
|
||||
|
||||
## Consequences
|
||||
|
||||
- The spec is larger than the architecture agent's scope spec, and the engine gives the agreed spec to every Specialist. The inventory should travel as a data file to the Specialists that use it; until the engine's digest step is built it travels with the rest.
|
||||
- The digest step is not built. Today an Execution starts from an operating spec that is already written.
|
||||
@@ -0,0 +1,29 @@
|
||||
---
|
||||
status: proposed
|
||||
---
|
||||
|
||||
# Cost is computed from the service provider's rate card
|
||||
|
||||
An operating model that names Roles invites the question of what they cost. A model asked for a cost will give one, and it will be a guess. We decide that staffing and cost are computed by code from the graph and one file, the **rate card**, and that no agent writes a number of people, a rate or a cost.
|
||||
|
||||
## Decisions
|
||||
|
||||
Taken by the owner on 2026-10-13:
|
||||
|
||||
1. **The rate card is a Skill, and it is the actual card.** Its prices are the service provider's: set by the owner, changed in review like any other Skill. A caller cannot name another card and an operating spec cannot state a rate.
|
||||
2. **Rates are by country and level.** A Role has both. The Specialist chooses them and says why; it never chooses to reach a cost.
|
||||
3. **The operating spec states the currency.** The card is written in one currency and carries fixed conversion rates with their date. No exchange rate is looked up. A currency the card cannot convert to refuses the start.
|
||||
4. **The customer's own staff are counted and never priced,** whatever the card says.
|
||||
|
||||
## How it is computed
|
||||
|
||||
- **People** are the effort of the Capabilities a Role is responsible for, over the productive hours of its country. Where the Role covers hours wider than a working day, never fewer than the smallest team that can cover them: a rota is sized by the hours it fills.
|
||||
- **Cost** is people at the day rate for the Role's country and level, with the uplift for its hours, converted into the spec's currency.
|
||||
- A cost over a stated budget is shown as over. It is not refused.
|
||||
|
||||
## Consequences
|
||||
|
||||
- Every cost is traceable: the card's version is recorded with the Execution and printed beside every total.
|
||||
- The card in the repository holds placeholder figures until the owner sets them. Its version says `draft`, and every cost worked from it says it is an illustration.
|
||||
- The owner's real prices will be readable by whoever can read the repository.
|
||||
- The effort a Specialist states for a Capability is the one estimate in the chain. The validator judges whether it is plausible; nothing computes it.
|
||||
Reference in New Issue
Block a user