Skip to content

state-machine.mdx

Latest commit

 

History

History
228 lines (189 loc) · 16.1 KB

state-machine.mdx

File metadata and controls

228 lines (189 loc) · 16.1 KB
Raw
title State machine — record lifecycle spec
navTitle State Machine (Lifecycle)
description Define strict business logic constraints prevents AI hallucinations by enforcing valid transitions.

State Machine Protocol

The State Machine (state_machine validation rule) lets you define the "Constitution" of a record's lifecycle: the legal status transitions a record may take. It is a flat, textbook finite-state-machine transition table that the write path enforces.

Per [ADR-0020](https://github.com/objectstack-ai/objectstack/blob/main/docs/adr/0020-state-machine-converge-and-enforce.md), a state machine is **one of the object's `validations`** — a rule with `type: 'state_machine'`. There is no top-level `stateMachine`/`stateMachines` property on an object, and the older XState-style shape (hierarchical states, `on`/`cond`/`actions`, `meta.aiInstructions`) was retired **as a record-lifecycle declaration** — nothing on the write path reads it. A flat `{ from: [to] }` transition table is the only enforced shape.

Why State Machines?

In the era of AI Agents, field-level validation is not enough. Large Language Models (LLMs) can "hallucinate" and attempt illogical data updates (e.g., moving a contract from draft directly to paid without going through approval).

The transition table provides a hard constraint layer:

  • Deterministic: On update, if the state field changed and the new value is not listed under the current value, the write is rejected.
  • Self-Documenting: The whole legal graph lives in one place, as data.
  • Introspectable: UIs can grey out illegal buttons and an Agent can ask "from here, what's legal next?" instead of parsing a formula.

Definition

Add a state_machine rule to the object's validations array. The rule names the state field and declares a transitions map of { currentValue: [allowedNextValues] }.

{/* os:check */}

import { ObjectSchema, Field } from '@objectstack/spec/data';

export const PurchaseRequest = ObjectSchema.create({
  name: 'purchase_request',
  sharingModel: 'private',
  label: 'Purchase Request',

  fields: {
    status: Field.select({
      label: 'Status',
      required: true,
      options: [
        { label: 'Draft', value: 'draft', default: true },
        { label: 'Pending', value: 'pending' },
        { label: 'Approved', value: 'approved' },
        { label: 'Rejected', value: 'rejected' },
      ],
    }),
    amount: Field.number(),
  },

  validations: [
    {
      type: 'state_machine',
      name: 'purchase_status_flow',
      label: 'Purchase Status Flow',
      field: 'status',
      // This rule governs UPDATE transitions only (`events: ['update']`
      // below). `initialStates` — the optional FSM entry point that locks
      // which states a record may be CREATED in (#3165) — is checked on
      // INSERT, so to use it here you'd also add 'insert' to `events`.
      // Otherwise INSERT is unchecked by this rule and ANY declared option is
      // a legal starting value. (An option's `default: true` is an authoring
      // hint — only a field-level `defaultValue` fills an omitted field.)
      events: ['update'],
      message: 'Invalid purchase status transition.',
      transitions: {
        draft: ['pending'],
        pending: ['approved', 'rejected'],
        // Terminal states need an EXPLICIT empty array to be enforced: a
        // state with no key at all is one the checker cannot reason about,
        // so it does not block moves out of it (see "Enforcement semantics").
        approved: [],
        rejected: [],
      },
    },
  ],
});
`ObjectSchema.create()` rejects unknown top-level keys (ADR-0032). A typo'd `stateMachine:` or `stateMachines:` key throws a build error rather than being silently ignored.

Structure

A state_machine rule shares the common validation-rule fields (name, label, message, severity, events, priority, active) with its siblings, plus its own:

Field Type Meaning
field string The state field this rule governs (e.g. status).
transitions Record<string, string[]> Map of { currentValue: [allowedNextValues] } — the legal edges (enforced on update).
initialStates string[] (optional) States a record may be created in. When set, an insert whose field value falls outside this list is rejected (invalid_initial_state) — the FSM entry point (#3165). Omit to keep the legacy no-check-on-insert behavior.

Transitions

transitions is the whole legal graph. The keys are current state values; each value is the list of states the record may move to. An explicit empty array is an enforced dead-end — every move to another state is rejected. A state with no key at all is different: the checker has nothing to reason about there and lets the write through (see Enforcement semantics), so declare [] when you mean "terminal".

transitions: {
  draft: ['pending'],
  pending: ['approved', 'rejected'],
  approved: [], // terminal — enforced
  rejected: [], // terminal — enforced
}

Enforcement semantics

  • On insert, the transitions table is not consulted — there is no prior state to transition from. If the rule declares initialStates, the created value must be one of them, or the write is rejected with invalid_initial_state (the FSM entry point). Without initialStates, insert is a no-op and the starting value is constrained only by the field-level select option-membership check (invalid_option), so any declared option is a legal start.
  • On update, if the state field changed and the new value is not in transitions[oldValue], the write is rejected. Clearing the field (writing null) is exempt.
  • The check is lenient where it cannot reason: if the prior state has no key in transitions (legacy or externally-written data, or a state you simply forgot to declare), it does not block. Only an explicit [] makes a state a hard dead-end.
  • Only a rule with severity: 'error' (the default) blocks the write. warning/info are advisory: logged per write on an ordinary write, folded into one summary line per rule on a seed/boot load, and not evaluated at all on an update whose payload touches only platform-injected system columns (#13889).
  • Seed writes are exempt (#3433). Curated seed data — package bootstrap fixtures, marketplace templates, per-org replay, all loaded by SeedLoaderService — is a snapshot of established facts, not a record walking its lifecycle, so it bypasses the state_machine rule entirely: a seed may be born mid-lifecycle (a completed project, a closed_won opportunity) and neither initialStates (insert) nor transitions (update) is enforced. Every other validation still runs, so a seed must still satisfy field shape, format, script, and the rest. os lint warns when a seeded value is not a state the machine declares, so a typo is still caught before boot.
  • A "historical" data import is exempt too (#3479). Migrating established facts — a batch of already-closed tickets, closed_won deals — is the same "snapshot, not a lifecycle event" situation. Set treatAsHistorical: true on the import request (default off) and the runner puts skipStateMachine on the write context, so initialStates doesn't reject those mid-lifecycle rows. A normal import leaves it off and still walks the FSM — the strict behavior is the default, so the exemption is always an explicit opt-in. Only the system principal may send it: the REST import doors (POST /api/v1/data/:object/import and /import/jobs) refuse treatAsHistorical: true from every other caller, every signed-in user and administrator included, with 403 PERMISSION_DENIED before a row is read, because it takes the object's state_machine rule and its read-only and audit protection off the write. An in-process import run as the system principal (a seed or migration job) keeps it.
  • treatAsHistorical also preserves the original audit timeline (#3493) — on the rows an import UPDATES (#6640). Skipping the FSM is only half of migrating established facts; the other half is keeping when they happened and who did them. Under the same flag the write context also carries preserveAudit, which (1) makes updated_at / updated_by client-preferred — a supplied historical last-modified survives instead of being stamped with the import instant — and (2) admits a whitelist through the static-readonly write strip: the audit/timestamp family plus author-declared business readonly fields (closed_at, resolved_by, …) — but never the record's own primary key (id), which is the address of the write rather than a fact being restored (#8215). Platform-managed system columns outside that family (organization_id and other tenancy/generated columns) stay stripped — a historical import reinstates facts, it does not forge tenancy. Like the FSM exemption this is opt-in: a normal write still auto-stamps updated_at/updated_by and strips readonly exactly as before, and permissions / RLS / field-level security are unchanged.
  • …but a historical upsert still drops those columns from the rows it CREATES (#6640). preserveAudit is an UPDATE-path exemption and nothing else reads it: both write paths run the same in-engine strip (stripReadonlyFields, #14147), but only the UPDATE call site passes preserveAudit — the create side is given no such flag and keeps context.isSystem as its only exemption. So a single treatAsHistorical upsert keeps closed_at on the rows it matches and strips it — together with a supplied created_at / updated_at, which the injected audit columns also declare readonly — from the rows it inserts. The asymmetry is deliberate, not an oversight: treatAsHistorical arrives on an ordinary (non-system) import request, so honouring it on create would let any caller seed the approval/status columns that create-side strip exists to protect. The ignored request is at least no longer silent — the server logs a WARN naming the object, the stripped fields and this UPDATE-only rule — but the strip still applies. To replay archival read-only facts on the rows an import creates, write from a system context (isSystem). Over HTTP this case no longer arises: the import doors admit treatAsHistorical: true from the system principal alone, whose creates the strip exempts; it still holds for an in-process, non-system caller of the import runner. Full rule and rationale: Security & Access Control.
  • Undoing a historical import is symmetric (#3549 / #3556). The import undo (POST /api/v1/data/import/jobs/:jobId/undo) logically rolls back a finished job — deleting the rows it created and restoring the captured pre-import snapshot on the rows it updated. That restore write now carries preserveAudit too, but only when the job was flagged treatAsHistorical, so the snapshotted updated_at / updated_by and business readonly fields (closed_at, …) are reinstated verbatim instead of being re-stamped to the undo instant. The undo is unaffected by the create-side carve-out above: it only ever deletes the rows the import created and updates the rows it touched, so every write it makes is on the path where the exemption is real. Without it the undo would silently overwrite the very timeline the historical import preserved; a normal (non-historical) import's undo keeps the default stamp/strip.

Conditional transitions

A state_machine rule has no per-transition guard. To gate a transition on a predicate, add a sibling script or conditional validation rule (CEL) in the same validations array.

Introspection

Because the transition table is data, both UIs and Agents can ask "from this state, what's legal next?" instead of parsing a formula.

  • In code, legalNextStates(objectSchema, field, currentState) from @objectstack/objectql returns the declared next states, [] when the state has no outgoing edges (an explicit [] or a state the table never mentions — introspection does not distinguish the two, enforcement does), or null when no state_machine rule governs the field.
  • Over HTTP, GET /api/v1/meta/object/:name/state/:field?from=:state returns { object, field, from, next }, where next is the legal-next list, or null for either of two reasons: no state_machine rule governs the field, or ?from= was omitted (no from ⇒ no transition table to answer with). A null from a call that passed no from is therefore not evidence that the field has no state machine — re-ask with ?from=.
When an AI Agent or Flow tries to update `status` to `approved` while the record is in `draft`, the write fails with a `ValidationError` (field error code `invalid_transition`) — the AI-mistake protection the rule exists for. There is no automatic injection of per-state AI instructions into the prompt; the guardrail is the enforced transition table.

Best Practices: File Structure

For complex business objects (like Lead, Opportunity, or Order), a transition table can grow large. To keep your object definition readable, extract the rule(s) into a plain TypeScript constant — there is no special *.state.ts format or StateMachineConfig type for object lifecycles; it is just an array of validation rules.

{/* os:check */}

// src/objects/lead.validations.ts
export const leadValidations = [
  {
    type: 'state_machine' as const,
    name: 'lead_status_flow',
    field: 'status',
    message: 'Invalid lead status transition.',
    events: ['update'] as const,
    transitions: {
      new: ['qualified', 'unqualified'],
      qualified: ['converted', 'unqualified'],
    },
  },
];
// src/objects/lead.object.ts
import { ObjectSchema } from '@objectstack/spec/data';
import { leadValidations } from './lead.validations';

export const Lead = ObjectSchema.create({
  name: 'lead',
  // ... fields ...
  validations: leadValidations,
});

Multiple Lifecycles (Parallel State Lines)

In real enterprise systems, a single object often has multiple independent state lines. For example, an Order has:

  • lifecycle — draft → submitted → confirmed → shipped → delivered
  • payment — unpaid → partial → paid → refunded
  • approval — pending → approved → rejected

These are N flat state_machine rules, one per state field, in the same validations array — not hierarchical or parallel statechart regions.

// src/objects/order.object.ts
import { ObjectSchema, Field } from '@objectstack/spec/data';

export const Order = ObjectSchema.create({
  name: 'order',
  fields: {
    status: Field.select({ options: ['draft', 'submitted', 'confirmed', 'shipped', 'delivered'] }),
    payment_status: Field.select({ options: ['unpaid', 'partial', 'paid', 'refunded'] }),
    approval_status: Field.select({ options: ['pending', 'approved', 'rejected'] }),
  },
  validations: [
    {
      type: 'state_machine',
      name: 'order_lifecycle',
      field: 'status',
      message: 'Invalid order status transition.',
      transitions: {
        draft: ['submitted'],
        submitted: ['confirmed'],
        confirmed: ['shipped'],
        shipped: ['delivered'],
      },
    },
    {
      type: 'state_machine',
      name: 'order_payment',
      field: 'payment_status',
      message: 'Invalid payment status transition.',
      transitions: {
        unpaid: ['partial', 'paid'],
        partial: ['paid', 'refunded'],
        paid: ['refunded'],
      },
    },
    {
      type: 'state_machine',
      name: 'order_approval',
      field: 'approval_status',
      message: 'Invalid approval status transition.',
      transitions: {
        pending: ['approved', 'rejected'],
      },
    },
  ],
});
"Do something when the state changes" (emails, webhooks, downstream record updates) is **not** part of the transition table. Express side effects as a record-triggered [Flow](/docs/automation/flows) — the state machine only locks the legal edges.