LA.001Docs · Status
LA.001Docs · Status

Five labels. Plain BUILT is the only one that means you could use it.

State of the world as at 29 August 2026, re-read after the 1 September sitting. Sixteen migrations that were written and applied to nothing are now cloud-applied; what separates the platform from working is a build, the run driver, not a window at a database.

LA.002The labels
LA.002The labels

Five words, and they are not interchangeable.

  • BuiltIn the main line, its tests run, and a person or a job actually reaches it.
  • Built, not switched onIn the product with its tests running, and nothing turns it on: a flag with no setting row behind it, or a caller that hasn't adopted it. Real, and unusable.
  • Written, not appliedMerged as a database migration and applied to no database. Needs a person at a keyboard with a database open. Since 1 September this label is mostly spent.
  • Not builtNo code. Some are gaps. Some are recorded decisions, and those say so.
  • Ruled out for nowA decision, not backlog. Named so nobody counts it as coming.
LA.003The matrix
LA.003The matrix

Thirty capabilities, one label each.

CapabilityStateNote
Agent-facing API, workflow definitions, run control loop, skill contractBuilt
Scheduled workflow dispatcher; inbound signal intake (form, webhook)Built
Draft-before-operate enforcement: a workflow must be promoted before it can actBuilt
Landing-zone data fence; sync metering and freshness; proven sync workersBuilt
Credential custody; session grants (scoped, short-TTL) and tripwireBuilt
Row-level security and client data isolationBuilt
Approvals permission floor: agents mint, humans decide, enforced in the databaseBuilt
Scoped agent identities replacing the master key (19 of them)Built
BYOK model-credential resolver with four named refusalsBuilt
Client portal on real records through row-level securityBuilt
Credential broker, the sole door to client platformsBuilt, not switched onNo live client credential activated
Ten-check gate chainBuilt, not switched onMaster flag off by default
Run and job ledger: tables, claim, settle, cost columnsBuiltCloud-applied 1 September; nothing drives it yet
A job running start to finishNot builtThe run driver. A four-to-seven day build, and the highest-weighted open risk on the board
Cheap planning pass and Gate 1 on the planBuilt, not switched onWaits on the run driver
Memory step: a run recap the next run readsBuilt, not switched onRoutes landed; no run has produced one
Per-client budget capsBuilt, not switched onSpend ceilings in cents set nowhere yet
Per-client fairness in the schedulerBuilt, not switched on
Fleet and per-client roll-upsBuilt, not switched onSeeded shut: an open flag over an empty ledger would present zeroes as fact
Starting work from a page or doc commentBuilt, not switched on
Chat panelBuilt, not switched onNeeds a metered key
Handlers for 36 client-platform actionsBuilt, not switched on17 read, 4 stage, 15 never. Unregistered
Worker sandbox boundaryBuilt, not switched onRefuses to arm until the host half exists
Skill release controlBuilt, not switched onInstalled service, four refusals hold it shut
Automations registry, library, silence alarmBuilt, not switched onRegistry live; first automation awaiting approval
Alerting that reaches a humanNot builtThe watchers fire. No device is subscribed
Portability and self-hosting toolingNot builtFive design papers, no code. The one thing that hasn't moved
Scheduled sessions on the provider's cloudNot built
A warehouse layerRuled out for now
Skills catalogue and studio surfacesRuled out for nowA ready plan, deliberately not started

The single best question to ask isn't "is it built". The answer is mostly yes and it's the less informative half. Ask: has a job ever run from start to finish, and can I see the receipt? As at 1 September the answer is no, and the board carries that as its highest-weighted open risk in those words. Everything on this page is downstream of it.