Skip to content

ROADMAP.md

Latest commit

 

History

History
77 lines (57 loc) · 3.47 KB

ROADMAP.md

File metadata and controls

77 lines (57 loc) · 3.47 KB
Raw

Dirac roadmap

Current as of 2026-08-17. This roadmap starts from the source-derived state in STATUS.md; it does not repeat completed implementation history.

Direction

Dirac grows as one molecular-discovery operating system. New capability should add a canonical ObjectKind or relation only when necessary, a Method for computation, a Command for application behavior, durable state and evidence, and then one or more projections. It must not create a private API, cache, Job state machine, error vocabulary, context store or mol* scene.

Current priorities

1. Finish the authoritative HCI action spine

  • Replace route-local and form-local mutations with offer, preview, commit and receipt.
  • Carry versioned context, selection, authorization and idempotency through browser, CLI and agent projections.
  • Make stale state, duplicate retry, partial saga and recovery first-class outcomes.
  • Complete a failure-complete Structures → Design → Campaigns vertical slice before declaring broader HCI readiness.

Source: docs/product/HCI_MIGRATION_PLAN.md.

2. Connect the remaining product Views

  • Preserve the 8-Workspace / 30-View information architecture.
  • Connect Views only when their Commands, durable objects, permissions, errors and recovery paths exist.
  • Keep shell-only Views visibly partial; navigation completeness is not capability.

3. Harden Program as the scientific operating root

  • Finish work graphs, handoffs, stage gates, evidence admission and lineage.
  • Preserve one identity chain from Program intent through Run, Job, Artifact and Decision.
  • Make material, sample, protocol, experiment and external-evidence invariants transactional rather than UI conventions.

4. Validate Motif scientifically, not only architecturally

  • Keep dataset versions, runtime locks, model gates and routing decisions reproducible.
  • Benchmark models and acquisition policies against declared controls and uncertainty.
  • Treat docking, MD, RBFE and OpenFE outcomes as evidence with explicit failure and applicability boundaries.
  • Separate executable workflow coverage from method validation and prospective utility.

5. Production readiness

  • Add a supported environment/bootstrap path for the full scientific stack.
  • Exercise backup, restore, migration rollback boundaries and disaster recovery.
  • Complete remote authentication, HTTPS, scopes, quotas, audit redaction and artifact authorization before any WAN exposure.
  • Add observability and operator runbooks that do not depend on the application being up.

Rewrite tripwires

A change is off-path if it:

  • makes an adapter own scientific logic;
  • embeds a large scientific result in a response instead of an Artifact;
  • introduces another Job or error state machine;
  • gives a Workspace its own scientific context or mol* instance;
  • treats a route transition as a scientific handoff;
  • stores presentation state as evidence;
  • equates a registered Method or passing transport test with scientific validation;
  • starts a duplicate Dirac service instead of rebuilding the intended supervised bundle (full shell on 1360 or Discovery Lab on 1370).

Deliberate non-goals for the current phase

  • public unauthenticated hosting;
  • claiming all 30 Views are scientifically implemented;
  • fabricating browser-only approximations for unavailable compute;
  • replacing upstream mol* internals without a demonstrated Dirac requirement;
  • autonomous architecture mutation by the observed twin.