Platform/Work in Elidyne/Workbench

Every change, tested on the twin and approved in the open.

Elidyne is the control pane where every loop is watched, tested, approved and traced. Workbench is where engineers plan and test the work: Elidyne’s specialised agents draft a plan across study stages and try each change on the twin, you approve it and reserve the compute, and every run, version and review decision lands in one thread you can follow back to its source.

app.elidyne.ai / buoy-a3 / workbench / planOpen full screen ↗
Loading the workspace…

The planner for a deep-water tsunami buoy anchor: study stages chained from site basis to design basis, with gates and compute estimates on every step.

  1. 01

    Plans chained across stages

    Site, hazards, capacity, installation. Each study feeds the next, and a change upstream marks what downstream has gone stale.

  2. 02

    Gates where decisions matter

    A named engineer accepts each basis before the next stage runs. Requests for changes are recorded, not lost in email.

  3. 03

    Priced before it runs

    Every step carries a compute estimate, so the plan is costed before anything is reserved or spent.

  4. 04

    Versions, branches and forks

    A revised plan is a new version with a visible diff. Try a larger caisson on a branch and keep the original for comparison.

How an investigation runs

From a brief to an accepted basis.

Five steps, the same every time. Each one leaves a record the next one can build on.

  1. 01

    Brief: state the objective and attach what you have.

    Core logs, catalogues, hindcasts, drawings, certificates. Elidyne reads them and lists the inputs they do not give you, instead of quietly assuming them.

    Brief · Investigation 1
    Objective
    Anchor survives Mw 9, then a 100-year cyclone
    Site
    3,850 m water depth · soft clay
    Attached
    5 files · core log, catalogue, hindcast, drawing, certificate
    Missing inputs
    9 flagged, none assumed
    Owner
    Engineering lead
  2. 02

    Plan: Elidyne drafts it, you edit it.

    The plan names the studies, the methods and their assumptions, the gates and the compute each step needs. Change a step and Elidyne shows the diff before anything moves.

    Plan v3 · 6 studies · 4 gates
    Study 1 · G1Site basisAccepted
    Study 5 · G3Anchor designWaiting on sign-off
    Study 6 · G4Design basisNot started
  3. 03

    Approve and reserve: nothing runs before you say yes.

    Approving a plan version reserves its compute. Runs are placed by policy and report their spend against the estimate as they go.

    Run status
    • Plan v3Approved · 2 Nov
    • Compute reserved7.1 of 10 CU
    • Run 5Running · 1.2 of 2.4 CU
    • Run 4 · step 14Stale · new certificate
  4. 04

    Review at the gate: accept, or ask for changes.

    Each gate belongs to a named engineer. Assumptions accepted at a gate are carried forward as assumptions, so nobody mistakes them for results later.

    Gates
    • G1 · Site basisAccepted after changes
    • G2 · Hazard basisAccepted · 1 assumption recorded
    • G3 · Anchor designWaiting on sign-off
    • G4 · Design basisNot reached
  5. 05

    Trace: accepted work becomes the basis for what comes next.

    Approved plans become reusable recipes, accepted results become the design basis, and the whole trace is ready for Reports to draft from.

The thread

One thread holds the whole investigation.

Messages, plan versions, runs, drawings and review decisions sit on a single timeline. When a reviewer asks for changes, the request, the revised plan and the re-run stay linked, and the inspector shows the inputs and settings behind any result.

app.elidyne.ai / buoy-a3 / workbench / threadOpen full screen ↗
Loading the workspace…
  • Review gates with accept, request changes or refuse, each one recorded
  • Engineering drawings and source files attached where they were used
  • An inspector panel with the inputs, settings and cost behind every result
  • Results flagged as stale the moment an input, such as a certificate, changes
Why it holds together

Built for work that has to be defended.

01

Context survives the handoff

Analysis knows what the site study established; the design knows which hazards it was checked against. Nothing is re-derived from a filename.

02

Cheap methods run before expensive ones

Screening and simple checks run first, so high-fidelity compute is spent where it changes the decision.

03

The write-up is a by-product

Reports draw on the same record the runs already carry, so a defensible document is not a separate project.

A result without its provenance is a number. With its provenance, it is engineering.
Workbench is built for work that still holds up in a design review, an audit or a licensing submission, long after the run has finished.
How it connects

Workbench and the other five.

Six products, one control pane. Workbench mainly carries Twin, Test and Limits; the rest of the loop is already connected when you need it.

  1. 01Measure
  2. 02Twin
  3. 03Test
  4. 04Limits
  5. 05Act
  6. 06Trace

Measure → Twin → Test → Limits → Act → Trace. Every product serves the same six steps, in every loop.

Start a mission

Bring us the system that matters.

Bring us the investigation that takes six weeks. We will show you what it looks like planned, tested on the twin, priced and gated in Workbench.