Problem How it works Capabilities Founder Trust & data Book a demo

A building isn’t a facilityuntil someone proves it.

Perceptix derives the dependency graph behind a commissioning program, computes the real critical path across it, and hands operations a verified functional state instead of a folder of scanned forms.

Send us a finished project See how it works
<3,000 certified Cx professionals in the US
$2.8M monthly carrying cost, 50 MW site
18–24 mo hyperscale ready-for-service
Perceptix — change impact PROJ IAD-14 · BLOCK B · 2N
CxA

The 2500 kVA unit substation on Block B was swapped for a different manufacturer after the lead-time slip. Tell me which completed tests that invalidates — and show me the dependency, not just the flag. Reference TX-B-02.

  • @graph-builder Re-derives the affected subgraph from the revised one-line, the submittal and the controls points list 142 nodes touched
  • @dependency-tracer Walks every downstream edge from TX-B-02 to the test nodes that assumed its rated impedance 6 invalidated
  • @evidence-linker Attaches the spec clause and sequence of operations that established each edge, with confidence 6 of 6 cited
  • @schedule-solver Re-solves the critical path against certified witness availability and Level 4 hold points RFS +3 days
6 tests invalidated, 14 still valid, 2 require re-witness. Protective relay coordination and secondary injection must be re-run; upstream medium-voltage acceptance is unaffected. Every call is evidence-linked. Nothing is auto-approved — a certified professional signs.
Integrated systems test Level 4 Re-witness list Ask about a change, a hold point, or a test…
The problem

Fewer than 3,000 certified professionals must witness every test that turns a finished building into a facility — and they are allocated by calendar and relationship, not by critical path.

The industry treats commissioning as a phase that starts once the building is substantially complete. It is not a phase. It is a dependency graph — thousands of tests, each requiring specific systems energized and functional in a specific order — and nobody has ever drawn it. Because the graph is invisible, everything is serialised for safety, latent parallelism goes unused, and when anything changes, nobody can say which completed tests are still valid.

How it works

Draw the graph.
Then solve it.

Constructing a systems dependency model by hand is weeks of senior engineering time, which is exactly why it has never been done. Reading sequences of operation, one-lines, P&IDs, submittals and points lists is now a computation.

What it does with the graph
Ingest — design intentParsing
SOOahu-sequence-rev-C.pdf — 46 states
ONE-LINEE-201 — 2N, 14 unit substations
P&IDM-410 — CHW primary/secondary
POINTSBMS export — 8,412 points
SPECDiv 01 91 13 — Cx requirements

Read what the engineers already wrote

Sequences of operation, electrical one-lines, mechanical P&IDs, equipment schedules, submittals, controls points lists, the owner’s project requirements and the commissioning specification. Contradictions between them surface immediately — several design teams get value from that alone.

Systems graphDerived
Nodes / systems1,284
Test nodes3,916
Prerequisite edges11,402
Evidence-linked98.1%
Low confidence → blocking214

Every test becomes a node with prerequisites

Which systems must be energized, which must be functional, which prior tests must have passed. Each edge is linked to the document clause that established it. Anything unresolved or low-confidence is treated as blocking, never as absent — and that bias is published, not hidden.

Critical path solveRe-solving
As-planned duration112 days
Dependency-derived93 days
Safely parallel branches11

Schedule the scarcest people against the path

A resource-constrained solve over the graph, with certified specialist availability as the binding constraint. Two outputs: the real critical path, which is almost never what the bar chart says, and where the sequence can safely run in parallel — with the dependency evidence for why it is safe.

What we believe

Speed is not a claim we make on its own. Every sequence we propose arrives with the dependency evidence behind it, and a certified human decides.

Capabilities

What the graph
makes possible.

Checklist software documents the sequence. A graph computes it, propagates change through it, and keeps it alive after handover — which is when the record stops being paperwork and starts being infrastructure.

Run it on a completed project
Change propagation6 invalid
FT-4.12Relay coordination — re-witness
FT-4.14Secondary injection — re-witness
FT-3.07MV acceptance — unaffected
IST-2.1Load bank @ 100% — blocked
Evidence chainshown per call

Change impact, answered instead of argued

Equipment substituted, a controls sequence modified, a valve replaced. A graph query returns exactly which completed tests are invalidated, which remain valid, and what must be re-witnessed — with the dependency chain as justification. Today that question is settled by whoever is most senior in the room.

Field executionLive
09:14FT-2.09 pass — witnessed, CxA-04
09:41Issue #318 → CRAH-12 linked
10:02Graph re-solved — 2 branches freed
10:03FT-5.02 released to Tuesday crew
Open issues on path4

Results captured structurally, not scanned

Test outcomes are recorded against the node they belong to and issues auto-link to the systems they affect. The graph updates as the program actually unfolds, so the critical path re-solves against reality rather than against the plan someone drew in month three.

Verified state — queryPersistent
>was CB-2A tested at current settings?
Last witnessed2026-04-18
Settings at testLTPU 0.8 / STD 0.2s
Changed sinceyes — 2026-07-02
Deferred items3 open

Operations receives a model, not a binder

At turnover: every system, every test, every result, every open issue, every deferred item — queryable. It persists through retro-commissioning, capacity expansion, phase two and equipment replacement. The most expensive document in the building’s life stops being a filing cabinet.

How we stay conservative

In mission-critical work, restraint is the product.

We never authorise a sequence

Perceptix proposes, with the dependency evidence attached. A qualified commissioning professional accepts or rejects. The certified human stays the authority, legally and operationally.

Every edge cites its source

Each dependency is linked to the document that established it — a sequence of operations, a one-line, a specification clause — so a reviewer can check the reasoning rather than trust it.

Unproven means blocking

Unresolved or low-confidence dependencies are treated as blocking, never as absent. The system is biased toward saying “we cannot prove these are independent” — and that bias is published.

Life safety is a separate class

Fire alarm, smoke control, egress and emergency power are handled as their own class, with no automated parallelisation proposals at all. Not conservatively — categorically.

Founder
VST
Vaibhav Singh Tomar Founder & CEO · Perceptix

You have to have stood in the switchroom to model it.

Perceptix exists because the sequence, not the paperwork, is the expensive part of commissioning — and because the people who know that best are the ones with the least time to prove it. Vaibhav started the company after watching finished buildings sit idle waiting on a dependency nobody could name.

The company is built around one discipline: model the facility the way a commissioning professional already reasons about it, then make that model computable. Everything else — the scheduling solve, the change propagation, the verified state handed to operations — falls out of getting the graph right.

We spent eleven weeks commissioning a building that was finished in seven. Not because anything failed — because nobody could prove which tests could run at the same time. Every day of that was a scheduling problem we solved with a whiteboard. Vaibhav Singh Tomar, Founder
founder@perceptix.cloud LinkedIn
Trust & data

Your drawings are the most sensitive thing you own.

Design documents, one-lines and commissioning records describe exactly how a facility can be attacked as well as how it works. Every project is isolated by default, access is scoped to the people on that project, and derived dependencies carry an immutable trail back to the source clause — so an auditor can reconstruct any conclusion the system reached.

  • Project-level isolation
  • Role-based access
  • SSO / SAML
  • Immutable audit trail
  • No training on customer documents
  • Customer-managed retention
  • SOC 2 Type II — in progress
Document scopePer project, per phase
Derivation recordClause-level, immutable
Model authorityCertified human
Data residencyCustomer region
ExportFull model, any time
Book a demo Request the security brief
Get started

Send us a project you already finished.
We’ll tell you how much faster it could have been.

Start a retrospective Book a demo