Asset Nova · Product 1

Shield

Profit Protection Through Claims Defense & Recovery

Reads every document on a megaproject. Spots incoming claims before the owner's defense window closes. Builds the evidence. Drafts the response. A human approves. Money protected.

Built on NovaOS Sovereign · MENA-first

Alexandre Cela · Case Study · August 2026

Outline

What we'll cover

1
What it does
Live demo + value proposition
2
Why this, why now
Research-backed wedge decision + platform advantage
3
How it works
Architecture, self-improving loop, success criteria
4
How we ship
Delivery roadmap, OKRs, 3-horizon portfolio
5
Risks and next steps
Honest assessment + what to de-risk first
See it working

Before I explain why, let me show you what

Jubail Gas Processing Expansion · $4.8B · EPC Package 3

400K+ documents. One agent reads them all. It detected a $42.3M claim building and found 3 weaknesses before it arrived.

1
Detect threat

Spots claim signals in site reports, letters, schedule changes

2
Assess defense

Pulls 6 docs, finds procedural gaps in contractor's position

3
Draft response

Defense letter with every sentence linked to source

4
Approve

Named human reviews and signs off

Open live prototype
Without Shield
  • Claims arrive as a surprise, defense scrambled under pressure
  • Evidence scattered across 6+ systems, assembled too late
  • Procedural weaknesses in contractor claims go unnoticed
  • Result: owners pay claims they could have defended
With Shield
  • Incoming claims detected before they formally arrive
  • Evidence assembled automatically, every sentence sourced
  • Procedural flaws identified and documented
  • Result: $30-80M protected per project
Full demo scenario: Contractor hit unexpected rock on EPC Package 3. They're preparing a $42.3M claim for extra costs. Shield spotted the signals in site reports, letters, and schedule changes. Agent pulled 6 documents and found: the contractor's letter doesn't meet formal notice requirements, and an amendment shortened their deadline from 28 to 21 days. Defense response written for the owner, every sentence links to its source document, protects the owner's position on procedural grounds. A named human reviews and approves. Nothing is sent without sign-off.

Before/After detail: Without Shield, contract amendments and deadline changes are buried in documents nobody re-reads. With Shield, contract amendments caught automatically, finds deadline changes the contractor may have missed. Owner is always prepared, $30-80M protected per project.
Value proposition

Why this is a must-have, not a nice-to-have

Dream Outcome

Protect $30-80M per project

Sources: HKA CRUX (2,200 projects, $95B); Arcadis ($91M avg ME dispute)
Likelihood of success

Proven AI on messy enterprise data

FlowNova: 60-90% faster decisions. ComplyNova: 99.8% compliance
Time & Effort

Always-on, plugs into existing systems

Read-only ingestion. Results in days on historical archives
Sacrifice

Zero. Sovereignty and alpha built in.

In-region VPC or on-prem. SACS-002, Saudi PDPL. Weights, not wrappers
Dream Outcome: 33% of capex ends in dispute. Average Middle East dispute: $91M. Shield catches incoming claims before the owner's defense window closes and finds weaknesses in the contractor's position. Not productivity -- money with a dollar sign. $30-80M = modeled estimate: $5B capex x 20% disputed (HKA CRUX haircut) x 3-8% value protected.

Likelihood: Orbitron already processes messy, multi-system enterprise data with FlowNova (supply chain) and ComplyNova (GRC). The reading infrastructure exists. Shield applies it to a new corpus (contracts + correspondence) on a proven substrate (NovaOS).

Time & Effort: Read-only ingestion from Aconex, P6 and SAP exports. No ERP integration, no re-platforming, no change management. The agent runs in the background. The review ritual replaces work that already happens -- it just happens late and manually today.

Sacrifice: In-region VPC or on-prem (SACS-002, Saudi PDPL). Customer data never trains shared models -- your project history stays yours. Sovereignty is guaranteed at the source via NovaOS: Glass Box audit trail, full data residency control, no compromise.
Research-backed

Why claims defense delivers the highest value in the shortest time

89%of ME projects over budgetEY, 365 O&G projects
33%of capex ends in disputeHKA CRUX, 2,200 projects
$91Mavg Middle East disputeArcadis, 12.5 months to resolve
~2xcontractors claim vs settlementIPA via National Academies
"Over 75% of respondents said the PM or Engineer's conduct was at the heart of how disputes crystallized -- most often a lack of understanding of the procedural aspects of the contract." Arcadis, Global Construction Disputes Report
⚡

We don't have to reinvent the wheel

Governance and sovereignty are rooted in NovaOS. Infrastructure from ComplyNova and FlowNova can be reused. The new work is the domain model.

All four quotes:
1. "Over 75% of respondents said the PM or Engineer's conduct was at the heart of how disputes crystallized -- most often a lack of understanding of the procedural aspects of the contract." -- Arcadis, Global Construction Disputes Report
2. "What most projects lose to poor claims management is not a dispute they should have won, but entitlement they never assembled the evidence to claim." -- Kairos, Construction Claims Management, 2026
3. "Failure to send notices in the correct form is one of the biggest mistakes that the parties to the FIDIC contracts make." -- Institute of Construction Claims Practitioners (8,145+ views)
4. "The difference between the amount submitted for a claim versus the amount actually paid at settlement is as much as half." -- IPA (Independent Project Analysis) via National Academies

Platform detail: NovaOS provides the agent studio, governance kernel, sovereignty controls, and human-in-the-loop gates. ComplyNova's audit evidence collection, case management, and risk scoring infrastructure can be reused for claims defense workflows. FlowNova's document ingestion, procurement agents, and decision automation provide reusable building blocks. What's new is the CPM domain model: contract clause extraction, notice deadline tracking, defense drafting. The pain is procedural, not strategic. Missed deadlines, scattered evidence, unread correspondence. These are reading problems -- and reading at scale is what LLM agents do.
Architecture

How Shield works

Inputs (read-only)
Contracts + amendments
Correspondence
P6 schedule files
Site reports
→
Governance kernel
Scope check
Action allowlist
Audit log
Sovereignty tag
→
LLMDetect threats
Read documents, find incoming claims
LLMLink evidence
Connect documents to build the case
CodeCompute deadlines
Date math, never LLM
CodeVerify citations
Every reference checked
LLMDraft response
Write defense letter with sources
→
Human gate
Named approver
Nothing leaves without sign-off
Sovereignty: guaranteed via NovaOS -- in-region, full audit trail. Customer data never leaves. Design rule: LLM where language is the problem. Code where being wrong is unrecoverable.
Judgment where needed: Event detection, clause interpretation, notice drafting use the LLM to read and reason across thousands of documents.

Determinism where stakes are absolute: Deadline math is code. Citation verification is a resolver. No LLM ever computes a date or fabricates a document reference.

Governance as the core: Scope enforcement, action allowlist, human approval gate, full audit trail. No code path acts first and checks later.

For the detailed pipeline walkthrough, eval gates, and self-improving loop, open the prototype and navigate to the System Design tab.
Gets smarter over time

Shield improves on your project -- without your data leaving

1

Agent flags an event

Detection, evidence, draft notice

→
2

Human reviews

Approve, edit, or dismiss

→
3

Decision logged

Stored as labeled example

→
4

Agent retunes

Better thresholds, fewer false alarms

↑ Repeats every day. The more your team uses it, the better it gets at your project. ↑

Dismissals reduce noise

False alarms raise threshold for that pattern

Edits sharpen drafts

Rewrites captured, draft quality matches your team's voice

Approvals confirm what matters

Confirmed events become regression tests for every update

Dismissals detail: When the team dismisses a false alarm, Shield raises its threshold for that pattern. Fewer irrelevant flags over time.

Edits detail: When the team rewrites a notice before sending, the edit is captured. Draft quality improves to match how your team actually writes.

Approvals detail: Confirmed events grow the set of known-good examples used to test every update. No change ships unless it passes on all confirmed events.

Sovereignty: All retraining happens inside the customer's own environment (in-region VPC or on-prem). Dispositions never leave. Your project data never improves someone else's agent. No shared model, no data export, no cross-tenant learning. The agent gets smarter on your projects and only your projects. Guaranteed via NovaOS.
Three layers of success

How we measure if Shield works

Business success
  • $20M+ validated exposures per pilot
  • Zero notice deadlines missed
  • $10-12M ARR at 24 months
Product success
  • 90% daily review ritual adoption
  • 80% events actioned (not just seen)
  • 70% draft notices approved with minor edits
AI Engineering success
  • 100% threat detection recall
  • 99.5% citation accuracy
  • 0 hallucinations in approved notices

Kill criteria

Recall drops below 100% (missed a real threat), or precision <60% after 8 weeks of tuning, or the contracts team stops the daily review for 4 consecutive weeks: pause deployment, return to backtest mode. Every disposition (approve/edit/dismiss) feeds the self-improving loop -- the agent gets sharper on your project without your data improving anyone else's.

Full Business success: $20M+ validated exposures surfaced per pilot (customer-countersigned). Zero notice deadlines missed on covered projects. 60% pilot-to-license conversion rate. $10-12M ARR at 24 months across 5-6 program licenses.

Full Product success: 90% daily review ritual adoption (the habit IS the product). 80% events actioned (reviewed, not just seen). 70% draft notices approved with minor/no edits. Less than 7 days from triggering event to notice sent.

Full AI Engineering success: 100% threat detection recall (never miss a real threat); maximize precision progressively. 99.5% citation accuracy (deterministic verifier). 0 hallucinations in approved notices. 100% deadline accuracy (rules engine). Less than $0.05/doc blended cost.
Delivery plan

How we ship Shield, and when

Phase 0 -- Backtest

Run on 1-2 closed projects. Expert audit. Produces "here's what you missed" sales artifact.

Weeks 0-6

Phase 1 -- Live pilot

1-2 active projects, owner-side. Daily review with contracts team. In-region VPC.

Months 2-6

Phase 2 -- Program license

Expand across EPC packages. Add contractor-side variant. Schedule Interrogator cross-sell.

Months 6-12
Phase 0 detail: No live risk, no political cost. The backtest on closed projects is the proof point.

Phase 1 detail: Success = customer-countersigned value memo. Document-based deployment.

Pricing: Outcome-anchored: ~0.1-0.3% of project value (industry benchmark). On a $5B program = $5-15M. Consistent with Orbitron's "ROI guaranteed" commercial motion. Pilot: $300-500K flat, below mega-procurement thresholds.

Go-to-market: Sell to: Capital Projects Director (feels the pain). Validate with: Digital office (architecture + security). Land: one project team, document upload, results in days. Expand: across program. Defend: the cross-system project record is the moat.
Vision & objectives

Where Shield takes us

Vision

No owner should pay a claim they could have defended. Shield reads the full record, spots claims early, and prepares the defense before deadlines make it permanent.

Objective 1: Prove AI claims defense protects real dollars (Months 0-6)
KR1: $20M+ validated exposures on each of 3 pilots
KR2: Zero notice deadlines missed
KR3: 100% recall, precision improving QoQ
Objective 2: Build the commercial engine (Months 6-24)
KR1: 3 of 5 pilots convert to program licenses
KR2: $10M ARR within 24 months
KR3: Cross-system project record enables Horizon 2
Full vision: No owner should pay a contractor claim they could have defended against. Shield reads the full project record, spots incoming claims early, finds the weaknesses in the contractor's position, and prepares the defense -- before deadlines make it permanent.

KR rationale: KR1 (exposures) validates the value prop with real dollar signs. KR2 (deadlines) is the zero-tolerance metric -- if we miss one, we've failed at the core promise. KR3 (recall) ensures we never miss a real threat while progressively reducing noise. On the commercial side, 60% conversion proves the pilot model works, $10M ARR proves the business, and the project record substrate is the strategic asset that unlocks Design Gate and later products.
Product portfolio

Where Shield fits -- and what comes next

FEL-1

Concept screening

FEL-2

Feasibility

FEED / FEL-3

Design & estimate

Design Gate
FID

Board approval

Budget Red Team
EPC Execution

Build phase

Shield
Commissioning

Handover

O&M

Operations

Now -- Horizon 1

Shield

Claims defense during EPC execution. Protects $30-80M/project.

Composes: ComplyNova + FlowNova on NovaOS
Next -- Horizon 2

Design Gate

AI validation of FEED revisions. Catches $35-70M rework per project.

70-80% of cost deviations trace to design (Arcadis)
Later -- Horizon 3

Budget Red Team

Reference-class forecasting at FID against portfolio history.

59% avg O&G overrun vs FID (EY, 365 projects)
Shield full: Claims defense & recovery during EPC execution. Reads contracts against the live project record. Protects $30-80M/project (modeled estimate, HKA + Arcadis). Composes: ComplyNova (evidence packs) + FlowNova (downstream impact).

Design Gate full: AI validation of every FEED revision against building codes, cost and buildability. Catches $35-70M rework per project while it costs thousands, not millions. Evidence: 70-80% of cost deviations trace to design (Arcadis). Dubai Municipality building it for permits (WhiteHelmet / Buildcheck). 4:1-30:1 ROI (nPlan, McKinsey).

Budget Red Team full: Reference-class forecasting at FID. Pressure-tests the board's billion-dollar approval against the customer's own portfolio history. Prerequisite: owners must trust Orbitron enough to share historical budget estimates vs actuals and timeline expectations vs outcomes. That trust is earned through Shield and Design Gate first. 59% avg O&G overrun vs FID (EY, 365 projects). 0.5% of megaprojects hit budget + time + benefits (Flyvbjerg, Oxford).
Platform advantage

Why Orbitron. Why Asset Nova. Why now.

Where existing players focus

ALICE TechnologiesSchedule optimization
nPlanSchedule risk from P6 files
Document Crunch (Trimble)Contract review at signing
SuperHive (AI71)Design studio, site monitoring
Nodes & LinksSchedule intelligence
ProcoreCommercial construction management

Why Orbitron wins this seat

  • NovaOS is already the substrate.
  • FlowNova + ComplyNova are already deployed.
  • MENA-first, sovereign by design.
  • Outcome-based pricing proven.
  • Glass Box + Trust Centre.
NovaOS: Composable agent studio with human gates, audit trails, agent registry. Shield doesn't build infrastructure -- it uses it.

FlowNova + ComplyNova: Messy enterprise data across supply chain and GRC. Proven at scale. Shield reuses their capabilities.

MENA-first: In-region deployment, IKTVA-ready, SACS-002 path. Sovereignty guaranteed via NovaOS.

Outcome-based pricing: FlowNova already sells on "ROI guaranteed -- commercial models structured around measurable outcomes." Shield follows the same motion.

Glass Box + Trust Centre: Auditability and transparency are existing brand pillars. Shield inherits them, not builds them.

TAM: Owner-side claims defense across contracts, correspondence, schedule, and cost -- the cross-system commercial layer. $95B is disputed globally across 2,200 projects annually (HKA). ME megaprojects alone represent an estimated $150-300M addressable market for claims intelligence software, based on 0.1-0.3% of ~$100B+ annual ME process-industry capex.
Honest assessment

Four risks to address, with clear next steps

$
Value Risk

$30-80M claim stacks two low-confidence assumptions. Customers may bless big numbers costlessly.

De-risk: 5-8 customer discovery interviews via existing relationships

⚙
Feasibility Risk

100% recall across heterogeneous corpora (scanned Arabic + English, mixed contract forms).

De-risk: Phase 0 backtest on closed project answers this before any commitment

⚖
Viability Risk

Sovereign GPU at scale. 9-18 month MENA sales cycles. IKTVA and SACS-002 certification.

De-risk: Bottoms-up cost model. Test contractor wedge as faster revenue path

✎
Usability Risk

Daily review ritual is new work. High precision required or contracts managers tune out.

De-risk: Prototype-test review ritual with 2-3 prospect contracts managers
Value Risk full: The $30-80M value claim stacks two low-confidence assumptions. "Validated exposures" is soft -- customers can bless big numbers costlessly. Contracts teams may fear the exposure register surfacing their own misses. De-risk: 5-8 customer discovery interviews (owner PMO leads, contracts managers in MENA) to validate willingness to share archives, daily review ritual appetite, and what "value" means to them.

Feasibility Risk full: 100% recall (never miss a real threat) across heterogeneous corpora (scanned Arabic + English, different contract forms). OCR pipeline, XER parsing complexity, per-contract-form deadline engine. Open-weights model performance on domain-specific legal language. De-risk: Phase 0 backtest on a closed project answers this before any commercial commitment. The eval harness (built -- see prototype) runs offline with zero dependencies.

Viability Risk full: In-region sovereign GPU at scale. Cost per document at 500K docs/project. 9-18 month sales cycles in MENA. IKTVA localization, SACS-002 certification, tech E&O insurance for deadline liability. Team investment vs other NovaOS priorities. De-risk: Bottoms-up cost model (GPU provisioning, team, certification timeline). Test EPC contractor wedge as faster revenue path if owner cycles stall.

Usability Risk full: The daily review ritual is new work on top of a contracts manager's day. High precision required or they tune out. Dispositions create a discoverable record (political risk). Governance of visibility (who sees what was flagged) is design-critical. De-risk: Prototype-test the review ritual with 2-3 prospect contracts managers. Team dispositions (not individual). Triage volume cap. Design for their existing artifact (event register), not a new dashboard.
Let's build it

Shield protects profits while saving costs.

The data already exists. The reading capacity now exists. The platform exists. The market is the most painful on Earth -- and it's Orbitron's home turf.

$30-80Mprotected per projectModeled: HKA + Arcadis data
Daysto first value
Zerosacrifice on sovereignty
Replay the demo

For the agent walkthrough, open the prototype and navigate to the System Design tab. For all PM artifacts, navigate to the How This Was Built tab.

Alexandre Cela · August 2026

Appendix -- The bigger picture

Shield can make the whole relationship healthier

Today, claims are adversarial by default. Disputes drag on for months, cost millions in legal fees, and damage relationships for both sides. Shield can change that -- for owners first, and eventually for contractors too.

Today's reality

Disputes are slow, expensive, and lose-lose

  • Average ME dispute takes 12.5 months to resolve (Arcadis)
  • Contractors overstate claims by ~2x because evidence is weak and negotiations are long (IPA)
  • Owners reject claims they should partially accept because they can't find the proof
  • Both sides spend millions on external claims consultants assembling the same evidence separately
  • Relationships deteriorate -- the next project starts with less trust
What Shield enables

Fast, precise, substantiated claims -- on both sides

  • For owners (now): defend against inflated claims with evidence, not guesswork. Respond in days, not months
  • For contractors (Phase 2): build substantiated, well-evidenced claims faster. Claims that are part of the job, not a fight
  • When both sides have good evidence, disputes resolve faster and fairer
  • Less money spent on external consultants and arbitration
  • Better relationships. The next project starts with more trust, not less
⚖

The win-win: better evidence on both sides means faster, fairer outcomes for everyone

Start with the owner (the buyer with the clearest pain and the biggest budget). Once Shield proves the model works, offer a contractor variant. Same technology, same evidence standard. When claims are fast, precise, and substantiated, they stop being disputes and start being business as usual.

Appendix

Supporting artifacts

Full PM deliverables produced during this case study.

Discovery & Research
  • Competitive Analysis -- SuperHive, Nodes & Links, full landscape
  • Impact Sizing -- driver tree, three scenarios, revenue model
  • Sub-Agent Review Synthesis -- 5-reviewer stress test, 12 findings
  • Orbitron Context Synthesis -- product, founder, platform analysis
Product Specs
  • Product PRD -- solution review stage, post sub-agent review
  • AI Engineering PRD -- DAG pipeline, governance kernel, eval suites
  • Strategy & Roadmap -- thesis, wedge decision, 3-horizon plan
Prototypes & Code
  • Live Interactive Prototype -- 5-stage detection-to-approval flow
  • System Design -- DAG pipeline, eval gates, self-improving loop
  • Eval Harness & Governance Kernel -- runnable code with synthetic data
Validation
  • Validation Quotes by Opportunity -- sourced quotes mapped to personas and JTBD
  • Discovery Report (Plain English) -- full report without jargon
  • Full Discovery Research -- 48 sources, 7 sections

All artifacts are clickable and downloadable in the prototype -- navigate to the How This Was Built tab.

1 /