Narrates the position
What moved this window and where the exposure sits — with every stated figure reconciled against the engine, and unverified numbers flagged on screen.
Zepth Rx · Schedule Intelligence
Claim Management & Delay Analysis
The delay happened months ago. The argument about whose it was starts now — with the reason for every attribution stored.
Last updated
Zepth Rx module
Claim Management
AI agent built into the module
Windows, not a single global sumEvery window, or no moneyFour buckets that reconcileOwnership, with the reason kept
Overview
Zepth Rx runs a windows analysis across every revision your contractor submits, continuously, as they arrive — and attributes each window’s slip to the parties, with the reasoning kept and openable. Built as the project runs, it costs nothing at the point of dispute. Reconstructed two years later from an archive, it is the single most expensive part of a claim, and the reconstruction is itself attackable.
Everything the module produces is a signal for review. Nothing it produces is a determination of entitlement — that depends on the contract, the facts, and the judgement of the people appointed to decide it. The module computes what the schedules show, records why it reached each conclusion, and presents it for a qualified person to assess.
Read this first
Everything this module produces is a signal for review. Nothing it produces is a determination of entitlement.
That framing is not a disclaimer bolted on — it is how the module is built, and it appears on every screen in the product. The declared method is AACE RP 29R-03, MIP 3.3: observational, dynamic, contemporaneous as-is. In plain terms, windows analysis over the programme’s own successive updates. It does not implement a time impact analysis, an impacted as-planned, or a collapsed as-built; those require a modelled counterfactual, which is a different method and would have to be declared as one.
“Signals for review, not determinations of entitlement. The method is declared: AACE RP 29R-03 MIP 3.3 — windows analysis, and nothing else.”
Why each activity is whose — and the reason is kept
Ownership resolves through a five-layer, first-match cascade, and at every level the reason is recorded. So “why is this the Employer’s?” is answered from stored data, per activity, rather than in aggregate: which layer decided it, which rule, which pattern it matched, at what priority.
That is the difference between “the employer owns 34 days” and a table a reviewer can work through line by line. A limitation we state rather than hide: ownership is resolved once against the latest submission and applied across every window, because responsibility rules have no history of their own.
Concurrency, measured rather than argued
Concurrent delay is the single most contested question in an extension-of-time dispute, so it is measured explicitly. The engine builds the intervals during which employer-caused critical delay was running and those during which contractor-caused critical delay was running, merges each side, and takes the intersection — the days on which both were genuinely active. It measures against every contractor span in the window, not only the ones a claim happens to name.
Net extension of time is then the sum over windows of the employer days less the concurrent days, floored at zero — following the Society of Construction Law Protocol’s preferred approach, in which a concurrent period gives time but not money.
A basis we declare rather than bury
Where a prelims day-rate is not configured, it is derived from the tender prelims allowance divided by the baseline duration. That derivation does not meet the SCL Protocol’s guidance on evaluating prolongation, which calls for the actual cost incurred in the period of delay.
This cannot be fixed by arithmetic — the actual cost is not in the schedule. So every figure declares which basis produced it, and a tender allowance is labelled as a tender allowance rather than presented as an incurred cost.
How it looks
The instruments this module produces, drawn on illustrative data so the method reads clearly.
The four buckets sum to the window’s slip by construction. Unattributed is the residual the analysis could not assign — published, never distributed across the named parties to tidy the arithmetic.
Measured as an interval intersection, not asserted. Net extension of time sums the employer days less the concurrent days over the windows: a concurrent period gives time, but not money.
Delay-damages exposure is integrated over the finish distribution and reported at P80 — a defensible planning figure rather than a single point that will certainly be wrong. The net position can be negative; the owner can be a net payer.
The value
Why it matters
A contemporaneous windows analysis maintained continuously, instead of reconstructed after the fact.
Concurrency measured from the schedule record as an interval intersection, rather than asserted.
Attribution you can explain per activity — which rule matched, which pattern, at what priority.
Notice obligations derived continuously, so they surface before the period expires rather than after.
Capabilities
What you can do
01
Windows, not a single global sum
The programme is divided into consecutive windows, one per snapshot-to-snapshot step, and the slip that occurred in that window is attributed against the programme as it stood at the time — rather than smeared across the whole job from a single end-of-project vantage point.
02
Every window, or no money
There is no sampling and no cap. If any window cannot be compared, the analysis fails and writes nothing rather than publishing a position computed from part of the record. A deliberately expensive choice: a partial windows analysis carries a figure that looks exactly like a complete one.
03
Four buckets that reconcile
Each driving activity’s telescoped day contribution is bucketed as employer, contractor, shared or unattributed, and the buckets sum to the window’s completion slip by construction. Unattributed is published rather than distributed, because a plug spread across the named parties is an invented number wearing the clothes of a measured one.
04
Ownership, with the reason kept
A five-layer first-match cascade — manual override, project responsibility rule, the contractor’s own responsibility activity code, a built-in keyword pass, then default. At every level the reason is recorded: which layer decided, which rule, which pattern it matched, at what priority.
05
Rules snapshotted with the analysis
The exact rule set, overrides and contractual parameters used are stored verbatim on the result — so a figure produced in March stays reproducible after somebody edits the rules in June. Without that, every historical analysis silently becomes wrong the moment the rules change.
06
The signal library
Around eighteen typed signals, each categorised and severity-rated: fragnet insertion, the review-duration trap, pacing, float consumption and sequestration, out-of-sequence trend, trade stacking, productivity degradation, constructive acceleration, critical-path ownership shift, prolongation accrual and more. Each expands to the named driving activities behind it.
07
The lifecycle of a served claim
Eleven stages, each reachable only by the act that earns it: notices with their contractual clocks, completeness against the mandatory heads, entitlement against a clause library, causation, quantum, assigned reviews, negotiation and determination.
The workflow
How it actually runs
- 1
Set the rules and the parameters
Responsibility rules appropriate to the project, plus the delay-damages rate, prelims rate, cap, currency and review periods.
- 2
Run the windows
Every snapshot-to-snapshot step is analysed. If any window cannot be compared, the analysis writes nothing rather than publishing a partial position.
- 3
Review the attribution
Four buckets reconciling to each window’s slip, with the unattributed residual published rather than distributed, and every activity’s ownership reason openable.
- 4
Triage the detected events
Engine-raised events sit in an anticipated register for review, acceptance, or dismissal with a mandatory reason. Decisions survive re-analysis.
- 5
Manage the served claim
Notices and their clocks, completeness, entitlement, causation, quantum, assigned reviews, negotiation and determination — with determined figures set only through a controlled action limited to the project’s approvers.
AI that does the work
How AI changes Claim Management management.
Cannot invent an event
Detected events are raised by the engine from the schedule record. A user, and the model, can only triage them. What a person contributes is judgement, not entries.
Never assesses entitlement
Attribution runs through a deterministic ownership cascade from rules a human wrote. Entitlement is decided by the people appointed to decide it, on the contract and the facts.
The engineer’s judgment stays in charge; the AI removes the latency and the blind spots.
Best practices
- Invest time in the responsibility rules. The defaults get a long way, but the rules are where project-specific knowledge enters, and time spent there improves every subsequent window.
- Set the contractual parameters early — delay-damages day rate, prelims day rate, the cap, currency and review periods. Without them the analysis runs but the money position cannot.
- Import every revision. Windows analysis works on the chain; a gap in the chain is a gap in the evidence.
- Read the unattributed bucket rather than ignoring it. It is the honest measure of how much the record could not explain.
Dashboards & reporting
A print-ready claim justification document — executive summary, delay attribution window by window, the signals register with its evidence, a Basis of Calculation appendix, and the responsibility rules as applied. Two CSV exports accompany it: the money position with a row per sourced input, and the signals register with a row per driving activity carrying its attribution reason.
Live dashboards
Drill-down & filters
Export to Excel / PDF
FAQ
Common questions
Does Zepth Rx decide who is entitled to an extension of time?
No. Everything it produces is a signal for review. Entitlement depends on the contract, the facts and the judgement of the people appointed to decide it — the module computes what the schedules show, records why it reached each conclusion, and presents it for a qualified person to assess.
Which delay method does it implement?
AACE RP 29R-03, MIP 3.3 — observational, dynamic, contemporaneous as-is, which is windows analysis over the programme’s own successive updates. It does not implement a time impact analysis, an impacted as-planned, or a collapsed as-built.
How is concurrency established?
As an interval intersection. The engine merges the periods of employer-caused critical delay and contractor-caused critical delay and takes the days on which both were genuinely active — measured against every contractor span in the window, not only the ones a claim names.
What happens to delay the analysis cannot attribute?
It is published as an unattributed residual. Distributing it across the named parties would produce an invented number that looks exactly like a measured one.
Will an analysis produced today still be reproducible after we edit the rules?
Yes. The exact rule set, overrides and contractual parameters used are stored verbatim on the result, so a figure produced in March survives somebody editing the rules in June.
What can we take away from it?
A print-ready claim justification document with a full Basis of Calculation appendix and the responsibility rules as applied, plus two CSV exports. There is no separate submission bundle.
Sources
- AACE International RP 29R-03 — Forensic Schedule Analysis, MIP 3.3 (observational, dynamic, contemporaneous as-is)
- Society of Construction Law — Delay and Disruption Protocol, 2nd edition
- FIDIC Conditions of Contract for Construction, 2017 (Red Book) — clause 20 claims procedure
Related modules
Zepth is the construction project delivery platform — it runs construction, procurement and asset management on one record, and does the work: reading the drawings, reviewing the submittals, matching the invoices and flagging the risks, with a human sign-off on anything consequential.
See it on your project.
A short, tailored walkthrough on your real workflow — no generic demo.