Structures a written site report
Turns pasted free text into quantities, trades, manpower and delay reasons — extraction and structuring, never judgement.
Zepth Rx · Schedule Intelligence
Field Data & Cost Actuals
A programme without a site record is a plan. With one, it becomes a measurement.
Last updated
Zepth Rx module
Field Data
AI agent built into the module
Structured daily reportsDrafts separated from publishedA live work-front checkThree ways in
Overview
Every analysis in Zepth Rx gets sharper with evidence from the ground. The forecast learns your uncertainty from completed work, and learns it better from measured productivity. The disruption and trade-stacking signals are computed from manpower on site. The credibility grade on a recovery programme asks whether a duration cut is supported by the resource actually deployed. Earned value has no cost performance at all until somebody records what was spent.
Field Data is where that evidence enters. It is not a reporting module — it is the input side of every other module in the product. Daily reports and non-conformance records are captured or imported here, in Zepth Rx itself.
Progress inferred from the site, not just reported
Where quantities are tracked, Zepth Rx compares progress inferred from installed quantity against the percent complete the contractor has reported in the programme — and can write the inferred figure back.
This is one of the few places a platform can catch optimistic progress reporting with arithmetic rather than argument. An activity reported 70% complete whose installed quantity says 45% is a discrepancy with a number attached.
The measured mile
Comparing a trade’s achieved rate in a disrupted period against its own rate in an undisrupted one is the most respected disruption method precisely because the comparison is internal. The trade is its own control, so the argument does not depend on an industry benchmark anyone can contest.
It only works if the undisrupted period was recorded while it was happening. That is the whole case for capturing quantities from day one, on a project where nobody yet believes there will be a dispute.
Coverage, declared
Where field data feeds an engine, the coverage is published: how many activities had quantities, how many had reports, how many observations resulted, and how many reports could not be matched. An unmatched report is reported as unmatched.
This matters more than it sounds. The absence of field evidence and the presence of no delay look identical in a result — only a coverage count tells them apart.
How it looks
The instruments this module produces, drawn on illustrative data so the method reads clearly.
- Concrete
- Blockwork
- MEP
- Finishes
The programme assumed a crew. The site recorded what turned up.
Non-conformances cross-tabulated by trade and zone, so a concentration reads as a pattern rather than a list. Repeated non-conformance in one cell is a rework problem and usually a productivity problem too.
The trade is its own control. Comparing a trade’s achieved rate in a disrupted period against its own rate in an undisrupted one — which is why the method does not depend on an industry benchmark anyone can contest.
The value
Why it matters
Disruption claims are won on productivity records and lost without them — and the undisrupted period has to have been recorded at the time.
Contemporaneous site records are the highest-grade evidence in a construction dispute, and their value depends almost entirely on having been captured contemporaneously.
Inferred progress catches optimistic reporting while it is still a discussion rather than a variance.
Quality hot-spots turn scattered non-conformances into a pattern you can act on before the rework lands on the critical path.
Capabilities
What you can do
01
Structured daily reports
A filed record rather than a free-text log — auto-numbered, with shift, working hours, subcontractor, permit number, work area, and the weather a delay analysis will later need. Each carries manpower, work summary lines with quantity installed and its unit, materials, equipment, photographs and attachments.
02
Drafts separated from published
Every engine reads only what has been published, so a half-finished report cannot leak into an analysis — a distinction enforced in the code rather than by convention.
03
A live work-front check
As a report is filled in, the form resolves the chosen trade and location against the active schedule and warns when the combination would credit no activity — at the point of entry, while the person who knows the answer is still looking at the screen.
04
Three ways in
Enter directly, import from Excel or CSV with duplicate detection and location-alias resolution, or paste free text and let an extraction path structure it. The model extracts and structures; it never decides whose delay it was.
05
Non-conformance records
Severity, zone, trade and date, entered directly or imported — cross-tabulated by trade and zone so a concentration reads as a pattern rather than a list, and feeding rework days into the mined risk register.
06
Productivity and the measured mile
Where quantities are recorded, weekly achieved rates are derived per trade — units installed per man-day. That underpins the measured-mile comparison of a trade’s rate in a disrupted period against its own rate in an undisrupted one.
07
Cost actuals
Imported, or recorded as certified amounts against bill pay items through a payment-certificate round-trip. Installed quantity is split by quantity, not by money — and anything that cannot be attributed is reported, never dropped.
The workflow
How it actually runs
- 1
File the report
Directly, by Excel or CSV import, or by pasting free text for extraction. The work-front check warns at entry when a trade and location combination would credit no activity.
- 2
Publish it
Engines read only published reports, so a draft cannot reach an analysis.
- 3
Categorise the delay reasons
Free text is mapped to one of eleven fixed conditions. The model never indicates who is responsible.
- 4
Record cost and certified quantity
Import actual cost, or run the payment-certificate round-trip — export a blank certificate, fill in certified amounts and installed quantities, import it back.
- 5
Feed the engines
Productivity, measured mile, inferred progress, deployment against plan, and the mined risk register — each publishing its own coverage.
AI that does the work
How AI changes Field Data management.
Categorises the condition, never the party
Maps “waiting on approval” to a fixed condition. The instruction is absolute: it must never indicate who is responsible.
Mined drivers arrive disabled
Categorised conditions and rework days feed the mined risk register, and every mined driver arrives switched off until a person enables it.
The engineer’s judgment stays in charge; the AI removes the latency and the blind spots.
Best practices
- Record quantities against work. It is the highest-value field to capture and the one most often skipped — without it, productivity, measured mile and inferred progress all report zero coverage instead of contributing.
- Map locations to the schedule’s work breakdown or location structure, so a report credits the right activity. A mis-filed report discovered three months later during an analysis is unrecoverable data.
- Publish reports rather than leaving them in draft. Engines read only what has been published, by design.
- Capture the undisrupted period as diligently as the disrupted one. A measured mile needs a baseline that was recorded at the time, and nobody can reconstruct it afterwards.
Dashboards & reporting
The daily-report register with date, zone and trade filters and the full report editor; the non-conformance register; and the cost-data screen with actual-cost import and the payment-certificate round-trip. Field data also surfaces as labour deployed and quality hot-spots, and as actual against planned progress sliceable by work breakdown, location or trade.
Live dashboards
Drill-down & filters
Export to Excel / PDF
FAQ
Common questions
Where do daily reports and NCRs live?
In Zepth Rx itself. They are captured or imported here — Rx is a standalone platform and does not read a site record from anywhere else.
Can the AI decide whose delay it was?
No, and it is instructed never to indicate responsibility. It maps a free-text note to a fixed condition. Attribution is done by a deterministic ownership cascade, from rules a human wrote, with the reason stored.
What is the single most valuable field to capture?
Quantity installed, with its unit. Productivity, the measured mile and inferred progress all depend on it, and without it they report zero coverage rather than contributing.
Can a half-finished report affect an analysis?
No. Drafts are separated from published reports and every engine reads only what has been published — enforced in the code, not by convention.
How do certified amounts get onto activities?
Through the same bill-to-programme bridge that carries budget, by the same proportional rule. Installed quantity is split by quantity rather than by money, and anything that cannot be attributed is reported rather than dropped.
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.