Analysis

every engine, run against the real inputs

Peterson Technology PartnersAnsco & Associates · proof of concept
11 ENGINES · 4 ON REAL DOCUMENTS
RULE SETS: PROVISIONAL — PENDING AT&T SPEC

Every card below runs the actual engine against the actual inputs and prints what it printed — no recordings, no mock-ups. Where the inputs are synthetic they say so, and where an engine refuses to answer, the refusal is the output. The first group is the point: those answers come out of the close-out capture itself, with nothing extra collected.

the review desk → what the technician sees →
From the capture, with nothing new collected

These read the records close-out already created. No extra visit, no extra form, nothing asked of a crew that is not already being asked.

Claims tie-back

“A pipe failed. Were we there, and was it us?”

synthetic data, real shape

Places a subcontractor 7 m from the reported point two years earlier, working outside their locate ticket, with a bore 6 inches under spec.

~1s

Qualified person

“Was whoever did this work ticketed for it, on the day?”

synthetic data, real shape

A climber working 21 days after his climbing ticket lapsed; a flagger 268 days past.

~1s

Phone as MFA / presence

“Who was on site, where, and for how long?”

real documents

Begin and end of day derived from close-out captures — evidence the crew creates by doing the job, not a button pressed in the van.

~2s

Safety screening

“What do the close-out photos already show about safety?”

real documents

Reads safety conditions off photographs the crew is already taking, so it costs the field nothing. Never identifies anyone — counts people, reports conditions.

~60s
From AT&T's paperwork

Drawings against contracts and invoices, the wireless document set against itself, and CD PDFs into GIS features.

Contract agent

“Does the invoice match the drawing and the price sheet?”

real drawing, synthetic rates

Finds a rate written per 100 LF but invoiced per LF, and four handholes built and billed to nobody.

~1s

Splice and test validation

“Do the fibre test results actually meet the budget?”

synthetic data, real shape

Catches a link measuring better than physically possible — the two documents were taken on different fibres.

~1s

PDF to GIS

“Can an AT&T CD PDF become GIS features?”

real documents

118 features placed from a drawing set holding no coordinates, self-checked at 1.27% drift. Refuses the second set, in words, because it carries no station equations.

~25s

Completed work vs invoiced

“Did we bill everything the drawing says we built?”

real documents

$22,100 unbilled on one job, read off a BOM sheet with four words of text.

~40s
The wider operation

The questions Chris and dispatch asked for, and the format problem that sits between Ansco and every customer.

Fleet utilisation

“Does Chris need more trucks?”

synthetic data, real shape

Savannah carries three more than its work needs while Atlanta North is three short. A transfer, not a purchase.

~1s

Routing and dispatch

“What fits in the day, around the access windows?”

synthetic data, real shape

Exact search for the largest feasible set, and it names what will not fit rather than quietly dropping it.

~3s

FieldSense translation

“Can one capture land in every format we owe?”

real drawing, synthetic rates

92 captures into two destination formats from declarative mappings. A new customer is a JSON file. Fails loudly and names the field when one cannot be filled.

~2s

The rule sets behind these are our best reading of AT&T’s requirements, not their published spec. Ansco’s own rejection history is what sets where the line actually falls — it calibrates the judgement, it does not create it.