Skip to main content
Quanta Meridian logo

Quanta Meridian Builds · one tested Build

Runnable tools with the human decision left visible.

This example includes its code, purpose-built test documents, workflow checks, reviewer action and operating limits.

Inspect Document and Request Triage

One record through the boundary

The tool prepares. The reviewer releases.

Hold the credit note until a reviewer confirms the posting sign and specialist queue.

  1. 01

    Generated record

    DOC-CRN-000417credit note · 9 retained fields
  2. 02

    Suggestion retained

    Credit noteOriginal queue: intake review
  3. 03

    Release held

    2 blocking rulesAmount sign and mixed document heading require a person.
  4. 04

    Reviewer edits

    -£10,946.65accounts payable analyst
  5. 05

    Audit retained

    9 trace eventsFinal queue: accounts payable adjustments

Runnable build, visible judgement

Let the tool prepare the record without taking the decision away from the reviewer.

A runnable demonstration is useful evidence, but it is not a claim of operational deployment. These four controls define what the evidence establishes.

  1. 01Runnable

    The retained Python package, fixed environment, generator, rules and verification commands rebuild the contained workflow.

  2. 02Evaluated

    900 generated test documents retain classification, extraction, abstention and known policy failures.

  3. 03Reviewable

    4 cases retain 4 human decisions without replacing the original suggestion.

  4. 04Bounded

    The data-generation, calibration, policy and production limitations remain beside the decision to publish or hold the result.

Build evidence comparison

One Build is documented across every evidence dimension.

The comparison register shows the evidence dimensions needed to assess a runnable Build. It currently contains one inspectable implementation.

Evidence dimensionDocument and Request Triage
Business routineMixed Finance and Operations document intake with a named reviewer before specialist routing.
Generated population6,000 documents · 6 classes · 68 template families
Retained evaluation900 test documents · 770 accepted suggestions · 130 held for review
Human review record4 cases · 4 decisions · 31 checksum-chained events
Featured traceDOC-CRN-000417 · edit · accounts payable adjustments
Method boundaryOperational pilot not approved · 0 external API calls

01Current Build

Contained Python workflow

One contained Python build, from generated page to reviewer decision.

The build prepares a class, fields and specialist queue for mixed Finance and Operations documents. Contradictory or incomplete records are held for an intake analyst, and the original suggestion remains visible after correction.

Inspect Document and Request Triage
Generated corpus
6,000
documents
Generated test
900
documents
Held by policy
130
records
Audit history
31
events
Generated non-client credit note DOC-CRN-000417 showing the positive displayed amount £10,946.65
Generated source pageThe generated page deliberately combines invoice and credit-note wording and displays £10,946.65 as a positive amount.Open generated credit-note evidence

Held suggestion · reviewer edit

DOC-CRN-000417

The page displays £10,946.65. The reviewer retains the credit-note class and records -£10,946.65 before the record enters the adjustments queue.

rule-credit-amount-sign
The credit amount is missing or does not have a negative sign.
rule-document-type-ambiguous
The document heading deliberately combines two business-record types.

3 workflow checks passed · 31 audit events independently reconciled

Contained local reviewer screen showing the generated mixed-document queue, received page, suggested class and fields to check
Contained reviewer workflowThe retained local screen shows the mixed intake, received document, original suggestion and fields a reviewer must inspect. It is a contained workflow, not a hosted service.Open reviewer-screen evidence

Retained generated evaluation

Strong synthetic results, with two known policy escapes.

These measurements describe the deterministic generated test set. They do not establish accuracy on real supplier files, scans or handwriting.

Class accuracy
100%
Policy coverage
85.6%
Document extraction exact match
98.6%
Review-decision accuracy
98.9%
Risk-labelled items accepted
2

Two invalid-date reporting requests passed the retained policy. A stronger chronological rule is required before any pilot.

Wrong accepted routes
0

Method boundary

A contained Build is not a production service.

Built by Quanta Meridian using deterministic non-client documents. The generator, seed, model, rules, test outputs and reviewer decisions are retained.

Operational pilot
not approved
External API calls
0

Known limits retained with the release

  1. 01

    All documents are deterministic generated records; the results do not measure performance on real supplier files, scans or handwriting.

  2. 02

    The retained scores are not calibrated probabilities. Calibration was measured but not selected because no class error occurred in the retained validation partitions.

  3. 03

    Two invalid-date reporting requests passed the retained review policy and require a stronger chronological rule before any pilot.

  4. 04

    The local service and reviewer workflow do not prove production access control, retention, monitoring or cloud deployment.

Related Solutions

SolutionAutomation

Design controlled routines that retain failures, retries and human decisions.

SolutionData Quality

Make release-blocking rules and failed records visible before reporting or routing.

SolutionAnalytics Engineering

Retain data contracts, transformation logic, validation and rebuild instructions.

Start with the routine

Bring one intake routine that depends on manual sorting.

Start with the documents, the fields people copy, the queues they choose and the conditions that should always require human review.

Contact Quanta Meridian