CNEN
Contact us

DELIVERY & QUALITY

Delivery & Quality

Quality with evidence.Delivery you can trace.

Agree acceptance criteria before production, then verify files, versions and review records at handover. Explore responsibilities, checks by data type and how to read a quality report.

Discuss delivery needsExplore the delivery workflow
Road images · Batch review example

QC-CAM-01

Image-annotation batch

Guideline
GUIDE-1.2
Checked units
120 frames
Initial review passed114 / 12095.0%
Rework recheck passed6 / 6100.0%
Current sample passed120 / 120100.0%

QC-CAM-01 / 120 sampled frames / GUIDE-1.2

01 / DELIVERY WORKFLOW

Eight steps from requirements to acceptance.

Each stage has a defined action and output. Review coverage, responsibilities and delivery cadence are agreed for the project.

  1. Discovery
  2. Assessment
  3. Guidelines
  4. Calibration
  5. Qualification
  6. Production
  7. Quality control
  8. Acceptance

STAGE 01 / 08

Discovery

Understand objectives; Define data types

THE OUTPUTScope and deliverable list
  1. Discovery
  2. Assessment
  3. Guidelines
  4. Calibration
  5. Qualification
  6. Production
  7. Quality control
  8. Acceptance

STAGE 02 / 08

Assessment

Review data volume; Evaluate data quality

THE OUTPUTData assessment and risk record
  1. Discovery
  2. Assessment
  3. Guidelines
  4. Calibration
  5. Qualification
  6. Production
  7. Quality control
  8. Acceptance

STAGE 03 / 08

Guidelines

Define processing rules; Break down data requirements

THE OUTPUTGuideline and exception version
  1. Discovery
  2. Assessment
  3. Guidelines
  4. Calibration
  5. Qualification
  6. Production
  7. Quality control
  8. Acceptance

STAGE 04 / 08

Calibration

Run a pilot batch; Align interpretations

THE OUTPUTPilot and guideline review record
  1. Discovery
  2. Assessment
  3. Guidelines
  4. Calibration
  5. Qualification
  6. Production
  7. Quality control
  8. Acceptance

STAGE 05 / 08

Qualification

Screen specialists; Train and assess

THE OUTPUTTraining and qualification record
  1. Discovery
  2. Assessment
  3. Guidelines
  4. Calibration
  5. Qualification
  6. Production
  7. Quality control
  8. Acceptance

STAGE 06 / 08

Production

Plan production; Assign and schedule

THE OUTPUTBatch plan and production record
  1. Discovery
  2. Assessment
  3. Guidelines
  4. Calibration
  5. Qualification
  6. Production
  7. Quality control
  8. Acceptance

STAGE 07 / 08

Quality control

Layered quality checks; Sample and full reviews

THE OUTPUTReview report and issue list
  1. Discovery
  2. Assessment
  3. Guidelines
  4. Calibration
  5. Qualification
  6. Production
  7. Quality control
  8. Acceptance

STAGE 08 / 08

Acceptance

Prepare deliverables; Check against criteria

THE OUTPUTAcceptance and version manifest
Every step leaves a record.

02 / QUALITY CONTROLS

Who reviews what.

Separate review responsibilities while linking results to the same guideline and data version. Use automated checks and human judgment for the questions each can answer.

01

Production review

Check completeness, classes and attributes against the guide. Record and address obvious issues first.

Record: self-review and open questions
02

Team review

Cross-check ambiguous classes and boundary cases. Feed unclear definitions back into the guide.

Record: review notes and rule clarifications
03

Specialist QC

Review the agreed scope and dimensions. Record issue location, type, severity and affected units.

Record: issues and recheck results
04

Handover review

Check files, versions, formats, sample scope and open issues before a delivery decision.

Record: acceptance checklist and manifest

QUALITY BY DATA TYPE

A shared process. Task-specific checks.

Agree on review units, sampling and acceptance criteria before production. Image counts, audio duration, text entries and paired samples require different checks.

Images

Class consistency, boundary accuracy, omissions, text accuracy and duplicate images.

View task

Video

Identity switches, broken tracks, event boundaries, timelines and clip completeness.

View task

Speech & audio

Transcription accuracy, speaker consistency, time boundaries, clipping, silence and noise.

View task

Text

Entity boundaries, label consistency, complete relations, encoding and duplicate content.

View task

3D & point clouds

Position, dimensions, orientation, point classes and coordinate consistency.

View task

Multimodal

Pair completeness, semantic consistency, timing, missing modalities and shared IDs.

View task

Each delivery states the data version, review scope, findings, rework and unresolved items. A review rate applies only to the stated units and sample.

03 / REVIEW REPORT

Read the result. See the denominator.

The following three road-vision and sensor batches illustrate initial, rework and current results. These rates describe the listed review units, not model accuracy or the status of unchecked data.

QUALITY REVIEWBatch review record

QC-CAM-01

Road-image annotation batch

Sample recheck complete
Review guideline
GUIDE-1.2
Checked units
120 image frames
Review scope
120 frames sampled from a batch of 600. Initial and current results refer to the same checked sample.
Sample composition
Review composition: 40 frames each from urban, junction and tunnel scenes.

Initial pass rate

95.0%

114 / 120

Initial passes / checked units

Rework recheck pass rate

100.0%

6 / 6

Recheck passes / initial failures

Current sample pass rate

100.0%

120 / 120

(Initial + recheck passes) / checked units

Review result counts

  • Passed initially114
  • Passed after rework6
  • Still unresolved0

Each unit is grouped by one primary issue and counted once. Current sample pass rate = (initial passes + recheck passes) / checked units.

Sample recheck completeAll 120 checked units currently pass. The remaining 480 frames are outside this result. Delivery still requires agreed acceptance.

Issues and recheck records

Issue groups summarize scope and corrections. Detailed records link each unit to its location, guideline clause and revision.

Road-image annotation batch — Issues and recheck records
Primary issueAffected unitsCorrectionRecheck result
Missing objects3 framesAdd objects and classes3 / 3 passed
Loose boundaries2 framesCorrect box boundaries2 / 2 passed
Occlusion inconsistency1 frameAlign the attribute to the guide1 / 1 passed

From the issue to the revision

Keep a reference chain for every correction so a reviewer can return to the same data unit. List unresolved issues separately and agree whether to rework, exclude or defer them.

  1. 01

    Locate

    Unit ID + issue location

    CAM-018
  2. 02

    Record

    Issue type + guideline clause

    ISSUE-006 / §4.2
  3. 03

    Correct

    Change + data revision

    r01 → r02
  4. 04

    Recheck

    Result + open issues

    RECHECK-006

04 / ACCEPTANCE & HANDOVER

Prepare the evidence with the data.

A handover includes more than data files. Bring guidelines, versions, review scope and outstanding items into one manifest for verification, reuse and traceability.

Suggested acceptance checks

  1. 01

    Scope and counts

    Check batch IDs, file counts, valid units and exclusions.

  2. 02

    Guidelines and versions

    Align task rules, class lists, revisions and export versions.

  3. 03

    Format and usability

    Check directories, fields, references and sample readability.

  4. 04

    Quality and issues

    State the review scope, calculations, rechecks and open issues.

  5. 05

    Receipt and acceptance

    Record receipt, outstanding items and the agreed acceptance result.

PROJECT ARRANGEMENTS

Agree access and retention at kickoff.

These are proposed topics for project planning. The actual arrangements depend on the agreement and the operating environment.

Access scope

Define project roles, accessible data, allowed actions and authorization periods.

Data handling

Confirm provenance and permitted use; agree handling and masking of sensitive fields.

Retention and handover

Agree transfer, retention, receipt checks and recorded disposition at project close.

QUESTIONS

Delivery questions, answered

01What if a batch still has unresolved issues after recheck?

Keep the open issues and affected scope visible. Agree further rework, exclusions or staged delivery. The LiDAR example on this page remains on hold.

02Can we use our own guidelines and quality tools?

Assess the guidelines, fields, checks and export formats, then validate them on a small batch. Confirm platform access and permissions separately.

03What happens when guidelines change during production?

Record the reason, applicable version and affected batches. Agree whether existing data needs rework and assess the impact before continuing. Keep review results tied to their guideline version.

START WITH THE ACCEPTANCE CRITERIA

Align on acceptance before production.

Start with the data type, expected volume and an existing guideline to shape review criteria and batch delivery.

Discuss delivery needs
Explore data annotationExplore customer cases