Migration field report
How 44pixels migrated 1,370 agent reports without forcing the final 127
44pixels inventoried 1,497 reports in its existing Reporter system and verified 1,370 as Handover records, or 91.5%. The migration deliberately held 127 reports for review, kept Reporter available, and left cutover blocked rather than guessing through ambiguous mappings.

Measured result
The cutover stayed blocked on purpose.
- Verified
- 1,370 · 91.5%
- Review hold
- 127 · 8.5%
The migration created a large verified corpus without silently assigning uncertain records. Reporter remains available while the review set is resolved.
01
The starting point
Reporter was 44pixels' internal system for agent-created analysis, slide decks, research, SQL, and supporting files. Its production inventory contained nine product areas, seven folders, 1,497 reports, 12,979 report files, and about 399 MiB of source report bytes. Two folders were restricted and needed separate access review.
- The team needed product-level spaces and folders without flattening existing organization.
- HTML decks had to keep their sibling assets and file navigation.
- Agent authorship and human access needed to resolve to Handover identities.
- Reporter had to remain available until pull, rendering, access, and rollback checks passed.
02
The migration method
The migration used Handover as the canonical destination while treating Reporter as a reversible source. Products mapped to Spaces, folder relationships were retained, and each report became a versioned handover with its files attached to an immutable revision.
- Capture a read-only source inventory before moving content.
- Create deterministic source-to-target mapping rows with exact, inferred, or review confidence.
- Verify the target record exists before marking a report complete.
- Compare SHA-256 hashes for source files where both sides expose a digest.
- Keep uncertain records in review and block cutover while unresolved work remains.
- Inventory
- Map
- Import
- Verify
- Hold
03
What production verification found
As measured on 5 August 2026, 1,370 source reports had verified target handovers. Those handovers contained 1,435 immutable revisions and 41,459 artifacts totaling about 613 MiB in the current Handover corpus. The corpus is larger than the original report-byte inventory because Handover retains revision-scoped artifacts and generated metadata rather than replacing prior state.
- The verified set includes HTML, Markdown, SQL, JSON, JavaScript, CSV, images, PDFs, and other files.
- All 400 product files with comparable source and target digests matched.
- No compared file digest mismatched.
- The 127 unresolved reports remained marked for review.
04
Why the migration is still blocked
Blocked is the correct safety state, not a failed outcome. The migration cannot be called complete while 127 report mappings, ten backlog items, and two restricted folder records still require review. Handover keeps these exceptions visible and Reporter remains available, so the team can resolve them without losing access or inventing provenance.
05
What other teams can reuse
The reusable pattern is to make ambiguity a first-class result. A safe migration needs an inventory, deterministic mappings, content verification, access review, parallel operation, and a cutover gate. Import volume alone is not evidence that people or agents can safely continue the work.
- Publish aggregate progress, but keep private report titles and contents out of the case study.
- Separate imported, verified, review, failed, and skipped states.
- Test both agent pull and human rendering before changing the default destination.
- Retain rollback until every restricted area and high-value legacy link is checked.
Reusable evidence
Inspect the method, not private report content
Disclosure and limitations
First-party evidence, not a customer endorsement
This is a first-party case study. 44pixels Limited operates Handover and operated the Reporter system being migrated.
The publisher and case-study subject are the same organization. No third-party endorsement is implied.
Only aggregate operational counts are published. Report titles, report contents, customer names, individual identities, and restricted folder names are excluded.
- The migration is not complete and remains blocked pending review.
- The results describe one internal production corpus and are not a general migration-speed benchmark.
- Artifact totals include immutable revisions and generated metadata, so they are not directly comparable with source report-file counts.
- Verification proves mapped record and digest checks covered by the migration, not semantic equivalence of every report.
Continue from evidence
Test the workflow before migrating your corpus.
Inspect a public multi-file handover, then use the migration checklist to inventory and verify one bounded report folder.