IBM TechXchange 2026 / Dev Day Hackathon

AI can write the change.
Prove it deserves to ship.

ChangeProof turns a deceptively simple brownfield maintenance request into a reviewable chain of evidence — across Node.js, RPGLE, CLLE, DDS, Db2, tests, and operational documentation.

IBM Bob 2.040.12 BobcoinsIBM i brownfieldSolo build
IBM iRPGLECLLEDDSDb2Node.jsJestEvidence engineering

01 / The problem

The ticket is not
the blast radius.

Brownfield changes hide duplicated rules, stale assumptions, downstream schedules, incomplete tests, and target-platform validation that a code generator cannot honestly claim from a laptop.

01

Requested

Give Preferred customers two additional hours for expedited orders.

16:00 → 18:00
02

Discovered

The fulfillment batch begins at the new cutoff itself — a timing collision the ticket never mentions.

SCDTIME(180000)
03

Proved

Local behavior is re-tested; IBM i execution remains explicitly bounded until target validation occurs.

18:15 / TARGET_VALIDATION

02 / Before → after

A change lifecycle you can inspect.

The preserved baseline captures the pre-change failure state. The final source tree is the remediated state. ChangeProof keeps both stories visible without pretending they are the same thing.

BASELINE

Before CHG-0042

OPEN
14passing
2failing
3IBM i skips
12blast radius
  • Preferred customer @ 17:00 rejected
  • 18:00 batch collision inferred
  • Operational docs stale
  • RPG / CL runtime unverified
Inspect preserved baseline ↗
POST-CHANGE

Evidence after remediation

REVIEWABLE
16passing
0failing
3IBM i skips
11blast radius
  • Preferred customer @ 17:00 accepted
  • Batch moved to 18:15
  • Docs + customer class corrected
  • IBM i claims remain bounded
Inspect final evidence pack ↗

03 / Evidence model

Three questions.
No confidence theater.

ChangeProof deliberately separates how a finding was established, where remediation stands, and where final validation must occur. That keeps source observation from masquerading as runtime proof.

HOW DO WE KNOW?

evidenceBasis

EXECUTED_LOCAL

Actually ran here.

OBSERVED_SOURCE

Directly visible in code or docs.

INFERRED

Derived across observations.

WHERE IS THE FIX?

status

OPEN

Still unresolved.

RESOLVED

Current target proves it.

TARGET_VALIDATION_REQUIRED

Edited, but not target-proven.

WHERE MUST IT RUN?

validationTarget

LOCAL✓Node / Jest / SQLite surrogate
IBM_I○RPG compile / CL execution / Db2 runtime

“An RPGLE source edit can be observed without being falsely presented as production-validated.”

ChangeProof design principle

04 / From the green screen out

Built for systems that
aren’t becoming React apps tomorrow.

ORDERPRO is fictional, but the maintenance problem is not. ChangeProof treats IBM i as a first-class target: RPGLE and CLLE are analyzed alongside the modern API, Db2 for i DDL stays separate from the local SQLite surrogate, and target-only claims remain explicitly unverified.

RPGLECLLEDDSDb2 for iQSYS2XMLSERVICE
CHANGEProofEVIDENCE RUNTIME
CHANGE_REQUEST.mdsemantic lens
↓
ANALYZE · EXECUTE · CORRELATE · DIFFgeneric analyzers + test evidence
↓
LOCALNode · Jest · SQLite
IBM_IRPG · CL · Db2
↓
EVIDENCE PACKHTML · Markdown · traceability JSON

Built with IBM Bob 2.0

Not autocomplete.
The build environment.

40.12Bobcoins consumed
1continuous retained task
50+files touched in Bob session

Bob was used for planning, polyglot authoring, test generation, document understanding, analysis-engine implementation, evidence generation, and iterative remediation. The repository includes the retained Bob task-session summary and supporting context.

Inspect Bob session evidence ↗

Release question

What changed?
What did we prove?
What still needs the target?

That’s the job ChangeProof is built to make answerable.