GitHub/ODC migration demo: Start at demo/README.md for the new customer-profile implementation, three releases, validation, native ODC handoff and reports. Demo evaluation separates implementation, local checks, GitHub Actions and live ODC/Oracle evidence. Historical research below retains its original conclusions.
Latest product comparison (7 October 2026): ODC versus Bytebase for practical adoption. The new research mission benchmarks Bytebase's evidenced daily workflow: ODC scores 80/100, has a moderate product gap and merits a focused adoption POC; production hardening risk remains HIGH. This documentation-only recommendation uses the practical workflow rather than the earlier custom-controller threshold. Earlier decisions and the separate CI implementation/runtime task below remain preserved; no SQL or new product test was run.
Current implementation task (7 October 2026): OSS-only single-target POC. The user has ruled out commercial software and selected reuse of the existing CI plus one Oracle engine. The runner/tests/reports are implemented; runtime results and blockers govern this task. The research decision records below remain historical evidence, with no result rewrites.
Start with the current baseline. It is the authoritative goal, requirements, architecture classification, candidate status and next decision, consolidated on 7 October 2026. Read it before extending research or interpreting an older shortlist.
For the actionable decision, read final-decision.md. The decision-closing round recommends a gated Bytebase EE native regression when commercial terms and supported controls are viable; otherwise select the existing CI + Oracle-engine contract POC. ODC remains HOLD, and AccessFlow remains a reuse-dependent reserve without a demonstrated simplicity advantage. The document contains the decision tree, exact entry/stop gates and seven internal facts that can change the branch. It grants no runtime or purchase authorization.
The project seeks the simplest maintainable centralized Oracle workflow improving DBeaver/manual SQL and standalone Git/Flyway CLI for approximately 50 instances and 20 users, preferably free OSS and self-hosted. CI/CD is an integration requirement; Jenkins and Flyway are optional.
An integrated platform is preferred, but no OSS adoption winner is proved. Four decision options remain: ODC native (HOLD), AccessFlow deployment API + engine (RESERVE), Bytebase EE (COMMERCIAL_CHECK), and CI/Flyway (FALLBACK). No new runtime POC is currently eligible. Next establish whether suitable self-hosted EE entitlement/budget and an exact licensed build are available for the actual estate; FREE failures do not decide EE correctness.
- Current baseline: problem, workflow, evidence, requirements, pain-to-capability mapping, decision model and timeline.
- Rejection register: why proposals were stopped/deprioritized and what evidence would reopen them.
- Decision questions: public facts closed and all original questions classified as public, internal, commercial, runtime or no longer relevant.
- Evaluation evidence index: detailed reports, source reviews, licensing, preserved acceptance and history.
| Record | Exact scope | Retained results |
|---|---|---|
| ODC API/UI POC, 4 October 2026 | ODC 4.4.1-20260116; Oracle 26ai Enterprise Edition 23.26.4.1.0; four environment schemas on one database; local TCPS wrapper | 12 PASS / 20 PARTIAL / 5 FAIL / 8 NOT_RUN |
| Bytebase API/UI POC, 4 October 2026 | Bytebase 3.22.1/FREE, commit a85f6cb4195299995e8554303550d672d5093e1d; same lab, not EE regression | 17 PASS / 14 PARTIAL / 4 FAIL / 3 BLOCKED / 7 NOT_RUN; extra SQL-review FAIL outside 45 totals |
Runtime comparison, 45-case Oracle contract, criteria, workload hashes and engine/source evidence retain their original case-level meaning. DOC, SOURCE, RUNTIME, DESIGN, UNKNOWN and NOT_RUN are separate labels. There are no new benchmarks or verified 50-instance load results.
These are dated supporting records. Their “current,” “next,” Category A/B/C and POC A/B wording belongs to the stage in which they were written; the baseline reconciles it. Older next-step recommendations are not additional active work.
| Generation | Documents |
|---|---|
| Initial research and product model, 3–4 October | Original report, brief, product constraint, product comparison, UI evidence, forks, licenses, Oracle coverage, maturity |
| Platform evaluation/T00–T04, 4 October | Evaluation report, old shortlist, frozen history, platform plan |
| Deep source/license/artifacts, 4–7 October | Round-two index, CloudDM artifact decision, ODC survey |
| Catalog/composition, 7 October | Catalog, solution shortlist, reference architecture, POC plans |
| Validation/gates/reassessment, 7 October | Design validation, decision gates, architecture reassessment |
The consolidation and final decision are documentation only. Historical evidence, expected results, workloads and source working files remain unchanged. Run the existing read-only document check with python tools/verify_research_documents.py; it checks research artifacts, not product acceptance.
The repository is hosted at devsecopslonghn/DB-Research. Upstream source snapshots under evidence/source/, evidence/additional-source/ and evidence/product-model/{sources,forks}/ are Git submodules pinned to the existing evidence commits. Restore them before checking source references:
git clone --recurse-submodules https://github.com/devsecopslonghn/DB-Research.git
cd DB-Research
python3 tools/verify_research_documents.pyFor an existing checkout, run git submodule update --init --recursive. Keep the recorded submodule commits; git submodule update --remote would replace the evaluated snapshots. Local credentials, environment overrides, caches and generated runtime output are excluded by .gitignore. Historical reports and retained evidence keep their existing contents.