# Research map source trace

Purpose: distinguish the shared approach from the specific public examples used to illustrate it.

The author describes question, scope, novelty checking and analysis as the architecture underlying the portfolio. The map summarises that approach. The retained Deliverable Filter and IDEACHECK method documents support route selection, prior-work checking, source assessment, reproducible calculations and dated follow-up. These method instructions do not by themselves prove historical execution. The public project links illustrate particular stages. The continuing loop also includes the question queue, record alignment, monitoring, revision and retained archive branches. No public queue-entry or alignment-receipt example is claimed.

| Stage | Public evidence | What the example supports |
| --- | --- | --- |
| Selection: question | [Power-demand entry](index.html#work-power), [SW case](sovereign-watch-case-study.html) | Named research and operational questions. |
| Selection: scope and method | [Forecast entry](index.html#work-forecast), [Article 50 entry](index.html#work-article50) | Comparable units and horizons; a historical public-evidence boundary. |
| Selection: novelty and usefulness | [FineWeb-Edu entry](index.html#work-fineweb) | A question about a specific threshold mechanism. This is not a completed novelty-search record. |
| Research: source evidence | [Infrastructure source catalogue](https://github.com/NMAIResearch/ai-constraint-relay/blob/6a557031b8a97181d6e19586574979342bbde3aa/v7.1/SOURCE_CATALOGUE.json), [SW source trace](SOURCE_TRACE.md) | Source identities, inspected locations and capture-state limits. |
| Research: analysis | [Forecast scorecard release](https://doi.org/10.5281/zenodo.20572928), [FineWeb-Edu entry](index.html#work-fineweb) | Recalculation of a scorecard and chart from one CSV, and inspectable threshold effects. |
| Research and revision | [IPO Ledger release](https://doi.org/10.5281/zenodo.20672581), [Model Dependency entry](index.html#work-model) | A prediction graded as failed with its earlier specification defect stated, and correction of claims. |
| Publication and supporting records | [Infrastructure v7.1](https://zenodo.org/records/22913848), [CEO Pay scorecard release](https://doi.org/10.5281/zenodo.20680108), [Registry documentation](https://github.com/NMAIResearch/divergence-registry/blob/ad2ca5f07db62dfebaa84f890cd48b6457a959bd/README.md) | Versioned releases and a component for recording methodological decisions. Registry is not evidence of release alignment. |
| Monitor and reuse | [Infrastructure v7.1](https://zenodo.org/records/22913848), [IPO Ledger release](https://doi.org/10.5281/zenodo.20672581) | Reuse materials are supplied, and a schedule of dated checkpoints is published. Neither establishes a later reuse or grading run. |
| Archive branches | [Selected route decisions](RESEARCH_ROUTE_EXAMPLES.md) | Retained decisions distinguish a prior-work stop, a folded publication, a parked skeleton and a superseded design. The private records are summarised without publishing their working trails. |

Homepage descriptions are retained from [site revision 452f6ed242617865109ca2e9dc97123c5566405b](https://github.com/NMAIResearch/NMAIResearch.github.io/blob/452f6ed242617865109ca2e9dc97123c5566405b/index.html). They are reused navigation evidence, not a fresh audit of the underlying papers. Source-code tracing for SW and the infrastructure release's review records are also reused.

The Forecast scorecard, IPO Ledger and CEO Pay scorecard examples were added on 28 September 2026. Their latest Zenodo records were downloaded with matching MD5 checksums, and their documented commands were run: the Forecast chart and the CEO Pay scorecard regenerated byte-identical to the shipped files, and the IPO Ledger grading ran from the bundled prices without network access. These checks cover the scripts, not the underlying sources.

The diagram shows a main route with return paths for rescoping and continuation. A study can revisit earlier stages and span several of the linked examples. Alignment means comparing the repository and deposit with the project-status record, not treating a methodological decision register as proof of agreement. Reframing or parking follows insufficient purpose, feasibility or evidence; a disconfirming finding is not a reason to suppress a result. The method allows descriptive work, replication and synthesis. The novelty stage is not a claim that every worthwhile study must discover something new.

No comparative workflow baseline, new source survey, fresh model experiment or full Lifecycle run was performed for this map. The script reproduces the presentation only. The author's choices and model/script assistance are distinct in the guide's disclosure.