# Reference-Figure Mapping and Neutral Background Use [scientific-figure-workflow.md](scientific-figure-workflow.md) for reference selection and construction. Keep the mapping below in existing notes/sources; Master attestations, frozen JSON and readiness flags apply only to the legacy compatibility section. ## Map the Scientific Job Before Generation After question/evidence planning, open the reference and caption, resolve its actual figure label and compare its structure before fixing plot types. Record: - source identity, venue/year, exact page/figure locator, retained viewable asset and access/use basis; distinguish a knowledge source from a design exemplar; - the reference's panel roles, marks/encodings, hierarchy, reading order and useful annotations; use panel bounds when needed to identify a crop or mapped region; - the current question, supported claim, boundary, manuscript/Results units, intended conclusion and hero panel; - each current panel's evidence anchor, data fields, units, transforms, role, intended reading, semantic conversion and retained domain differences; - the concrete structural features that will transfer, where they belong in the current drawing and which source content must not transfer. Cover each current panel and identify each exemplar's structural contribution; a bibliography entry or similar chart mark alone is not a mapping. Transfer linked comparisons, uncertainty, scales or process layers where they serve the question, retaining a simple design when it suffices. Never present reference data, values, results, wording or patient information as this study's own. Actually read, attributed literature may support conceptual entities/relationships, distinct from observations and hypotheses. Preserve user results and `materials_only`; learning does not authorize analysis or reruns. If no suitable lawful reference can be viewed, record original design and the specific limitation; continue without claiming reference-guided completion or backfilling guidance. Reuse valid learning for minor edits and revise the affected mapping openly when the question or evidence changes. ## Trace the Chosen Figure and Compare the Right Sources Connect the selected export/source to actual transfers, panel evidence, caption and manuscript use. Repair lost explanatory layers before cosmetic polishing. Keep these comparison purposes distinct: | Comparison | What it can establish | | --- | --- | | External published reference / current figure | Whether the viewed exemplar's relevant scientific structure and visual grammar were learned; label the external source and exact figure locator | | Whole concept prototype / reconstructed final | Whether supported structure survived and prototype errors were corrected | | Previous study figure / revised study figure | Whether the requested edit improved the figure without regressions | Do not label this study's earlier figure "Reference / existing figure" or claim it demonstrates published-reference learning. Reference sheets and prototypes are learning aids, not selectable results. Show complete sides simultaneously and legibly, adding attributed detail crops when needed. A toggle or context-hiding crop does not suffice. Inspect the sheet for clipped titles/labels and retain direct views for zoom. Check the selected figure at actual manuscript size in each requested format. ## Background Rule Prefer transparent editable canvases and pure-white publication backgrounds. Another background can be justified by the user, venue or scientific encoding; inspect actual contrast and publication-size readability. Avoid decorative tints, gradients and textures. Semantic marks, heatmap cells, mechanism regions and informative highlights may use color. A pixel/SVG probe can assist this inspection but cannot establish scientific quality; do not claim a probe that was not run. ## Reuse of Point/Line Grammar Point, line, interval, ROC, forest, heatmap, flow and network marks are neither advanced nor primitive by themselves. Reuse a family when the questions and data structure warrant it, retaining stable meanings and noting the reason when useful. Do not manufacture variety by changing statistical semantics or repeat the same result merely to create more figures. ## Recheck Affected Work Recheck affected mapping/comparisons when source, reference, evidence, story, panel map, candidate, selected asset, editable source, caption, Results link or placement changes. Preserve unaffected checks/choices. Dependency changes matter even with unchanged filenames; unrelated task/installation versions do not invalidate the figure set. ## Legacy compatibility: explicit historical engine maintenance only The old engine contracts below retain their own schemas and dependency checks. Use them only for an explicitly maintained historical task; they do not create current-host approval or file-work prerequisites. - `paperspine5.figure-reference-plan/1.0`: before candidate generation, the Master owned the plan, authority attestation and self-hash, bound from `figure_requests.json` by relative path/SHA-256. It identified task, material snapshot/revision, `reference_guided` or explicit `original_no_reference`, source/image bytes and lawful-use locators, content/preview hashes, reference panel bboxes/grammar and the current scientific mapping described above. Workers consumed that frozen plan rather than authoring or backfilling it. - `paperspine5.figure-reference-mapping/1.0`: after independent selection it bound the unchanged plan, exact current/candidate/winner/publication/editable assets, independent review, background probe and simultaneous comparison. The publication hash matched `figure_body_contract.json`. Its historical left side allowed a reference, original figure or explicit original-design sentinel; that compatibility allowance does not equate old/new comparison with external reference learning. - Historical original design used an empty reference array and concrete rationale; missing fields did not imply it, and it could not pass reference-guided review. Probe records included version/render hash and every panel; data plots declared complete axes, while mechanisms could mark axes `not_applicable`. - `FIGURE_MAPPING_READY` combined plan, mapping, probe, independent review, body binding, pixels and user comparison. Explicit engine hash invalidation remains local to those dependencies. This flag never authorizes submission or another external action and is not the current host's scientific acceptance rule.