# Alchemy in motion · Twelve Gemini Deep Research prompts

Choose one investigation and paste its prompt into Gemini Deep Research. Each works independently. When the report returns, open its cited figure or folio before using a claim.

## 1. Where Egyptian blue still glows

**Deliverable**  A measurement-to-object evidence map

Run a Google Gemini Deep Research investigation for the Alchemy Research Catalog.

Can a near-infrared signal identify ancient paint on a sculpture?

Build a dossier for a pigment chemist and a museum editor linking cuprorivaite spectroscopy to an actual interpretation of the Parthenon sculptures. Start with Accorsi et al. (2009), https://doi.org/10.1039/B902563D, and Verri et al. (2023), https://doi.org/10.15184/aqy.2023.130. Retrieve the original figures or openly accessible author versions. Verify the reported peak, units, excitation conditions, specimen and uncertainty or reporting limit. Distinguish reflected visible light from near-infrared luminescence after excitation. Follow one patch of pigment from instrument or image to the authors’ interpretation of the object. Name alternative explanations a conservator would test, and the evidence required to infer an original painted design. Produce a four-row map from signal to identification to pattern to reconstruction, a museum caption of at most 70 words, and one precise question that remains open. Label each row by the strongest evidence actually available.

Date your evidence check. For every central claim, include the primary source, the exact figure, page or folio, what was observed or computed, and a concrete limitation. Label historical interpretation and invention. Flag any requested observation you cannot substantiate.

## 2. The chemistry behind blue cloth

**Deliverable**  A molecular state table and a falsifiable mechanism

Run a Google Gemini Deep Research investigation for the Alchemy Research Catalog.

What changes in indigo between a dye bath, a fibre and exposure to air?

Trace the chemical identities that connect indigo, reduced leuco forms and the blue material present after oxidation. Work from published observations. Start with Roessler et al., Direct electrochemical reduction of indigo, https://doi.org/10.1016/S0013-4686(02)00028-2. Find an independently authored measurement on a textile or fibre. For each system link oxidation state, acid–base speciation, solubility and optical response, distinguishing what an author measured from a mechanism inferred later. Compare a plant-derived material with a purified indigotin sample only where a source documents the comparison. Select one reported spectrum or voltammogram and explain the control that makes it interpretable. Deliver a state table, an atom/charge-balanced conceptual redox equation with the assumed protonation state, one contested claim and the single observation that could resolve it. Keep the deliverable a desk study of published chemistry.

Date your evidence check. For every central claim, include the primary source, the exact figure, page or folio, what was observed or computed, and a concrete limitation. Label historical interpretation and invention. Flag any requested observation you cannot substantiate.

## 3. One formula, two crystals

**Deliverable**  A phase-selection comparison with a testable prediction

Run a Google Gemini Deep Research investigation for the Alchemy Research Catalog.

When does a calcium carbonate specimen become calcite or aragonite?

Use CaCO₃ to investigate how two solids with the same chemical formula acquire different structures. Begin with Sun and colleagues’ study of nucleation in seawater, following https://news.mit.edu/2015/why-seashell-mineral-forms-differently-in-seawater-0302 to the primary paper. Compare it with a separate experimental diffraction or microscopy study. Report how phase identity was established, and separate bulk thermodynamic stability from nucleation barriers and growth rate. For each comparison record temperature, pressure, solution context, magnesium content if reported and whether the system was abiotic or biological. Avoid using seawater simulations as evidence for organism-level control. Deliver a two-system comparison, a causal diagram in words and one prediction with a named observable that could overturn it. Provide the evidence needed to identify an unknown specimen; visual appearance alone leaves the question unresolved.

Date your evidence check. For every central claim, include the primary source, the exact figure, page or folio, what was observed or computed, and a concrete limitation. Label historical interpretation and invention. Flag any requested observation you cannot substantiate.

## 4. Colour built from cellulose

**Deliverable**  A curator’s guide to reading a reflectance spectrum

Run a Google Gemini Deep Research investigation for the Alchemy Research Catalog.

Which optical measurements distinguish structural colour from a conventional dye?

Compare cellulose nanocrystal photonic pigments with a conventional absorbing colorant of similar apparent hue. Begin with Parker et al., Cellulose photonic pigments, https://doi.org/10.1038/s41467-022-31079-9. Locate two independent primary studies with reflectance, angle-resolved or polarization-sensitive data and record sample format, pitch definition, optical geometry and matrix. Explain what produces selective reflection, where the simple pitch-to-wavelength approximation applies, and what changes when viewing angle, particle orientation or surrounding medium varies. Audit one claim about scale, colour stability or environmental benefit against a measured endpoint and system boundary. Deliver a spectrum-reading key, a pairing of structure and observable and a small brief for an educational visual. Mark any drawn curve as schematic unless you can cite reusable data and its licence.

Date your evidence check. For every central claim, include the primary source, the exact figure, page or folio, what was observed or computed, and a concrete limitation. Label historical interpretation and invention. Flag any requested observation you cannot substantiate.

## 5. What a metal remembers

**Deliverable**  An annotated hysteresis reading and use-limit card

Run a Google Gemini Deep Research investigation for the Alchemy Research Catalog.

Which evidence shows a reversible crystal transformation in a shape-memory alloy?

Interpret one published nickel–titanium dataset with a traceable specimen composition and thermomechanical history. Use the Sandia constitutive model guide as vocabulary, then find original experimental data with calorimetry, diffraction or strain measurements. For one specimen record transformation start and finish markers on heating and cooling, units, extraction method, uncertainty and cycling history where supplied. Tie one feature of the curve to a directly observed change or state explicitly where the interpretation is model-based. Distinguish heat-driven shape recovery from stress-induced superelasticity. Identify one published case where composition, processing or cycling shifts the response, keeping cross-study conditions visible. Deliver an annotated reading of the curve, a short design limit for a hypothetical actuator and the missing evidence needed to trust lifetime performance.

Date your evidence check. For every central claim, include the primary source, the exact figure, page or folio, what was observed or computed, and a concrete limitation. Label historical interpretation and invention. Flag any requested observation you cannot substantiate.

## 6. From record cell to rooftop

**Deliverable**  A cell-to-module transfer and durability scorecard

Run a Google Gemini Deep Research investigation for the Alchemy Research Catalog.

What would a laboratory tandem result have to demonstrate before it belongs in an installer’s brief?

Assess perovskite–silicon tandem photovoltaics for a German solar installer explaining research to technically literate customers. Consult the current confirmed research-cell chart at https://www.nlr.gov/pv/cell-efficiency and the ISOS-based reporting consensus, https://doi.org/10.1038/s41560-019-0529-5. Select one traceable independently measured cell result and one independently documented module result, recording test date, effective or aperture area, architecture, efficiency, stabilization or scan procedure and certifying body if available. Compare like quantities; preserve missing values. Find one operational stress test with elapsed time, conditions, encapsulation, tracking mode and initial reference for any retention metric. Deliver an evidence matrix, three sentences an installer can defend, and the separate field or warranty evidence required for a yield or lifetime claim. Date volatile records. State any commercial relationship of a source where disclosed.

Date your evidence check. For every central claim, include the primary source, the exact figure, page or folio, what was observed or computed, and a concrete limitation. Label historical interpretation and invention. Flag any requested observation you cannot substantiate.

## 7. How a workshop made imitation stone

**Deliverable**  A folio-level object biography

Run a Google Gemini Deep Research investigation for the Alchemy Research Catalog.

What can one early modern manuscript actually establish about imitation gems?

Read a documented imitation-stone or imitation-gem entry in BnF Ms. Fr. 640 using the Making and Knowing digital critical edition, https://edition640.makingandknowing.org/content/about. Identify one entry and its exact recto/verso folio; locate the facsimile, original transcription, translation and a relevant editorial essay. Pick a material term whose translation changes the interpretation. Trace which details come from the author-practitioner, editorial inference, later reconstruction or your own reading. Compare one independent historical study or museum object with documented provenance; explain how far the analogy travels. Deliver a catalogue caption, a short attribution chain, two plausible readings of the term and a precise evidence gap. Retain period vocabulary with a gloss and credit translators. Do not assign the manuscript an alchemical identity merely because its craft concerns transformation.

Date your evidence check. For every central claim, include the primary source, the exact figure, page or folio, what was observed or computed, and a concrete limitation. Label historical interpretation and invention. Flag any requested observation you cannot substantiate.

## 8. A notebook travels from Fez to Damascus

**Deliverable**  A provenance timeline anchored to individual folios

Run a Google Gemini Deep Research investigation for the Alchemy Research Catalog.

What can the layers of one multilingual codex tell us about the circulation of alchemical knowledge?

Study British Library Or 13006 through the Qatar Digital Library facsimile and catalogue, https://www.qdl.qa/archive/81055/vdc_100000003591.0x000001. Verify the catalogue’s identification of the earlier notebook, the later copied portion, their places, dates, named copyists and languages. Inspect at least two relevant folio images, recording stable folio URLs and distinguishing colophon, marginalia and editorial catalogue description. Preserve both Hijri and Common Era dates where documented. Attribute direct transcriptions, translations and uncertain readings to their sources; record where specialist paleographic review remains necessary. Use one scholarly interpretation of alchemical writing in the region for context, and avoid projecting it onto a folio without textual evidence. Deliver a four-event provenance timeline, a small term note and a paragraph on what a physical copy date does and does not establish about an underlying work’s composition.

Date your evidence check. For every central claim, include the primary source, the exact figure, page or folio, what was observed or computed, and a concrete limitation. Label historical interpretation and invention. Flag any requested observation you cannot substantiate.

## 9. When lead really became gold

**Deliverable**  A detection-to-inference ledger with a unit check

Run a Google Gemini Deep Research investigation for the Alchemy Research Catalog.

What did ALICE detect, and what did researchers infer from the signal?

Reconstruct the 2025 ALICE result announced at https://home.cern/alice-detects-conversion-lead-gold-lhc/ using the collaboration’s original paper and any public analysis data. Keep the account conceptual. Specify the precursor nucleus, charge change and signal channel, what the detectors directly counted and what depended on a model or extrapolation. Establish which quantities refer to a collision point, run, integrated estimate or sample and whether gold nuclei were recovered as a material. Recalculate one published nuclei-to-mass conversion using explicitly stated mass-number assumptions and units; identify the dominant uncertainty and the precision justified by the source. Give a compact comparison with a chemical change such as plating, where surface composition can change without conversion of one element into another. Deliver a measurement ledger, a dimensional check and a headline suitable for a science editor.

Date your evidence check. For every central claim, include the primary source, the exact figure, page or folio, what was observed or computed, and a concrete limitation. Label historical interpretation and invention. Flag any requested observation you cannot substantiate.

## 10. Audit an AI crystal discovery

**Deliverable**  A candidate-level replication ledger

Run a Google Gemini Deep Research investigation for the Alchemy Research Catalog.

How far does one predicted crystal advance from a model output to a characterized material?

Audit two specific AI materials-discovery claims, including one candidate from MatterGen (https://doi.org/10.1038/s41586-025-08628-5) and one comparator chosen for accessible original evidence. Use the primary papers and Matbench Discovery (https://doi.org/10.1038/s42256-025-01055-1). Trace each candidate through generation, duplicate and novelty checks, stability calculation, experimental phase identification, experimental property measurement and independent replication. Record database/model versions and phase disorder; leave a stage empty when evidence is unavailable. Include one simple baseline, a split or training-overlap risk and a condition under which the target property comparison is fair. Deliver two candidate rows, a verdict on each transition, and the single highest-value next check. Evaluate the status on the research date.

Date your evidence check. For every central claim, include the primary source, the exact figure, page or folio, what was observed or computed, and a concrete limitation. Label historical interpretation and invention. Flag any requested observation you cannot substantiate.

## 11. Can a model identify a pigment?

**Deliverable**  A benchmark design with an abstention rule

Run a Google Gemini Deep Research investigation for the Alchemy Research Catalog.

Would a pigment classifier still work on a spectrum from another instrument or collection?

Design a desk-based evidence audit for classifying non-hazardous artist pigments from Raman spectra. Inspect the IRUG database at https://irug.org/about-us/the-database and one independent, openly documented reference collection. Verify access, redistribution and training rights separately; never infer a licence from a download button. Choose one pigment pair or mixture that confounds identification and establish a reference-label hierarchy. Compare a nearest-reference or spectral-angle baseline with one learned model. Specify sample-, acquisition- and institution-grouped held-out splits to prevent repeat-scan leakage. Include the effects of fluorescence, mixtures and unrelated materials. Report macro-F1 and per-class confusion together with coverage and error among predictions retained after abstention. Deliver a benchmark card, an empty results table for unrun tests, and the evidence needed to approve a new collection or instrument. Cite measured performance only from genuinely matched evaluations.

Date your evidence check. For every central claim, include the primary source, the exact figure, page or folio, what was observed or computed, and a concrete limitation. Label historical interpretation and invention. Flag any requested observation you cannot substantiate.

## 12. A glaze that remembers a brushstroke

**Deliverable**  An original non-canon worldbuilding sheet

Run a Google Gemini Deep Research investigation for the Alchemy Research Catalog.

What would a memory-bearing glaze cost the people who depend on it?

Write a research-backed concept for an original fictional material called Night Glaze. Begin with two plausible inspirations drawn from published photoluminescence and material hysteresis. Cite one primary source for each and say exactly which observable property inspired the design. Develop this fictional rule: one firing stores the potter’s final continuous brushstroke; the first reading consumes it forever. Preserve that rule across both honest and deceptive uses of the vessel. Specify the storage unit, scarcity, failure mode, recovery limit and the observable sign that a vessel has already been read. Show how a kiln keeper and a witness disagree about a specific vessel in an original 120-word scene. Attach a three-panel art-direction note and a continuity check that would invalidate a future story use. Treat every invented effect and passage as fiction. Use no clinical claims, chemical recipes, or attributions to an established fictional canon; source the inspirations, credit the invention as a proposal and keep the two origins legible.

Date your evidence check. For every central claim, include the primary source, the exact figure, page or folio, what was observed or computed, and a concrete limitation. Label historical interpretation and invention. Flag any requested observation you cannot substantiate.
