QSTVE

Ideas
for a more
human economy

100 Ideas for Volunteer-Powered AI Science

100 ideas for volunteer-powered AI science. Pick an idea, copy the prompt, and paste it into your AI.

Ideas only. No pilot or study was run.

Stay with nonclinical, offline research: no diagnosis, patient matching or medical advice.

Inclusion is attribution, not endorsement.

Download the PDF (107 pages)

Medical evidence without medical advice

001

Trial eligibility into auditable logic

Turn one study's rules on who can join into a clear checklist, quoting the exact text.

Limit: Never match a real person to a trial.

Data: ClinicalTrials.gov study search

Show prompt
Help me start “Trial eligibility into auditable logic” as a beginner research exercise. Choose one registry study and turn five eligibility clauses into a table of exact quotes, inclusion/exclusion labels, negation and unknowns. Include an ambiguous clause and a human correction column; do not match any real person to a trial.

Candidate input: ClinicalTrials.gov public registry records; public API available. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Clinician-curated clause gold set; count unsupported clauses and negation errors.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

002

Outcome-unit mismatch finder

Check that one study result uses consistent units, direction and follow-up time.

Limit: Flag outcomes that don't compare; never convert them by guesswork or claim a treatment works.

Data: ClinicalTrials.gov study search

Show prompt
Help me start “Outcome-unit mismatch finder” as a beginner research exercise. Select five outcome fields from two studies on the same broad topic. Make a worksheet separating measurement, unit, direction and follow-up window; flag incomparable outcomes instead of converting them by guesswork.

Candidate input: Public ClinicalTrials.gov outcome fields; larger sample requires retrieval. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Double-annotated records; exact unit and time agreement; abstain on ambiguity.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

003

Registered-result completeness audit

Check whether one trial's registry entry matches the paper it links to.

Limit: A missing publication link is unknown, not proof of misconduct.

Data: ClinicalTrials.gov study search

Show prompt
Help me start “Registered-result completeness audit” as a beginner research exercise. Start with five registry IDs and only publications explicitly linked from their records. Design a dated presence/absence/unknown checklist for result fields; treat an absent publication link as unknown, not proof of nonpublication or misconduct.

Candidate input: ClinicalTrials.gov records and Crossref publication metadata. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Manual sample of links; report missingness rather than misconduct.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

004

Duplicate trial identity resolver

Decide whether two trial records describe the same study, using IDs, dates and wording.

Limit: Never merge or delete records automatically.

Data: ClinicalTrials.gov study search

Show prompt
Help me start “Duplicate trial identity resolver” as a beginner research exercise. Create ten candidate pairs using public study identifiers and titles. Begin with an exact-ID rule, keep a separate review queue for near matches, and never merge or delete records automatically.

Candidate input: ClinicalTrials.gov IDs and Crossref titles/identifiers. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Held-out pair labels; prioritize precision and preserve separate records.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

005

Adverse-event duplicate sensitivity study

Test how different duplicate-removal rules change one slice of drug side-effect reports.

Limit: Side-effect reports can't show how common a harm is or what caused it.

Data: openFDA drug adverse events

Show prompt
Help me start “Adverse-event duplicate sensitivity study” as a beginner research exercise. Use one small public FAERS slice to outline two documented deduplication rules. Preserve raw counts, show which records each rule would group, and explain why spontaneous reports cannot establish incidence or causality.

Candidate input: openFDA publicly releasable FAERS records and documented caveats. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Reproduce baseline counts; bootstrap differences; audit matched reports.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

006

Adverse-event field harmonizer

Suggest groups for one set of differently spelled drug names, keeping the original spellings.

Limit: A drug-data curator approves every name grouping before use.

Data: openFDA drug adverse events

Show prompt
Help me start “Adverse-event field harmonizer” as a beginner research exercise. Prepare ten permitted product-name strings and a conservative alias table with original values intact. Include similar-looking names that must remain separate; require a drug-data curator to approve mappings before use.

Candidate input: openFDA FAERS structured fields; sampled export still required. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Expert-labeled alias pairs and conservative abstention thresholds.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

007

Trial geography coverage atlas

Roll up one study's sites into broad regions and flag any missing locations.

Limit: Site counts are not disease burden; leave out contact details.

Data: ClinicalTrials.gov study search

Show prompt
Help me start “Trial geography coverage atlas” as a beginner research exercise. Pick five studies and aggregate their published site fields to country or broad region. Track missing sites separately and reconcile totals; do not interpret site coverage as population access to treatment.

Candidate input: Public ClinicalTrials.gov facility and study metadata. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Manual location sample and reconciliation of aggregate totals.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

008

Medical citation-update sentinel

Check whether one paper has corrections or retractions, without judging the paper itself.

Limit: Report notices only; never judge the research itself.

Data: Crossref REST API

Show prompt
Help me start “Medical citation-update sentinel” as a beginner research exercise. Use five public DOIs to check correction or retraction relationship fields and linked publisher notices. Draft a dated alert table that distinguishes confirmed notices, missing metadata and lookup failures without judging the underlying research.

Candidate input: Crossref metadata, including update relations where deposited. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Cross-check publisher links; report relation availability and false alerts.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

009

Trial plain-language fidelity test

Write one short plain explanation of a study where every sentence points to its source.

Limit: Research text only; never medical advice, trial recruitment or a patient service.

Data: ClinicalTrials.gov study search

Show prompt
Help me start “Trial plain-language fidelity test” as a beginner research exercise. Choose one public trial summary and draft a 100-word explanation with every sentence linked to an exact source span. Build a checklist for altered numbers, omitted caveats and unsupported benefit language; do not offer medical advice.

Candidate input: Public ClinicalTrials.gov study descriptions. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Independent raters check omissions, changed numbers and unsupported benefits.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

010

Study-design retrieval benchmark

Rank a fixed set of studies by how well they answer one researcher's question.

Limit: Public registry records only, never personal health records.

Data: ClinicalTrials.gov study search

Show prompt
Help me start “Study-design retrieval benchmark” as a beginner research exercise. Write one narrow study-design question and manually label ten public registry records for relevance. Compare a simple keyword ranking with one optional local model using the same fixed review budget and a held-out subset.

Candidate input: ClinicalTrials.gov public protocol descriptions and Crossref metadata. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Blinded relevance labels; recall at fixed review budget versus keyword search.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

Cells, proteins, and reproducible biomedicine

011

Cell-count reproducibility challenge

Count the cells in one image with fixed settings and record how far off the count is.

Data: BBBC005 synthetic cells

Show prompt
Help me start “Cell-count reproducibility challenge” as a beginner research exercise. Begin with five licensed microscopy images with reference cell counts. Outline a fixed segmentation baseline and an error table across original and blurred copies; use established images only, not new biological experiments.

Candidate input: BBBC005 synthetic microscopy with stated ground truth. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Compare true counts and analytic baselines across blur conditions.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

012

Microscopy focus triage

Score how sharp one microscope image is, and how sure that score is.

Data: BBBC005 synthetic cells

Show prompt
Help me start “Microscopy focus triage” as a beginner research exercise. Take ten licensed microscopy images and compare a simple focus score with blinded human sharpness labels. Hold back several blur conditions and define an uncertainty bucket instead of discarding low-scoring images automatically.

Candidate input: BBBC005 simulated image conditions and ground truth. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Hold out blur levels; compare a sharpness metric and blinded reviewers.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

013

Cell-boundary disagreement atlas

Compare two outlines of the same cells and mark where they disagree.

Data: BBBC image collection

Show prompt
Help me start “Cell-boundary disagreement atlas” as a beginner research exercise. Select three licensed images with paired reference masks. Sketch an overlay workflow that separates boundary shifts, splits and merges; build a five-case expert-review sheet before proposing any automatic correction.

Candidate input: BBBC annotated image sets; choose a set with matching masks. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Boundary overlap plus expert review of sampled splits/merges.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

014

Plate illumination correction bake-off

Try one lighting-correction method on one batch of microscope images.

Data: BBBC image collection

Show prompt
Help me start “Plate illumination correction bake-off” as a beginner research exercise. Use a tiny licensed image batch with documented controls to compare an uncorrected baseline and one illumination correction. Track whether control variation improves without changing the ordering of object intensities.

Candidate input: BBBC image collections; candidate plate-structured subset requires checking. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Known-control stability and preserved object-intensity ranking.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

015

Synthetic-cell uncertainty calibration

Test one cell-counting model on deliberately blurred or noisy images, and record its error ranges.

Data: BBBC005 synthetic cells

Show prompt
Help me start “Synthetic-cell uncertainty calibration” as a beginner research exercise. Use an existing permitted synthetic-cell generator or released synthetic examples, not new biological protocols. Plan a small blur/noise grid with held-out conditions and an interval-coverage worksheet; keep simulated evidence clearly labeled.

Candidate input: BBBC005 synthetic-cell images and expected counts. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Coverage versus interval width on withheld conditions.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

016

Protein structure provenance cards

Write down how one protein structure was determined (experiment or computed model), how detailed it is, and its caveats.

Data: RCSB PDB data

Show prompt
Help me start “Protein structure provenance cards” as a beginner research exercise. Pick three public structure accessions and build provenance cards from explicit deposited fields. Separate experimental method, applicable resolution, model status and missing facts; make no functional or therapeutic inference.

Candidate input: RCSB PDB entry metadata and linked validation information. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Schema checks and curator comparison; every statement links to fields.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

017

Missing-residue bias census

Measure how much of one protein structure is missing from the published model.

Data: RCSB PDB data

Show prompt
Help me start “Missing-residue bias census” as a beginner research exercise. Choose three public structure chains and compare deposited sequences with resolved residues. Build a coverage table that distinguishes missing coordinates from sequence differences and flags alignment uncertainty for curator review.

Candidate input: RCSB PDB coordinate and sequence records. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Cross-check sequence alignment and deposited annotations.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

018

Structure-alignment reproducibility grid

Line up two protein structures with one fixed method and check the result repeats.

Data: RCSB PDB data

Show prompt
Help me start “Structure-alignment reproducibility grid” as a beginner research exercise. Select one known homolog pair from public structures and outline a documented alignment baseline. Add a harmless rigid translation/rotation control and two repeated runs; record software versions, parameters and agreement tolerances.

Candidate input: Public RCSB PDB structures in a curator-selected benign benchmark. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Known homolog controls, rigid-transform invariance and replicated outputs.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

019

Protein-motion teaching replication

Run one short protein-motion simulation, only after an expert has reviewed the plan.

Data: RCSB PDB data

Show prompt
Help me start “Protein-motion teaching replication” as a beginner research exercise. Find a published teaching protocol and a small permitted example system. Draft only a reproduction plan with seed, timestep, resource cap and energy checks; stop before running until a qualified molecular-simulation expert approves the protocol.

Candidate input: Benign curated PDB structures; approved force-field inputs remain a candidate. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Energy conservation checks, independent seeds and published reference comparison.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

020

Bioimage annotation adjudication club

Have several trained volunteers settle one disputed cell outline.

Data: BBBC image collection

Show prompt
Help me start “Bioimage annotation adjudication club” as a beginner research exercise. Prepare five already licensed cell-boundary disputes with reference masks. Draft volunteer instructions, blind duplicates and an expert escalation form; collect no new human or biological samples and make no diagnostic judgments.

Candidate input: BBBC image sets with documented annotations. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Blind duplicate cases, expert arbitration and inter-rater agreement.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

Signals from the cosmos

021

Gravitational-wave tutorial replication swarm

Rerun one published gravitational-wave tutorial on a small slice of data.

Data: Gravitational Wave Open Science Center

Show prompt
Help me start “Gravitational-wave tutorial replication swarm” as a beginner research exercise. Choose one official open gravitational-wave tutorial and its bounded event segment. Make a local reproduction checklist with pinned versions, expected peak-time tolerance and a CPU/time cap before attempting any batch distribution.

Candidate input: GWOSC public tutorials, event data and software links. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Compare recovered peak time and numerical tolerance with a reference run.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

022

Detector-glitch uncertainty queue

Rank one batch of detector-noise images for an expert to review.

Data: Gravitational Wave Open Science Center

Show prompt
Help me start “Detector-glitch uncertainty queue” as a beginner research exercise. Select ten permitted labeled glitch spectrograms and propose a baseline ranking for human review. Reserve examples for a blind check and measure useful labels per reviewer minute rather than claiming new astrophysical events.

Candidate input: Gravity Spy public project and GWOSC data; labeled export needs checking. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Held-out glitch labels, calibration and reviewer time saved.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

023

Continuous-wave search parameter shards

Run one small, project-assigned piece of a long search for signals from spinning neutron stars.

Data: Einstein@Home

Show prompt
Help me start “Continuous-wave search parameter shards” as a beginner research exercise. Locate the official project's documented volunteer entry point and validator requirements. Draft a readiness checklist for a project-issued parameter shard; do not invent shards, submit fabricated work or run an independent all-sky search.

Candidate input: Einstein@Home documented neutron-star search programs; work only through project authorization. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Project validators, independent replicas and vetted numerical tolerances.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

024

Radio interference morphology audit

Look at one patch of radio data and tag the shape of the interference.

Data: Breakthrough Listen open data

Show prompt
Help me start “Radio interference morphology audit” as a beginner research exercise. Start with five licensed time-frequency tiles and a documented interference taxonomy. Prepare a review sheet of visible morphology, uncertainty and expert labels; do not call unexplained features extraterrestrial signals.

Candidate input: Breakthrough Listen public open-data archive; manageable labeled subset requires checking. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Known interference examples and blinded expert review of anomalies.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

025

SETI candidate ranking reproducibility

Recheck one signal ranking, only if suitable public data tables are released.

Data: SETI@home current analysis

Show prompt
Help me start “SETI candidate ranking reproducibility” as a beginner research exercise. First check whether a suitable public candidate table and ranking method actually exist. If either is unavailable, produce a blocker list; otherwise plan a ten-row threshold-sensitivity replay with no discovery claim.

Candidate input: SETI@home analysis descriptions; candidate-level downloadable tables are unverified. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Compare documented rankings and sensitivity to thresholds.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

026

TESS transit injection recovery

Hide a fake planet signal in one star's brightness data and see if it can be found.

Data: MAST TESS data

Show prompt
Help me start “TESS transit injection recovery” as a beginner research exercise. Use one permitted TESS light curve and a documented transit-injection tutorial. Plan three explicitly synthetic injections, a baseline recovery method and a false-alarm check on an uninjected control.

Candidate input: MAST TESS public time-series products; selected light-curve files require retrieval. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Known injected parameters; completeness and false-alarm curves.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

027

Exoplanet parameter contradiction cards

Compare one planet's reported size, orbit and other values across sources, with units.

Data: NASA Exoplanet Archive

Show prompt
Help me start “Exoplanet parameter contradiction cards” as a beginner research exercise. Choose three planets in a public archive and tabulate two reported values for one parameter with units, uncertainties and provenance. Distinguish different definitions or epochs from actual contradictions and route unresolved cases to an astronomer.

Candidate input: NASA Exoplanet Archive tables and cited references. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Numeric/unit checks and astronomer adjudication of discrepancies.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

028

Gaia astrometric selection audit

Test how one set of data-quality filters changes a fixed sample of star positions.

Data: ESA Gaia archive

Show prompt
Help me start “Gaia astrometric selection audit” as a beginner research exercise. Select one small Gaia sky sample through the documented query service. Write two explicit quality-cut recipes and a count-reconciliation table; preserve query text and explain why selected stars are not a representative census.

Candidate input: ESA Gaia released catalogs and documented quality fields. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Reproduce counts; injected selection tests and independent scripts.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

029

Stellar spectral outlier explanations

Rank one batch of star spectra for oddities, then describe the differences with citations.

Data: Sloan Digital Sky Survey data

Show prompt
Help me start “Stellar spectral outlier explanations” as a beginner research exercise. Choose a tiny licensed labeled stellar-spectrum sample and a simple numeric distance baseline. Plan a held-out outlier ranking with measured wavelength deviations and an astronomer review sheet, not invented physical explanations.

Candidate input: SDSS public spectral data release; selected spectra and labels need checking. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Known-class holdouts and astronomer review; report rejection rate.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

030

Astronomy archive metadata repair suggestions

Check one downloaded data file's labels against its published format description.

Data: MAST TESS data

Show prompt
Help me start “Astronomy archive metadata repair suggestions” as a beginner research exercise. Download metadata for three permitted astronomy products and their official schemas. Produce a read-only linter checklist with original values and suggested repairs in a separate file; never edit the archive or infer missing observations.

Candidate input: MAST TESS and NASA Exoplanet Archive product documentation. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Deterministic schema baseline; expert review of every suggested change.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

Planets and the changing sky

031

Galaxy morphology confidence audit

Run one batch of galaxy images through a fixed shape-sorting model and record its confidence.

Data: Sloan Digital Sky Survey data

Show prompt
Help me start “Galaxy morphology confidence audit” as a beginner research exercise. Use twenty licensed galaxy images and a permitted fixed morphology model. Split by sky field, compare with a simple baseline, and draft a confusion/calibration worksheet with an explicit human-disagreement category.

Candidate input: Zoobot documentation and SDSS public images; compatible labeled subset must be selected. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Hold out sky fields; calibration, confusion matrix and human disagreement.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

032

Faint galaxy artifact tribunal

Have several volunteers judge one model-flagged galaxy feature that may be an image flaw.

Data: Sloan Digital Sky Survey data

Show prompt
Help me start “Faint galaxy artifact tribunal” as a beginner research exercise. Prepare five licensed galaxy-image features, including known imaging artifacts and blind duplicates. Draft a volunteer adjudication card and astronomer escalation rule; preserve uncertain cases instead of announcing faint-object discoveries.

Candidate input: SDSS public image cutouts; candidate artifact labels need expert preparation. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Blind duplicates and expert consensus; measure artifact rejection.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

033

Planetary crater edge benchmark

Trace crater edges in one image patch and mark the edges that are unclear.

Data: NASA Planetary Data System

Show prompt
Help me start “Planetary crater edge benchmark” as a beginner research exercise. Select two licensed planetary image tiles with expert crater-rim annotations. Plan an edge baseline, uncertain-rim overlays and diameter-error measurement on held-out terrain; do not treat every circular feature as a crater.

Candidate input: NASA PDS planetary imagery; curated crater masks remain a candidate. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Expert rim masks, diameter error and held-out terrains.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

034

Planetary image registration challenge

Line up two images of the same planet surface and estimate how uncertain the match is.

Data: NASA Planetary Data System

Show prompt
Help me start “Planetary image registration challenge” as a beginner research exercise. Choose one permitted image pair with known landmarks and create a copy with a documented synthetic shift. Plan registration residual vectors and uncertainty checks before comparing any naturally changed surfaces.

Candidate input: PDS image pairs with suitable geometry; exact pairs require screening. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Known landmarks, synthetic shifts and residual vector maps.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

035

Mars surface-change review cards

Make one before-and-after image card of possible Mars surface change for a person to judge.

Data: NASA Planetary Data System

Show prompt
Help me start “Mars surface-change review cards” as a beginner research exercise. Start with a licensed, well-documented Mars before/after pair and one stable control area. Draft a card showing acquisition geometry, shadows and alignment limits; require independent review before labeling a surface change.

Candidate input: Public PDS repeat-imaging collections; licensed pair selection untested. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Stable-area controls and independent reviewers; reject shadow artifacts.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

036

Planetary spectral mineral uncertainty map

Estimate the minerals in one spectrum with a documented method, including uncertainty.

Data: NASA Planetary Data System

Show prompt
Help me start “Planetary spectral mineral uncertainty map” as a beginner research exercise. Find a documented public reference spectrum and permitted mixture examples. Outline a numeric fit with residuals and calibration sensitivity; label uncertain components and avoid treating a model coefficient as confirmed mineral identification.

Candidate input: PDS spectral products; matching reference library must be verified. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Reference mixtures, residuals and sensitivity to calibration choices.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

037

Asteroid tracklet consistency review

Check whether one set of asteroid sightings fits a single consistent path and timing.

Data: NASA Planetary Data System

Show prompt
Help me start “Asteroid tracklet consistency review” as a beginner research exercise. Use public teaching tracklets or explicitly synthetic positions with known times. Plan unit/timestamp checks and residual plots against a documented motion baseline; ask an astronomer to set thresholds and make no operational hazard claim.

Candidate input: PDS small-body products; time-tagged detection subset is a candidate. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Injected tracks and astronomer-approved residual thresholds.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

038

TESS variable-star shape atlas

Find stars whose brightness curves have shapes similar to one chosen star.

Data: MAST TESS data

Show prompt
Help me start “TESS variable-star shape atlas” as a beginner research exercise. Choose ten permitted labeled TESS variable-star curves. Plan a normalization recipe and simple Fourier-feature retrieval baseline, keeping labeled stars separate for evaluation and recording gaps rather than inventing measurements.

Candidate input: MAST TESS light curves; curated variability labels require checking. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Held-out labeled stars and simple Fourier-feature baseline.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

039

Planetary archive accessibility captions

Write one factual image description for screen-reader users, based only on the archive's records.

Data: NASA Planetary Data System

Show prompt
Help me start “Planetary archive accessibility captions” as a beginner research exercise. Take three permitted planetary images with official metadata and draft concise accessibility descriptions. Separate visible features from metadata facts and mark uncertain spatial language; prepare questions for an accessibility reviewer and planetary scientist.

Candidate input: PDS public images with captions and metadata. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Blind accessibility review; count unsupported objects and spatial errors.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

040

Survey footprint coverage reconciler

Compare star counts from two surveys within one small patch of sky.

Data: ESA Gaia archive

Show prompt
Help me start “Survey footprint coverage reconciler” as a beginner research exercise. Pick one coarse sky tile and two documented public catalog queries. Reconcile raw and filtered counts with footprints, magnitude limits and selection rules; record query differences before interpreting gaps as missing objects.

Candidate input: Gaia and SDSS release coverage documentation and catalogs. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Catalog query replication and selection-function sanity checks.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

Atmosphere and climate

041

Weather-station discontinuity screen

Check one weather station's record for sudden jumps that may come from equipment changes.

Data: NOAA Global Historical Climatology Network daily

Show prompt
Help me start “Weather-station discontinuity screen” as a beginner research exercise. Use one public station-variable series and a neighboring control. Plan a simple breakpoint screen plus a copy with a labeled synthetic break; hold back time periods and do not call every discontinuity climate change.

Candidate input: NOAA GHCN-Daily station series and quality flags. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Synthetic breaks, neighbor controls and withheld periods.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

042

Weather flag explanation assistant

Explain in plain words why one weather record was flagged, citing the documentation.

Data: NOAA Global Historical Climatology Network daily

Show prompt
Help me start “Weather flag explanation assistant” as a beginner research exercise. Choose five flagged weather records and the exact version of their flag documentation. Create a quote-backed decoder table with unknown values left unresolved, then compare a rules-only explanation with any model-generated wording.

Candidate input: GHCN-Daily documentation, flags and public station files. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Rule-based decoder baseline; zero invented flag meanings.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

043

Heatwave-definition sensitivity grid

Count one station's heatwaves under one clearly stated definition of a heatwave.

Data: NOAA Global Historical Climatology Network daily

Show prompt
Help me start “Heatwave-definition sensitivity grid” as a beginner research exercise. Select one public station-year and write two explicit heatwave definitions with baseline periods and missing-day rules. Plan a count comparison and a small hand-check example; do not issue public-health guidance or real-time warnings.

Candidate input: GHCN-Daily maximum/minimum temperature series. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Reference implementation, missing-data tests and alternate baselines.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

044

Rainfall missingness stress test

Hide one block of rainfall readings and compare ways of filling the gap.

Data: NOAA Global Historical Climatology Network daily

Show prompt
Help me start “Rainfall missingness stress test” as a beginner research exercise. Use a public rainfall segment with enough complete observations and mask one known block in a copy. Compare a simple baseline with one alternative, preserving withheld truth and checking errors on extreme rain separately.

Candidate input: GHCN-Daily precipitation records and flags. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Known withheld observations and extreme-tail preservation.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

045

Air-monitor unit and cadence audit

Check one air monitor's readings for mixed units, uneven timing and missing values.

Data: OpenAQ

Show prompt
Help me start “Air-monitor unit and cadence audit” as a beginner research exercise. Select one public monitor-week and its official unit/time documentation. Draft a read-only audit of units, timezone, cadence and gaps using deterministic checks; never interpret a clean file as evidence that air is safe.

Candidate input: OpenAQ public platform; downloadable API/export authorization must be checked. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Deterministic conversions and manual sample; never impute safety assurances.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

046

Air-pollution gap-filling uncertainty

Hide one day of air-quality readings and estimate them, with an error range.

Data: OpenAQ

Show prompt
Help me start “Air-pollution gap-filling uncertainty” as a beginner research exercise. Choose a small permitted monitor dataset and hold out whole days rather than random rows. Plan a neighbor-based baseline, leakage checks and interval coverage; label estimates clearly and do not replace official safety measurements.

Candidate input: OpenAQ historical monitor series; access and benchmark subset unverified. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Time-blocked holdouts, neighbor leakage checks and interval coverage.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

047

Reanalysis downsampling error atlas

Compare one area's climate data before and after it is averaged into larger squares.

Data: ERA5 single levels

Show prompt
Help me start “Reanalysis downsampling error atlas” as a beginner research exercise. Use one small public reanalysis tile for one short period. Specify area weights and units, compare original versus aggregated values, and include a conservation check before making any regional representativeness claim.

Candidate input: ERA5 single-level reanalysis documentation; data retrieval may require an account. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Conservation checks and withheld resolution; report representativeness error.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

048

Fire-detection persistence audit

Group one region's daily satellite fire detections to show which hotspots persisted.

Data: NASA FIRMS

Show prompt
Help me start “Fire-detection persistence audit” as a beginner research exercise. Select one region-day of public fire detections and available cross-sensor documentation. Plan persistence groups with timestamps and overpass gaps; distinguish satellite detection summaries from verified fire boundaries or emergency advice.

Candidate input: NASA FIRMS documented active-fire products; historical export needs checking. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Cross-sensor agreement and uncertainty from overpass timing.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

049

Carbon-trend method robustness lab

Compare a few standard ways of finding the carbon dioxide trend in one time window.

Data: NOAA carbon dioxide trends

Show prompt
Help me start “Carbon-trend method robustness lab” as a beginner research exercise. Choose one published carbon-series segment and reproduce a documented seasonal/trend baseline. Plan a second estimator, endpoint sensitivity and appropriate resampling intervals; report method dependence rather than attributing a cause.

Candidate input: NOAA GML published carbon-dioxide trend records. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Reference-series reproduction, bootstrap intervals and endpoint sensitivity.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

050

Climate ensemble checkpoint donation

Run one authorized piece of a volunteer climate model on your own computer.

Data: Climateprediction.net

Show prompt
Help me start “Climate ensemble checkpoint donation” as a beginner research exercise. Find an established climate project's authorized volunteer or ensemble workflow. Draft a consent, resource-budget and checkpoint-validation checklist; if no permitted task is available, stop at a partner inquiry rather than running an invented climate model.

Candidate input: climateprediction.net project-issued simulation tasks, when available. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Project scientific controls and ensemble consistency, not LLM plausibility.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

Ocean life and moving water

051

Argo profile quality-control replay

Recheck one ocean temperature and saltiness profile without changing the official record.

Data: Argo data

Show prompt
Help me start “Argo profile quality-control replay” as a beginner research exercise. Pick three public Argo profiles with documented quality flags. Outline a read-only replay of published checks and a disagreement log against delayed-mode flags; preserve archive values and seek an oceanographer's review.

Candidate input: Argo public profile data and documented quality-control system. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Compare delayed-mode flags and manually review disagreements.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

052

Ocean heat-content integration sensitivity

Estimate the heat stored in one ocean profile using one stated calculation method.

Data: Argo data

Show prompt
Help me start “Ocean heat-content integration sensitivity” as a beginner research exercise. Start with an analytic teaching profile whose integral is known, then one public ocean profile. Write down depth bounds, units and interpolation choices before integration; carry uncertainty and avoid presenting a small sample as global ocean heat content.

Candidate input: Argo temperature profiles; reference climatology choice requires verification. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Analytic profiles, published-method replication and uncertainty propagation.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

053

Mixed-layer depth definition comparison

Compare how several standard definitions place the depth of one ocean profile's mixed surface layer.

Data: Argo data

Show prompt
Help me start “Mixed-layer depth definition comparison” as a beginner research exercise. Choose two public profiles and a synthetic profile with an obvious mixed layer. Define two published threshold conventions, then plan depth comparisons with missing-value handling and an oceanographer-reviewed interpretation.

Candidate input: Argo temperature/salinity profiles. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Known synthetic profiles and expert-reviewed seasonal contrasts.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

054

Surface-current interpolation sensitivity

Track simulated floating particles through ocean currents and see how calculation settings change their paths.

Data: OSCAR ocean surface currents

Show prompt
Help me start “Surface-current interpolation sensitivity” as a beginner research exercise. Begin with an analytic flow and explicitly synthetic particles. Plan two timesteps and interpolation schemes with a convergence plot and coast-mask test before considering a small permitted current product; do not make navigation predictions.

Candidate input: NASA PO.DAAC OSCAR final surface-current dataset documentation; data download may require Earthdata Login. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Analytic-flow controls, step-size convergence and coast-mask checks.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

055

Marine occurrence taxonomy repair queue

Flag one confusing species-name match in ocean records and note where it came from.

Data: Ocean Biodiversity Information System

Show prompt
Help me start “Marine occurrence taxonomy repair queue” as a beginner research exercise. Choose ten public occurrence-name strings and their cited taxonomic authorities. Build an original-name/accepted-ID/uncertain mapping table; flag homonyms and require taxonomist approval before changing any downstream record.

Candidate input: OBIS marine occurrence records and documented quality-control resources. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Expert taxonomist approval and accepted-name identifier checks.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

056

Marine sampling-effort bias cards

Summarize how many ocean records and datasets cover one broad region and time period.

Data: Ocean Biodiversity Information System

Show prompt
Help me start “Marine sampling-effort bias cards” as a beginner research exercise. Use public occurrence metadata from one coarse region and season, suppressing sensitive coordinates. Draft coverage summaries stratified by contributing dataset, noting where sampling effort is unknown; do not equate more records with more animals.

Candidate input: OBIS occurrence collections; event/effort completeness must be checked. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Collection-stratified resampling and explicit missing-effort flags.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

057

Coral annotation disagreement workshop

Have volunteers settle one disputed label on a coral reef seafloor photo.

Data: CoralNet

Show prompt
Help me start “Coral annotation disagreement workshop” as a beginner research exercise. Prepare five permitted benthic-image points with expert labels and blind repeats. Draft a short volunteer annotation guide and disagreement form, retaining uncertain classes and expert arbitration rather than inventing coral-health diagnoses.

Candidate input: CoralNet public-source collections; reuse permission and label export require checking. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Expert consensus, repeated points and inter-rater agreement.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

058

Coral classifier drift watch

Test one existing image model on coral photos from a source it has not seen.

Data: CoralNet

Show prompt
Help me start “Coral classifier drift watch” as a beginner research exercise. Choose two licensed coral-image sources and a permitted pretrained model. Keep one source completely held out and outline class accuracy, calibration and reviewed failure categories; check label compatibility before comparing results.

Candidate input: Permission-cleared CoralNet public images and annotations, subset unverified. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Cross-source accuracy, calibration and human-reviewed failure categories.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

059

Whale-call candidate triage

Score one short underwater recording for sounds that may be whale calls worth a listen.

Data: NOAA passive acoustic data

Show prompt
Help me start “Whale-call candidate triage” as a beginner research exercise. Use a few licensed audio clips with expert whale-call annotations and no sensitive location disclosure. Plan a simple detection baseline and blind review queue with precision per listening minute; do not interpret candidates as confirmed population counts.

Candidate input: NOAA NCEI passive-acoustic archive; labeled licensed subset requires selection. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Held-out expert detections and precision at fixed listening budget.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

060

Ocean noise-band reproducibility mesh

Measure the noise levels in one calibrated block of underwater sound.

Data: NOAA passive acoustic data

Show prompt
Help me start “Ocean noise-band reproducibility mesh” as a beginner research exercise. Find one permitted calibrated underwater-audio example and its unit documentation, plus a known synthetic tone. Plan frequency-band summaries with an independent FFT check; if calibration is absent, stop short of absolute sound-level claims.

Candidate input: NCEI passive-acoustic recordings and instrument metadata. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Known tones, unit checks and independent FFT implementation.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

Biodiversity and conservation

061

Camera-trap empty-frame filter

Sort one batch of camera-trap photos, keeping any animal sightings the model is unsure about.

Data: Snapshot Serengeti

Show prompt
Help me start “Camera-trap empty-frame filter” as a beginner research exercise. Select twenty licensed camera-trap frames from distinct locations with expert empty/animal labels. Plan a simple baseline and uncertainty-preserving threshold; explicitly count missed rare animals and protect species locations and incidental people.

Candidate input: LILA Snapshot Serengeti images and labels. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Held-out locations; rare-animal false-negative rate versus baseline.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

062

Wildlife count uncertainty benchmark

Estimate the number of animals in one photo, and say when hidden animals make it unclear.

Data: Snapshot Serengeti

Show prompt
Help me start “Wildlife count uncertainty benchmark” as a beginner research exercise. Choose ten licensed wildlife images spanning sparse and crowded scenes. Prepare expert count intervals, an occlusion/abstention rule and a calibration worksheet; avoid converting image counts into population estimates.

Candidate input: Snapshot Serengeti images and documented annotation resources. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Expert count intervals, crowding strata and calibration.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

063

Camera-trap sequence deduplication

Group burst photos from one camera trap that show the same moment.

Data: LILA biodiversity datasets

Show prompt
Help me start “Camera-trap sequence deduplication” as a beginner research exercise. Use a small licensed camera-trap sequence with timestamps and privacy-sensitive content excluded. Plan duplicate grouping with a simple timestamp rule and manual pair labels, then ensure related frames cannot leak across evaluation splits.

Candidate input: LILA camera-trap datasets; sequence timestamps must be verified. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Manual pair labels and cross-split leakage audit.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

064

Nocturnal image degradation challenge

Test one model on night photos that have been deliberately darkened or blurred.

Data: LILA biodiversity datasets

Show prompt
Help me start “Nocturnal image degradation challenge” as a beginner research exercise. Select permitted day/night wildlife examples and create clearly labeled degraded copies. Plan a small exposure/noise grid, holding out natural night images so synthetic degradation is not mistaken for real-world evidence.

Candidate input: LILA camera-trap collections with night-image subsets requiring selection. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Separate natural night holdouts; accuracy and uncertainty by condition.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

065

Rare-species abstention benchmark

Test whether one confidence cutoff lets a model hold back on rare species it may get wrong.

Data: LILA biodiversity datasets

Show prompt
Help me start “Rare-species abstention benchmark” as a beginner research exercise. Choose a licensed wildlife benchmark with a species or region left out of training. Plan confidence thresholds, precision-recall and abstention coverage with a taxonomist review; suppress sensitive coordinates and do not claim reliable rare-species discovery.

Candidate input: LILA species-labeled image collections. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Precision-recall, abstention coverage and taxonomist review.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

066

Birdsong model disagreement queue

Compare models' guesses on one bird recording and send disagreements to human listeners.

Data: BirdNET

Show prompt
Help me start “Birdsong model disagreement queue” as a beginner research exercise. Start with ten licensed birdsong clips and compatible permitted models. Make a model-disagreement listening queue with blind expert checks; track verified calls per minute and distinguish overlapping or unknown calls.

Candidate input: BirdNET open-source tools; openly licensed expert-labeled audio subset is a candidate. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Blind expert labels and verified calls per listening minute.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

067

Conservation annotation training cards

Make one short training card that shows volunteers a checked way to tell two animals apart.

Data: Snapshot Serengeti

Show prompt
Help me start “Conservation annotation training cards” as a beginner research exercise. Pick one expert-verified visual distinction using licensed wildlife images. Draft a beginner training card and five pre/post questions, then request an independent specialist review before involving volunteers; do not reveal sensitive locations.

Candidate input: Snapshot Serengeti labeled images, subject to item rights. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Volunteer pre/post tests and independent expert review of every card.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

068

Biodiversity dataset license inventory

Write down one dataset's access, credit and reuse rules, quoting the exact text.

Data: LILA biodiversity datasets

Show prompt
Help me start “Biodiversity dataset license inventory” as a beginner research exercise. Choose five biodiversity datasets and inspect their actual license/access pages. Build a quote-backed inventory of attribution, redistribution, sensitive-data restrictions and unknowns; public visibility alone is not permission to reuse.

Candidate input: LILA dataset descriptions and linked licenses; each underlying license must be fetched. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Human rights review and exact-source entailment; unknown is allowed.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

069

Habitat background leakage test

Check whether a model identifies animals or just the background, by hiding the animals.

Data: LILA biodiversity datasets

Show prompt
Help me start “Habitat background leakage test” as a beginner research exercise. Use licensed wildlife images with approved animal-region masks and site labels. Plan a controlled masked-versus-retained comparison with site-held-out evaluation, noting that masking artifacts can confound conclusions about background reliance.

Candidate input: LILA camera-trap images with authorized site splits. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Site-held-out performance and controlled masks; expert sanity check.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

070

Wetland wildlife seasonal coverage atlas

Summarize how well one site and season are covered by photos and labels.

Data: LILA biodiversity datasets

Show prompt
Help me start “Wetland wildlife seasonal coverage atlas” as a beginner research exercise. Take one permitted wetland camera dataset and aggregate timestamps to coarse site-seasons. Reconcile image totals, missing dates and label coverage without exposing sensitive sites; do not confuse seasonal camera effort with wildlife abundance.

Candidate input: LILA catalog's wetland camera datasets; timestamp availability requires checking. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Reconcile image totals, missing timestamps and sampling intervals.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

Renewable energy and resilient grids

071

Solar resource seasonal bias audit

Compare one location's seasonal sunshine data under clearly stated averaging rules.

Data: NASA POWER

Show prompt
Help me start “Solar resource seasonal bias audit” as a beginner research exercise. Choose one public solar-resource location-year with documented units and time basis. Write explicit seasonal aggregation and missing-hour rules, then compare with a reference calculation rather than implying a project-specific energy forecast.

Candidate input: NASA POWER solar and meteorological datasets. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Reference calculations, missing-data checks and unit consistency.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

072

Solar-data parameter translator

Match one plain-language solar data request to the right variables, units and time basis.

Data: NASA POWER

Show prompt
Help me start “Solar-data parameter translator” as a beginner research exercise. Write five beginner solar-analysis requests and map each to exact documented variables, units and timezone conventions. Include an unsupported request that must be rejected and have a solar analyst check the mapping before code generation.

Candidate input: NASA POWER documentation and parameter dictionary links. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Expert gold mappings; reject unsupported variables and invented conversions.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

073

Grid load calendar anomaly scout

Check one country's electricity demand data for duplicate times, gaps and calendar shifts.

Data: Open Power System Data

Show prompt
Help me start “Grid load calendar anomaly scout” as a beginner research exercise. Use one public country-year load series and a documented calendar. Draft checks for duplicated hours, gaps and daylight-saving transitions; retain original timestamps and avoid filling operational grid data by guesswork.

Candidate input: Open Power System Data time-series packages. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Known calendar transitions, package schema and independent parsing.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

074

Heating-demand assumption cards

Pull out one scenario's temperatures, units and conversion assumptions, with citations.

Data: Open Power System Data

Show prompt
Help me start “Heating-demand assumption cards” as a beginner research exercise. Choose one open heating-demand scenario and extract quoted assumptions about temperatures, units and conversions. Make an assumptions card with unknowns and a dimensional check, then request an energy modeler's review rather than recommending building changes.

Candidate input: OPSD When2Heat and weather package documentation. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Source-span agreement and energy-modeler review.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

075

Grid-model constraint explanation bench

Explain one rule in a power-grid model in plain words, and link it to its inputs.

Data: PyPSA documentation

Show prompt
Help me start “Grid-model constraint explanation bench” as a beginner research exercise. Choose one constraint from a small open grid-model tutorial. Draft a plain-language explanation linked to exact variables and units, include a symbolic/dimensional check, and prepare a power-engineer rubric without implying operational advice.

Candidate input: PyPSA public documentation and small published examples. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Power engineer rubric plus symbolic/dimensional checks.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

076

Renewable dispatch replication grid

Rerun one small power-grid scenario with fixed software versions and settings.

Data: PyPSA documentation

Show prompt
Help me start “Renewable dispatch replication grid” as a beginner research exercise. Find a licensed tiny dispatch example with a documented baseline solver. Pin inputs, solver and limits; draft a reproduction notebook with feasibility residuals and objective tolerance before attempting any distributed solve or live-grid connection.

Candidate input: PyPSA public example networks and OPSD time series. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Feasibility residuals, objective tolerance and independent reruns.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

077

Battery-material provenance filter

Summarize how one battery-material property was measured or computed, with units and source.

Data: Materials Project open data

Show prompt
Help me start “Battery-material provenance filter” as a beginner research exercise. Select five public battery-material property records with explicit licenses and methods. Build provenance cards preserving units, measurement/computation distinctions and missing methods; do not propose synthesis, cell construction or safety testing.

Candidate input: Materials Project documentation and public data-access routes; subset authorization unverified. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Curator checks against original records; abstain on missing methods.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

078

Materials-model domain-shift audit

Test one permitted materials model on a family of materials it was not trained on.

Data: Materials Project open data

Show prompt
Help me start “Materials-model domain-shift audit” as a beginner research exercise. Find a permitted pretrained material-property model and a compatible licensed benchmark. Plan one composition-family holdout against a numeric baseline, checking domain and uncertainty limits; do not turn predicted properties into manufacturing instructions.

Candidate input: Materials Project public datasets; licensed property/structure export needs checking. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Family-disjoint splits, uncertainty calibration and numerical baseline.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

079

Energy-scenario claim verifier

Check one written energy claim against the linked numbers and assumptions.

Data: PyPSA documentation

Show prompt
Help me start “Energy-scenario claim verifier” as a beginner research exercise. Pick three claims from an open energy-scenario report and their linked numeric tables. Plan exact arithmetic, units and assumption checks with supported/unsupported/unclear labels; separate scenario assumptions from observed outcomes.

Candidate input: Public PyPSA example outputs generated by authorized researchers, candidate files not retrieved. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Exact arithmetic checks and expert entailment labels.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

080

Grid benchmark simplification audit

Compare a simplified power-grid model with the full version it came from.

Data: PyPSA documentation

Show prompt
Help me start “Grid benchmark simplification audit” as a beginner research exercise. Use a licensed small grid benchmark and one documented reduced version. Plan a paired offline solve checking balance, congestion and objective error with fixed tolerances; never connect the experiment to grid controls or operating decisions.

Candidate input: PyPSA public example networks. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Power balance, congestion patterns and objective-error limits.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

Public knowledge and humanitarian infrastructure

081

Map accessibility-tag completeness review

Flag one mapped place's missing accessibility details without guessing them.

Data: OpenStreetMap copyright

Show prompt
Help me start “Map accessibility-tag completeness review” as a beginner research exercise. Choose five mapped public amenities and the relevant accessibility-tag documentation. Build a completeness-only worksheet preserving unknowns; draft questions for local mappers instead of guessing accessibility or editing live map records.

Candidate input: OpenStreetMap openly licensed data; bounded extract not yet downloaded. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Rule-based completeness baseline and local-mapper adjudication.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

082

Humanitarian route-topology linting

Check one small map area for road lines that may break off or cross by mistake.

Data: OpenStreetMap copyright

Show prompt
Help me start “Humanitarian route-topology linting” as a beginner research exercise. Use a permitted small offline road-graph extract and inject a few clearly synthetic dangling-edge defects in a copy. Plan a topology linter and local-mapper review sheet; do not offer navigation or emergency-routing assurances.

Candidate input: OpenStreetMap public road graph under its license. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Synthetic graph defects and local mapper validation.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

083

Building-map disagreement adjudication

Compare one computer-drawn building outline with the photo and the existing map.

Data: Humanitarian OpenStreetMap Team

Show prompt
Help me start “Building-map disagreement adjudication” as a beginner research exercise. Choose one small area with imagery explicitly licensed for this mapping use. Prepare a disagreement overlay between permitted building footprints and existing geometry, with uncertain cases for local review; make no live edits or inferences about occupants.

Candidate input: HOT public tools and OpenAerialMap imagery; exact licensed imagery subset requires checking. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Expert-reviewed polygons, overlap metrics and local mapper review.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

084

Public-domain OCR error restoration

Suggest fixes for scanning errors in one paragraph while keeping the original text.

Data: Project Gutenberg permissions

Show prompt
Help me start “Public-domain OCR error restoration” as a beginner research exercise. Select one verified public-domain scanned page with its original image. Draft a paragraph-level OCR correction workflow that preserves the raw transcript and marks uncertain characters; compare corrections against the scan rather than inventing missing text.

Candidate input: Project Gutenberg texts with item-level rights; paired scan source still a candidate. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Human scan comparison; character error rate and false-correction rate.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

085

Scientific glossary translation fidelity

Translate one short science definition, keeping key terms matched.

Data: Gravitational Wave Open Science Center

Show prompt
Help me start “Scientific glossary translation fidelity” as a beginner research exercise. Choose one openly licensed technical definition and a target language. Plan aligned source/translation terms, numbers and units plus a glossary; require bilingual domain review and mark phrases without a secure equivalent.

Candidate input: Public GWOSC/PyPSA documentation; reuse license and target-language gold set require checking. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Bilingual domain experts score meaning and unit preservation.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

086

Citation-graph broken-link census

Check a small batch of paper links for broken ones, without overloading the service.

Data: Crossref REST API

Show prompt
Help me start “Citation-graph broken-link census” as a beginner research exercise. Start with ten public DOI links and the resolver's published usage rules. Draft a rate-limited checker distinguishing temporary errors, redirects and genuinely unresolved IDs; preserve dated logs and avoid treating one failed request as permanent breakage.

Candidate input: Crossref public DOI metadata and OpenAlex research index. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: HTTP outcome logs, duplicate suppression and sampled manual checks.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

087

Scientific abstract claim-span benchmark

Match one short AI-written claim to the exact source text that supports it.

Data: OpenAlex documentation

Show prompt
Help me start “Scientific abstract claim-span benchmark” as a beginner research exercise. Use one openly licensed scientific abstract and write three modest candidate claims, including one unsupported claim. Plan exact span alignment and blinded entailment labels with an abstain option; make no claim stronger than the source.

Candidate input: OpenAlex/Crossref metadata; reusable abstract availability and rights require checking. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Blinded entailment labels; unsupported-claim and abstention rates.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

088

Data-availability statement parser

Turn one paper's data-availability statement into clear access terms and links.

Data: Crossref REST API

Show prompt
Help me start “Data-availability statement parser” as a beginner research exercise. Select five openly licensed data-availability statements and quote access conditions verbatim. Build a table of identifiers, public/request-only/restricted/unknown status and dated URL checks; do not attempt access-control workarounds.

Candidate input: Publisher-linked openly licensed articles from Crossref; full-text corpus not yet verified. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Manual statement gold set and URL verification; distinguish request-only data.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

089

Citizen-science instruction ambiguity test

Write different readings of one volunteer instruction so people can test whether it is clear.

Data: Zooniverse about

Show prompt
Help me start “Citizen-science instruction ambiguity test” as a beginner research exercise. Choose one licensed citizen-science instruction and draft two plausible interpretations plus clarifying wording. Prepare a small opt-in comprehension exercise for project-owner approval, with consent and minimal data collection before recruiting anyone.

Candidate input: Zooniverse public project instructions; licensed text subset requires confirmation. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Controlled volunteer comprehension study with expert-approved examples.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

090

Science-plot accessible description audit

Write one plain description of a science chart, based on its axes and data.

Data: NOAA carbon dioxide trends

Show prompt
Help me start “Science-plot accessible description audit” as a beginner research exercise. Pick one openly licensed chart with accessible source data. Draft a description covering axes, units, directions and intervals, then check each numeric statement against the data; prepare an accessibility-review checklist rather than guessing hidden values.

Candidate input: NOAA GML public trend plots and accompanying descriptions. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Blind accessibility review and checks of signs, units and intervals.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

Making volunteer science trustworthy

091

Scientific abstention challenge set

Answer one test question, or say it cannot be answered, citing the exact text.

Data: RCSB PDB data

Show prompt
Help me start “Scientific abstention challenge set” as a beginner research exercise. Use a short openly licensed scientific passage to write five answerable and five deliberately unanswerable questions. Keep a separate expert-checkable answer key and source spans; score unsupported answers and appropriate abstentions, not fluency.

Candidate input: Public PDB, GWOSC and NOAA documentation excerpts. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Expert answerability labels, unsupported claims and calibrated refusal rates.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

092

Numerical result explanation cross-check

Explain one fixed result, naming its units, assumptions and uncertainty.

Data: PyPSA documentation

Show prompt
Help me start “Numerical result explanation cross-check” as a beginner research exercise. Choose one published teaching calculation with a known reference value. Draft an explanation of units, assumptions and uncertainty, independently recompute it, and prepare a blinded expert checklist; do not infer new experimental findings.

Candidate input: PyPSA tutorials and GWOSC worked examples. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Reference values, dimensional checks and blinded expert grading.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

093

Small-model energy-per-accepted-record trial

Run one fixed labeled batch through a permitted small model on your own computer.

Data: ClinicalTrials.gov public study sample

Show prompt
Help me start “Small-model energy-per-accepted-record trial” as a beginner research exercise. Choose ten licensed labeled records and a permitted local model, plus a rules-only baseline. Plan a capped measurement run logging accepted records, idle-subtracted energy and reviewer minutes; if energy cannot be measured, report that limit rather than estimate a result.

Candidate input: The public registry and archive inputs of ideas 1, 42 and 55. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Accepted records per incremental Wh and reviewer minute versus rules.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

094

Volunteer-result corruption challenge

Send deliberately wrong results to an offline test checker, never a live project, to see if it catches them.

Data: BBBC005 synthetic cells

Show prompt
Help me start “Volunteer-result corruption challenge” as a beginner research exercise. Build a tiny offline validator fixture with clearly synthetic correct results and seeded rounding, missing-field and checksum faults. Plan detection and false-rejection checks; never send corrupt submissions to a real volunteer project or other people's systems.

Candidate input: BOINC-style task design and public BBBC005 reference inputs. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Measure detection, honest rejection and replication overhead on seeded faults.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

095

Scientific-document prompt-injection benchmark

Pull one fact from a document while ignoring any instructions hidden inside it.

Data: PyPSA documentation

Show prompt
Help me start “Scientific-document prompt-injection benchmark” as a beginner research exercise. Create a local toy scientific document containing harmless canary text and an instruction to ignore the extraction task. Plan a fixed-field extractor that treats document instructions as data; test clean accuracy and rejection without real secrets, network actions or third-party targets.

Candidate input: Public documentation excerpts and locally authored benign adversarial strings. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Canary instruction rejection, schema adherence and clean-task accuracy.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

096

Internet-sharded inference suitability study

Replay a recorded or simulated slow connection locally to see how a model split across computers holds up.

Data: Petals repository

Show prompt
Help me start “Internet-sharded inference suitability study” as a beginner research exercise. Use a local simulator with a documented measured trace or clearly labeled synthetic latency/bandwidth trace. Plan a small independent-task replay against complete local-model execution, including transfer and failure overhead; do not deploy to public machines.

Candidate input: Petals and Hivemind public software documentation; public text fixtures. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Compare accepted-task throughput against complete local-model execution.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

097

Mixed-hardware numerical drift census

Run one fixed image batch on one hardware setup and check whether the answers drift.

Data: Zoobot galaxy morphology

Show prompt
Help me start “Mixed-hardware numerical drift census” as a beginner research exercise. Choose one fixed licensed image batch and permitted model. Plan repeated runs on hardware you own or are authorized to use, recording precision, seeds and versions; evaluate prediction differences against task tolerances rather than assuming bitwise equality.

Candidate input: Zoobot public weights/documentation and licensed image subset requiring checking. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Prediction disagreement, calibration and task-level tolerance checks.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

098

Scientific unit-extraction adversary set

Pull one number with its unit and source from a document, including tricky cases.

Data: NASA POWER

Show prompt
Help me start “Scientific unit-extraction adversary set” as a beginner research exercise. Write ten small source-attributed numeric examples spanning compatible, incompatible and ambiguous units. Keep a separate exact-field gold set and dimensional checks; require abstention when the source cannot support a conversion.

Candidate input: NASA POWER, PDB and GHCN documentation. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Exact-field gold set, dimensional constraints and explicit abstention.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

099

Model-card scientific reproducibility curator

Build one cited checklist of what a model needs: weights, input format, license and known caveats.

Data: BirdNET

Show prompt
Help me start “Model-card scientific reproducibility curator” as a beginner research exercise. Choose one public scientific model card and its actual weight/license pages. Draft a cited checklist of input format, availability, permitted use, evaluations and missing facts; do not infer reproducibility from a downloadable file alone.

Candidate input: BirdNET and Zoobot public documentation and model links. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Maintainer-style manual review; mark absent facts unknown rather than infer.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.

100

Volunteer science error adjudication commons

Have trained reviewers settle one conflicting result and record why they decided.

Data: BBBC005 synthetic cells

Show prompt
Help me start “Volunteer science error adjudication commons” as a beginner research exercise. Prepare five non-sensitive, licensed or explicitly synthetic conflicting-result cases. Draft an adjudication form with evidence, rationale, hidden gold checks and expert appeal, then outline an opt-in reviewer test plan without recruiting or publishing scores yet.

Candidate input: BBBC005 ground truth plus public GWOSC tutorial reference tasks. First verify access, license, attribution and data suitability; if unavailable, list the blocker, not invented data. Build the validation plan around: Hidden gold tasks, inter-rater agreement and expert appeals.

Return a small, resource-capped first-session plan, a worksheet/schema and a qualified-review checkpoint. Cite source URLs; distinguish planned steps from executed results. Use no confidential employer material, personal records or unsupported scientific/medical claims.
About this list

On hold, not abandoned. Other priorities come first.

Check the permissions and start with a baseline, not a claim of discovery. A prompt is a planning aid: verify citations, arithmetic, code and outputs yourself. An assistant may not have the data or the tools to run it.

Public access is not a reuse license. Use public or explicitly permitted data. No confidential employer files or personal records. Stay with nonclinical, offline research. No diagnosis, patient matching, biological experimentation, live operational decisions or unsupported findings.

Candidate datasets and existing programs are sources or analogs, not proof that a proposed service works, exists or has a committed partner. Citations identify inputs, documentation or enabling analogs, not proof of feasibility. Inclusion is attribution, not endorsement.

The prompts are starting aids, not tested protocols or scientific results. Sources were retrieved November 15 and 16, 2026.

Moonshots, episode 311

The discussion that inspired this collection was recorded November 13, 2026. Its official description names the four participants. [49]

Watch or listen to episode 311

Jensen’s Departure, NVIDIA’s Succession Plan, AI Doomer Response, and Demis Talks RSI | #311 (Live)

The SETI@home reference is at about 1:42:00. This is a source check, not a check of the episode’s claims. [50]

Jump to the SETI@home discussion at about 1:42:00

Participant identities and LinkedIn destinations come from the official episode description. Website identities were retrieved separately. Intelligent Internet is the organization linked for Emad. [49][51][52][53][54]

Sources
  1. [1] BOINC overview

    https://boinc.berkeley.edu

  2. [2] Folding@home science

    https://foldingathome.org/science

  3. [3] Einstein@Home

    https://einsteinathome.org

  4. [4] Climateprediction.net

    https://main.cpdn.org

  5. [5] SETI@home current analysis

    https://news.berkeley.edu/2026/01/12/for-21-years-enthusiasts-used-their-home-computers-to-search-for-et-uc-berkeley-scientists-are-homing-in-on-100-signals-they-found/

  6. [6] SETI@home

    https://setiathome.berkeley.edu

  7. [7] Zooniverse about

    https://www.zooniverse.org/about

  8. [8] Petals repository

    https://github.com/bigscience-workshop/petals

  9. [9] Hivemind repository

    https://github.com/learning-at-home/hivemind

  10. [10] exo repository

    https://github.com/exo-explore/exo

  11. [11] openFDA drug adverse events

    https://open.fda.gov/apis/drug/event

  12. [12] BBBC image collection

    https://bbbc.broadinstitute.org

  13. [13] RCSB PDB data

    https://www.rcsb.org/pages/about-us/index

  14. [14] Gravitational Wave Open Science Center

    https://gwosc.org

  15. [15] MAST TESS data

    https://archive.stsci.edu/missions-and-data/tess

  16. [16] ESA Gaia archive

    https://www.cosmos.esa.int/web/gaia/data

  17. [17] NASA Exoplanet Archive

    https://exoplanetarchive.ipac.caltech.edu

  18. [18] NASA Planetary Data System

    https://pds.nasa.gov

  19. [19] Breakthrough Listen open data

    https://seti.berkeley.edu/opendata

  20. [20] Sloan Digital Sky Survey data

    https://www.sdss.org/dr18

  21. [21] OpenAQ

    https://openaq.org

  22. [22] NOAA Global Historical Climatology Network daily

    https://www.ncei.noaa.gov/products/land-based-station/global-historical-climatology-network-daily

  23. [23] ERA5 single levels

    https://cds.climate.copernicus.eu/datasets/reanalysis-era5-single-levels?tab=overview

  24. [24] NASA FIRMS

    https://firms.modaps.eosdis.nasa.gov

  25. [25] NOAA carbon dioxide trends

    https://gml.noaa.gov/ccgg/trends

  26. [26] Argo data

    https://argo.ucsd.edu/data

  27. [27] Ocean Biodiversity Information System

    https://obis.org

  28. [28] CoralNet

    https://coralnet.ucsd.edu

  29. [29] NOAA passive acoustic data

    https://www.ncei.noaa.gov/products/passive-acoustic-data

  30. [30] LILA biodiversity datasets

    https://lila.science/datasets

  31. [31] BirdNET

    https://birdnet.cornell.edu

  32. [32] NASA POWER

    https://power.larc.nasa.gov

  33. [33] Open Power System Data

    https://open-power-system-data.org

  34. [34] PyPSA documentation

    https://docs.pypsa.org/latest

  35. [35] Materials Project open data

    https://docs.materialsproject.org

  36. [36] OpenAlex documentation

    https://docs.openalex.org

  37. [37] Crossref REST API

    https://www.crossref.org/documentation/retrieve-metadata/rest-api

  38. [38] OpenStreetMap copyright

    https://www.openstreetmap.org/copyright

  39. [39] Humanitarian OpenStreetMap Team

    https://www.hotosm.org

  40. [40] Project Gutenberg permissions

    https://www.gutenberg.org/policy/permission.html

  41. [41] Zoobot galaxy morphology

    https://github.com/mwalmsley/zoobot

  42. [42] Gravity Spy

    https://www.zooniverse.org/projects/zooniverse/gravity-spy

  43. [43] ClinicalTrials.gov public study sample

    https://clinicaltrials.gov/api/v2/studies?pageSize=1&format=json

  44. [44] Snapshot Serengeti

    https://lila.science/datasets/snapshot-serengeti

  45. [45] BBBC005 synthetic cells

    https://bbbc.broadinstitute.org/BBBC005

  46. [46] GHCN daily public directory

    https://www.ncei.noaa.gov/pub/data/ghcn/daily

  47. [47] Gravity Spy machine learning and citizen science

    https://arxiv.org/abs/1611.04596

  48. [48] OSCAR ocean surface currents

    https://podaac.jpl.nasa.gov/dataset/OSCAR_L4_OC_FINAL_V2.0

  49. [49] Moonshots episode 311: official recording and description

    https://www.youtube.com/watch?v=GZrDapFXk0I

    https://www.youtube.com/watch?v=GZrDapFXk0I&t=6120s

    Timestamp for the SETI@home discussion, at about 1:42:00.

  50. [50] Real Alpha: Moonshots SETI@home transcript excerpt

    https://realalpha.app/s/ai-can-surface-objects-hidden-in-existing-astronomy-data

  51. [51] Peter H. Diamandis: official website

    https://www.diamandis.com

  52. [52] Salim Ismail: official website

    https://salimismail.com

  53. [53] Alexander Wissner-Gross: official website

    https://www.alexwg.org

  54. [54] Intelligent Internet: organization linked for Emad Mostaque

    https://www.ii.inc

  55. [55] Salim Ismail: LinkedIn

    https://www.linkedin.com/in/salimismail

  56. [56] Alexander Wissner-Gross: LinkedIn

    https://www.linkedin.com/in/alexwg

  57. [57] Emad Mostaque: LinkedIn

    https://www.linkedin.com/in/emad-mostaque-9840ba274