README.md

ATLAS Data Observatory

See the values. Read the assumptions. Follow every figure to its table.

Scientific figure gallery · Complete table archive · Machine-readable inventory · Proposed data contracts · Research register

Observatory profile	Recorded status
On the lab bench	12 actual CSVs with 98,759 stored rows and 61 exact column definitions. Rows combine different record types and are not independent physical observations.
Public archive window	200 published catalog rows, 193 host names, five discovery-method categories, and nine blank metallicity fields.
Numerical workbench	11 synthetic/reference CSVs from eight reduced-model demonstrations; named assumptions, parameters and units accompany them.
On the drawing board	117 proposed data contracts and header-only acquisition templates, distinct from acquired or generated scientific tables.
Figure wall	Nine additional diagnostic spreads, each as SVG and PNG with an evidence ledger; the original nine numerical figures remain intact.
Chain of custody	The inventory generator verifies the original 40-asset model manifest and records exact source hashes before making a table profile.

Current reading route: open a spread → inspect its exact CSV → check the units and missingness → read the assumptions → follow the corresponding engineering record.

A real 200-row NASA Exoplanet Archive snapshot: period and radius values, discovery methods, and missing host-metallicity fields. Its recorded request uses a 200-row limit and name ordering among rows with period and radius; it supports descriptive exploration rather than population occurrence inference.

Open the CSV · Read acquisition provenance · Inspect figure provenance

Open the catalog acquisition card — query, included fields and limits

The saved snapshot was retrieved on 2026-10-02 at 09:01:19 UTC from the NASA Exoplanet Archive pscomppars table:

SELECT TOP 200 pl_name, hostname, pl_orbper, pl_rade, st_met, discoverymethod
FROM pscomppars
WHERE pl_orbper IS NOT NULL AND pl_rade IS NOT NULL
ORDER BY pl_name

The six fields identify the planet and host, orbital period in days, radius in Earth radii, host metallicity in dex relative to solar, and discovery-method category. The saved extract omits per-parameter error bars, reference links, metallicity abundance basis and detection completeness. The column-by-column inventory preserves exact headers and displays a compact preview. Repeating the query now could return different values or rows; the checked-in snapshot is identified by its original hash and retrieval record.

The unchanged acquisition sidecar and original numerical view describe an alphabetically truncated selection. Their wording remains historical acquisition metadata. The recorded request uses TOP 200 and ORDER BY pl_name; only the returned 200 rows were retained, so this repository does not independently verify their ranking against the full archive. Current diagnostic captions describe the saved query-ordered extract. Use the exact request and preserved response together when assessing selection behavior.

Discovery categories in this extract are Transit (149), Radial Velocity (46), Pulsar Timing (3), Astrometry (1) and Transit Timing Variations (1). These counts describe this selected file, not the archive population or relative effectiveness of detection methods.

Three evidence classes

Visible label	What is actually present	Where to read it
	Published archive values with a saved query, access time, row selection and original-file hash. Composite catalog values may include estimates or combine publications; no error bars were acquired in this small extract.	200-row exoplanet snapshot and acquisition record
	Eight reduced-model demonstrations, with invented teaching parameters, generated numerical fields, simulated noise draws or a formula-based reference curve.	Model definitions and limits, CSV outputs and immutable asset manifest
	117 proposed record schemas, field dictionaries and header-only acquisition CSVs covering 877 defined fields. These describe future evidence and contain no original project measurements.	Contract guide and the data companions in each research project folder

Read a scientific spread

Nine additional figure pairs expose relationships that are easier to assess visually: a heat budget, a water ledger, actuator saturation, orbital conservation, inverse-problem information, phase regions, finite numerical resolution and approximation error. Each pairs a vector SVG with a shareable PNG and an evidence ledger. The original nine model figures remain available in the Model Foundry.

Explore	Data to inspect	Engineering record
Catalog values and coverage	200 rows, discovery-method categories and nine missing metallicity values	Session C
Thermal power and response	Signed component powers and two node temperatures	Session E
Hydrologic water accounting	Effective recharge, outflow, storage and conservation residual	Session B
Control authority	Angle/rate states, requested torque and actuator bounds	Session I
Orbital numerical accuracy	Energy error, angular momentum and three timestep resolutions	Session I
Spectral information	2,000 synthetic noise-realization fraction estimates	Sessions C and H
Ideal-binary phase regions	A hypothetical liquidus and eutectic under declared assumptions	Session A
Fractal numerical resolution	Finite escape iteration fields and unresolved sampled points	Session A
Drag reference departures	A drag correlation compared with its creeping-flow asymptote	Session D

The laboratory wall

The spreads are built from the same source tables, with additional views that reveal signs, balances, units, solver behavior and missing information. Each image opens at full resolution; its title and footer identify the evidence class.

Heat flows and thermal lag	Water accounting and response
Signed powers expose energy entering or leaving the wall. Compare their sum with how node temperatures evolve. Synthetic thermal model.	Input, delayed release and cumulative storage/discharge expose a closed conceptual water balance. Synthetic reservoir model.
Fields and units · Full spread	Fields and units · Full spread

State evolution and actuator authority	Conservation and refinement
Follow the time-colored phase portrait, then compare requested torque with the applied actuator-limited value. Synthetic one-axis model.	Read bounded energy error beside the measured response to timestep refinement. Synthetic normalized two-body model.
Fields and units · Full spread	Orbit fields · Refinement fields

Information in an inverse problem	Regions of a hypothetical phase diagram
Nearly identical endmembers magnify noise-driven fraction variability. Unphysical estimates remain visible. Synthetic noise experiment.	Read liquid, liquid-plus-solid and two-solid regions under ideal-liquid/pure-solid assumptions. Invented teaching constants.
Spectral fields · Noise-draw fields	Phase-diagram fields · Full spread

Finite numerical resolution	A reference approximation's departure
Distinguish the first-escape iteration from a point that remains unresolved after the iteration budget. Numerical fields.	Follow the difference between a sphere-drag reference formula and its creeping-flow asymptote. Correlation evaluations.
Mandelbrot fields · Julia fields	Drag fields · Full spread

Every actual table, in view

The complete table archive expands this shelf into all 61 field definitions, recorded/verified units, finite stored ranges, column missingness, compact first-row previews, parameters, assumptions and unit-source links. Each CSV remains in its single original location under models/data.

Exact CSV	Rows × columns	Evidence	Profile / source
exoplanet_sample.csv	200 × 6	Real public catalog	Fields · Acquisition record
balloon_thermal.csv	1,441 × 10	Synthetic	Fields · Sidecar
hydrologic_reservoir.csv	241 × 7	Synthetic	Fields · Sidecar
one_axis_attitude.csv	2,401 × 4	Synthetic	Fields · Sidecar
two_body_orbit.csv	4,001 × 7	Synthetic	Fields · Sidecar
two_body_convergence.csv	3 × 2	Synthetic	Fields · Shared orbit sidecar
spectral_mixture.csv	180 × 7	Synthetic	Fields · Sidecar
spectral_monte_carlo.csv	2,000 × 3	Synthetic	Fields · Shared spectral sidecar
ideal_binary_liquidus.csv	600 × 4	Synthetic	Fields · Sidecar
mandelbrot.csv	43,621 × 4	Synthetic numerical field	Fields · Sidecar
julia.csv	43,621 × 4	Synthetic numerical field	Fields · Sidecar
sphere_drag.csv	450 × 3	Formula reference	Fields · Sidecar

Pick a diagnostic route

flowchart LR
    A[Choose a scientific question] --> B{What needs checking?}
    B --> C[Energy and water accounting]
    B --> D[Numerical approximation]
    B --> E[Information and missingness]
    C --> F[Thermal powers and reservoir ledger]
    D --> G[Orbit refinement, drag and fractal budget]
    E --> H[Catalog coverage and spectral noise draws]
    F --> I[Inspect CSV fields and source sidecar]
    G --> I
    H --> I
    I --> J[Read assumptions and engineering proposal]

Loading

Route	A careful sequence	Read before making a claim
Energy detective	Inspect absorbed_W, convection_W, radiation_W and wall_to_payload_W; follow their signs in the thermal spread; compare resulting node trends with the prescribed boundaries.	Positive enters the named node. A trend generated by imposed conditions is not a measured thermal limit.
Water-accounting auditor	Open cumulative recharge, cumulative discharge and storage in the reservoir profile; check their ledger; then inspect effective recharge versus outflow.	A small accounting residual checks conservation, not the accuracy of a flood forecast.
Numerical-method reviewer	Read the three-row refinement table with the orbit record; compare resolutions; inspect the absolute relative-error definition and normalized scales.	Order estimated from three resolutions is evidence about this integration experiment, not universal solver accuracy.
Information hunter	Inspect both fraction-estimate columns before the histograms; compare the different horizontal scales; read the injected fraction and known noise model.	Central draw intervals are not posterior credible intervals or measured abundance uncertainty.
Catalog curator	Inspect six exact catalog fields; locate the nine missing metallicities; read the query and omitted uncertainty/reference columns.	A selected composite snapshot provides a descriptive view, not a survey denominator or selection-corrected occurrence rate.
Boundary skeptic	Compare fractal escape counts with distance-estimator validity; inspect the Julia profile and its critical-point caveat; check the iteration budget.	Undefined distance and unresolved escape are different conditions. A finite-grid image cannot certify every membership claim.

Missingness bulletin

The actual CSVs contain nine blank cells, 10,683 NaN cells, and zero other nonfinite numeric tokens. These counts are storage diagnostics with subject-specific meanings.

Field / table	Stored missing or undefined values	Interpretation
st_met / catalog	9 blank cells	Missing host-metallicity values in this saved extract. Values and uncertainty/reference fields must be obtained before an appropriate metallicity analysis.
distance_estimator / Mandelbrot	9,873 NaN cells	No exterior estimate for the sampled points that remain unresolved under this run's iteration budget.
distance_estimator / Julia	809 NaN cells	808 unresolved sampled points plus the critical starting point (0,0), which escapes at iteration 29 but has a vanishing derivative.
next_bin_recharge_mm_day / reservoir	1 NaN cell	The final time boundary has no following forcing bin; its future-bin input is intentionally undefined.

Reading rule: retain missingness reasons, use the explicit status/count field when one exists, and never replace a missing or undefined quantity with zero merely to make a plot look complete.

Follow the evidence

The tables in models/data are the single copy of the included numerical inputs and outputs. The new data/figures folder contains presentation assets and their source/output hashes; it duplicates no raw datasets. The figure manifest connects each plot to its exact source files, captions, transformations and linked projects.

For an analysis to become empirical evidence, retain dataset version, calibration applicability, units, coordinate/time frame, uncertainty, missingness, exclusions and transformation history as specified in the data management plan. The uncertainty rules distinguish model uncertainty, numerical error and noise-draw intervals. Specified verification work is visible in the case register; it remains separate from the checks already executed on software.

Reproduce the gallery

From the repository root, with the existing model plotting dependencies available:

python -m pip install -r models/requirements.txt
python tools/build_data_inventory.py
python tools/render_data_figures.py

The renderer reads the checked-in CSVs and sidecars, writes the nine additional SVG/PNG pairs and provenance records, and records dependency versions and its own hash. It uses no network access and does not regenerate model outputs or overwrite the original asset manifest. Exact rendered bytes can differ across plotting-library versions or platforms; numerical inputs remain identified by their hashes.

The standard-library inventory generator writes DATA_INVENTORY.json and TABLES.md. Its Python API is generate(repo_root: Path) -> dict. Counts, ranges and previews are calculated from the original tables; units retain their recorded sidecar entry or cite a verified producing-model definition. The fallback for an undocumented field is explicitly unit not recorded / TBD. Generation is deterministic for unchanged inputs and generator bytes.