General Research

Peer-reviewed (and tiered-authoritative) sources for the economicspace asteroid-mining profitability pipeline, its Stage 3 reference package spacecost, and the Stage 1 package AsteroidCatalog. Every source here is one that can actually be accessed and analyzed: full text where legally redistributable, otherwise metadata, abstract and extracted key tables. Since Round 74 the registry also lists every source those three repos cite, including the grey ones (tier T4), so each upstream dependency is visible even before it has been read. The exceptions are sources the owner has rejected, which are listed only under INDEX.md "Rejected sources".

Why this repo exists

economicspace/CITATIONS.md records what the pipeline already cites (SBDB, ssoBFT, NEOWISE V2.0, USGS/LME/yfinance prices, Izzo 2015 for Lambert). This repo backs the cells that citation list leaves weakly sourced, one domain per group of cells. Domains 1–5 were the weakest-sourced; the rest were added as research rounds found cells no domain covered (the research log records when). Each domain's FINDINGS.md lists its target cells, their current values and what a source has to supply.

#	domain	pipeline cell(s) it could back or replace
1	Bulk density by spectral type, PGM factors, population statistics	AsteroidCatalog taxonomy_composition.csv density/PGM columns; ranking quality
2	Composition → commodity value mapping (what an S/C/X/K/Q body actually contains)	taxonomy_composition.csv material fractions, mineral_value.py mineralogy
3	Δv budgets and propulsion performance tables	spacecost/reference/delta_v_segments.csv, propellants.csv Isp rows
4	Launch $/kg to LEO/GTO — history and projections	spacecost/reference/launch_vehicles.csv price rows, cost cascade
5	In-space storage (boil-off), ISRU, operational costs	storage_systems.csv; the volatile-ISRU, power and electric-propulsion rows of operational_costs.csv (power and EP overlap domain 11; the sources for its finance, licensing, operations, mining and hardware-rate rows moved to 16–21 in Round 77); environment penalties
6	Commodity market prices (the value side)	mineral_value.py ref_price_usd_per_kg, which feeds every profitability row
7	Mission program-cost benchmarks (NRE / development)	operational_costs.csv NRE rows (spacecraft development, autonomous mining control), sample-recovery ops envelope
8	NEA accessibility oracles (round-trip Δv ground truth)	distribution/rank reference for the NEA rows of delta_v_segments.csv (economicspace probe_nhats.py)
9	In-space delivery mass fractions & EDL survival	spacecost delivery.py constants such as MARS_LANDED_MASS_FRACTION
10	Launch vehicle & engine hardware / propellant performance	engine-level ground truth for propellants.csv and launch_vehicles.csv
11	Spacecraft power & electric propulsion	power and electric-propulsion rows of operational_costs.csv
12	Destination environments & physical constants	spacecost environments.csv (flux, gravity, escape velocity, light time, eclipse) and units.py constants
13	Astrodynamics methods, ephemerides & audited software	economicspace research/starred-repos/ (Lambert and Kepler solvers, ephemeris oracles, the 17-repository licence audit)
14	Propellant & consumable prices	propellants.csv ref_cost_usd_per_kg and its component prices
15	In-space demand & market absorption	economicspace mineral_value.py in-space utility (market side; the engineering basis of each row is in 28–30 since Round 161), annual-demand and class-share tables; calc.py demand_elasticity, surplus_price_fraction
16	Reliability, learning curves & hardware service life	operational_costs.csv launch reliability, spacecraft MTBF, mining success and growth, rig life, trips and salvage rows; calc.py learning_curve_rate
17	Cost of capital, contingency & insurance	operational_costs.csv WACC, contingency, launch and third-party insurance rows; calc.py contingency_fraction
18	Licensing, planetary protection & space-resource law	operational_costs.csv FAA Part 450 licensing rows and the MPL behind third-party liability; the Earth-return and right-to-sell premises (context-only)
19	Asteroid excavation throughput & beneficiation	calc.py mining rate, rig mass, beneficiation recovery and concentration cap; operational_costs.csv drilling and beneficiation energy, plant throughput; mineral_value.py in-space refining energy
20	Mission operations, communications & ground segment	operational_costs.csv mission operations, DSN time, downlink and depot berthing rows
21	Spacecraft mass & recurring-cost estimating relationships	operational_costs.csv recurring $/kg rows by hardware class; calc.py return-structure and heat-shield fractions, NRE/recurring overlap
22	Prior-art techno-economic benchmarks	economicspace's headline answer (campaign/results.csv best_obj, winner): an external oracle for profitability, as domain 8 is for Δv
23	Cost indices, dollar years & exchange rates	the reference_year of every spacecost table, the CPI factors in row notes, and the exchange rates behind non-US prices
24	Aerocapture & aerobraking	calc.py aero-assisted returns (LEO/Mars trim, GEO aerocapture, fallback saving); delta_v_segments.csv aerocapture and aerobraked rows
25	Net smelter return: refining charges & payable terms	calc.py _mineral_implied_value (100% of contained metal at refined prices) and the refined-metal basis of element prices
26	Spacecraft storage & energy-system physical properties (propellant tankage / COPV / MLI / cryocooler; RTG / fission surface power)	storage_systems.csv propellant-storage + energy-storage cells (COPV burst performance factor, MLI passive, orbital-depot boil-off, RTG specific power, Kilopower-class fission W-elec/kg); the cargo-containment cells stay upstream judgement rows for now
27	Orbital debris & collision risk	the debris environment behind operational_costs.csv launch-insurance premium drivers (domain 17), economicspace's satellite-lifetime and disposal assumptions, and the depot mission durations of domain 26
28	Radiation-shielding value of in-space materials	the rock rows of economicspace mineral_value.py IN_SPACE_UTILITY (silicates, oxides, rock-valued carbonates and phosphates), the shielding part of water's row, and their destination overrides
29	Structural metals, alloys & in-space manufacturing	the structural-class rows of IN_SPACE_UTILITY (Fe, Ni, Co, Cu, Cr, Ti, Fe-Ni alloy phases and their ores): the manufacturing-gap discount and its destination overrides
30	Volatiles, process chemicals & catalyst metals in space	the propellant, chemical and trace-class rows of IN_SPACE_UTILITY (water, ammonia, CO2, carbon, organics, S, P, PGMs and other trace metals) and the rejected PGM catalyst utility

Layout

README.md                                This file: what it is and how to read it
INDEX.md                                 Master index: every source → tier, access, pipeline mapping; rejected sources and research log at the bottom
sources.csv                              Machine-readable registry, one row per source (GENERATED — see "Tools")
full_texts_manifest.csv                  Every hosted file → its source id, size, sha256 and licence
revision_candidates.csv                  Upstream cells the evidence says should change, with current status
upstream_commit_map.csv                  Upstream commits cited by checked_against that a history rewrite took off main, mapped to their main equivalents
DOWNLOADS.md                             Checklist of the non-hosted sources still to fetch for a full extraction, with exact links
AGENTS.md                                Working rules for agents running research rounds here
tools/                                   build_registry.py (regenerate derived files), validate.py (consistency checks) and tests/ (planted-defect tests of both, a coverage gate and a debug script)
01_density_and_population/               Domain 1: density by spectral type, PGM factors, population stats
02_composition_value/                    Domain 2: composition → commodity value mapping
03_dv_propulsion/                        Domain 3: Δv budgets, propulsion performance tables
04_launch_economics/                     Domain 4: launch cost per kg — history and projections
05_inspace_operations/                   Domain 5: storage/boil-off, ISRU, in-space operational costs
06_commodity_market_pricing/             Domain 6: commodity/mineral market prices
07_mission_program_costs/                Domain 7: mission program-cost benchmarks (NRE / development anchors)
08_nea_accessibility_oracles/            Domain 8: NEA accessibility oracles (round-trip Δv ground truth)
09_delivery_mass_fractions_edl_survival/ Domain 9: in-space delivery mass fractions & EDL survival
10_launch_vehicle_engine_hardware/       Domain 10: launch vehicle & engine hardware / propellant performance
11_spacecraft_power_electric_propulsion/ Domain 11: spacecraft power & electric propulsion
12_destination_environments_physical_constants/ Domain 12: destination environments & physical constants
13_astrodynamics_methods_ephemerides_software/  Domain 13: astrodynamics methods, ephemerides & audited software
14_propellant_consumable_prices/          Domain 14: propellant & consumable prices
15_inspace_demand_market_absorption/      Domain 15: in-space demand & market absorption
16_reliability_learning_service_life/     Domain 16: reliability, learning curves & hardware service life
17_cost_of_capital_contingency_insurance/ Domain 17: cost of capital, contingency & insurance
18_licensing_planetary_protection_space_law/ Domain 18: licensing, planetary protection & space-resource law
19_asteroid_excavation_beneficiation/     Domain 19: asteroid excavation throughput & beneficiation
20_mission_operations_communications/     Domain 20: mission operations, communications & ground segment
21_mass_cost_estimating_relationships/    Domain 21: spacecraft mass & recurring-cost estimating relationships
22_prior_art_techno_economic_benchmarks/   Domain 22: prior-art techno-economic benchmarks
23_cost_indices_dollar_years/             Domain 23: cost indices, dollar years & exchange rates
24_aerocapture_aerobraking/               Domain 24: aerocapture & aerobraking
25_net_smelter_return_refining_terms/     Domain 25: net smelter return: refining charges & payable terms
26_spacecraft_storage_energy_properties/     Domain 26: spacecraft storage & energy-system physical properties
27_orbital_debris_collision_risk/          Domain 27: orbital debris & collision risk
28_radiation_shielding_value_of_materials/ Domain 28: radiation-shielding value of in-space materials
29_structural_metals_inspace_manufacturing/ Domain 29: structural metals, alloys & in-space manufacturing
30_volatiles_chemicals_catalysts_inspace_use/ Domain 30: volatiles, process chemicals & catalyst metals in space

<domain>/sources_domain.csv              The registry rows for that domain (the file you edit)
<domain>/FINDINGS.md                     Prose write-ups and per-number comparisons, one block per round
<domain>/full_texts/                     Legally redistributable full texts (arXiv PDFs, NASA public-domain docs, CC-BY articles)
<domain>/extracted_data/                 CSVs of the numbers pulled out of each source, with page/table locations

Source tiers

• T1 — Peer-reviewed journal articles.
• T2 — NASA / ESA technical reports and AIAA / IAC proceedings (institutionally reviewed, not peer-reviewed in the journal sense).
• T3 — Authoritative government or dataset publications with a DOI (e.g., USGS MCS, PDS releases, NEOWISE V2.0).
• T4 — Secondary / grey sources that upstream rows cite: company documents and price pages, news reporting, encyclopedias, vendor or market-research posts, textbooks, software and market-data APIs (added in Round 74). A T4 row marks a cell whose evidence is weak; replacing it with a T1–T3 source is the usual next step.

Each row of sources.csv carries its tier label explicitly.

Access classes

The first word of every access_status is one of these; the rest of the field says how it was verified or why it could not be.

class	meaning
full_text_hosted	full text committed under <domain>/full_texts/ and listed in full_texts_manifest.csv
public_domain_excerpt_hosted	public-domain document too large to commit; the relevant passage is quoted in FINDINGS.md
verified_live_not_pulled	fetched and read from this machine, but not committed
open_not_pulled	open access, but the publisher blocks this machine or the licence forbids hosting; metadata + abstract recorded
open_service	a live database or API, verified from this machine; derived numbers committed, not the service itself
skipped	deliberately not hosted; the user decision is recorded in the row
registered_not_pulled	an upstream citation registered so the dependency is visible; its DOI was checked against Crossref or its landing page against a live request, but the full text has not been sought and nothing is extracted. Re-class the row when it is pulled

Access rules (this repo is public)

1. Full texts are committed only when legally redistributable: arXiv preprints under their licenses, NASA/ESA documents in the public domain or CC-licensed, and journal articles that are genuinely open access with a license permitting redistribution. Each hosted file's licence is recorded in the license column of full_texts_manifest.csv, read from the file or its record (PDF licence statement, NTRS copyright determination, arXiv abs page). arXiv's default licence (nonexclusive-distrib/1.0) lets arXiv distribute a paper but does not let anyone else, so only CC-licensed arXiv versions qualify. validate.py warns on every hosted file whose licence does not permit redistribution.
2. Paywalled items get: full metadata, abstract (as published), and any key tables/numbers extractable from the accessible portion — recorded in extracted_data/ with page/table references so they can be verified later.
3. Every extracted number carries its source id + location (page / table) so a claim downstream is traceable to the paper, not to this repo's summary.

How an item earns a place here

A candidate must satisfy all of:

1. It directly relates to one of the domains above (i.e., it can back or replace a specific pipeline table/cell — "space mining is interesting" does not qualify).
2. Its access status was actually verified from this machine (a URL that 403s at search time but serves full text passes; one that only ever shows an abstract bar does not get full_text_hosted).
3. The numbers it carries were extracted into a CSV, or the item is recorded as context-only with the reason stated in sources.csv.

A registered_not_pulled row is the one exception to 2 and 3: it records that an upstream cell cites the source, and it moves to one of the other classes when an extraction round reads it.

Adding a domain

Open a domain when upstream cells have no domain behind them, not for a topic. A new domain needs:

1. A folder NN_short_name/ (the next two-digit number; lowercase letters, digits and underscores) with sources_domain.csv holding only the header line, and FINDINGS.md holding a title, a Backings: line naming the upstream files and cells, and a first block that tables those cells with their current values and the upstream commit they were read from.
2. A ## Domain N — title section in INDEX.md, above ## Rejected sources, with a _Backings:_ line and the five-column table header and separator. A table with no rows is valid until a round registers a source.
3. A row in the domain table at the top of this file and a line in "Layout" (validate.py fails if a domain folder is not mentioned in this file).
4. A research-log entry for the round.

A source already registered in another domain moves to the new one when the cells it backs are the new domain's (next section); otherwise the new domain's FINDINGS.md refers to it by id.

Moving a source between domains

Move a source when the cells its pipeline_mapping names belong to another domain. Its id never changes. Round 77 set the procedure:

1. Move its row from one sources_domain.csv to the other, and its INDEX row to the end of the new domain's table.
2. git mv its hosted files (and change domain_dir in full_texts_manifest.csv) and every extracted-data CSV that cites only this source. A CSV that also cites sources staying behind stays where it is; if the moved row names it, write the path with the old domain's folder.
3. Append a block to the old domain's FINDINGS.md listing each moved id and the new paths of its files, and one to the new domain's saying what arrived and which old blocks hold its history. Older blocks and the research log stay as written, even where their relative paths no longer resolve.
4. Leave domain_dir in revision_candidates.csv alone: it names the FINDINGS.md that holds the write-up.

Rejecting a source

The owner decides which sources are rejected. A rejected source leaves the registry, and a row in INDEX.md "Rejected sources" becomes its only record: id, source (authors, year, title), link (DOI or URL) and reason (why, and in which round). No data from it is kept. The procedure, set 2026-09-28:

1. Add its row to "Rejected sources". The id stays reserved, so no new source may reuse it.
2. Delete its row from <domain>/sources_domain.csv and from its INDEX domain table.
3. git rm its hosted files and delete their full_texts_manifest.csv rows. The rule on extracting every number before un-hosting a file does not apply.
4. Delete every extracted-data CSV that cites only this source, and in CSVs shared with other sources delete each row that cites it, a;b rows included.
5. Delete its FINDINGS.md write-ups, its DOWNLOADS.md entry, and each mention of its id in other sources' registry rows and FINDINGS blocks. This is the one case where older FINDINGS blocks are edited.
6. Delete each revision candidate whose source_ids is this source alone. For one that also cites others, remove the id from source_ids and rewrite the evidence without it. A deleted rc-NNN number is not reused.
7. Leave past research-log entries as written, and add one naming the rejected id, the reason and any deleted rc-NNN.

validate.py fails if a rejected id is still registered, or if any file path or any CSV or Markdown file other than the research log names it.

Revision candidates

This repo never edits the pipeline repos. When a source contradicts a cell in spacecost, AsteroidCatalog or economicspace, the finding goes into that domain's FINDINGS.md and a row in revision_candidates.csv: target file and row, current value, proposed change, evidence and status (open, applied, declined, superseded, blocked, or withdrawn when this repo's own evidence turns out to be wrong). Each row records the upstream commit its current value was read from, so a stale status is visible. Re-check the open rows against the upstream repos whenever they release.

Tools

python tools/build_registry.py          # regenerate sources.csv; fill manifest sizes + hashes
python tools/build_registry.py --check  # fail if either is out of date
python tools/validate.py                # consistency checks; exit 1 on any error
python tools/validate.py --strict       # exit 1 on warnings too
python tools/validate.py --verbose      # also print how many items each check examined
python tools/validate.py --upstream     # also check cited commits in the sibling upstream clones
python -m unittest discover -s tools/tests   # the tools' own tests
python tools/tests/measure_coverage.py  # which lines and branches of the tools no test reaches (needs coverage.py)
python tools/tests/debug_case.py CODE   # plant a code's test cases and show each full report (--keep, --list)
python tools/tests/debug_case.py --build CODE   # the same for build_registry.py's cases; --accepted runs the valid variants

validate.py checks:

• that every CSV, Markdown, JSON and Python file, and .gitignore, is UTF-8 with no byte-order mark, no stray carriage returns and a final newline;
• that the root and the domain folders hold only what "Layout" lists, that tools/ holds only Python modules, that domains are numbered 01..N with no gap or repeat, and that .gitignore hides nothing but __pycache__/ (a hidden file is invisible to every check);
• that every sources_domain.csv is rectangular with no empty or padded cell and a well-formed year, and that sources.csv matches them;
• that each INDEX.md table lists exactly its domain's sources in registry order, with five non-empty cells per row;
• that the research log runs from Round 0 to the latest with no gaps, real dates and no mislabelled entries, and that the committed log survives unchanged as its tail (entries are only ever added at the top; this compares with HEAD, so it catches a rewrite before it is committed; during a merge the merge base's log must be the tail and every line of both sides must be kept, in either order);
• that every hosted file is in the manifest, by bare file name, with the right hash and a licence, and that .gitattributes marks it -text, so a checkout with core.autocrlf keeps its bytes (and hash) unchanged;
• that every extracted-data CSV is rectangular, has a source_id column, holds rows and no all-blank row, and cites registered ids, that every JSON file parses, and that every read source is cited by an extracted-data file of its domain or marked "Context-only: ";
• that revision_candidates.csv uses known statuses, sources, repos and dates, that a candidate names its upstream file (only a method candidate may name a model stage instead), that open and blocked candidates were checked against their own target repo, and that upstream_commit_map.csv agrees with it and holds no unused row;
• that every data file cited as extracted_data/<file> or full_texts/<file> (bare in its own domain, or <domain>/...) by a FINDINGS.md, README.md, DOWNLOADS.md, INDEX.md above the research log, a registry cell or a candidate exists, and that a <sha>:<domain>/... reference exists in that commit (a file no longer here is named without its directory);
• that no rejected source is still registered or named anywhere but the research log.

It warns, without failing, only where the decision is the owner's: hosted files whose licence does not permit redistribution, files over GitHub's 50 MB warning size, and an access_status sha256 that matches none of the source's hosted files (it may record an earlier, removed copy). registered_not_pulled rows are exempt from the citation check and counted in the summary line instead. --upstream reads the clones of AsteroidCatalog, spacecost and economicspace next to this repo. It checks that every commit a candidate cites exists, that the file it names exists there, that open and blocked candidates were checked against origin/main, and that an off-main commit on a closed candidate has a row in upstream_commit_map.csv whose commits have identical trees, whose tree and subject columns are right and whose old commit is really off main. --verbose names the origin/main commit of each clone it compared with; fetch the clones first.

Nothing fails silently. The readers stop on a malformed file instead of reading around it. A check that crashes is reported as check-crashed, one that examined nothing as nothing-checked, and git that cannot be run as git-unavailable; if git cannot list the repository's files, no check runs. Every finding carries a code. When a file another check depends on is unreadable, that failure is reported and the dependent cross-checks are skipped rather than reported a second time as false errors: while a sources_domain.csv is unreadable, no id is reported as unknown, since it may be in that file.

tools/tests builds a throwaway fixture repository (with sibling upstream clones), plants a defect for every code (one case for each place that reports it) and fails if a code has no case. Each case's report must hold exactly its code plus the codes the case declares the defect also causes, so a check that misfires on another check's input fails the suite. Cases run in parallel. It also checks that valid variants pass (CRLF line ends, a lazy-continuation paragraph in a log entry, a blank source_id), that no planted defect crashes a check, that every check function is registered, and that the tools hold no except that swallows an error and none of the fallback idioms that hid defects here before (dict.get with a default, hasattr, utf-8-sig, errors="replace"). measure_coverage.py runs the suite under coverage.py and fails unless every line and branch of the three tools is reached; run it after changing tools/. test_scripts.py runs the pass and fail paths of debug_case.py and measure_coverage.py themselves. Standard library and git only (coverage.py for measure_coverage.py alone); Python 3.8+.

Provenance of this process

Research rounds are logged at the bottom of INDEX.md (round number, date, what was added), newest first. Extraction was done interactively during each round and is not committed as scripts. Instead, every extracted number records the page or table it came from, so it can be re-checked against the hosted full text or the publisher's copy.

License

MIT, see LICENSE. That covers this repository's own work: the index, the registry and tracking CSVs, the FINDINGS.md write-ups, the records in extracted_data/ and the scripts in tools/. It does not cover the documents committed under full_texts/. Those are third-party papers and reports redistributed under the terms set out in "Access rules" above, and each keeps the license its row in sources.csv identifies.