Source registry · methodology v0.1
Where numbers are allowed to come from.
A publishable HSI value has to name its source, date, transformation, and license. HSI uses a governed hierarchy of public statistics, reproducible benchmarks, primary HSI measurements, market observations, proxies, and explicit scenarios. Every constituent in a published HSI snapshot keeps its source, vintage, classification, confidence, transformation history, and license.
Being registered does not mean a source appears in every release. Each immutable snapshot records exactly which observations were used. The explorer currently publishes research-beta snapshot research-beta-2024-acs-ce-scenario-production-v1. That snapshot is scenario-classified and not an official national result.
Source registry v0.1.0 · hsi-methodology-0.1 · Research beta
This registry defines eligible HSI evidence sources and their methodological role. Calculator scenarios and test data never become published source observations. A source enters an HSI release only through validated ingest, provenance, licensing, and publication controls.
Source hierarchy
When more than one source can measure the same phenomenon, HSI prefers the higher tier. A lower-tier source may be used only when a higher-tier source does not measure the required quantity. The tier is provenance, not a verdict on truth.
- 01
Tier 1 — Authoritative public measurement
- 02
Tier 2 — Reproducible independent benchmark
- 03
Tier 3 — Primary manufacturer/model source
- 04
Tier 4 — Independent commercial market data
- 05
Tier 5 — HSI direct measurement
- 06
Tier 6 — Proxy
- 07
Tier 7 — Scenario
Households & needs
Household structure and residual need rest on federal statistical programs: Census ACS PUMS, BLS Consumer Expenditure and CPI, and BEA regional price parities.
| Registry ID | Source | Evidence type | Current HSI status | License / terms |
|---|---|---|---|---|
DS-HH-001 | ACS PUMSU.S. Census Bureau | Observed | National production ingest in progress | U.S. Census ACS PUMS (public use)Household weights require a hashed release manifest before any national claim. |
DS-RECR-001 | Consumer Expenditure SurveyU.S. Bureau of Labor Statistics | Observed | Production basket in progress | BLS Consumer Expenditure Survey (public-domain publication)Expenditure is not identical to residual need. Basket rules remain methodological. |
DS-RECR-002 | Consumer Price IndexU.S. Bureau of Labor Statistics | Observed | Production CPI mapping in progress | BLS CPI public data, API terms applyCategory series only. Headline CPI is not applied indiscriminately. |
DS-RECR-003 | Regional Price ParitiesU.S. Bureau of Economic Analysis | Observed | Same-year spatial rule exists; production ingest in progress | BEA Regional Price Parities (public data)Broad regional price levels. Do not double-adjust costs already measured locally. |
Energy
Location-specific generation is modeled with National Laboratory of the Rockies (NLR) PVWatts V8 on NSRDB resource data. EIA supplies residential electricity context. NLR is the current name of the laboratory formerly published as NREL.
| Registry ID | Source | Evidence type | Current HSI status | License / terms |
|---|---|---|---|---|
DS-ENERGY-001 | PVWatts V8National Laboratory of the Rockies | Derived | Modeled; not a published national snapshot | National Laboratory of the Rockies PVWatts V8 / NSRDB termsModeled irradiance and PV yield. Site-specific production is not observed from this preview. |
DS-ENERGY-002 | National Solar Radiation DatabaseNational Laboratory of the Rockies | Observed | Resource data supporting PVWatts; not a published snapshot | National Laboratory of the Rockies NSRDB termsResource data supporting PVWatts. Redistribution requires review. |
DS-ENERGY-010 | EIA Residential Electricity Retail SalesU.S. Energy Information Administration | Observed | Connector exists; not a published national extract | U.S. EIA public-domain data, API terms applyRetail electricity measures. Connector fixtures are not a published national extract. |
Capital & hardware
Solar capital draws on DOE PVSCM and Berkeley Lab project data. Hardware-price research remains a proxy. Treasury yields anchor capital-rate scenarios.
| Registry ID | Source | Evidence type | Current HSI status | License / terms |
|---|---|---|---|---|
DS-CAPITAL-001 | DOE PV System Cost ModelU.S. Department of Energy | Derived | Model ingest not in a published snapshot | DOE PVSCM / benchmark publication termsDerived cost-model outputs only. Do not hard-code a benchmark vintage into the index. |
DS-CAPITAL-002 | U.S. Distributed Solar and StorageLawrence Berkeley National Laboratory | Observed | Project data not in a published snapshot | Berkeley Lab Tracking the Sun termsProject-level installed prices after cleaning. Redistribution requires review. |
DS-HW-002 | Household Hardware Acquisition PriceMultiple | Proxy | Sampling methodology still research | Derived street-price research onlyNo single authoritative public hardware-price index currently exists. |
DS-FIN-002 | U.S. Treasury Daily Treasury Yield Curve RatesU.S. Department of the Treasury | Observed | Rate series not in a published snapshot | U.S. Treasury public data with attributionDaily yield-curve reference rates for capital scenarios. |
Compute & intelligence
Two different kinds of evidence sit in this family. External benchmarks and model releases tell us where the capability frontier is and what can execute on household hardware. Only HSI’s own measurements — the Autonomous Productive Work Benchmark and the Compute Harness — can qualify a household system as sovereign productive capacity.
| Registry ID | Source | Evidence type | Current HSI status | License / terms |
|---|---|---|---|---|
DS-COMPUTE-001 | MLPerf Inference PowerMLCommons | Measured | External capability-frontier reference; not an HSI-certified result | MLPerf license-specific termsExternal full-system power reference. Not an APWB productive-work unit. |
DS-COMPUTE-003 | HSI Compute HarnessHousehold Sovereignty Index | Measured | HSI-certified measurement not yet released. External benchmark evidence is tracked while the HSI Compute Harness matures. | HSI original measurement · not yet releasedClassification is measured. No HSI-certified run has been published yet. |
DS-MODEL-001 | Model Publisher and Hugging Face RegistryMultiple | Observed | Release, licensing and memory evidence; not a capability claim | Publisher and repository terms; metadata onlyModel identity and license text. Commercial permission is not inferred from the word “open.” |
Two kinds of evidence
External benchmarks, model releases and community reports are how we read the capability frontier. They are genuinely informative, and they are not sufficient: a score published elsewhere does not establish that a household-controlled machine completed economically useful work. Two HSI measurements are designed to establish exactly that — the APWB (Autonomous Productive Work Benchmark) and the HSI Compute Harness.
Qualifying productive intelligence has to execute on household-controlled hardware under SCI (Sovereign Compute and Intelligence). Work that clears the benchmark is then priced by the MVO (Market Value Oracle) and scaled by PU (Paid Utilization — share of qualifying available productive capacity that finds paying demand) before any of it reaches modeled income.
Productive workloads that require rented external intelligence do not qualify toward sovereign productive capacity. Frontier proprietary capability shows where intelligence is going; open, locally executable capability tells us when that capability becomes household-sovereign.
The evidence ladder
Evidence moves in one direction, and nothing skips a rung. Each step answers a question the step above it cannot, and only the last rung — the SCF (Sovereign Compute Frontier) — decides what qualifies economically. Benchmark scores, community excitement and parameter counts sit at the top of this ladder precisely because none of them is economically productive sovereign intelligence on its own.
- 01
Community and release signals
Weights are published, licenses change, and credible practitioners report what they got working on real hardware. This detects an emerging capability. It can nominate a configuration for HSI testing and it never enters an HSI calculation.
- 02
Independent benchmarks
Reproducible external measurement of general, coding and agentic capability, of cost per successful task, and of throughput and power. This tells us where the frontier is, not whether a household can bank it.
- 03
HSI APWB
Does the locally executed system complete an economically useful productive intention end to end — counting the human interventions and the failures?
- 04
HSI Compute Harness
What the household-controlled system actually costs to run: wall energy, reliability, throughput and productive work per kWh on the machine that did the work.
- 05
Sovereign Compute Frontier
Which locally controlled configurations qualify economically at the current state of hardware, energy and model availability. Only evidence that survives every rung above reaches this one.
Capability evidence we track
Registered sources appear in the table above with their evidence class, current status and license. The table below states what each class of intelligence evidence is for, and how far each one currently sits from an HSI release.
| Source | Source role | What it tells HSI | Current HSI status |
|---|---|---|---|
| MLPerf Inference — edge, datacenter and powerMLCommons · registered as DS-COMPUTE-001 | Independent benchmark | Reproducible throughput, latency and full-system power for inference workloads on named hardware. | Registered. Efficiency and frontier reference only; it is not an HSI-certified productive-work result. |
| Artificial AnalysisArtificial Analysis | Independent benchmark | Independent comparison of model and agentic capability, latency and cost, including the open-weight models that can be executed locally. | Tracked as external evidence. Not a registered HSI source, not ingested, and not used in any published HSI value. |
| LiveBenchLiveBench — public research leaderboard | Independent benchmark | Regularly refreshed capability testing across coding, agentic coding, data analysis and reasoning, with cost per successful task where the leaderboard reports it. | Tracked as external evidence. Not a registered HSI source, not ingested, and not used in any published HSI value. |
| METR time horizonsModel Evaluation and Threat Research | Independent benchmark | How long and how complex a task an autonomous system can complete without human intervention — the closest external analogue to the question APWB asks. | Tracked as external evidence. Not a registered HSI source, not ingested, and not used in any published HSI value. |
| Model publishers and Hugging FaceMultiple · registered as DS-MODEL-001 | Release evidence | Release availability, weights, license terms and memory requirements — the evidence that decides whether a model can execute on household hardware at all. | Registered as metadata. A published weight is an availability fact, not a capability claim and not a commercial permission. |
| HSI Capability RadarHousehold Sovereignty Index · planned | Discovery | A future discovery layer reading public technical-community evidence for configurations — model, hardware, autonomy level, task, corroboration — worth putting in front of the benchmark. | Not built. No adapter, no ingestion, and no HSI value depends on it. Nomination only, never certification. |
| HSI Autonomous Productive Work BenchmarkHousehold Sovereignty Index · registered as DS-TASK-002 | HSI measurement | Whether a locally executed system completes an economically useful productive intention end to end, with interventions and failures counted against it. | HSI-certified measurement not yet released. External capability evidence is tracked while the benchmark matures. |
| HSI Compute HarnessHousehold Sovereignty Index · registered as DS-COMPUTE-003 | HSI measurement | Wall energy, cost, reliability, throughput and productive work per kWh for the household-controlled system that performed the work. | HSI-certified measurement not yet released. External benchmark evidence is tracked while the harness matures. |
Source role says what a source is for. It is a reading aid on this page, not an evidence classification: the canonical evidence states remain Observed, Measured, Derived, Proxy, Scenario and Unobserved.
Discovery signals may nominate a configuration for HSI testing. They never certify one. No popularity, sentiment or enthusiasm measure is converted into an HSI constituent, and parameter count is not capability.
This external evidence is published by its owners at MLPerf Inference, Artificial Analysis, LiveBench, METR, Hugging Face models. HSI reads it. HSI does not restate it as an HSI measurement.
Autonomous work & market value
O*NET seeds task coverage. APWB is HSI’s own productive-work benchmark. Comparable-output research stands in until realized market-clearing prices exist.
| Registry ID | Source | Evidence type | Current HSI status | License / terms |
|---|---|---|---|---|
DS-TASK-001 | O*NET Web ServicesNational Center for O*NET Development | Observed | Taxonomy ingest not in a published snapshot | O*NET attribution licenseTask taxonomy for APWB coverage. O*NET does not price autonomous output or human worth. |
DS-TASK-002 | HSI Autonomous Productive Work BenchmarkHousehold Sovereignty Index | Measured | HSI-certified measurement not yet released. External capability evidence is tracked while the benchmark matures. | HSI original measurement · summaries onlySummaries only. No HSI-certified APWB result has been released yet. |
DS-MARKET-001 | Autonomous Output Market-Clearing PriceFuture qualified transaction systems | Observed | Not yet available | Aggregated transaction data only, when availableThe designed evidence type is observed transactions. None are available yet. |
DS-MARKET-002 | Comparable Output Price ResearchHousehold Sovereignty Index | Proxy | Comparable-output research; not realized clearing prices | HSI comparable-output research; derived products onlyTemporary Market Value Oracle until autonomous-market clearing prices exist. |
Demand & realized performance
Paid Utilization is an explicit sensitivity scenario until qualified household transactions exist. Realized household machine revenue remains unavailable; HSI-R stays unpublished.
| Registry ID | Source | Evidence type | Current HSI status | License / terms |
|---|---|---|---|---|
DS-UTIL-001 | Paid UtilizationFuture qualified transaction systems | Scenario | PU10 / PU25 / PU50 / PU75 sensitivity scenarios | HSI methodology scenario (PU10 / PU25 / PU50 / PU75)Paid Utilization is a scenario, not a measured market share. |
DS-REV-001 | Realized Qualifying Household Machine RevenueFuture opt-in household transaction network | Observed | Not yet available. HSI-R stays unavailable. | Not establishedRealized household machine revenue is not available. HSI-R stays unavailable. |
Evidence states
Evidence type describes the kind of evidence a source is designed to supply. It is not the same as whether that evidence is currently in an HSI release.
Directly measured by an eligible external source or an actual transaction.
Produced by a documented HSI experiment or benchmark.
Calculated entirely from observed and/or measured inputs.
An observable stand-in for a target that is not yet directly measurable.
An explicit hypothetical assumption chosen for analysis.
A required concept with no defensible measurement yet.
These six states are the canonical evidence taxonomy. The source-role labels used in Compute & intelligence — Independent benchmark, Release evidence, Discovery and HSI measurement — describe what a source is for, and are not evidence classes.
Publication rule
A registered source does not enter HSI merely because data exist. It must pass source validation, licensing, freshness, reconciliation, and provenance requirements. Published snapshots keep their original vintages; later source revisions create new observations and, when appropriate, separately identified restatements.
What this registry refuses to do
- It does not convert a connector fixture into a published national result.
- It does not treat PU scenarios as observed paid utilization.
- It does not fill HSI-R from HSI-C, NAPI, or any modeled envelope.
- It does not treat U.S. government public-use files as a completed HSI release until a hashed manifest exists.
Next
Inspect the current release, then the method.
Source registry v0.1.0 · methodology hsi-methodology-0.1