DwellMetry
DwellMetry is a live home-information and evidence system I built to represent a home as structured records — its spaces, equipment, documents, maintenance history, utilities, and relationships — while keeping track of how each piece of information is known. The product includes a public interactive demo, authenticated household workspaces, evidence-aware records, and measurement-ingestion workflows. Thermal and behavior modeling remain a separate scientific layer that requires sufficient measurements, weather context, calibration, and validation before any model or diagnostic conclusion is presented.
Current scope
The record system and measurement-ingestion infrastructure are implemented. Thermal and behavior models remain a separate scientific layer: DwellMetry does not present a calibrated physical twin, validated diagnosis, or field-validated model without sufficient measurements, weather context, calibration, and chronological validation.
Summary
- Physics & Modeling
- Flagship
- Continuing
Evidence on file
My contribution
I designed and built DwellMetry end to end: the household and role model, home/equipment/document data model, evidence-provenance system, record and maintenance workflows, utility-history handling, Connected Home interface, telemetry-ingestion architecture, and the separation between ordinary household records and scientific analysis. I also built the public fictional demo so product behavior can be inspected without exposing real household data.
For the scientific layer, I designed workflows for measurement ingestion, normalization, data-quality review, weather alignment, and thermal-model readiness. I do not treat those workflows as proof of a calibrated model or validated diagnosis; the application requires additional evidence and validation before those claims are allowed.
Project overview
Why I built this
Information about a home is usually split across equipment labels, manuals, inspection reports, service records, warranties, and utility bills. I wanted to build a structured record that could keep those sources connected without turning missing or conflicting information into false certainty.
The same foundation also gave me a place to investigate a harder question: what additional measurements, models, and validation would software need before it could say anything responsible about how a home is behaving?
What I built
I built and deployed DwellMetry as a home-information and evidence system. Its implemented product layer includes authenticated household workspaces, role-aware access, structured home, floor, room, equipment, document, maintenance, service, warranty, and utility records, and an interactive Connected Home schematic. Records preserve sources, revisions, conflicts, and unknowns instead of flattening them into one inferred answer.
I also built measurement infrastructure for telemetry sources, mapped devices and channels, raw observation ingestion, normalized readings, import batches, explicit CSV mapping, data-quality review, weather-source provenance, chronological validation, and thermal-model readiness. Those workflows prepare evidence for analysis; they do not establish a calibrated home model or a validated diagnosis.
Design approach
The implemented application treats records and evidence as the foundation. DwellMetry uses a connected-home / digital-twin architecture to organize canonical records, relationships, evidence, and measurements. Its current structural view is an interactive architectural schematic, not measured geometry, a scan, a to-scale model, or a calibrated physical twin.
Measurement ingestion sits between ordinary records and scientific analysis. Raw source readings remain separate from normalized values, and missing observations remain missing rather than being silently converted to zero. A scientific result must identify its inputs, assumptions, data-quality limits, calibration state, and validation status before the interface can present it as a conclusion.
Product interface evidence
- 01Public fictional-demo overview connecting home structure, sources, maintenance, utilities, documents, service history, and role-aware access without presenting scientific diagnosis.
- 02Historical utility, equipment, and service records remain separate from diagnostic claims; missing information and conflicts stay visible.
- 03Evidence-aware records and Home History preserve how information is known, what changed, and which conflicts remain instead of flattening uncertainty into one inferred answer.
Testing and validation
The live application and public resettable fictional demo establish the implemented product workflows without exposing a real household. The demo is non-measured and its Connected Home presentation is illustrative.
The measurement-ingestion and review workflows are implemented infrastructure. They do not establish that a thermal model is calibrated or that a diagnostic conclusion is accurate in a real home. Those claims require sufficient measurements, weather context, applicable physical constraints, calibration, chronological validation, and purpose-specific evaluation.
Failures and debugging
The research work exposed several ways software can produce a convincing but unsupported answer: treating missing telemetry as zero, learning a home’s past behavior and calling it healthy, or fitting a thermal curve and assuming the fit identifies each physical parameter.
Those are research and modeling failures, not field results. They led me to make unknowns, provenance, identifiability, and abstention explicit parts of the design.
What I learned
The strongest part of DwellMetry is the boundary between what the records establish and what a model is still trying to infer. A useful home-information product does not need to pretend that every stored value supports a physical conclusion.
Keeping product capability, research implementation, and future direction separate made the technical work more credible, not less ambitious.
Next iteration
The next scientific step is to take homes with sufficient, well-curated measurements through weather alignment, calibration, chronological validation, and explicit applicability checks before presenting any thermal or behavior conclusion.
A calibrated physics-informed twin, validated HVAC conclusions, automatic equipment or unusual-usage diagnosis, and safety or condition assessment remain conditional directions rather than current product claims.
Implemented product
These capabilities are visible in the current product and its public fictional demo. They describe records and workflows, not a measured building model or scientific diagnosis.
- Authenticated household workspaces with role-aware access for the people connected to a home.
- Home, floor, and room records connected through an interactive architectural schematic. The view is illustrative, not measured geometry, a scan, a to-scale model, or a calibrated building model.
- Source-guided equipment records linked to rooms, systems, documents, warranties, service events, and maintenance history.
- Private documents and warranties, service and maintenance history, and revision-aware records that keep corrections connected to their sources.
- Manual or imported utility history with original periods and units preserved, plus historical arithmetic, per-day, and unit-cost comparisons where the records support them.
- Deterministic search over a household’s private records, without sending private household prose into a public index.
- A public, resettable, fictional, non-measured household demonstrates product behavior without exposing a real address, household identifier, filename, or personal record.
Implemented measurement infrastructure
IMPLEMENTED WORKFLOW / INFRASTRUCTURE
These capabilities collect and prepare observations for review and later analysis. Their implementation does not establish field-validated monitoring, a calibrated home model, anomaly detection, or automatic diagnosis.
- A telemetry-source model, mapped devices, and measurement channels with explicit semantics and units.
- Raw source observations remain separate from normalized values. Missing readings stay missing rather than being silently converted to zero.
- Import batches, explicit CSV mapping, data-quality review, and curation preserve how a reading arrived and how it was interpreted.
- Weather context is attached with its source and alignment rather than treated as an unrecorded model input.
- Chronological-validation and thermal-model readiness/configuration workflows check whether the evidence is suitable for a proposed analysis before a conclusion is allowed.
Scientific modeling — conditional / not yet established
These conclusions require sufficient measurements, weather context, calibration, chronological validation, and applicable physical constraints. They are not established by the current product evidence.
- No calibrated thermal behavior model or calibrated physics-informed digital twin is claimed.
- No validated HVAC performance, efficiency, or equipment-condition conclusion is claimed.
- No automatic equipment diagnosis, unusual-usage diagnosis, or anomaly-to-cause conclusion is claimed.
- The absence of a warning is not a safety assessment or proof of a home’s condition.
Evidence provenance
DwellMetry keeps the status of a claim visible instead of collapsing every value into a single answer. Records can be user confirmed, document supported, device reported, measured, derived, estimated, unverified, or conflicting. A correction creates a revision connected to its source rather than silently replacing the earlier assertion.
Utility history without diagnosis
Utility records preserve their original billing periods and units. DwellMetry can make transparent historical, per-day, and unit-cost comparisons where the data supports them, but those comparisons are not weather normalized unless a separate analysis establishes that context. Historical change does not prove efficiency, and a bill comparison does not diagnose equipment. A gas bill expressed in volume remains volume unless a valid heating-value basis is available; an unsupported conversion remains unknown.
Technical notes
Implemented product and measurement architecture, with conditional scientific models labeled separately throughout.
Read the technical notesEvidence
Log entries for this project
- More fitting cannot recover information the sensors never measuredThe thermal model reproduced the data well and had not learned what I thought it had. The limit is structural, not numerical.PhysicsIdentifiability
- An anomaly is not a diagnosisDetecting that behaviour changed and explaining why are different problems, and merging them produces confident causes the evidence does not support.Fault ReasoningUncertainty
Related work

Solar photovoltaic performance intelligence: a deployed application for comparing measured and expected production and investigating persistent underperformance. Public demonstrations use fictional/reference data while the analysis pipeline continues to be validated and expanded.

Live quantitative-research software for turning market hypotheses into versioned evidence records with backtesting, walk-forward validation, calibration, cost/stress testing, research lineage, and explicit abstention when evidence requirements are not met.