What was supposed to happen?
Start with the formal decision: a law, regulation, executive action, program, budget choice, or institutional commitment. What did the decision actually require?
METHOD · IN PLAIN LANGUAGE
Loom follows the chain between a decision and an outcome. The goal is not to produce a more convincing story. It is to make the mechanism inspectable — and to make it difficult for unsupported conclusions to pass as established research.
IF YOU ARE NEW HERE
Most accounts of policy begin with a decision and jump to an outcome. Loom deliberately studies the middle: the institutions, authorities, dependencies, capacity, information, technical systems, contracts, timing, and adaptation that determine whether a decision becomes implementation.
The result is a research record you can read at different depths. Start with the public case. Follow the evidence if you want to check a claim. Inspect the machine-readable artifacts if you want to audit how the record was governed.
Start with the formal decision: a law, regulation, executive action, program, budget choice, or institutional commitment. What did the decision actually require?
Establish the observed outcome without treating the outcome as an explanation. A policy announcement is not evidence that implementation occurred.
Map the institutions, authorities, people, resources, dependencies, information flows, and timing required to move from decision to implementation.
Look for mechanisms: missing authority, capacity constraints, dependency failures, contradictory incentives, information failures, timing problems, or breakdowns in self-correction.
Trace each material claim back to evidence. Preserve contradiction. Record verification. Separate an observation from a claim, and a claim from a diagnostic explanation.
THE ARCHITECTURE
Loom is not built from the premise that every component is unprecedented. Several research traditions already do parts of this work well. What is distinctive is their combination around primary empirical governance research, with a mechanical gate between the research record and what can be promoted or published.
Like investigative research platforms, Loom links claims to documents and specific evidence. Unlike a document viewer alone, the research record carries formal epistemic state and provenance rules.
Like reproducibility systems, Loom preserves versions, provenance, and inspectable research artifacts. Its focus is not primarily computational reproducibility; it is evidentiary reproducibility for qualitative governance research.
Like claim-review and argument-mapping systems, Loom represents claims and contradictions explicitly. It additionally tracks source provenance, evidence, verification, and promotion state.
Like rigorous open-source investigations, Loom cross-checks records and documents uncertainty. The difference is that the rules are encoded into the research artifact and validator rather than living only in analyst practice.
Tools such as Obsidian and Quartz are useful raw material for linking and publishing research. Loom adds schemas, validation, provenance, and state rules so a flexible documentation layer cannot silently become the authority for a research claim.
WHAT MAKES THE SYSTEM DISTINCTIVE
Research systems are often fail-open: an analyst can publish a conclusion and correct it later. Loom instead uses validation rules to block promotion or publication when required provenance, verification, contradiction handling, or artifact relationships are missing.
Terms such as epistemic_status, verification_level, promotion_stage, and contradiction_status are not merely labels in prose. They are structured fields subject to rules.
The public layer is intentionally plain. Behind it sits the research record: source files, evidence extractions, claims, gaps, verification, hashes, schemas, and validator results. Readers can stop at the level they need or continue downward.
A CONCRETE EXAMPLE
Loom's Flint work illustrates why this architecture matters. The record contains original and revised lead-and-copper sampling materials, later federal review, and a 2015 FOIA response describing the rationale for removing two samples as handwritten notes on a draft report plus lab-report pages.
That does not mean Loom fills the missing notes in from later conclusions. The missing annotations remain an evidence gap until the underlying record is found, reviewed, or the search protocol is genuinely exhausted.
“Not enough evidence” is a research result.
See the Flint case record →THE CORE RULE
This is not a slogan about responsible AI. It is a governance rule for the research system. Generated text can help locate, organize, compare, or propose. It cannot promote itself to an established claim. Evidence has to establish the proposition, and the research process has to permit promotion.