Methodology

Methodology designed for decision confidence.

RepoTruth explains not only what it observes, but also the limits of those observations.

Methodology transparency is fundamental to trust. You should be able to see how every output is reached, how confident it is, and where its knowledge ends, so your people can rely on the evidence and challenge it. This page explains, in plain terms, how RepoTruth arrives at what it produces.

  • Evidence, not verdicts
  • Confidence-scoped
  • Unknowns preserved
  • Explainable
  • Human review required

How RepoTruth works

From a snapshot to a decision-ready package, in five steps.

Each step is observable, explainable, and read-only.

  1. Snapshot

    A read-only, point-in-time copy of the codebase is provided. Nothing is connected and nothing runs.

  2. Repository Reality Extraction

    The system is observed as it actually is, structure, dependencies, history, and the boundaries of what can be seen are recorded.

  3. Decision Intelligence

    From the evidence, RepoTruth derives where to investigate first, how confident each observation is, and what remains unknown.

  4. Diagnostic Package Generation

    Everything is assembled into one structured deliverable, with its own explicit statement of limits.

  5. Human Review

    Your people read, interpret, and decide. RepoTruth concludes nothing on their behalf.

The foundational distinction

Interpretation ≠ Authority.

RepoTruth is designed to interpret evidence. RepoTruth is not permitted to exercise authority.

Interpretation is mandatory, without it, evidence is only data. Authority is prohibited, the decision, and the responsibility, remain your people’s. Everything RepoTruth produces lives inside that single distinction.

Interpretation is mandatory. Authority is prohibited.

Decision Intelligence Ontology

RepoTruth produces intelligence, not decisions.

The platform generates a fixed set of evidence-based outputs, and nothing that resembles a verdict.

  • Evidence

    Observations of the system, cited to their source and scoped by confidence.

  • Concern candidates

    Structural observations raised for an architect’s judgment, candidates for review, never verdicts.

  • Investigation corridors

    Suggested paths through the system for human review, a way to look, not a way to act.

  • Review priorities

    Where to investigate first, ordered by investigation urgency and paired with confidence.

  • Unknown inventories

    An explicit record of what could not be determined, surfaced, not assumed away.

RepoTruth interprets the evidence and orders where to look. Decision authority always remains human.

Score Constitution

Scores indicate investigation urgency, not truth.

Every score answers one question, where should humans look first?, and nothing more.

  • Score ≠ Truth

    A score reflects where attention is best spent, not a fact about the system.

  • Score ≠ Severity

    It measures urgency of investigation, never the seriousness or danger of a finding.

  • Score ≠ Recommendation

    A high score invites a closer look. It never instructs anyone to act.

  • No global health score

    There is no single number for a repository. A one-figure verdict is deliberately not produced.

The one number a buyer or an automated gate would most like to act on is the number this methodology refuses to compute.

Confidence Model

Every observation carries confidence.

Confidence scopes each observation, communicates its limitations, and remains explainable, never a black box.

  1. Directly Observed

    Read straight from the snapshot, with no inference.

  2. Strongly Derived

    Inferred from multiple, consistent observations.

  3. Moderately Derived

    Inferred from limited but consistent evidence.

  4. Weakly Derived

    Inferred from sparse or partial evidence, with the gaps stated.

  5. Historically Inferred

    Derived from repository history, where it is present and readable.

  6. Ambiguous

    The evidence supports more than one reading, surfaced as ambiguity, not resolved by guess.

  7. Unverifiable

    A question a static snapshot structurally cannot answer.

  8. Unknown

    No evidence either way, recorded explicitly, never inferred shut.

Every confidence value can be explained: what was observed, what was inferred, and why it is held no more firmly than that.

Unknown Preservation

Unknowns are preserved, never hidden.

Where the evidence runs out, RepoTruth records the question rather than filling in an answer.

Most tools quietly fill in what they cannot determine, or omit it entirely. RepoTruth treats the unknown as a first-class entity, named, categorized, and surfaced with the same prominence as any finding.

An unknown is not a failure. It is not noise. It is not low risk. It is an un-assessed question, and often the most useful thing in the package, because it tells your people exactly what to ask next.

  1. Unknown

    Recorded explicitly where evidence is absent, with its category and who could resolve it.

  2. Reviewed

    Surfaced to your people as a question worth investigating, never inferred away.

  3. Resolved

    Retired only when evidence or a human supplies the answer, never by RepoTruth filling the gap.

Human review workflow

Humans remain terminal.

RepoTruth supports review. Your people remain responsible for the decision.

  1. Observe

    Your people read the evidence, the priorities, the confidence, and the unknowns.

  2. Interpret

    They bring context RepoTruth cannot have, and form a judgment from the evidence.

  3. Challenge

    Any finding can be questioned. Because outputs cite their evidence and confidence, they can be checked.

  4. Decide

    The decision is made by people with the authority to make it, never by the platform.

Constitutional boundaries

The methodology runs inside permanent boundaries.

These are properties of how RepoTruth is built, not options it can be configured past.

  • Read-only

    It writes to nothing in your environment.

  • Snapshot-based

    It analyzes an inert, point-in-time copy, never a live system.

  • No credentials

    It stores, requests, and reuses none.

  • No production access

    It never connects to a running system.

  • No execution

    It runs no code from your repository.

  • No remediation

    It changes, fixes, and patches nothing.

  • No authority

    It approves, certifies, and decides nothing.

  • Human review required

    Every output is built for a human to decide on.

Diagnostic generation flow

From evidence to deliverable, one explainable stage at a time.

Each stage builds on the last, and nothing is asserted more firmly than the evidence allows.

  1. Evidence

    What was directly observed, cited to its source.

  2. Signals

    Patterns derived from the evidence, readings a human can reason about.

  3. Scores

    Investigation urgency over a region, where to look first, never how bad it is.

  4. Confidence

    How firmly each observation is held, and why, explainable, never opaque.

  5. Unknowns

    What could not be determined, recorded explicitly and carried forward.

  6. Review Priorities

    An ordering of where human attention is best spent first.

  7. Diagnostic Package

    Everything assembled into one decision-ready deliverable, with its own statement of limits.

Methodology FAQ

How the method holds, in plain terms.

How are scores calculated?

Scores are derived deterministically from observed evidence, for example, how strongly other code depends on a region, or how thin the available knowledge is. They express investigation urgency only, and each carries its own confidence. The same snapshot always yields the same scores, and every score can be traced back to the evidence behind it.

What if confidence is low?

Low confidence is reported, never hidden. A low-confidence observation is uncertain, not wrong, it is a call to look closer, alongside an honest statement of why we hold it no more firmly. A finding is never asserted beyond what the evidence supports.

How should unknowns be interpreted?

As questions, not as gaps. An unknown is un-assessed, not low risk and not safe. A region dense with unknowns is precisely where human investigation is most valuable. Unknowns are resolved only by evidence or a human, never inferred shut.

Can outputs be automated?

RepoTruth’s outputs are built for human review and are non-actionable by construction. They are not designed to gate a pipeline, trigger an action, or stand in for a decision. The package informs people; the people decide.

Can RepoTruth approve architectural decisions?

No. RepoTruth produces evidence and concern candidates for an architect’s judgment. It approves nothing and decides nothing. Architectural decisions are made by people with the authority to make them.

Why does RepoTruth avoid recommendations?

Because interpretation is not recommendation. A recommendation embeds a decision RepoTruth has no standing to make, it lacks your context, your constraints, and your authority. By producing evidence, priorities, and unknowns instead, RepoTruth keeps the judgment, and the responsibility, with your people.

Understand how RepoTruth reaches its conclusions, and its limits.

A read-only diagnostic, scoped to your decision.

Tell us the decision you’re facing. We’ll scope a Diagnostic Package to it, produced by the method on this page, read-only and snapshot-based, with every finding carrying its evidence, its confidence, and its unknowns.

Evidence, not verdicts. Human review remains terminal.