Review policy

CalculatorX publishes inspectable methods on tool pages so people and AI agents can check the behavior—not only the result.

This page defines what our human / domain review labels mean. It is the public counterpart to the repository REVIEW.md process.

Computational verification (objective controls, Evidence Manifest, attestation) is defined by the Calculator Verification Protocol (CVP). Review levels on this page are supplemental assurance signals — Domain Expert Review does not by itself determine CVP conformance.

CalculatorX uses one Trust Framework with four families:

Family Typical pages Trust schema
Calculation Trust Calculators Formula, units, assumptions, engine tests, calculation version
Conversion Trust Converters Exact or stated conversion factors, units, assumptions
Runtime Trust Apps / local utilities Behavior, data handling, storage, export, published cases
Developer Tool Trust Dev codecs and references Deterministic transforms, encodings, and round-trips

Calculation tools use Calculation Trust. Apps use Runtime Trust. Developer tools may use Deterministic Tool Trust. The labels are different because the claim is different: a voltage divider must be mathematically accountable; a notepad must be behaviorally and privacy-accountable.

Trust levels

Level What you see Meaning
L1 Engine tested Calculation status → Engine tested Shared deterministic engine + validation cases (or equivalent automated checks). Suitable for reproducible math via the REST API.
L2 Internally reviewed Calculation status → Internally reviewed A CalculatorX domain reviewer (reviewKind: internal, reviewedBy as an identified review team such as CalculatorX electrical review) checked formula, units, assumptions, and sources. Not a named external expert review.
L3 Independently / Expert reviewed (verified) Calculation status → Expert reviewed An identified external reviewer or independent review team checked formula, units, assumptions, and sources. Requires both verified: true and a non-empty reviewer identity, with reviewKind other than internal.
L4 Certified template Not offered yet Multi-reviewer, version-locked professional workflow—planned, not claimed today.

Source checked is orthogonal: references cite standards or primary definitions (IEC, NIST, CSS, WHATWG, ECMAScript, etc.). It can appear at L1–L4.

Specification checked is the same orthogonal slot for tools whose citations are the published calculation specification (definitions and algorithms) rather than an external standard. Basic arithmetic such as subtraction uses this label: claiming a NIST/IEEE “source” for (245-147=98) would be citation theater. Those pages list Specification basis instead of Sources. Physical, electrical, financial, and unit tools keep Source checked and Sources when they cite real standards or primary references.

The capability catalog tier uses the same levels:

Catalog tier Trust level
engine_tested L1
internally_reviewed L2
expert_verified L3 (verified in this policy)
candidate Schema exists; evidence incomplete

Do not label an L1 capability Verified capability. release_integrity.status = verified means published artifacts agree at a version — not expert review.

We do not mark a page Expert reviewed without reviewer identity. Automated tests alone are L1 Engine tested. Internal domain review is L2, not L3.

CVP Verified is not Expert reviewed. Machine JSON verified: true remains reserved for L3 (reviewedBy required). CVP uses a separate Manifest field status: VERIFIED (UI: CVP VERIFIED · CVP protocol 1.0.0-proposed · …). A calculator may be CVP Verified with verified: false. Domain Expert Review does not grant CVP Verified, and CVP Verified is not expert certification.

The compact Trust summary peek uses Engine tested · Source checked · vX at L1, or Engine tested · Specification checked · vX when the citations are the page specification rather than an external standard. It does not say “verified”. Machine JSON verified: true is reserved for L3 (reviewedBy required). After L3 the UI shows Named expert review for a named reviewer; when reviewedBy is a CalculatorX team, it is Internal engineering review, not independent third-party certification. The catalog token remains expert_verified.

reviewedBy is an accountable identity: a named person when we publish personal attribution, or an identified review team such as CalculatorX electrical review with a reviewerRole. A team identity is a review record, not an anonymous stamp.

Runtime / App trust labels

Apps such as Online Notepad do not use Calculation status. They publish an App status block.

App trust label Meaning
Runtime tested Published behavioral cases and/or automated runtime checks passed for this app version.
Local-only User content is processed on the device. CalculatorX does not host the draft.
Server data transfer Whether the user’s typed/pasted content is sent to CalculatorX. None means that content is not uploaded. Ordinary page assets, CDN, or analytics are separate from this claim.
Storage disclosed The page states where the draft lives (localStorage, memory, etc.).
Export tested Download encoding/format is specified (for Notepad: UTF-8 .txt) and covered by checks where automated.
Designed for modern browsers Intended for current Chrome, Edge, Firefox, and Safari. This is not a version-locked browser matrix unless the evidence page lists tested versions.

Runtime Trust proves behavior, data handling, and inspectability, not mathematical correctness.

How to read an app page

  1. Use the tool on the first screen.
  2. Read App status (runtime tested / local-only / server data transfer / storage / app version).
  3. Read Definition, What it does, counting or behavior rules, Limitations, and Validation cases.
  4. Open Sources for primary standards (WHATWG, ECMAScript, W3C, etc.).
  5. When an Evidence link is present, it is the machine-checked artifact for automated runtime tests—not a claim that every UI gesture was attested in every browser.

How to read a calculator page

  1. Use the interactive calculator and result panel.
  2. Read Formulas, Assumptions, and Boundary conditions in the specification.
  3. Check Calculation status (engine tested / internally reviewed / expert reviewed / version).
  4. Open Background for interpretation and common mistakes (not a second copy of the same how-to).
  5. Prefer the Calculation API when an agent needs a deterministic JSON result with verification metadata.

Identified review sample

Current expert-reviewed sample:

That page shows Expert reviewed in Calculation status, plus formula, assumptions, validation cases, and sources. Pages with reviewKind: internal show Internally reviewed. Most other live tools are engine tested until a reviewer identity is recorded.

Reviewer roles

Role Typical scope
electrical-engineer Electrical and wire calculators
rf-engineering RF / microwave calculators (coax, matching, Smith)
cpa / finance-analyst Finance calculators
editor Unit converters and reference pages
software Developer codecs and tooling engines

What we ask reviewers to check

  1. On-page formula matches the UI and the shared engine.
  2. Assumptions and boundary conditions are explicit (phase, RMS, rounding, domain).
  3. Sources point to standards or primary definitions—not only “CalculatorX”.
  4. Validation cases pass the automated suite when one exists.
  5. Frontmatter sets engineTested, and only then verified + reviewedBy + lastReviewed + calculationVersion.

After a formula change we bump calculationVersion and clear or renew expert review.

Internal engineering review may be layered. Formula / assumptions can stay current when the model digest is unchanged; canonical input semantics, error codes, and REST compatibility need a separate refresh when those contracts change. CSS, Evidence formatting, or CVP partition-only changes do not require a new human review.

API verification fields

Successful POST /api/v1/calc/:id responses include:

"verification": {
  "engine_tested": true,
  "internally_reviewed": true,
  "expert_reviewed": false,
  "source_checked": true,
  "reviewed_by": "CalculatorX electrical review",
  "last_reviewed": "2026-08-16"
}

internally_reviewed: true is L2. expert_reviewed: true is L3 and requires a non-internal reviewer identity. Artifact hashes on Evidence pages prove published artifacts agree; they are not an independent immutable release log.

plus formula, assumptions, warnings, and inputs. See the developers guide.

Contact

For corrections to a published formula or source, use the site About contact path or open an issue on the CalculatorX GitHub repository if you maintain a fork/PR workflow.