Reads the decision records a system exports in STD-07, the Institute's draft record format for work delegated to machines. It checks them against the schema, recomputes the hashes, and compares the conformance level the records earn with the level the system claims. Use it when a system already produces records; use the delegation audit when it does not.
Paste or load a stream, set the level the system claims, and see the level it earns. Nothing is uploaded.
Readout: The earned level, what holds it below the next one, and findings graded blocking, finding, or note with the clause each comes from.
Read STD-07
Set the level by hand only when the system publishes no manifest. A level read from what the system itself publishes is the one worth checking. Leave the time blank to use now. Set it to when the stream was exported, so deadlines are judged against the system's lateness, not yours.
Paste a stream and run the audit. It checks what the records say about each other: whether the hashes hold, whether references resolve, whether each record says what would make it invalid, and whether a discrepancy was ever answered. It does not judge whether a belief was correct or an authorization wise, because a stream cannot show that.
What the checker verifies from a stream, what it cannot tell you, and how the earned level is computed.
Show inputs, procedure, scoring, assumptions, and validation
Inputs
A record stream as newline-delimited JSON, a JSON array, or an object with a records array.
Optionally the emitter's manifest, whatever its shape: the declaration is found by its fields rather than by a fixed path.
The conformance level the emitter declares, if it declares one.
The moment to judge clocks against, defaulting to now.
Procedure
Paste or load the stream. Nothing is uploaded; the audit runs in the page.
Set the declared level so an overclaim can be contradicted.
Read the findings blocking first, then the note-level observations.
Copy the readout and keep it with whatever the emitter published.
Outputs
The level the stream earns, and what holds it below the next one.
Findings graded blocking, finding, or note, each citing the clause it comes from.
The record ids each finding applies to.
A contradiction when the declared level is higher than the earned level.
Measures
Whether every record validates against the published STD-07 schema.
Whether the hashes recompute and the prior_hash chain links, so a removed or edited record shows.
Whether beliefs, authorizations, and actions state what would end them, and whether a discrepancy was ever answered inside its clock.
Whether the emitter's own manifest agrees with its stream: the level it declares, and the record kinds it claims to emit.
Does not measure
It cannot tell whether a belief was correct or an authorization wise. A stream does not contain that, and scoring it anyway would make this tool the safeguard-on-paper it exists to catch.
It reads what the records say about each other, not what the system did. A conforming log can describe a badly run institution.
A dangling reference is usually a partial export, not a defect.
Assumptions
The stream is the emitter's own output, exported whole rather than filtered.
Clocks are judged against the export time, not against when the page is opened.
The declared level is what the emitter publishes in its own manifest, read from the manifest itself when one is supplied.
Instrument prompts
Does every record validate against the schema the standard publishes?
Do the hashes recompute, and does each record chain to the one before it?
Does every belief, authorization, and action say what would end it?
Was every discrepancy answered by a revision or an objection, inside the clock the record declared?
Does every record say who has standing to object, and has an objection ever been accepted?
Is the level the emitter declares the level this stream supports?
Rubric
Level 0: every record validates, with as_of kept apart from recorded_at.
Level 1: Level 0, and nothing edited — every record hashed, and chained through prior_hash.
Level 2: Level 1, and beliefs, authorizations, and actions carry invalidated_by, with every discrepancy answered inside its clock.
Level 3: Level 2, and standing declared on every record, with at least one objection accepted and answered.
Scoring logic
Blocking: an invalid record, a hash that does not recompute, a broken chain, a discrepancy never answered, or a declared level above the earned one.
Finding: an ungrounded grant, an answer that arrived after the clock, or records hashed but not chained.
Note: a reference resolving outside the stream, a root record resting on nothing, or a condition with no clock to be judged against.
The earned level is the highest whose requirements carry no blocking finding.
Collection cost
None. This instrument asks no one anything, and its readout does not move when a person is tired.
Record-supplied. It reads an exported stream and the time the export was taken.
Schema validity, hash recomputation, and chain linkage are read from the stream itself.
The conformance level a system publishes is compared against what its records show.
Nothing in the readout depends on a person's judgment or memory.
Not applicable. The same stream and the same as-of time give the same readout, so no respondent can change it.
Validation notes
Built against the two systems that emit the shape today, using the exported output of one of them as the worked example rather than a fixture written to pass.
Deterministic. The same stream and the same as-of time give the same readout, because every check is mechanical and none of them asks for a judgment.
Auditing a filtered export and reading the resulting chain break as tampering.
Judging clocks against now rather than the export time, which fails a record for the reviewer's lateness.
Reading a clean readout as evidence the institution is well run. It is evidence the log is well formed.
Treating a note as a defect. Most notes are properties of the export, not of the system.
Replicability
Export the stream from the emitting system, whole rather than filtered.
Load it here and set the level that system declares.
Set the as-of time to the export time if any clock is close.
Copy the readout and file it beside the emitter's own conformance claim.
Example outputs
Level 2 earned, Level 2 declared, and no blocking findings.
Level 3 blocked because no objection appears in the stream, so acceptance from outside cannot be observed.
A contradiction naming the earned level when the declaration is higher.
Cite this page
· v1.6.0 · updated Oct 2, 2026 · CC BY-SA 4.0
APA
Ethotechnics Institute Diagnostics Lab. (2026). Record conformance checker. Ethotechnics Institute. https://ethotechnics.org/diagnostics/record-conformance
MLA
Ethotechnics Institute Diagnostics Lab. "Record conformance checker." Ethotechnics Institute, 2026, https://ethotechnics.org/diagnostics/record-conformance.
Chicago
Ethotechnics Institute Diagnostics Lab. "Record conformance checker." Ethotechnics Institute. Oct 2, 2026. https://ethotechnics.org/diagnostics/record-conformance.
TY - WEB
TI - Record conformance checker
AU - Ethotechnics Institute Diagnostics Lab
PY - 2026
UR - https://ethotechnics.org/diagnostics/record-conformance
ER -
Authorship
Ethotechnics Institute Diagnostics Lab · Ethotechnics Institute
· hello@ethotechnics.org
v1.6.0 ·
Oct 2, 2026 — Added checks for unrecorded repair work and staff workarounds to the workload modeler and corrective capacity self-assessment.
Earlier versions (6)
v1.5.0 ·
Oct 1, 2026 — Each of the four instruments now states the respondent labor it consumes before the respondent begins, what the operator's own retained records already answer without asking a person, and how non-response is represented in the readout. Three instruments take respondent time. Record Conformance reads a record stream and asks no one, and now says so on the page.
v1.4.0 ·
Sep 29, 2026 — Added two optional context questions to the corrective capacity self-assessment: who holds authority over the conditions that produce workarounds, and whether the institution has priced a fix. They record a possible conflict of interest and do not change the score.
v1.3.0 ·
Sep 18, 2026 — Added a self-assessment of action-capacity growth, challenge intake, reversal, revision, and workarounds using fixed categorical weights.
v1.2.0 ·
Sep 6, 2026 — Added the delegation audit, which takes one workflow through six questions.
v1.1.0 ·
Jan 9, 2026 — Published method cards, transparency notes, and replicability guidance for each diagnostic.
v1.0.0 ·
Dec 3, 2025 — Initial diagnostics suite release.