Reference

An index, not a second copy of the method. The methodology lives in skill/ux-mizan/ — the same files Claude loads — so it cannot drift from what actually runs.

Where each topic lives

Topic	File
Triggering, the two layers, the gate DAG, modes, risks, anti-patterns	`SKILL.md`
Gate question sets, defaults, the re-opening protocol	`references/gates.md`
Cognitive-walkthrough template and its blind spots	`references/walkthrough.md`
Metric battery, applicability matrix, lostness, baselines, instrumentation plan	`references/metrics.md`
Gate 3, redesign, spec handoff, what a two-model handoff cannot promise	`references/handoff.md`
Recovery ramps R-00…R-13, the model failure classes, the closing scorecard	`references/recovery.md`
Registry format and the rules in comment form	`schemas/ux-registry.yaml`

Evidence tiers

Tag	TR	EN	Meaning
`[K]`	Kanıtlanmış	Proven	Direct evidence; source cited; threshold met
`[H]`	Makul Hipotez	Plausible hypothesis	Grounded; empirical support missing or below threshold
`[S]`	Spekülatif	Speculative	Interesting; not testable, or no test designed
`[R]`	Reddedildi	Refuted	Tested and failed its own threshold — kept on record
`[KKE]`	Kritik Kontrol Eksik	Critical control missing	Result exists; the control that could flip it has not run
`[Y]`	Yanıltıcı	Misleading	Technically true; implies more than the evidence supports

The tags are labels, not prose: keep them bilingual in every language.

The rules the validator enforces

Rule	What it rejects
U1	`[K]` with no resolving `evidence_artifact_id`, or with model-side `source_provenance`; an evidence artifact with no honesty annexes
U2	A finding with no `parent_flow_id`, or one that does not resolve
U3	A `severity` that does not equal `failure_magnitude × priority_weight × frequency`; a finding that restates `priority_weight`
U4	A flow or finding present in the baseline and missing now; an entry moved out of `[R]` (needs `--against`)
U5	A finding with no `refutation_condition`, no `metric.name`, no named `metric.instrument`, or an invalid `metric.kind`
U6	A metric absent from the flow’s `applicable_metrics`; an `applicable_metrics` entry the flow’s `app_type` does not switch on
U7	A flow with no `gate_provenance`; `[K]`/`[R]` on a flow whose Gate 0 is not human-locked; a finding citing a stale premise version
U8	`[K]` on a behavioural metric whose flow has no measured baseline for it
U9	`[K]` with no `min_n.n` or no `min_n.decision_rule`
U10	A finding proposing a `fix` with no `self_check_homogenisation` answer
U11	A `[KKE]` finding with no `kke_kind`, an invalid one, or the field left on an entry that is no longer `[KKE]`
U12	An open finding with no `review_by`, or one whose `review_by` has passed with no decision recorded
U13	A `conjunction` finding missing its pair, its guarantee, or its safe order; a pair whose two halves are the same flow or whose parent is the lighter of the two

kke_kind says WHICH control is missing: control (a confound check that could flip it has not run) · independence (the producer of the claim is also its judge) · data (the measurement is designed, the data is not in) · validation (the instrument itself was never run against a known-positive case). A field, not four new tags — the six tiers are what the four sibling skills share, and Mizan already carries two of these reasons without naming them (R2 missing data, R8 missing independence).

Warning	What it flags (advisory; `--strict` promotes)
W1	Every tiered finding is `[K]`
W2	No `flow-level` finding exists
W3	A flow with no measured baseline
W4	An open finding with no owner and no scheduled measurement

python skill/ux-mizan/scripts/ux_validate.py --lang tr --strict ux-registry.yaml
python skill/ux-mizan/scripts/ux_validate.py --against HEAD ux-registry.yaml
python skill/ux-mizan/scripts/ux_validate.py --format github ux-registry.yaml  # PR annotations (json also); same exit code

The metric matrix

Each app_type earns its row by switching a metric off. A category that only adds is decoration.

app_type	switched OFF	why
`navigational-multiscreen`	—	the full battery
`form-heavy-transactional`	lostness	the path is prescribed; failures are field-level
`single-canvas-tool`	lostness, nav_depth	one screen; revisits are work, not wandering
`b2b-internal-dashboard`	sus	a captive user cannot leave; satisfaction is floor-censored
`content-consumption`	task_success	there is often no task
`guided-sequence`	lostness, nav_depth	the server enforces the order; a revisit is re-reading

The matrix is hard-coded in ux_validate.py (METRIC_MATRIX) as well as documented in the schema. If the two ever disagree, the validator is the one that runs, and the disagreement is itself a finding.

Scripts

Script	Produces	Tier of its output
`ux_validate.py`	U1–U14 verdicts, W1–W4 warnings	— (it checks, it does not claim)
`structural_checks.py`	State coverage, feedback gaps, generic labels, nav depth, orphan routes	`[KKE]` — what is missing
`layout_signals.py`	Arrangement traces + the behavioural hypothesis each licenses	trace `[KKE]`, claim `[H]`
`lostness.py`	Per-flow L, completed and abandoned reported apart	needs Layer-B data to mean anything
`ux_to_backlog.py`	Findings → İskele tasks: confirmed → fix, open [H] → measure	estimates are not guessed; every task is marked uncalibrated

Two rules that are not in any script

Both are in prose, and both are load-bearing. They are listed here because PROSE-SCHEMA-AUDIT.md names them as gaps rather than hiding them:

Open the file before writing a finding. location must be non-empty; nothing checks that it says what the mechanism claims. This rule caught every scanner defect this repo has shipped.
A new detector runs against a known-positive case before it is trusted on unknown ones. Every structural check here has shipped a silent under-report at least once, and an under-reporting checker is indistinguishable from a clean codebase. This one moved from the contributor docs into SKILL.md after an external review pointed out it was written everywhere except the file that travels into a host — still unscripted, but no longer invisible to the tool that must obey it.