Lemonfiber
← The specification

Quality-profile sync

Rendered from spec at76d46f3 (2026-10-25) · its source · Edit this page

Status: Accepted · Audience: Operator · Area: H — Ecosystem glue

Purpose

Getting Sonarr and Radarr to grab the right release means custom formats and quality profiles — dozens of them, scored just so — and the community already maintains the definitive catalogue (the TRaSH-Guides). Hand-copying that catalogue into each instance is a day of tedious, error-prone clicking that goes stale the moment the guides update. Quality-profile sync applies the community catalogue to the *arr apps declaratively, so the operator names the profile they want and the tool makes the instance match — then proves it matched, rather than trusting that the writes stuck.

Behaviour

It applies the community catalogue declaratively

The operator selects which community-maintained custom formats and quality profiles they want; the tool reconciles each instance to that selection — creating what is missing, updating what has drifted, and leaving unmanaged items alone unless the operator opts into pruning. The operator declares the desired state; the tool makes the instance match it.

It shows the diff before it writes

Before applying anything, the tool presents a preview: which custom formats and profiles would be created, which updated, which (if pruning is enabled) removed, and how scores would change. A dry run applies nothing. The operator sees the blast radius before a single write, because a sync that silently rewrites scoring is exactly what erodes trust in automation.

It proves the result, not the request

This is the crux. A competitor stops at “the API accepted the write” — but an accepted write is not a present, correctly-scored custom format. After a sync the tool queries the *arr API and asserts that the specific custom formats and profiles now exist and carry the expected scores, reading back the effect rather than trusting the acknowledgement. A sync whose read-back does not match the intended state is reported as failed, even if every write returned success.

It verifies the wiring and the source

The tool checks that each *arr instance is reachable and its credential accepted, and that the community catalogue source is reachable and readable, before it reconciles anything. It builds on the plain-language quality presets (D2): those choose what quality the household wants; this feature realises that choice as the concrete formats and scores inside each *arr.

It can explain a grab decision (stretch)

Given a release the *arr scored, the tool can explain why — which custom formats matched, what each contributed, and how the total compared to the profile’s cutoff — so a surprising grab (or a surprising skip) becomes legible instead of magic.

Every step is scriptable

Previewing a sync, applying it, and running the read-back proof are each reachable non-interactively, so the catalogue can be reconciled on a schedule.

States

State Meaning
unconfigured No *arr instances selected for sync, or no profiles chosen
in-sync Last sync applied and read-back confirmed the instance matches the selection
drift Instance diverges from the selection; a preview is available
previewed A dry run computed a diff that has not been applied
failed A sync applied but read-back did not confirm the expected formats or scores
source-unavailable The community catalogue source is unreachable

Edge cases

Situation Behaviour
Another tool writes the same instance and reverts changes Detect the divergence at the next read-back and report a fight over the instance rather than silently re-applying in a loop.
Pruning would remove formats the operator added by hand Never prune unless explicitly enabled; when enabled, list every unmanaged item that would be removed in the preview first.
Catalogue source unreachable Enter source-unavailable and leave the instance untouched; never apply a partial or empty catalogue over a working one.
Catalogue updated upstream since last sync Show the changed formats and scores in the diff so the operator sees what moved before applying.
Write accepted but format absent on read-back Report failed; an accepted write is not proof the format exists.
Score written but read back different Report failed with the expected and observed scores, not success.
Instance credential invalid Fail the credential check before any write; never write with an unverified credential.
Profile references a custom format the instance lacks Create the format before the profile that depends on it, or report the unmet dependency; never leave a profile referencing a missing format.
Two instances (Sonarr and Radarr) with divergent catalogues Reconcile each to its own selection independently; a failure on one must not block the other.
Operator applies without previewing Still compute and record the diff so the change is auditable after the fact.

Acceptance criteria

ID Requirement
H3-R1 The catalogue source MUST be an open, community-owned quality catalogue; the tool MUST NOT depend on a proprietary or closed quality source.
H3-R2 The tool MUST wire each *arr instance via its API and reconcile it to the operator’s selected custom formats and profiles declaratively.
H3-R3 The tool MUST present a preview diff — creates, updates, and (if enabled) removals with score changes — before applying, and a dry run MUST apply nothing.
H3-R4 After a sync, the tool MUST query the *arr API and assert that the intended custom formats and profiles exist and carry the expected scores; it MUST NOT treat an accepted write as proof.
H3-R5 A sync whose read-back does not confirm the expected formats and scores MUST be reported as failed, even if every write returned success.
H3-R6 The tool MUST verify each instance’s reachability and accepted credential, and the catalogue source’s reachability, before reconciling.
H3-R7 The tool MUST NOT prune unmanaged custom formats unless pruning is explicitly enabled, and when enabled MUST list every item that would be removed in the preview.
H3-R8 If the catalogue source is unreachable, the tool MUST leave the instance untouched and MUST NOT apply a partial or empty catalogue.
H3-R9 A custom format a profile depends on MUST be created before the profile, or the unmet dependency MUST be reported; a profile MUST NOT be left referencing a missing format.
H3-R10 Divergence caused by another tool writing the same instance MUST be reported as a conflict rather than driving a silent re-apply loop.
H3-R11 Each *arr instance MUST be reconciled independently, so a failure on one does not block the others.
H3-R12 The tool SHOULD be able to explain a release’s score by naming the custom formats that matched and their contributions relative to the profile cutoff.
H3-R13 Applying without an explicit preview MUST still compute and record the diff so the change is auditable.
H3-R14 Previewing, applying, and running the read-back proof MUST each be reachable non-interactively.