Design Resolution Collection

Overview

The tenant-scoped wf2des decision ledger — an autonomous quality ratchet for component selection. There is no human-correction loop, so quality must converge machine-side: the ledger maps a deterministic, versioned structural section signature to the best accepted component resolution for each structural KIND of section. generation-assemble OWNS this collection — it CONSULTS the ledger before model selection and WRITES BACK through a score ratchet after the deterministic guards; no other run touches it. One doc per resolved section KIND, keyed by a structural, LLM-free section_signature, so the same section KIND hits the same entry on every screen and every rerun. It is not a vector index or learning corpus.

Table Definition

Logical Name	Physical Name	Column Name	Data Type	Primary Key	Relation	Unique	Nullable	Default Value	Remarks
Design Resolution	design_resolution	_id	string	◯					{organization_id}_{project_id}_{section_signature} — this IS the identity; no separate surrogate
		organization_id	number						Tenant owner; mandatory tenancy scope on every read/write
		project_id	number		project:id				Project scope; mandatory tenancy scope on every read/write
		section_signature	string						STRUCTURAL section identity, versioned (sig@1:…) — sha256 over the section's kit-component names, normalized name, child-type shape, and text count. Deterministic and LLM-free (never the LLM-assigned roles), so the same section KIND hits the same entry on every screen and rerun
		case	string						instance | compose
		platform_design_id	string		design:id		◯		The resolved component (= design_component._id); null for a compose resolution
		component_name	string						Display/debug convenience
		variant_policy	object						axis → value; CLAMPED to registry axes before write
		provenance	string						mined | voted (see Type Codes); mined evidence has authority over votes
		score	float						DETERMINISTIC decision quality (base + modeled + variant-completeness + scope-fit; mined = 1.0) — the ratchet's currency
		prompt_version	string						Provenance of a voted decision
		updated_at	string						Last accepted ratchet write (ISO-8601 UTC timestamp)

Relations

• _id = {organization_id}_{project_id}_{section_signature}, by value — the sole identity, no surrogate.
• project_id → project.id
• platform_design_id → the platform design row id (type=component) = design_component._id, by value.
• No relation is enforced inside DocumentDB; relations are logical and validated by application code.

Indexes

• PRIMARY KEY (_id)
• SCORE-RATCHET upsert on _id: a voted write lands only where the held entry is NOT mined AND scores STRICTLY lower (server-side filter; a DuplicateKeyError on the fenced upsert means KEPT, returned as such — never an error). mined writes are unfenced. Identical evidence scores identically → reruns never churn; better decisions win → monotonic improvement.
• Every read and write filters organization_id + project_id (tenancy scope).

Type Codes

provenance:

• mined: read from designer-made designs already in the Figma file — the only above-vote authority (score = 1.0).
• voted: the K-sample self-consistency selection vote.

case:

• instance: a single registered component resolves the section.
• compose: the section is composed from multiple units (each unit re-selects).

Notes

• generation-assemble OWNS writes — assemble-only; no other run touches this collection. One doc per resolved section KIND (not per screen).
• CONSULT (assemble, BEFORE selection): a mined entry LOCKS its section and skips the selection call (no LLM). voted entries are retained as diagnostics and never override current-screen selection. Instance entries are revalidated at consult: a stale entry whose component is not among the section's candidates falls through to the vote, and default-variant entries are scope-revalidated (default_size speaks only for the default variant).
• WRITE-BACK (assemble, AFTER the guards): every non-unmatched section (and compose-child) decision upserts with its deterministic score, fenced by the ratchet.
• Evaluation runs may disable ledger reads and writes; production defaults keep both enabled.
• section_signature is the crux: because it is STRUCTURAL and LLM-free, it is stable across screens and reruns — the ledger accumulates per section KIND, not per instance.
• No lineage block — the ledger is a running decision cache keyed by structural identity, not a lineage-tracked artifact; updated_at records the last accepted write.
• See the wf2des I/O Definition for the assemble consult / write-back steps and the score-ratchet fence.