Specira sample artefact. Rendered from the governed default template on a fictional company. Names, figures and dates are illustrative.All artefacts →
SAMPLE
seeded demo data · specira.ai
Specira Embedding & Vector Strategy: Governed Template Rendering
Governed template rendering reference: Embedding & Vector Strategy default v2 (draft) definition b6e86011…d61e
Embedding & Vector Strategy · Project artifact SPECIRA

Embedding & Vector Strategy: Dispatch Modernization

Meridian Field Services: instructional example, not project evidence

Draft · watermark policy: draft_only template embedding_vector_strategy v2 · pack: specira_default_delivery embeddings of PII are PII; labels propagate; vectors rot silently
§1

Vectorized Content Inventory

mandatory 1 decision1 evidence rule validators: every_embedded_content_class_cites_a_classification_label · every_vector_source_field_cites_its_data_model_field · every_embedding_model_choice_documents_migration_cost

One index at pilot: override-note search for the exception review (the US-3 surface: "find past overrides with similar reasons"). The label propagates: the vectors and the index are as sensitive as their sources.

FieldValue
Source fields (cited) Assignment.reason_code · Assignment.note[1]
Class · propagated CONFIDENTIAL worker-conduct records (labels[2]); inversion reconstructs note text, so store access mirrors the source's: operations managers only (permission model[3])
Chunking One note, one vector; notes run 20 to 200 tokens; splitting buys nothing; rationale recorded
Metadata assignment id · hub · reason_code · created_at (the filter fields)
Model & store decisions

Model: the compact 384-dimension embedding model over the 1,536-dimension option. Quality on the held-out set differs by two points of Recall@10; storage and query cost by four times. Migration cost, stated: re-embedding the pilot corpus (60 thousand notes) is an afternoon and single-digit dollars, so the cheap model is reversible (embedding model[4]). Store: beside the relational data. The corpus sits far under the dedicated-store threshold and the filters are relational joins; the dedicated-store trigger is a ten-million-vector corpus or a hybrid-search need (vector store[4]).

§2

Retrieval Design & Baselines

mandatory 1 decision1 evidence rule validators: every_retrieval_config_declares_top_k_and_threshold · hybrid_and_rerank_stages_carry_triggers · every_retrieval_path_has_a_recall_or_mrr_baseline

Declared, not defaulted, and staged additions earn their place with triggers, so the future conversation has criteria. Anecdote is not a baseline.

Config (exception-review search)Value
Top-k · threshold 20 · similarity 0.72
Metadata filters · FIRST hub · date range · reason_code: relational, cheap, and they cut the candidate set before any scoring
Hybrid (dense + lexical) DEFERRED: reason codes are already a filter column, so the exact-identifier case is handled relationally; the trigger is free-text queries missing coded concepts past 10% of review sessions
Reranking DEFERRED: first-stage precision meets the baseline; the trigger is Recall@10 below the gate on the refreshed set

Baseline: Recall@10 0.91, MRR 0.78 on the golden query set (40 real review questions from M. Chen's team plus 60 synthetic, human-reviewed, register[5]); release gate: Recall@10 at 0.85. Retrieval-only, distinct from any generation eval, and none exists; the surface is retrieval-only at pilot (patterns[6]).

§3

Freshness & Invalidation

mandatory 1 decision1 evidence rule validators: re_embed_sla_is_numeric · every_deleted_source_has_a_tombstone_rule · rot_check_wired_to_observability

A stale note scores as high as a fresh one; similarity cannot tell. This section exists because of that.

RuleValue
Freshness · event-driven A new override note embeds on write (the queue the worker already consumes, building blocks[7]); re-embed SLA: 5 minutes; the nightly re-index is the backstop only
Tombstones The seven-year assignment archive transition (retention rules[2]) soft-deletes the vector (flagged, skipped at query time); asynchronous cleanup weekly; the retention rule drives the index, never the other way
Rot check Nightly comparison of source write timestamps against vector timestamps; a gap past the SLA raises a ticket (alerting[8])
§4

Open Questions

mandatory1 decision
QuestionOwnerAnswer byBlocks
Do synthetic golden queries need union-visible review? (The corpus is worker-conduct text; shared thread with security open questions[9].) N. DuvalSep 30, 2026 The golden set's refresh rule only
Refs

References & Package Contents

In the Specira workspace

specira [1] Data model: Assignment.reason_code, Assignment.note app.specira.ai/projects/dispatch-modernization/artifacts/data-model
specira [2] Data classification: labels, retention rules app.specira.ai/projects/dispatch-modernization/artifacts/data-classification
specira [3] Auth & authz policy: the permission model the store mirrors app.specira.ai/projects/dispatch-modernization/artifacts/auth-authz-policy#permission
specira [4] Decision log: embedding model, vector store decisions app.specira.ai/projects/dispatch-modernization/artifacts/decision-log
specira [5] Test cases: the golden query set as a registered asset app.specira.ai/projects/dispatch-modernization/artifacts/test-cases
specira [6] AI orchestration blueprint: the retrieval-only pattern row app.specira.ai/projects/dispatch-modernization/artifacts/ai-orchestration-blueprint
specira [7] Architecture: building blocks (the queue the embed hook rides) app.specira.ai/projects/dispatch-modernization/artifacts/architecture
specira [8] Observability: the rot-check ticket wiring app.specira.ai/projects/dispatch-modernization/artifacts/observability-monitoring#alerting
specira [9] Security requirements: the shared union-review thread app.specira.ai/projects/dispatch-modernization/artifacts/security-requirements
Generated by Specira · template embedding_vector_strategy v2 (draft) · pack specira_default_delivery lineage b6e86011…d61e · page 1 of 4