Back to dashboard
`docs/design/search-quality-eval.md` §14.2 Phase 1.4 (acronym / synonym), query class C from §1.4

Search-quality baseline: acronym / synonym recall (Phase 1.4, v1.2.0 candidate)

This audit tests framework_aliases.synonyms — the cupertino-specific table that maps colloquial / abbreviated names to canonical framework slugs (nfc → corenfc, wifi → corewlan, bluetooth → corebluetooth, ml → coreml, etc.). The expectation: when a developer types the synonym alone as a query, cupertino should route to the canonical framework root via the synonyms table.

Measured 2026-05-20·Weak

Headline result
4 / 22 (18.2%)

Read in detail

Each card opens its own page. The headline and charts above are all you need at a glance; the cards are for the why and how.

Aggregate

This is the worst-performing class baseline by a wide margin.

Read details →

The 4 Wins

All four wins are explainable by literal-substring or prefix-token matching, not by synonym lookup. The synonyms table may not be contributing to these results at all.

Read details →

The 18 Misses

Every miss is a literal-token match against a deeper path that happens to contain the query word. The canonical CoreX framework is either far down the ranking (rank 2-7) or absent from the top-10 entirely.

Read details →

What This Baseline Says

The framework_aliases table exists in the schema and contains 22 rows with a populated synonyms column (verified directly via SELECT identifier, synonyms FROM framework_aliases WHERE synonyms IS NOT NULL).

Read details →

Possible Future Directions (out of Scope for This Audit)

Following the feedback_code_changes_as_ideas_for_future rule, three candidate paths, in increasing complexity:

Read details →

Implications for Criterion 2 (anti-hallucination)

For the AI agent consumer: if the agent issues a bare NFC query expecting the CoreNFC framework reference, it gets an authentication-services deep page instead.

Read details →

Method Recap

22 (synonym, canonical_URI) pairs, all drawn from framework_aliases.synonyms rows in the v1.2.0 candidate DB (excluding data and text as too-generic).

Read details →

Combined Phase 1 Baseline Coverage on V1.2.0

Five of eight Phase 1.x classes from §1.4 now have documented baselines. Three remain: G (prose), H (symbol-attribute), Phase 1.7 (anti-hallucination agent-end-to-end).

Read details →

Sources cited in this measurement

Every metric and method this audit relies on, with a link to the foundational source. Auto-collected from the audit text.

Mean Reciprocal Rank

Voorhees (1999), TREC-8 QA Report

Open citation

Reciprocal Rank Fusion (k=60)

Cormack, Clarke, Büttcher (2009), SIGIR

Open citation

P@k (Precision at k)

Manning, Raghavan, Schütze (2008) IIR §8.4

Open citation