# Implementation Plan

## Milestone 0 — Source reconnaissance + repository foundation

- verify Python package/CLI setup
- detect both expected source files:
  - `data/raw/smk_all_da.json`
  - `data/raw/smk_all_en.json`
- inspect actual top-level JSON structures without guessing
- compute bytes + SHA-256
- count/summarize records with bounded memory
- compare DA/EN identity sets
- write `source-manifest.json`
- document any structural surprises

Done when we understand the actual dumps well enough to implement the parser from evidence rather than assumptions.

## Milestone 1 — Streaming source loading + canonical schema

- implement bounded-memory loader for actual observed structure
- join DA/EN by `object_number`, never by row position
- preserve localized text where useful
- map stable non-language fields into canonical columns
- preserve source provenance/disagreements
- parse physical dimensions/units defensively

Done when a representative fixture/sample becomes deterministic canonical rows.

Status: source loading and pre-canonical audits complete. Stable source-row identity, lossless
logical Parquet, duplicate taxonomy, dimension vocabulary, palette/visual quality and DA/EN
precedence recommendations are documented. Canonical entities are implemented in Milestone 2
without collapsing duplicate `object_number` groups.

## Milestone 2 — Canonical entities + first analyzable corpus

- pair language rows only by SMK `id`
- preserve `object_number` as non-unique grouping metadata
- canonicalize fields with explicit provenance and conflicts
- resolve image bundles atomically
- preserve and normalize dimension observations without choosing conflicts
- freeze `smk-public-visual-v1` with machine-readable exclusion reasons

Status: complete. See `CANONICALIZATION.md` and `CORPUS_V1.md`. Visual/colour feature extraction
has not started.

## Milestone 3 — Deterministic per-artwork visual features

Implemented for frozen `smk-public-visual-v1`:

- perceptual centroid
- explicitly uniform bulk-palette weights and duplicate-colour handling
- sRGB -> OKLab -> OKLCH with achromatic hue semantics
- palette entropy/effective size with documented uniform-weight limitation
- palette spread
- weighted lightness/chroma variation
- hue concentration
- SMK visual fields + provenance
- physical/image geometry + logs

Status: complete. Raw features remain corpus-independent; no corpus normalization or similarity is
implemented. See `FEATURE_EXTRACTION_V1.md` and `FEATURE_EXAMPLES.md`.

## Milestone 4 — Corpus statistics + normalization

Status: complete. Profile every candidate scalar over frozen `smk-public-visual-v1`, preserve raw
OKLab semantics and missingness, generate robust-01/midrank-percentile/robust-z variants, add log
aspect and circular-hue diagnostics, and publish separate unweighted candidate families. See
`NORMALIZATION_V1.md` and `FEATURE_VECTOR_V1.md`.

## Milestone 5 — Pilot build

Status: complete as deterministic exact similarity. Three separately versioned profiles use
sample-calibrated component scales, explicit weights, shared-component missingness and explainable
contributions. QA examples, stability diagnostics and performance measurements are published in
`SIMILARITY_V1.md` and `SIMILARITY_EXAMPLES.md`.

## Milestone 6 — Similarity preparation

Status: enrichment-backed weighted color layer implemented separately from bulk v1. It provides a
resumable append-only source cache, weighted palettes/features v2, bulk/enrichment diagnostics and
an exact palette-transport similarity profile. See `ENRICHMENT_V1.md`, `WEIGHTED_PALETTES_V1.md`
and `WEIGHTED_SIMILARITY_V1.md`.

## Milestone 7 — First usable application + sonification

Status: complete. A compact read-only web catalog feeds one FastAPI + React service with search,
artwork detail, existing exact similarity profiles, explicit cached weighted similarity, and three
versioned deterministic browser-audio mappings. See `SONIFICATION_V1.md` and `WEB_UI_V1.md`.

## Milestone 8 — Similarity

Implement separate profiles rather than one magic similarity score:

- `color`
- `visual`
- `composition`

Use fast feature-vector candidate retrieval followed optionally by weighted-palette optimal-transport reranking. Do not precompute all pairs.

## Milestone 9 — Art Tone v1

Implement versioned, explainable sonification mapping separate from objective features.

Initial mapping:

- centroid hue -> chromatic pitch class
- physical area percentile -> register
- chroma -> velocity/richness
- lightness -> timbral brightness
- palette entropy -> complexity
- contrast -> attack
- hue concentration -> tonal purity

Emit exact inputs alongside every mapped result.

## Milestone 10 — Exploration

Only after Dataset v1 is trustworthy, start experimental interfaces/notebooks/apps:

- nearest-neighbour browser
- colour-space explorer
- timelines
- artist palette fingerprints
- mosaics
- sonification player