Seven new aroma heads from unique chemistry, and a count sweep

[rvnminers-A-and-N]

Progress on #262.

Chasing blackberry produced nothing across three rebuilds, so this asks the better question instead: which descriptors deserve a head that don't have one? Ten sat below the ten-positive floor; seven had genuinely distinctive character chemistry and were given it.

	aroma heads	total heads
before	166	189
after	172	195

Five landed (figures from the final manifest, verified against the running app):

new head	AUROC	n_pos	precision	character chemistry
angelica	0.972	11	0.54 — confident	macrocyclic lactones
quince	0.955	12	0.02	marmelo lactone / oxathiane
galbanum	0.922	10	0.12	galbanum pyrazine, undecatriene
wormwood	0.807	10	0.21	thujones, artemisia ketone
immortelle	0.781	12	0.14	italidiones

lychee reached 12 positives but finished at 0.693 — just under the bar. It measured 0.831 in
an intermediate run, so it belongs in the marginal cluster described below rather than the win
column. tagetes stayed at 7 positives and never trained.

Deliberately not attempted: cucumber, watermelon, boronia. Their chemistry is the same C9 nonadienal/nonadienol set melon already owns and the ionones violet owns — entangled classes, which is the trap that cost blackberry 0.10 AUROC. Adding them would likely damage heads that already work.

quince is worth staring at

AUROC 0.955 with 0.02 precision. Near-perfect ranking; right one time in fifty when it fires. It's now the sharpest single example in ACCURACY.md of why AUROC alone is not a quality measure — and it replaces a caveat there that had gone stale.

The count sweep

Verifying the README turned up that we'd been publishing 190 heads when the app served 189. The figure predated blackberry dropping and had propagated into the README badge, the CHANGELOG, ACCURACY.md, the v0.2.0 release notes and code comments. Now 195 everywhere, checked against the manifests rather than carried forward.

ACCURACY.md's tables were stale too — precision distribution read 25/9/60/58 against a live 19/15/74/58, and the named "never wrong" heads were the wrong sixteen. Both regenerated from the manifest.

On the roster number — now documented

A cluster of heads sits at AUROC 0.69–0.71 on 11–15 positives, where 5-fold CV can't resolve better than ~±0.02. blackberry measured 0.696 → 0.707 → 0.687 → 0.720 across four consecutive rebuilds with its positive count unchanged at 13. The folds moved, not the head.

A head crossing the bar only means something if n_pos moved with it — which is exactly what separates that from chamomile: positives 11 → 15, AUROC 0.691 → 0.741, and it now makes confident calls at 1.00 precision.

MIN_AUROC stays at 0.70. Lowering it to 0.69 would hand back the marginal heads instantly and make the roster whatever we wanted it to be.

Verification

• ruff clean; suite 43 passed
• rebuilt: master_enrichment 8,869 (all named), profile_index 8,850 × 183, flavor_map
• audit: 154/172 generalise, 17 precision-limited, 1 memorising (turmeric)
• needs sudo systemctl restart flavormancer — every model and all three artifacts changed