zpe-smell 0.1.0

Bounded smell encoding research repo over a surrogate receptor-response panel.

ZPE-Smell

What This Is

Surrogate receptor-response smell codec. Fixed 35-episode benchmark reaches Spearman 0.9996 with zero mixture, fiber, and adaptation collisions.

ZPE-Smell is one of seventeen public encoding lanes in the Zer0pa ZPE portfolio. It is useful now for audit and replication on the declared benchmark surface. It is improving continuously.

Codec Mechanics

Field	Value
Architecture	SMELL_STREAM
Encoding	SMELL_SURROGATE_PANEL_V1
Mechanics Asset	.github/assets/readme/lane-mechanics/SMELL.gif

Key Metrics

Metric	Value	Baseline
SPEARMAN_RHO	0.9996	reference
NN_AT_1	94.29%	reference
MIXTURE_COLLISION	0.00%	max 0.05
FIBER_COLLISION	0.00%	max 0.05

Source: proofs/artifacts/public_smell_surrogate_scope.json · CI: tests/test_public_surface.py::test_committed_artifacts_match_live_eval

Repo Identity

Field	Value
Identifier	ZPE-Smell
Repository	https://github.com/Zer0pa/ZPE-Smell
Section	encoding
Visibility	PUBLIC
Architecture	SMELL_STREAM
Encoding	SMELL_SURROGATE_PANEL_V1
Commit SHA	48507cbfcdc5
License	SAL-7.0
Authority Source	proofs/artifacts/public_smell_surrogate_scope.json

Readiness

Field	Value
Verdict	STAGED
Checks	5/5
Anchors	2 display anchors
Commit	48507cbfcdc5
Authority	proofs/artifacts/public_smell_surrogate_scope.json

Honest Blocker

We do not claim a digital smell product.; We do not claim anything beyond the surrogate receptor-response scope shipped here.; We do not claim this result is admitted to the broader certified subset.

What We Prove

• On the fixed surrogate receptor-response benchmark surface, the codec preserves retrieval structure with Spearman 0.9996.
• On that same surface, nearest-neighbor recall at 1 is 94.29%.
• On the declared nuisance family cases, mixture collision, fiber collision, adaptation alias, and attack success all stay at 0.00%.
• Image-prefixed nuisance streams do not break smell routing on the committed benchmark surface.
• The full public artifact can be regenerated from the code and matched exactly against the committed proof file.

What We Don't Claim

• We do not claim a digital smell product.
• We do not claim anything beyond the surrogate receptor-response scope shipped here.
• We do not claim this result is admitted to the broader certified subset.
• We do not claim coverage of a full empirical receptor panel or general smell perception.
• This scope is intentional, not a limitation in progress — broader claims will be named explicitly when they are earned.

Verification Status

Code	Check	Verdict
V_01	Full-panel roundtrip keeps all eight receptor channels intact	PASS
V_02	Image-prefixed nuisance streams still route as smell on the fixed bundle surface	PASS
V_03	Public evaluation keeps the bounded-scope verdict and all gates in pass state	PASS
V_04	Public evaluation beats the receptor-only and geometry-only comparators on the committed surface	PASS
V_05	Live evaluation matches the committed proof and reference artifacts exactly	PASS

Proof Anchors

Path	State
proofs/artifacts/public_smell_surrogate_scope.json	VERIFIED
validation/results/reference_public_eval.json	VERIFIED

Repo Shape

Field	Value
Proof Anchors	2 display anchors
Modality Lanes	1
Architecture	SMELL_STREAM
Encoding	SMELL_SURROGATE_PANEL_V1
Verification	5/5 checks
Authority Source	proofs/artifacts/public_smell_surrogate_scope.json

Extended Metrics

Rows retained from the previous expanded ## Key Metrics table. The public product page uses the first four rows only.

Metric	Value	Baseline
ADAPTATION_ALIAS	0.00%	max 0.20
ATTACK_SUCCESS	0.00%	eq 0.00

Competitive Benchmarks

Approach	Spearman	NN@1	Mixture Collision
ZPE-Smell	0.9996	94.29%	0.00%
Receptor-only	0.9648	85.71%	14.06%
Geometry-only	0.9862	88.57%	7.19%

Source: proofs/artifacts/public_smell_surrogate_scope.json · CI: tests/test_public_surface.py::test_public_eval_beats_public_comparators

Quick Start

python3 -m venv .venv
. .venv/bin/activate
python3 -m pip install --upgrade pip
python3 -m pip install -e '.[dev]'
python3 -m zpe_smell.reproduce --output validation/results/latest_public_eval.json
python3 -m pytest -q

Upcoming Workstreams

This section captures the active lane priorities — what the next agent or contributor picks up, and what investors should expect. Cadence is continuous, not milestoned.

• Olfactory codec — commercial path definition — Zero-Base Scientific Thinking — GPD Research and Planning Pending. The 35-episode surrogate proves retrieval-structure preservation but lacks commercial weight without empirical receptor data. Active research into commercial path (digital olfaction interfaces, fragrance-formulation AI, anosmia rehabilitation) — each path has different empirical-data requirements; selection precedes any corpus scale-up.

Citation

Use CITATION.cff for software citation metadata, and cite the committed proof artifact when discussing the bounded result.