Paid local software · v1.0.0 · Private delivery

Freeze the feature list before outcomes can influence it.

Target-Blind Feature Freezer turns a strict, values-free column schema and an explicit selection policy into a deterministic candidate-feature contract, ordered manifest, exclusion ledger, and checksum set.

Request the $249 tool Check v1 fit Evaluator guide Download evaluation pack Inspect clean manifest Read exact offer Compare leakage boundaries

Already enforcing the same pre-target boundary with a feature store, schema contract, model signature, or reviewed internal gate? Keep it. This package is for teams that need a small standalone, values-free contract and retained exclusion evidence. No account is required to run it; the private implementation is not published to GitHub.

01 / Target-blindThe interface accepts no dataset values.

The schema must declare data_values_present: false; unknown fields and embedded row samples are rejected.

02 / ExplicitRoles and policy are declared twice.

Target, identity, and time names must exist, match their schema roles, remain disjoint, and cannot become candidates.

03 / VerifiableThe ordered choice is hash-bound.

Every governed output has a checksum, and the verifier detects changed, missing, or unexpected files.

Supported decision

Which declared fields satisfied this exact policy before target values were read?

Feature selection can absorb outcome knowledge long before model fitting begins—through manual curation, permissive naming, or an undocumented schema pass. The freezer creates a reviewable pre-target boundary from metadata alone and preserves both what entered and why every other column stayed out.

Workflow

Declare, freeze, verify.

01 / Declare

Name every boundary

Assign each column a normalized type and one of five roles, then separately declare target, identity, and time names.

Inspect column schema

02 / Freeze

Apply one exact policy

Use allowed types, ordered include and exclude expressions, forbidden exact names, and selected-count bounds.

Inspect contract schema

03 / Verify

Retain a checkable pack

Keep the ordered manifest, selected-feature CSV, exclusion ledger, summary, and exact file checksums together.

Inspect manifest schema

Best for: ML engineers, data scientists, model validators, and governance teams freezing pre-model feature contracts
Input: A strict JSON freeze contract and a strict values-free JSON column schema
Output: Manifest JSON, selected-feature CSV, exclusion ledger, Markdown summary, and checksums
License: Perpetual internal use by one purchasing legal entity; no redistribution, resale, publication, or service-bureau use
Distribution: Private delivery after accepted request and confirmed payment; no public package URL

Boundary

A frozen declaration is not proof that leakage is absent.

The tool cannot verify that operator-supplied roles are correct, that upstream feature computation was target-blind, that hidden proxies are harmless, or that a selected feature is useful. Hashes identify bytes; they are not signatures, attestations, or trusted timestamps.

Feature-freeze contractInspect accepted schema, freeze exits, release files, adversarial tests, and support.

Licensed runtime

Schema metadata in. Frozen contract out.

python -m target_blind_feature_freezer freeze contract.json --out frozen-features
python -m target_blind_feature_freezer verify frozen-features

freeze exits 0 after successful create-only publication and 1 for invalid input or operation. verify exits 0 when intact, 2 on integrity failure, and 1 when the pack cannot be inspected.

Licensed delivery

Implementation, schemas, tests, and synthetic evidence.

Source-available Python packageStandard-library CLI and library under the one-organization commercial license.
Wheel and checksumsInstallable v1.0.0 wheel plus SHA-256-bound commercial archive.
Three strict JSON SchemasDraft 2020-12 contracts for the freeze request, values-free column schema, and frozen manifest.
Thirty-five testsRole, pattern, limit, path, determinism, tamper, privacy, isolated-install, and CLI coverage; one Windows symlink case may skip without privilege.
Reference casesSynthetic clean and adversarial-exclusion inputs with deterministic expected packs.
One support roundOne consolidated installation or contract-format email received within 30 calendar days after delivery.

Public evaluation recordsOpen the fit check, evaluator guide, synthetic manifests, schemas, pack, checksum, and exact offer.

Request to buy

Confirm the licensee and schema scale before payment.

Send the legal-entity name, billing contact, country, intended Python version, and approximate column count. Do not send datasets, target values, credentials, private source, customer data, regulated data, or confidential feature names. A request is not an accepted order, and availability is not guaranteed until accepted in writing.

Request the $249 tool Read product terms