szl-experiment-replay
Replays a pinned, bounded CSV column mean against a frozen reference result and writes an explicit match, divergence or incomplete receipt. Use when a researcher needs to rerun a small tabular result offline and distinguish numeric disagreement from changed evidence.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-15
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- szl-experiment-replay
- license
- Apache-2.0
- description
- Replays a pinned, bounded CSV column mean against a frozen reference result and writes an explicit match, divergence or incomplete receipt. Use when a researcher needs to rerun a small tabular result offline and distinguish numeric disagreement from changed evidence.
02 资源文件 5
- 资产文件assets/declaration.json207 B
- 资产文件assets/expected.json62 B
- 资产文件assets/synthetic.csv51 B
- 脚本scripts/engine.py13.4 KB
- 脚本scripts/run.py2.8 KB
03 SKILL.md 正文
Experiment replay
Use Python 3.10+ and the standard library to replay one fixed operation on local CSV data. This
executes the skill's built-in csv_column_mean_v1; it does not execute a supplied script,
shell command, notebook, plugin or arbitrary Python. It uses no network or credentials.
Run
From this skill directory, use a writable copy of the synthetic assets or a study root of your own:
python -I -B scripts/run.py prepare assets/declaration.json --root . --output replay-pin.json
python -I -B scripts/run.py replay replay-pin.json --root . --receipt replay-receipt.json
Each output path must be new and relative to --root; the runner refuses overwrite. The
declaration names a CSV input, JSON reference result, column and absolute decimal tolerance.
prepare freezes the exact input, reference and built-in scripts/engine.py hashes. The CLI
loads that engine before replay compares its hash, then checks the input and reference hashes
before calculating the mean. This detects drift after loading a trusted package; it is not a
pre-execution integrity, authenticity or sandbox gate. MATCH means the observed mean
fell within the declared tolerance; DIVERGED means it did not. Changed or missing files yield
INCOMPLETE; invalid pins yield REFUSED. Exit codes: 0 match, 1 divergence, 2 incomplete or
refused. With a valid root and a safe, new receipt path, a missing, unreadable or malformed
on-disk pin also yields a retained REFUSED receipt. An invalid pin is not identified by its
unvalidated claimed digest: the receipt's pin_sha256 is null. If the root or receipt path is
unsafe or unwritable, the CLI exits 2 but cannot retain a receipt. A reference with a different
value must be deliberately prepared as a new pin to test divergence; editing a previously pinned
reference yields INCOMPLETE.
Retain the pin, receipt and input/reference bytes together. The self-digests catch accidental
changes but are not signatures. Receipts explicitly say SAME_HOST_LOCAL, signed: false,
independent_witness: false and scientific_claims_verified: false. The runner does not prove
the experiment design, data collection, author identity or another machine's reproduction.
Bounds and relationship to other skills
Input CSV: at most 256 KiB, 10,000 rows, 64 columns; reference JSON: at most 4 KiB. Paths are canonical relative paths under the selected root. Symlinks, junctions, other reparse points, special files, traversal, credential-like names and existing output files are refused. This is for a cooperative local filesystem, not an adversarial race-safe sandbox.
This is a CLI-only skill; scripts/engine.py is not a Claude Science sidecar.
Source tests and an SDK test double do not verify that the real Claude Science host can run
the CLI or attach this skill. Keep host registration and task evaluation NOT_VERIFIED until
that control plane supplies a gate, readback and pilot evidence.
Use szl-reproducibility-capsule to inventory broader source, environment and analysis plan
files. Its replay declaration remains inert; this skill independently executes only the fixed
CSV mean operation. For publication, arrange independent retention and a separate witness.
04 同仓库更多 Skills
szl-unit-invariants
Checks declared SI dimensions, allowlisted unit conversions, expected ranges and fixed pointwise invariants (equal, sum, difference, product, ratio) over selected measurement or simulation columns, flagging incompatible quantities, scale errors of 1000x and broken conservation checks. Use when merging instrument exports with different units, validating simulation output against conservation laws, or when a reviewer asks how unit errors were ruled out. Not an uncertainty-propagation or symbolic tool.
szl-uncertainty-lineage
Propagate declared measurement uncertainty through a small arithmetic dependency graph using a first-order delta approximation. Use when a researcher needs to see how corrected inputs or shared-source correlation change a reported standard uncertainty; require unit checks separately.
szl-typesafe-ai
TypeSafe/Jev: Choice/Noul/Score in code, not TypeScript types. System One models turn natural language and application state into probabilities that software can compose. Use when building routing, ranking, extraction, verification, or replacing an LLM prompt-and-parse step with structured decisions; also when brainstorming what TypeSafe could make possible. Do not use for TypeScript/Java/Rust static types, Zod/Pydantic/mypy/JSON Schema/protobuf codegen, or asking whether a pipeline is type-safe. At SZL, Jev is an optional second reader only.
demo
暂无简介
demo
暂无简介
szl-skill-update-review
Compare pinned local skill packages using inventories and a retained lock; report changed skills, files, declarations and reruns without approving safety or validity.