rewrite

README.md

@@ -58,25 +58,25 @@ configs:
 
 # Leanstral Mathlib calibration corpora
 
-The sample data is derived from the pinned Apache-2.0-licensed Mathlib source
-tree. This repository contains data and curated methods documentation, not the
-separately developed builder package; that package should not be redistributed
-until its standalone license file is resolved.
+The sample data comes from the pinned Apache-2.0-licensed
+[Mathlib source tree](https://github.com/leanprover-community/mathlib4). This
+repository holds the data and curated methods documentation—but not the
+separately developed builder package.
 
-This dataset contains the calibration corpora used to select a static FP8
+This dataset contains the calibration corpora used to pick a static FP8
 activation profile for an MXFP4 W4A8 conversion of Leanstral 1.5 119B-A6B. It
-publishes all candidate corpora, their manifests, and the shared
+publishes every candidate corpus, their manifests, the shared
 iterative-development pack, and the post-selection release-validation packs.
-Variant 1 is also exposed as the selected subset.
+Variant 1 also appears separately as the selected subset.
 
-The purpose is reproducibility: users can inspect the exact samples that shaped
-the activation profile, reproduce the corpus comparison, or test a different
-quantization method without reconstructing the data-selection pipeline.
+You can inspect the exact samples that shaped the activation profile, reproduce
+the corpus comparison yourself, or try a different quantization method without
+having to rebuild the data-selection pipeline from scratch.
 
 ## Loading with 🤗 Datasets
 
-Every corpus is a named Hugging Face configuration. `selected` is the default
-and is byte-identical to Variant 1:
+Every corpus is its own named Hugging Face configuration. `selected` is the
+default, and it's byte-identical to Variant 1:
 
 ```python
 from datasets import load_dataset
@@ -98,14 +98,14 @@ release_validation = load_dataset(
 
 The available configuration names are `selected`, `variant-1` through
 `variant-6`, `development`, `release-validation`, and
-`historical-validation`. They appear as separate subsets in the Hub Dataset
-Viewer and can be passed as the second argument to `load_dataset`.
+`historical-validation`. Each shows up as a separate subset in the Hub Dataset
+Viewer, and you can pass any of them as the second argument to `load_dataset`.
 
 ## Contents
 
 Every calibration variant contains 512 source-grounded records. The original
-four candidates hold the model, quantization rule, sample count, and category
-vocabulary fixed while changing the data distribution.
+four candidates keep the model, quantization rule, sample count, and category
+vocabulary fixed, and only vary the data distribution.
 
 | Split | Role | Summary | Release status |
 | --- | --- | --- | --- |
@@ -121,9 +121,10 @@ vocabulary fixed while changing the data distribution.
 | `selected` | convenience alias/copy | exact Variant 1 records | **Selected** |
 
 V5 and V6 are complete calibration corpora, not mixtures of activation-scale
-values. Each was recalibrated as one coherent static FP8 profile.
+values—each one was recalibrated from scratch as one coherent static FP8
+profile.
 
-## Suggested repository layout
+## Repository layout
 
 ```text
 README.md
@@ -155,13 +156,15 @@ schema.json
 SHA256SUMS
 ```
 
-The published manifest for each split should record its source revision,
-builder revision, tokenizer identity, sample count, token count, category and
-length distributions, deduplication inputs, leakage checks, and SHA-256.
+Each published split manifest records its source revision, tokenizer identity,
+sample count, token count, category and length distributions, provenance
+completeness, deduplication counts, and SHA-256. The records contain the
+original environment receipt, duplicate fingerprints, and leakage and
+admission results.
 
 ## Record format
 
-The quantization input is JSON Lines. At minimum, publication should preserve:
+The quantization input is JSON Lines. Each published record preserves:
 
 - a stable record identifier;
 - category and context-length stratum;
@@ -171,10 +174,9 @@ The quantization input is JSON Lines. At minimum, publication should preserve:
 - content/fingerprint fields used for deduplication; and
 - admission and leakage-check metadata needed to audit the split.
 
-The release preparation step should publish a generated schema from the actual
-JSONL files rather than relying on this prose alone. Private machine paths and
-internal operational metadata should be removed without changing sample text or
-stable identifiers.
+The checked-in schema is generated from the actual JSONL files rather than from
+this prose. Release preparation removed private machine paths and internal
+operational metadata without changing sample text or stable identifiers.
 
 ## Construction method
 
@@ -200,38 +202,38 @@ F  42
 total 512
 ```
 
-The publication methods document should define A–F from the builder's frozen
-category specification and cite the exact pinned Mathlib revision. Those values
-must be copied from the machine manifests during release assembly.
+The category definitions and exact pinned revisions are recorded in
+[`docs/corpus-construction.md`](docs/corpus-construction.md). Those values come
+from the machine manifests rather than being retyped for this card.
 
 ## Why Variant 1 is selected
 
 The candidates were evaluated as frozen static per-tensor FP8 activation
-profiles on the same MXFP4 weights. After FlashInfer startup autotuning was
-identified as the source of small cross-restart numerical drift, V1–V5 were
-compared in five independent normally autotuned cold starts. Every profile used
-the same dynamic reference within each start.
-
-V1 won candidate NLL, signed NLL delta, mean absolute per-example NLL delta,
-top-1 agreement, and top-1 flip count in all five starts. V5 consistently
-improved mean and maximum KL and repaired recurring V1 tail failures, but it
-lost central NLL/top-1 quality in every start and introduced stable failures of
-its own.
-
-Variant 6 tested the strongest remaining simple-composite hypothesis. It kept
-464 V1 records and made 48 targeted D/E substitutions, then was recalibrated as
-one complete profile. V6 improved mean KL in every start, but its V6-minus-V1
-candidate NLL averaged `+0.028728` and top-1 agreement averaged `-0.042`,
-outside the frozen `+0.01` and `-0.01` central-quality margins. The same
-decision holds using starts 2–5 alone.
-
-> Under normal FlashInfer startup variation, V1 consistently provides the best
-> ordinary predictive preservation. The available KL-tail improvements are
-> real, but neither V5 nor a targeted V1-centered composite achieves them
-> without an unacceptable central-quality cost.
-
-The experiment selects the exact frozen V1 corpus/profile pair. It does not
-claim that V1 is universally optimal for every model, quantizer, or future
+profiles on the same MXFP4 weights. Once FlashInfer startup autotuning turned
+out to be the source of small cross-restart numerical drift, I compared V1–V5
+across five independent, normally autotuned cold starts. Every profile used the
+same dynamic reference within each start.
+
+V1 came out ahead on candidate NLL, signed NLL delta, mean absolute per-example
+NLL delta, top-1 agreement, and top-1 flip count in all five starts. V5
+consistently improved mean and maximum KL and fixed recurring V1 tail failures,
+but it lost central NLL/top-1 quality in every start and brought its own stable
+failures along with it.
+
+Variant 6 tested the strongest remaining simple-composite hypothesis: keep 464
+V1 records, make 48 targeted D/E substitutions, then recalibrate the whole
+thing as one complete profile. V6 improved mean KL in every start, but its
+V6-minus-V1 candidate NLL averaged `+0.028728` and top-1 agreement averaged
+`-0.042`—both outside the frozen `+0.01` and `-0.01` central-quality margins.
+Dropping the first start and using 2–5 alone doesn't change that decision.
+
+> Under normal FlashInfer startup variation, V1 consistently gives the best
+> ordinary predictive preservation. The KL-tail improvements on the table are
+> real, but neither V5 nor a targeted V1-centered composite gets there without
+> an unacceptable hit to central quality.
+
+What this experiment selects is the exact frozen V1 corpus/profile pair—it's
+not a claim that V1 is universally optimal for every model, quantizer, or future
 recalibration.
 
 ## Selected artifact receipt
@@ -268,11 +270,11 @@ Appropriate uses include:
 - benchmarking alternative activation quantizers; and
 - auditing category, length, expert-coverage, and tail behavior.
 
-This is calibration and iterative-development data. It is not a general
-instruction-tuning corpus or a general model-quality benchmark. The
-`release-validation` split was untouched during profile selection, but it is
-published as the completed release-gate evidence rather than advertised as a
-reusable hidden test.
+This is calibration and iterative-development data—not a general
+instruction-tuning corpus, and not a general model-quality benchmark. The
+`release-validation` split stayed untouched during profile selection. It is
+published as the completed release-gate evidence, not advertised as a reusable
+hidden test.
 
 ## Limitations
 
@@ -285,15 +287,19 @@ reusable hidden test.
   worst teacher-forced token position.
 - Normally autotuned cold starts are the independent replication units.
 - Lower overflow-risk counts did not reliably predict better model behavior.
-- The Apache-2.0 declaration applies to the published Mathlib-derived data.
-  Model weights and tokenizer assets are distributed separately in the model
-  repository under the official source model's terms.
+- The separately developed builder package is not included. This repository
+  supports artifact-level audit and reuse, not an exact source-level builder
+  rerun.
+- The Apache-2.0 declaration covers the published Mathlib-derived data only.
+  Model weights and tokenizer assets are distributed separately, in the model
+  repository, under the official source model's terms.
 
 ## Reproducibility documents
 
-The dataset repository should include a concise construction document. The
-model repository should carry the detailed evaluation receipts:
+This dataset repository includes the construction document, while the model
+repository holds the detailed evaluation receipts:
 
 - activation-profile selection;
 - replicated normally autotuned profile study; and
 - constrained V1-centered composite result.
+