Upload K-12 math and coding dataset (1.0B math tokens, 1.02B code tokens)

.gitattributes

@@ -58,3 +58,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+code/final.jsonl filter=lfs diff=lfs merge=lfs -text
+math/final.jsonl filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,95 @@
+---
+license: odc-by
+task_categories:
+- text-generation
+language:
+- en
+tags:
+- math
+- code
+- education
+- k12
+size_categories:
+- 1M<n<10M
+---
+
+# K-12 Math & Coding Dataset
+
+A ~2 billion token dataset built for K-12 math and coding education use
+cases: **1,004,991,667 tokens** of math content and **1,020,505,114 tokens**
+of code content, both comfortably over the 1B-token target for each
+category.
+
+## Dataset Summary
+
+| Category | Tokens | Rows | File |
+|----------|-------:|-----:|------|
+| Math | 1,004,991,667 | 676,018 | `math/final.jsonl` |
+| Code | 1,020,505,114 | 937,248 | `code/final.jsonl` |
+
+Built via a 6-phase pipeline: source collection → cleaning (PII redaction,
+boilerplate/spam removal, encoding fixes) → deduplication (exact SHA-256 +
+fuzzy MinHash LSH) → heuristic quality filtering → budget-controlled final
+sampling → manual quality audit. Full methodology, per-source breakdowns,
+and known limitations are in `final_report.md`.
+
+## Schema
+
+Each line is a JSON object:
+
+```json
+{
+  "text": "...",
+  "source": "finemath4plus",
+  "category": "math",
+  "quality_score": 4.3,
+  "token_count": 512,
+  "language": "en",
+  "metadata": {}
+}
+```
+
+- `quality_score`: populated only for `finemath4plus` (has an upstream
+  classifier score); `null` for all other sources.
+- `language`: natural-language code (all records verified predominantly
+  English). For code records, the actual programming language is under
+  `metadata.programming_language`.
+- `metadata`: source-specific extra fields (URL, license, file path, repo
+  name, exercise name, difficulty, etc. — varies by source).
+
+## Sources
+
+**Math**: FineMath 4+, OpenWebMath, GSM8K, MATH (hendrycks_math).
+**Code**: GitHub Code (Python/JavaScript/Java/HTML/CSS via
+`codeparrot/github-code-clean`), freeCodeCamp, Exercism, APPS, CodeContests.
+
+## Token Counting
+
+All token counts use the `tiktoken` `cl100k_base` tokenizer, computed via
+exact `encode_ordinary()` calls (not estimates) at every pipeline stage.
+
+## Quality Audit
+
+Both categories were manually spot-checked twice (once pre-assembly, once
+on the final sampled corpus). Code passed cleanly at every check (~90-94%
+good, 0-2% bad). Math's final spot-check found a 15% "bad" rate (above the
+10% target threshold) driven by several low-frequency, hard-to-cheaply-filter
+patterns (AI-generated SEO filler, off-topic forum comment dumps, one
+essay-mill advertisement, thin template pages) — documented as an accepted
+residual limitation rather than silently delivered. **See `final_audit.md`
+and `final_report.md` Section 8 for full detail before using this dataset
+for anything quality-sensitive.**
+
+## Known Limitations
+
+See `final_report.md` Section 8 for the complete list, including:
+math corpus residual bad-rate, AMPS/OpenStax not sourced, The Stack v2
+substituted with GitHub Code due to gating, freeCodeCamp/Exercism volume
+shortfall, and HTML's residual auto-generated-doc contamination.
+
+## License
+
+Source licenses vary per record — see `metadata.license` for code records
+where available. FineMath 4+ and OpenWebMath are ODC-By. Aggregate dataset
+provided as-is for research/educational use; verify individual source
+licenses before redistribution.

code/final.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a4274d16123a3352a1633ce95d3fe0130e75645af0119f45c36202ff12975a1a
+size 4684965188

final_audit.md

@@ -0,0 +1,116 @@
+# Phase 5, Step 5.4 — Final Quality Spot-Check
+
+Samples drawn from the ASSEMBLED FINAL corpora (`data/final/{category}/final.jsonl`),
+seed=42, via `scripts/assemble/sample_final_audit.py` — distinct from Step 4.4's
+audit, which sampled from the pre-assembly `filtered.jsonl` files. This is the
+last line of defense before delivery.
+
+## Math (200-record sample, seed=42, from `data/final/math/final.jsonl`)
+
+Read 40 of 200 sampled records in full detail.
+
+### Ratings
+- ✅ Good: 29/40 (72.5%) — genuine, on-target K-12 through advanced math and
+  applied-STEM content: worked physics/EE/thermodynamics problems, forum
+  Q&A (PhysicsForums, MathHelpForum, MathOverflow, StackExchange), exam
+  content (CBSE, board exams, GMAT/GRE quant), logic puzzles, real analysis
+  and number theory, MATLAB/R applied-math code, K-12 arithmetic word
+  problems. Several records use a listicle/FAQ blog format but contain
+  factually correct, pedagogically useful content (e.g. "What is the
+  difference between real numbers and integers?", "How to calculate 2/3 of a
+  number") — style alone was not treated as disqualifying.
+- ⚠️ Borderline: 5/40 (12.5%) — legitimate but low-value: a bare index page
+  of NRICH problem titles/links with no actual problem content, a
+  commercial worksheet product page with a genuine explanatory tail, a
+  CourseHero paywall preview truncated mid-content, a WikiAnswers page with
+  messy comment-thread residue attached to an otherwise-fine answer.
+- ❌ Bad: 6/40 (15%) — **above pipeline.md's <10% threshold.** Specific
+  findings, each a distinct pattern:
+  1. **AI-generated clickbait filler** (2 instances, e.g. "Mind-Blowing
+     Result: You Won't Believe What Happens When You Divide 70 By 5!") —
+     sensationalized SEO content with no real mathematical depth. Already
+     documented as a known residual limitation in `quality_audit.md` (Step
+     4.4 found the same pattern, sample #25).
+  2. **Off-topic forum comment-thread dump** — a robot-automation/UBI/
+     politics discussion thread (source: i-am-bored.com) containing a couple
+     of incidental dollar-figure calculations, which was enough to pass
+     finemath4plus's classifier-trust shortcut (density/language checks are
+     skipped for this source per the Step 4.1 AMC-8 fix) despite having
+     almost no real educational math content. A new pattern — similar in
+     spirit to the Reddit-dump issue already fixed, but from a different
+     domain/format with no shared detectable signature.
+  3. **Content-aggregator dump** — a Q&A-farm page splicing together several
+     unrelated questions under an irrelevant title, including one entirely
+     off-topic, non-English question (Tagalog, about a typhoon's landfall).
+  4. **Thin SEO template page** — a "NCERT Solutions for Class X Maths"
+     listing that only describes chapter topics in generic boilerplate,
+     repeated near-verbatim across chapter headers, with no actual worked
+     content.
+  5. **Scraping artifact** — a factoring-trinomials page with unrelated
+     sidebar/related-links text (from completely different topics — pop
+     culture quotes, gaming guides, unrelated products) spliced into the
+     main text field, a scraper capturing page chrome alongside content.
+  6. **Essay-mill / homework-cheating-service advertisement** — a legitimate
+     econometrics/regression homework problem wrapped in explicit marketing
+     copy for a paid paper-writing service ("Zero-plagiarism guarantee",
+     "Money-back guarantee", "Free-revision policy"). The most concerning
+     individual finding: this doesn't just lower quality, it embeds
+     promotional content for an academic-dishonesty service inside an
+     educational corpus.
+
+### Conclusion and decision
+This sample's bad rate (15%, n=40) exceeds pipeline.md's 10% threshold and is
+higher than the earlier Step 4.4 pre-assembly audit (7.5%, n=40) — a real
+signal, not just sampling noise, though also not enormous in absolute terms.
+All six bad findings are DIFFERENT patterns from each other and from the
+already-fixed Reddit-dump issue; none has a cheap, low-false-positive
+detection signature the way the Reddit vote-marker did. Flagged to the user
+with the tradeoff (accept as a documented limitation vs. build one more
+targeted fix vs. broadly re-tighten finemath4plus's skipped density/language
+checks, at the cost of reintroducing the AMC-8 false-positive problem that
+motivated skipping them in the first place). **User decision: accept and
+proceed to delivery** — the corpus is still ~85% good+borderline (72.5%
+cleanly good), token targets are comfortably met (1.005B, 0.5% over target),
+and further tightening risks re-losing the legitimate terse/notation-heavy
+content the AMC-8 fix was specifically designed to recover. Documented here
+as a known, accepted residual limitation rather than silently ignored.
+
+## Code (200-record sample, seed=42, from `data/final/code/final.jsonl`)
+
+Read 46 of 200 sampled records in full detail, spanning all 9 sources
+(githubcode x5 across python/javascript/java/html/css, freecodecamp,
+exercism, apps, codecontests).
+
+### Ratings
+- ✅ Good: 43/46 (93.5%) — real, functional, well-formed code across all
+  languages and sources: production libraries (urllib3, IntelliJ, Quarkus,
+  Apache Aries), application code (Django/Angular/Android), freeCodeCamp
+  lesson content with solutions, Exercism canonical solutions, and
+  competitive-programming problems with correct multi-language solutions
+  (Codeforces via CodeContests).
+- ⚠️ Borderline: 2/46 (4.3%) — auto-generated artifacts that are real but
+  not educational: a Clover test-coverage-instrumentation JS file, a
+  Doxygen HTML search-index JS data file. Consistent with the same
+  "auto-generated docs slip through" gap already documented in
+  `quality_audit.md`'s code section (there found for Sphinx; here for
+  Clover/Doxygen — same category of gap, different tools).
+- ❌ Bad: 1/46 (2.2%) — a near-empty JS build-artifact stub containing only
+  `"use strict"` boilerplate and a base64-encoded sourcemap comment, with
+  zero actual logic. A transpiler/bundler output file that neither the
+  `is_minified` nor `is_autogenerated` heuristics were designed to catch
+  (it's not minified — it's just empty — and carries no "auto-generated"
+  marker text).
+
+### Conclusion
+Both thresholds (<10% bad, <30% borderline) are comfortably met — consistent
+with Step 4.4's pre-assembly code audit (~90% good, ~10% borderline, 0% bad).
+No further action needed for code; the two minor gaps found are the same
+class of low-frequency, low-severity issue already accepted for the code
+category.
+
+## Overall Step 5.4 conclusion
+Code passes cleanly. Math has a real, above-threshold bad rate (15%) driven
+by diffuse, low-frequency patterns with no cheap fix — flagged to the user,
+who chose to accept it as a documented residual limitation given the corpus
+is still ~85% good+borderline and both categories comfortably exceed their
+1B-token targets. Proceeding to Phase 6 (Documentation & Delivery).

final_report.md

@@ -0,0 +1,237 @@
+# Final Report — K-12 Math & Coding Dataset
+
+A 1B+ token, K-12-focused math corpus and a 1B+ token educational coding
+corpus, built per `pipeline.md`'s 6-phase spec. Both categories comfortably
+exceed the 1B-token target after cleaning, deduplication, quality filtering,
+and budget-controlled final sampling.
+
+## 1. Total tokens per category (final deliverable)
+
+| Category | Final Tokens | Final Rows | Target | Margin |
+|----------|-------------:|-----------:|-------:|-------:|
+| Math     | 1,004,991,667 | 676,018 | 1,000,000,000 | +0.5% |
+| Code     | 1,020,505,114 | 937,248 | 1,000,000,000 | +2.1% |
+
+Deliverables: `data/final/math/final.jsonl`, `data/final/code/final.jsonl`.
+Schema per record: `{text, source, category, quality_score, token_count,
+language, metadata}`. `quality_score` is populated only for finemath4plus
+(the one source with an upstream classifier score) — `null` elsewhere, since
+no equivalent score exists for structured benchmarks or code sources.
+`language` is the natural-language code (`"en"` throughout, all sources
+verified predominantly English); for code sources the programming language
+is preserved under `metadata.programming_language` instead.
+
+## 2. Per-source breakdown (final corpus)
+
+### Math
+
+| Source | Mode | Rows | Tokens | % of Category |
+|--------|------|-----:|-------:|---------------:|
+| finemath4plus | sampled to 700M | 508,357 | 700,003,574 | 69.7% |
+| openwebmath | sampled to 300M | 146,473 | 300,001,270 | 29.9% |
+| gsm8k | all | 8,792 | 1,417,863 | 0.14% |
+| hendrycks_math | all | 12,396 | 3,568,960 | 0.36% |
+
+### Code
+
+| Source | Mode | Rows | Tokens | % of Category |
+|--------|------|-----:|-------:|---------------:|
+| githubcode_python | all (under budget) | 236,531 | 287,901,844 | 28.2% |
+| githubcode_javascript | sampled to 300M | 333,786 | 300,000,630 | 29.4% |
+| githubcode_java | sampled to 240M | 236,242 | 240,000,466 | 23.5% |
+| githubcode_html | sampled to 90M | 51,764 | 90,006,022 | 8.8% |
+| githubcode_css | sampled to 60M | 50,152 | 60,000,074 | 5.9% |
+| freecodecamp | all | 10,258 | 5,175,953 | 0.5% |
+| exercism | all | 422 | 275,072 | 0.03% |
+| apps | all | 4,959 | 16,080,052 | 1.6% |
+| codecontests | all | 13,134 | 21,065,001 | 2.1% |
+
+`githubcode_python` came in under its original 400M config target (Step 4.1's
+`invalid_syntax` filter dropped 41,130 Python-2-syntax files — a deliberate,
+verified-beneficial filter, not a bug) and was included in full rather than
+backfilled by loosening quality thresholds. The shortfall was made up by
+increasing javascript/java/html/css's sampled targets above pipeline.md's
+original example percentages, since those sources had ample surplus.
+
+## 3. Pipeline stage summary — token counts
+
+| Category | Raw | After Cleaning | After Dedup | After Filtering | Final |
+|----------|----:|----------------:|------------:|-----------------:|------:|
+| Math | 3,248,617,940 | 3,231,640,208 | 3,215,245,918 | ~3,059,148,111 | 1,004,991,667 |
+| Code | 4,589,566,182 | 2,289,366,123 | 2,235,414,819 | ~2,109,766,319 | 1,020,505,114 |
+
+(Code's raw→cleaned drop is large because Step 1.4's `githubcode` source was
+deliberately over-downloaded well beyond the final target, per pipeline.md's
+"download ~10x headroom" guidance, then filtered down at every stage.)
+
+## 4. Cleaning pipeline summary (Phase 2)
+
+5 rule groups applied across 4 cleaning scripts (`clean_web_math.py`,
+`clean_structured_math.py`, `clean_code_stack.py`, `clean_code_educational.py`):
+- **Group A** (web-crawled math): boilerplate/nav stripping, spam/warez
+  pattern removal, encoding normalization (ftfy), PII redaction.
+- **Group B** (structured math): light validation only (already-clean
+  benchmark data).
+- **Group C** (GitHub code): vendored-code exclusion, autogenerated-file
+  exclusion (javadoc-specific + generic), minified-code exclusion, excessive
+  encoded-content exclusion, trivial-test exclusion.
+- **Group D** (educational code): freeCodeCamp non-coding block exclusion
+  (108 language-course superblocks removed via `freecodecamp_excluded_blocks.json`).
+- **Group E** (PII): email/phone/IPv4/SSN/API-key redaction with extensive
+  context-aware false-positive suppression (math arithmetic, version
+  strings, ISBNs, DOIs, test262 spec identifiers — see `config/cleaning_rules.md`).
+
+22 real bugs were found and fixed during Phase 2/4 via direct spot-checking
+of actual output (not just exit-code/summary-count trust), documented in full
+chronological detail in `progress.md`'s Decisions Log. The single most
+consequential: a math-density/language heuristic was wrongly rejecting
+excellent K-12 content (41 official AMC 8 competition problems, geometry
+"kite/rhombus" problems) because it lacked explicit math vocabulary or used
+non-ASCII notation (Greek letters) — fixed by trusting FineMath's own
+upstream quality classifier for that source instead of re-applying a cruder
+heuristic on top of it.
+
+## 5. Deduplication statistics (Phase 3)
+
+**Exact (SHA-256) dedup:**
+| Category | Rows In | Rows Out | Duplicates Removed |
+|----------|--------:|---------:|--------------------:|
+| Math | 2,181,776 | 2,172,722 | 9,054 (0.4%) |
+| Code | 1,974,655 | 1,973,991 | 664 (0.03%) |
+
+**Fuzzy (MinHash LSH, Jaccard 0.8 math / 0.85 code) intra-source dedup:**
+| Source | Kept | Removed |
+|--------|-----:|--------:|
+| finemath4plus | 1,878,340 | 5,128 |
+| openwebmath | 267,941 | 21 |
+| gsm8k | 8,792 | 0 |
+| hendrycks_math | 12,484 | 16 |
+| githubcode_python | 289,598 | 3,216 |
+| githubcode_javascript | 410,580 | 5,542 |
+| githubcode_java | 805,252 | 18,490 |
+| githubcode_html | 327,796 | 18,692 |
+| githubcode_css | 64,347 | 1,545 |
+| freecodecamp | 10,262 | 113 |
+| exercism | 422 | 2 |
+| apps / codecontests | 5,000 / 13,134 | 0 / 0 |
+
+**Fuzzy inter-source dedup:**
+| Category | Total Seen | Total Kept | Removed |
+|----------|-----------:|-----------:|--------:|
+| Math | 2,167,557 | 2,166,099 | 1,458 |
+| Code | 1,926,391 | 1,926,161 | 230 |
+
+A performance bug was found and fixed during this phase: per-shingle
+`MinHash.update()` cost scaled with document length (measured 13.6ms/doc,
+would have taken ~6+ hours on the largest sources) — fixed by capping
+shingles fed into MinHash to 150/doc via deterministic even-spacing
+(not random sampling, to stay reproducible), achieving a 4.6x speedup with
+`num_perm=128` kept unchanged per spec.
+
+## 6. Quality filter statistics (Phase 4, Step 4.1)
+
+**Math** (`filter_math.py`) drop reasons:
+| Source | Records In | Records Out | Top Drop Reasons |
+|--------|-----------:|-------------:|-------------------|
+| finemath4plus | 1,878,340 | 1,875,104 | reddit_comment_dump (1,108), length_out_of_bounds (1,716), repetition (401), encoding_quality (11) |
+| openwebmath | 266,495 | 232,018 | low_math_density (29,603), length_out_of_bounds (3,047), non_english (1,680), repetition (83), reddit_comment_dump (62), encoding_quality (2) |
+| gsm8k | 8,792 | 8,792 | none |
+| hendrycks_math | 12,472 | 12,396 | length_out_of_bounds (76) |
+
+**Code** (`filter_code.py`) drop reasons:
+| Source | Records In | Records Out | Top Drop Reasons |
+|--------|-----------:|-------------:|-------------------|
+| githubcode_python | 289,598 | 236,531 | invalid_syntax (41,130), comment_ratio_out_of_bounds (10,198), poor_identifier_names (905), length_out_of_bounds (834) |
+| githubcode_javascript | 410,571 | 400,453 | comment_ratio_out_of_bounds (6,703), poor_identifier_names (1,586), length_out_of_bounds (1,829) |
+| githubcode_java | 805,105 | 771,600 | comment_ratio_out_of_bounds (30,344), length_out_of_bounds (2,409), poor_identifier_names (752) |
+| githubcode_html | 327,765 | 324,900 | poor_identifier_names (943), length_out_of_bounds (1,002), comment_ratio_out_of_bounds (920) |
+| githubcode_css | 64,304 | 62,405 | comment_ratio_out_of_bounds (1,367), length_out_of_bounds (493), poor_identifier_names (39) |
+| freecodecamp / exercism / apps / codecontests | small, near-zero drops | | length_out_of_bounds only |
+
+`invalid_syntax` for Python was investigated directly (14.2% of raw files) and
+confirmed to be genuine Python 2 syntax (`print x`, `except X, e:`) — kept as
+a deliberate filter since excluding legacy syntax benefits a modern coding
+corpus. `comment_ratio` originally had a serious formula bug (a
+double-normalized calculation that returned meaningless values, plus a
+DOTALL regex bug that let one `#` comment "match" to end-of-file) — both
+fixed and the check simplified to a single upper bound (0.8) uniform across
+all 5 languages, after measuring that ~54% of real JavaScript files fall
+below any reasonable lower bound despite being genuinely good code.
+
+Step 4.2 (optional GPU classifier) was explicitly skipped per pipeline.md's
+"optional" designation — not requested by the user, and unnecessary given
+ample raw token surplus confirmed at every stage. Step 4.3 was effectively
+completed by Step 4.1's scripts.
+
+## 7. Audit results summary (Step 4.4 + Step 5.4)
+
+**Step 4.4 (pre-assembly filtered-data audit, sampled from `data/filtered/`):**
+- Code: 39/100 read — ~90% good, ~10% borderline (Sphinx docs, license
+  pages), 0% bad. Thresholds met.
+- Math: 40/100 read — ~80% good, ~12.5% borderline, 7.5% bad. Thresholds
+  met. Found and fixed the Reddit comment-dump content-appropriateness
+  issue (`is_reddit_comment_dump`, added despite low 0.065%/0.030% volume
+  on appropriateness grounds, not quality-volume grounds).
+
+**Step 5.4 (final post-assembly audit, sampled from `data/final/`):**
+- Code: 46/200 read — 93.5% good, 4.3% borderline, 2.2% bad. Thresholds met.
+- Math: 40/200 read — 72.5% good, 12.5% borderline, **15% bad (above the
+  10% threshold)**. Six distinct low-frequency patterns with no shared
+  cheap detection signature (AI-spam filler, off-topic forum dumps,
+  content-aggregator splicing, thin SEO templates, a scraping artifact, and
+  one essay-mill/homework-cheating-service ad). Flagged to the user with
+  three remediation options; **user decision: accept as a documented
+  residual limitation and proceed to delivery**, given the corpus is still
+  ~85% good+borderline and further tightening risks reintroducing the
+  AMC-8 false-positive problem. Full detail in `reports/final_audit.md`.
+
+Full detail for both audit rounds: `reports/quality_audit.md` (Step 4.4),
+`reports/final_audit.md` (Step 5.4).
+
+## 8. Known limitations and caveats
+
+- **Math corpus bad-rate (15% at final n=40)**: see above — accepted by the
+  user as a residual limitation. Content is genuinely low-quality/off-topic
+  in these cases, not corrupted or unsafe at scale, but a downstream
+  consumer doing further quality-sensitive work may want to apply additional
+  filtering for AI-generated SEO filler, off-topic forum dumps, and
+  commercial/service advertisements.
+- **AMPS and OpenStax** (pipeline.md's example math sources) were never
+  sourced — AMPS has no confirmed public HuggingFace dataset, and OpenStax
+  was not investigated given time constraints and ample surplus from the
+  sources actually used.
+- **The Stack v2** (pipeline.md's primary code source) was gated/inaccessible
+  (403 GatedRepoError, confirmed 2026-08-22) and was replaced with
+  `codeparrot/github-code-clean` per user decision — a non-gated substitute
+  with a different (row-level, not pre-sharded) structure.
+- **freeCodeCamp + Exercism combined (~5.5M final tokens)** fell well short
+  of pipeline.md's example ~150M "tutorial/educational" allocation — accepted
+  by the user, made up by drawing more from the primary GitHub code source
+  instead.
+- **HTML quality**: flagged early (Step 1.6) as having real quality issues
+  (OCR garbage, non-javadoc auto-generated docs slipping through). Kept per
+  user decision, relying on Phase 4's heuristic filters — final audit found
+  the residual auto-generated-doc gap persists at low frequency (Sphinx,
+  Clover, Doxygen artifacts), consistent across both audit rounds.
+- **quality_score is null for all non-finemath4plus sources** — no
+  equivalent upstream classifier score exists for those sources; this was a
+  deliberate choice over fabricating a placeholder value.
+- **Duplicate-block detector (built, not deployed)**: a generic
+  internal-duplication heuristic was built and tested during the Step 4.4
+  math audit (catching 1.5% of finemath4plus) but rejected after inspection
+  showed most matches were legitimate forum-quoting structure, not junk
+  duplication — documented as a rejected approach rather than shipped.
+
+## 9. Pipeline runtime and process notes
+
+The pipeline ran across 2026-08-22 through 2026-08-24. 22+ real bugs were
+found and fixed via direct spot-checking of actual output at every stage
+(never trusting exit codes or summary counts alone) — full chronological
+detail in `progress.md`. One data-integrity incident occurred late in
+Phase 5: two duplicate `filter_math.py --source finemath4plus` processes
+were found running concurrently, both writing to the same output file in
+`"w"` mode (a race condition from an unreconciled duplicate background
+launch), corrupting the file mid-write. Caught during Phase 5 token-summing
+(the file's row count changed between two consecutive reads), both
+processes were killed, the corrupted output deleted, and a single clean
+instance rerun and fully verified before any downstream work relied on it.

math/final.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f64b40cfd1df790e45c4b9cc8b491ae3614933e65a64e7f70099508d7448b7c
+size 3931184848