_Created: 03-07-2026 · Last updated: 11-11-2026 (H7975: registered tools/dead_lane_sweeper.py + tools/dead_lane_wt.py — ежесуточный мёртвец-свипер (dead PR / клеймы / ворктри → суточный дайджест + автоудаление merged+clean); prior 09-11-2026 (H7908: registered the OG-card pixel-gate pattern (stdlib PNG codec + golden per-pixel comparator + negative-control selftest); prior 01-11-2026, H7658: registered tools/zcode_cli_concurrency_baseline.py — per-plan in-flight/retry baseline over both ZCode log schemas; H7647: registered the venue-profile contract + fail-closed release receipt; H7645: registered the approved-freeze machinery — paper release contract per ADR-0007; prior 31-10-2026)_
_Created: 03-07-2026 · Last updated: 22-10-2026 (H7162 retrieval-surfaces census: registered the Systema FAQ-RAG measurement kit (§1 — previously absent from every registry), the retrieval-surfaces census hub (§4), the NotebookLM external answering stack (§5), and two new interlinks edges — ORS-FAQ→Systema faq-gold-question-seed, claude-config→SecondBrain ask-the-wiki-vault-qa; full census: reports/RETRIEVAL_SURFACES_CENSUS_20-10-2026.md); prior 17-10-2026 (DSH lane: registered magenta-duck-vps/tools/appstore_storefront_probe.py — App Store storefront availability probe, 6 country stores × 8 VPN/proxy clients, --selftest, 17-10-2026; H5275, OxAlpha (ollama-cloud/glm-5.3-flash): registered tools/jev_probe.py — the shared TypeSafe Jev «System One» client (choice/noul/score, $0.042/1M, selftest 16/16) for the jev pilot batch H5274–H5280; H3744, Opus 5 claude-opus-5: registered sense-alignment-pilot — the first ALIGNED cross-dictionary sense table, PWG↔MW↔Apte joined by shared literary witness; H2874, Opus 5 claude-opus-5: registered the PWG <ls> count tables — two families, work-family vs cleaned-string, never comparable; H2883, Opus 5 claude-opus-5: registered the Nīlakaṇṭha licence register — 151 tradition-attested Pāṇini deviations, join key deviation_term_sa)_
Purpose. A benefit-first, clickable index of the canonical reusable assets across the 163 repos (registry count, 24-09-2026) — the code you should import, the data you should query, the spine you build on, the hubs you read first, and the external stacks you call rather than clone. It unifies the two code/data hubs into one "consume, don't re-derive" catalogue so a fresh session spends its time on the *gap*, not the twentieth copy.
Read one altitude higher first: MEGABOOK.md explains which programme the task serves, what foundation already exists, and which work should be strengthened, maintained, or paused. This index then answers the narrower question: exactly which code or data asset should be consumed.
MEGABOOK.md §12 — the owner → asset → consumer → do-not-rebuild chain over this catalogue's key code/data/schema/workflow families, with rights/quarantine tiers (slot A3 of the MEGABOOK × Sangram plan, 11-07-2026).SHARED_CODE.md (census 2026-06-14).PROJECT_INTERLINKS.md.FEATURES_INDEX.md — the capability inventory of every dictionary, interface, dataset and tool, with a real example and stable ID per asset. This file says *"don't rebuild — consume this"*; that one says *"here's what there is."*SEO_STRUCTURED_DATA_PLAYBOOK.md — the Yandex-first JSON-LD recipe (P0 entity spine → P1 enrichment → P2 gated entity graph) generalized from the samskrte.ru build; follow it for any of our public sites instead of re-deriving the schema/tier/validation/deploy sequence.> Re-verify a specific dependency against its source repo before a large change. > Some linked repos are private (github-spine, Uprava).
Handoff completion and reuse: 4 September audit distinguishes registry closure from verified delivery and lists newly enabled consumers. Exact completed-task remints now pass through the existing mint collision guard, which reads both registry halves; use the existing dependency resolver and in-flight mint journal.
| Asset | What it is | Reuse instead of | |
|---|---|---|---|
H7975 tools/dead_lane_sweeper.py + tools/dead_lane_wt.py (11-11-2026) | Ежесуточный мёртвец-свипер: три детектора молча умерших линий — agent-authored PR гasyoun (red-ci / BEHIND>24ч / stale>24ч), expired/ownerless Active Claims + протухшие .claim-сайдкары (только чтение), и ворктри merged+clean → автоудаление (дисковое правило MG 10-11-2026: git worktree remove без --force + branch -d) с 14 гвардами (worktree_claim_guard H6328 · hid-в-имени FINDINGS §637 · правка<24ч · desktop-live · cwd-live H5883 · возраст<30м · не-disposable ignored H2673 · PII-скан свёрнутых каталогов H4699 · locked · главное дерево · своё); протухшие незанятые → salvage-отчёт. Один датированный дайджест на день (derive-don't-store, идемпотентен) + GTD-строка gtd_mint_row (стабильный текст на дату); лейн schtasks UpravaDeadLaneSweep. Тесты 24/24 fixture-based: удаление доказано только на фикстурных ворктри. | Писать новый свипер мёртвых PR/клейм/ворктри; ручной обход git worktree list перед чисткой; отдельный скрипт удаления merged-ворктри без гвардов | |
H7901 v02/makotemplates/web/og_meta.php (09-11-2026) | Autogenerated Open Graph include for every Cologne web surface — composes og:title/og:description/og:image/twitter:* from data only: per-work registry rows (name/author/year from csl-atlas dictionary_inventory.csv, entry counts from headwords.json L_count, dictparms.py name fallbacks), site-level constants, and ONE committed 1200x630 card og/site_card.png (csl-apidev, deterministic make_site_card.py). Consumers: every generated web/ tree (inventory.txt row *:web/og_meta.php:C), csl-apidev listview/simple-search/servepdf shells, the csl-homepage generator (Python twin block), and vendored byte-identical copies in csl-westergaard/csl-whitroot/csl-kale/csl-inflect. Contract: edit the canonical file, re-copy, open both PRs — csl-apidev twin-drift.yml guards its pair byte-identical (fifth pair); the four vendored copies are NOT CI-guarded. Surface matrix + provenance: csl-apidev WEB_BACKEND_MANUAL §14a. | Hand-writing og meta or description prose per page; giving pages their own image URLs; patching any og_meta.php copy alone | |
H7908 scripts/check-og-cards.mjs (09-11-2026) | OG-card pixel-gate pattern — self-contained stdlib-zlib PNG codec + committed-golden per-pixel comparator + negative-control selftest, a template for any image-rendering pipeline gate (csl-guides npm run check:og-cards): frozen card input (committed literals, never live data — a catalogue change never churns the golden), golden PNG + provenance README under scripts/fixtures/og-cards/, strict tolerance 0 justified by path-based deterministic rendering (satori emits glyph paths; the choice + its measurement are stated in the script header), fail-closed on dimension mismatch / missing golden / any delta (never SKIP-as-pass), and --selftest re-runs the gate as a child process against injected regressions (1-pixel golden tamper, the v3 tofu-mark reproduction) so a gate that silently stops comparing pixels fails its own controls. Cross-platform receipt: macOS-generated golden byte-identical on the ubuntu-latest CI toolchain. | Rolling a PNG decoder ad hoc without fail-closed shapes; comparing PNG bytes instead of decoded pixels; a golden over live/derived inputs (churns on every data change); a gate with no negative controls ("green because broken") | |
| Project map validator (H7470) | Repository-root CLI for navigation links, code paths, placeholders and required sections; usage, broken-path fixtures and Systema/Uprava pilot evidence | Rebuilding a repository-specific project-map path checker | |
H7658 tools/zcode_cli_concurrency_baseline.py (01-11-2026) | Per-plan in-flight concurrency + retry baseline from ZCode CLI JSONL logs, both schema variants (3.16 MSI model.request.*, 3.17 Air ttft.model.*): wall-clock-merge pairing (each close pops the oldest open — never negative, never drops; last close decides outcome so retry-then-success counts as completed; duplicate summary closes / cross-layer events never mint phantom requests), verified-integer status counts (string decoys ignored; nested alt-key ints count per H7656 semantics), TTFT p50/p90, three distinct in-flight levels (request / attempt / query — the H7658 "don't confuse faster requests with throughput" contract), unclosed requests excluded from series (no TTL phantoms), --merge cross-box minute series (upper-bound semantics stated), 28-check offline-replay self-test incl. reviewer-round-1 negatives. Privacy contract as H7656 (plan ids allowlisted; session ids counted, never printed; paths basename-only). Baseline + trial protocol: reports/H7658_CROSS_MACHINE_CONCURRENCY_TRIAL_2026-11-01.md. | Grepping "429" in JSONL; hand-deriving concurrency from event counts; treating session/turn counts as request in-flight; scoring retry-recovered requests as failures; rebuilding the H7656 aggregate patterns instead of extending them | |
csl-observatory/scripts/_common.py (H7360, 28-10-2026) | Shared subprocess contract for the observatory tooling: bounded run, GitHub API retries, fast no-sleep gh_cli_retry, and rate-limit-aware gh_api_harvest; 12 former private wrappers are consumers, with three genuinely different local exceptions documented in the module docstring. | Adding another private gh() / git() / run() retry wrapper; re-deriving pull_data's capped rate-limit wait and exponential 5xx backoff | |
tools/task_claim.py (19-10-2026) | Task-claim mutex — one atomic claim per TASK, shared by every window and harness on a box. O_EXCL take (no TOCTOU), box-local ledger <real home>/.claude/task-claims/, liveness = owner pid alive AND heartbeat fresh (the opencode server is ONE process hosting many windows, so the pid is the coarse signal and the heartbeat the per-window one), dead owner reapable, corrupt payload fails OPEN to 'held' (a doubted lock is refused, never silently reaped), release is owner-only (pid/sid fence) unless --force, lease classes (scope key 1 h, declared task 3 h), append-only _audit.jsonl. check = the ONE-SHOT allow/refuse decision an enforcement hook calls, so the decision logic lives in the tool and the hook cannot drift from it (the H932 lesson). Report + evidence: reports/TASK_CLAIM_MUTEX_2026-10-19.md. Consumers: the opencode-config plugin plugin/task-claim-gate.ts (PR #42/#43) and any harness that shells the CLI. | Writing a second O_EXCL claim lock per repo — the primitive is deceptively simple and two copies drift apart exactly the H932 way; rolling a per-window 'is this mine?' check by hand instead of an atomic ledger take | |
magenta-duck-vps/tools/appstore_storefront_probe.py (17-10-2026, GTD 0X2, v0.1.1) | App Store storefront availability probe: iTunes lookup by bundleId across 6 country stores × 8 VPN/proxy clients, --json, --selftest (3 no-network checks). Answers «does the RU (or any) storefront carry this app» in seconds instead of a browser session. Measured 17-10-2026: RU drops Tailscale/Hiddify/AmneziaVPN/Streisand/V2Box/Happ, carries AmneziaWG + WireGuard + VLESS clients (Karing, sing-box MT, OneXray, v2raytun, Npv Tunnel). Two traps: resultCount 0 means «not offered in this storefront», never «app unpublished»; and a brand is not a function — sweep by keyword too (vless, sing-box, wireguard), because the RU store carries layer-3 clients under brands nobody probed (FINDINGS §1603) | Concluding «app X is not on iOS» from one account's store; guessing storefront availability from memory; assuming the same brand list covers every country | |
| H6889 netpath probe, rotator, geo-split owner | Reuse the existing locked JSONL/notification probe, inode-validated copytruncate rotator and exact-table route ownership machinery; consumer contract and limits: verifier report and acceptance record. Periodic rotation is a threshold, not a hard cap; concurrent copytruncate writes can be lost; route mutations require ownership and crash recovery. | Building another network monitor, unsafe pathname-based rotator, or prefix-only route remover | |
| H7552 connectivity classifier + six-scenario fixtures | Reuse the gap/state-aware incident classifier before writing any streak-counter over probe logs: segments at heartbeat gaps and route/tunnel/DNS state changes, excludes planned A/B transitions and baseline-blocked targets, tunnels-absent means UNRECORDED (not inactive), verdicts name a layer or an ambiguity — never ISP/WiFi, output redacts public IPv4 to x.x.x.<last octet>. Contract + limits: acceptance record. Gap threshold derives from each stream's own cadence (config gap_multiplier/gap_min_seconds); IPv6 not auto-redacted; live deployment is propose-only under the H6889 owner-consent gate. | Writing another raw "N consecutive failed rows" joiner over connectivity CSV/JSONL, or attributing incidents to the ISP/WiFi from gateway-alive evidence | |
tools/windows_gitbash_env_drill.py (H5546, 02-10-2026) | Windows Git-Bash reliability drill for scheduled/CI jobs: env pin (which bash, PATH order, HOME, locale, git config per resolvable bash candidate), representative POSIX payload (set -euo pipefail + heredoc + head -c + C:/ read + git check) through both recorded invocation lanes (review-bot Actions argv shape, class_verifier bash -lc shape), PASS/FAIL evidence table where unpinned rows must REPRODUCE the PR #3766/#3870 failure modes and pinned/explicit rows must PASS; --selftest 18 logic checks, --report evidence file, stdlib-only, Py≥3.9. Invariant doc: docs/WINDOWS_GITBASH_DRILL_H5546.md | Wiring another scheduled Windows job that resolves bash by bare name (dies on System32 WSL launcher or runs the payload in the wrong universe); re-diagnosing the same defect family per incident | |
scripts/karta/gen_karta_index.py + karta_data.py (H5527, 27-09-2026) | Annotated domain-map generator for any Pages repo: disk scan + hand-written annotations → one static index page; coverage guard (unmapped disk content fails --emit/--check with the named list — already caught the diagrams/ and post-merge mindmaps/ waves), --emit/--check byte parity, --probe-islands live HTTP audit of every mapped link; unique stable K### IDs (never reused), status badges. Live output: gasyoun.github.io/karta | Hand-writing another index page that silently rots after the next wave lands; re-deriving the H3768 infographics pattern this generalizes beyond a single dir | |
tools/ocr_vlm/ (direct MG unit, 24-09-2026) | VLM OCR scanned books (RU + Sanskrit latin) → Markdown with italics: PyMuPDF 300-dpi render → qwen3-vl-plus (DashScope, DASHSCOPE_API_KEY from env) with the italics-first prompt v3 (negative rules: quotations are roman, IAST ending anchors Śūnyatā→-tā, [^N] footnotes, hyphen-merge), 4 parallel workers ≈ 4 min / 53 pages; assembler fixes [^N]: defs, page markers, prompt-echo on image-only pages. Measured on Stch.pdf (Shcherbatskoi 1988, 53 pp): 232 IAST diacritics + 377 italic spans vs gemini-2.5-flash 166/393; Tesseract rus/rus+eng and the embedded Acrobat layer ship zero diacritics (tess rus maps Latin to Cyrillic lookalikes); two-pass verify measured and REJECTED (fixes IAST, wipes body italics). 6 live-caught traps in the README; benchmark doc: Uprava_evidence/Школа 2026/stch-ocr-benchmark-sanskrit-markdown-24-09-2026.md; recipe RECIPES §25 | Re-running the model race and prompt tuning per scan; using Tesseract on mixed Cyrillic+IAST scans (Latin→Cyrillic lookalike garbling); trusting the embedded Acrobat OCR layer (0 diacritics book-wide); repeating the rejected two-pass verify | |
RussianTranslation/src/nkrya_client.py + nkrya_evidence_card.py (H5261, 23-09-2026) | The estate's only NKRYa (ruscorpora.ru) query tool, built on the official API: word sketch (top-10 collocates per relation, with dice), ipm plus the 1–6 category, and a lex-gramm pair concordance with a 19th-century slice. The Bearer token comes from the keychain ruscorpora-api or keyring, and the client fails closed without one. Responses are cached on disk by request hash, at 1 request per second, and the offline selftest uses doc-shaped fixtures. The card is advisory: a human votes. Command: /nkrya. Note: the sketch returns only the top 10 collocates, so an absence means "not in top 10", not "unattested"; participles are lemmatized to their verb. | Scraping ruscorpora.ru or using kunansy/rnc / kmike/ruscorpora-tools (ruled out 22-09-2026); choosing Russian collocations by intuition | |
tools/stepik_publish_h4817.py (H4817, 21-09-2026) | Stepik course-publisher with the hard-won platform contract: OAuth password/authorization_code/refresh grants (Basic auth per official StepicOrg/Stepik-API examples), token persisted to ~/.secrets/stepik.env; captcha write-gate contract — writes need a reCAPTCHA token from the page's own widget (lesson/step-source/unit → captcha inside the object, sections → none); /api/step-sources (NOT /api/steps, read-only) with block.source as dict and every option carrying is_correct; publication via the /course/<id>/publication wizard (checklist → copyright → promo → publish). Full dead-ends map: research/STEPIK_H4817_PUBLISH_AUTOMATION_FINDINGS_21-09-2026.md | Re-deriving Stepik OAuth per session (invalid_client on body-creds, grant-type traps); hitting the captcha wall blind (9 documented dead ends); assuming /api/steps accepts POST (405) | |
tools/sanskritism_guard.py + data/sanskritism_fixes.tsv (19-09-2026) | Deterministic check+fix of mangled Sanskritisms in RU text: explicit mishearing-stem table («Ромаины»→«Рамаяны», endings preserved), dry-run default, --check gate (exit 1), --fix requires --backup-dir outside the tree, --suggest fuzzy ≤1 vs glossary headwords (report-only, human adds confirmed rows); 152-FZ contract — reports carry paths/line numbers/term pairs/counts, never content; selftest 12/12 incl. «Роман Л. Толстого» negative. Runbook: docs/SANSKRITISM_GUARD_RUNBOOK_RU_2026-09-19.md | Hand-grepping each mishearing per course; letting n8n «таймкоды» LLM output (supadata→OpenRouter→Rutube/Drive/TG) publish «Ромаины» unchecked — first live scan: 75 mangles in 26 transcript files | |
| Sentinel-breaker guardian classes + host install (H4691, 14-09-2026) | --class kill·delete·service_restart·self_kill·drain_restart presets. The first two are 2/4/12h, service_restart 3/6/6h, self_kill 3/8/2h, drain_restart 4/12/6h; explicit limits win. SENTINEL_BREAKER_ALARM_CMD is the Linux scream hook (<cmd> <label> <msg>): .92 → hermes/lanes/sentinel_breaker_page.sh (page_or_queue P1), .91 → backup_alert.py, Systema app → php artisan guards:breaker-alarm. Installed at /usr/local/lib/sentinel-breaker/ with state in /var/lib/sentinel-breaker/ on .91 and .92. Reference ports: gateway_watchdog.sh (fail-open restart) and restic-stale-locks-clean.sh (fail-closed delete). | Hand-rolled restart counters (_restart_budget_left-style jsonl scans); an osascript-only alarm on a Linux host, which is silent; restarting a dead-on-arrival unit every tick forever | |
tools/worktree_extract_remove.py + tests (H5455, 24-09-2026) | One-pass extraction-then-removal for worktrees the salvage gates refused (H4699 PII-denylist / secret names), executing 0E7 «извлекать всегда»: re-runs the H5441 liveness gates (H3953 heartbeat, <24 h mtime, live claim, open PR), extracts committed+pending content as --binary patches with the denied pathspec (:(exclude,literal), the dirty_tree_rescue form) into worktree_attic extra_blobs custody (write→re-read verify), re-greps every patch path against pii_denylist + the tree's own denied path (header-only — content-level matching false-positives on findings files about the incident), then worktree remove --force + prune. Dry-run default, SE-lint clean, ledger JSONL, --only targeting; measured 24-09: 22 of 23 trees removed under custody (1 SKIP-LIVE), 0 escapes, C: +5.45 GiB. Report | Re-inventing per-tree extraction scripts; treating an H4699-refused worktree as disposable (PII commits stay on the kept local branches — branch deletion is a separate pass); grepping raw patch bytes for denied paths | |
tools/worktree_root.py + tools/worktree_orphan_reap.py (H5437, 24-09-2026) | Where a worktree lives, and what to do with one git can no longer open. worktree_root.worktree_path(repo, name) = ~/Documents/GitHub-worktrees/<repo>/<name> (env WORKTREE_ROOT); estate_github_root(p) follows a tree's .git pointer back to GitHub/ — use these instead of Path(__file__).parent^3 or <clone>.parent / name, which break once a tree sits outside GitHub/. worktree_orphan_reap.py (dry-run default, --apply, --owner <parked-clone>) extracts a dead-gitdir tree to the attic through the donor clone, then removes it; never pushes. Sibling: worktree_extract_remove.py (row above) for PII-refused trees that git CAN open. | hand-built GitHub/<repo>-<slug> sibling paths; rm -rf or worktree remove --force on a tree whose gitdir is gone (content lost unextracted) | |
tools/orphan_adjudicate.py (H5705, 03-10-2026) | One-time agent adjudication of worktree_gc ORPHAN stock (MG-ruled). Rubric verdicts (R-RECOVERABLE/R-REDUNDANT/R-JUNK/R-UNIQUE/R-KEEP) over a fresh census; bulk gh PR join (--limit 5000 — this gh rejects --paginate); live ls-remote evidence (H1181 unknown≠absent); the ONLY sanctioned git branch -D lane (PR-MERGED headRefOid==tip, ledgered, re-verified at apply+audit); breaker-gated apply on the shared worktree_gc sentinel gate; offline/live audit replay + quarantine hash re-verify. FINDINGS §1562. | hand-adjudicating orphan branches (the squash-survivor class never answers merge-base); per-branch gh pr view walks; any -D without a ledgered PR-MERGED head triple | |
tools/methodolog_weekly_report.py (H4757, 15-09-2026) | Weekly methodolog metrics report (E005 wave 1, charter §3): read-only stdlib over the EXISTING prod-DB dumps (data/custdev_db_slices_*.json + samskrte_payments/attendance/users_funnel_*.csv, no new adapters) → dated aggregate-only MD in reports/METHODOLOG_WEEKLY/ with WoW deltas (week vs prev-7d + vs previous report's metrics-json block), drop-off/trajectory section (<50%-coverage courses, pooled retention curve), rule-based GTD card drafts, honest NO INPUT sections per missing feed; built-in PII self-scan against the identity columns of its own inputs (2002 identities on the first live run, 0 hits, report deleted+exit 1 on leak); --selftest = full offline pipeline on synthetic fixtures incl. detector-negative control and NO-INPUT mode | Re-deriving the custdev/samskrte feed schemas per session; hand-building a weekly money/activity свод; skill-обвязка цикла /methodolog-weekly | |
tools/dualrun_postmortem_diff.py (H4633, 13-09-2026) | Automated post-mortem diff for rival dual-run lanes: two artifacts (drain-ledger JSONL or close-report .md) in → one structured diff table out (commit sets COMMON/NET-NEW per lane, per-unit verdict rows EQUIVALENT/CONFLICTING/A-only/B-only, superseded list, keep-best RECOMMENDATION SKELETON — every conflict resolves to «(human ruling)», the tool never picks); --selftest on committed fixtures (positive divergent → non-empty, negative identical → EMPTY, 0 net-new invented), stdlib-only, Py≥3.9. Smoke: reports/DUALRUN_POSTMORTEM_DIFF_SELFTEST_13-09-2026.md | Hand-building another adjudication table per dual-run salvage (H4303/H3919 shape); eyeballing two lane ledgers side-by-side in chat | docs/DUAL_RUN_SALVAGE_H4303_07-09-2026.md |
tools/omniroute_night_lane.py (H4559, 13-09-2026) | OmniRoute night-test-lane driver: daemon UP/DOWN probe, --ensure-up wake of the no-autostart daemon, free-stack smoke (POST /v1/chat/completions), named-outcome classifier (STREAM_OK/DAEMON_DOWN/ALL_ACCOUNTS_INACTIVE/AUTH_ERROR/SMOKE_ERROR) with a fixed exit-code contract, dated JSONL log to ~/.omniroute/night_lane.log; --selftest 10 offline positive+negative controls, stdlib-only, Py≥3.9. Lane doc + reconnect runbook + schtasks/launchd wiring: docs/OMNIROUTE_NIGHT_TEST_LANE_2026-09-13.md | Re-deriving the RECIPES §25 curl ladder per incident; hand-probing "is the free-stack alive tonight"; another silent cron that dies quietly | |
Systema-Sanscriticum scripts/n8n_tail_repair_builder.py (12-09-2026) | Build a repair workflow from a FAILED n8n execution: finds the failed node, replays upstream outputs as same-named stub Code nodes (so tail $('X') expressions resolve), copies the tail verbatim, enables retryOnFail on the failed/LLM nodes, writes an importable JSON with an explicit id | Re-implementing runData flatten-decoding + graph surgery per incident; a full re-run that duplicates un-deduplicated uploads (runbook §7) | RUNBOOK §9 |
tools/class_verifier_launch.py (H4555, chain H5125 18-09) | Auto-launch the cross-family verifier for a classed close (MG pairing ruling 11-09-2026: OxAlpha↔DeepSeek): deduces the executor family from the registry row / .claim sidecar, refuses loudly on ambiguity/Claude-tier/same-family, builds the H4430-compliant verification brief and fires opencode run -m <model> detached (dated log + done-signal). H5125 chain (MG 18-09): three-tier route cheapest-first — LOCAL ollama nights ($0) → FREE OpenRouter models on 7-key rotation (~/.openrouter_keys.env, graceful KEY_1-only via the auth-store key, values never logged) → PAID OMS_DEEPSEEK_MODEL reserve; money/security classes jump straight to PAID; --chain-probe = $0 tier-order evidence; degenerate scaffold-loop guard + frozen-log stall guard advance the chain instead of waiting (both observed live); --selftest 18/18, --unclassed-ok for H4235 triples. Wired into the /go close-out (claude-config e7f13c01/7c0305ad) | Hand-assembling a verifier brief or manually opening a model-switched window per classed close; re-deriving the mirror map per session; paying DeepSeek for mechanical night verifications a free/local tier can carry | |
tools/class_verifier_sweep.py (H4556, 13-09-2026) | Safety-net sweeper over the digest's «Ждёт верификатора» population (H4535 RCA: the rule landed after the row — nothing swept it): IMPORTS agent_daily_digest.verifier_pending() as the selector and class_verifier_launch's resolver/launcher (never forked), throttles per candidate (fired-before log glob = no-double-launch, claim-sidecar recency, git-truth mint age, --max N def 1, resolver refusal = OxAlpha↔DeepSeek lanes only), appends a dated census+launch report under tools/reports/ with an EFFORT_LOG-style line per launch; --dry-run default (report is the only write), --launch --max N real, --selftest 10/10 incl. the H4535/H4536 archived-verified skip-proof | Re-scanning «Ждёт верификатора» rows by hand each session and re-deriving the pairing/throttle logic; double-launching a window for a row that already fired | |
tools/vote_sheet_census.py (H4061) | Read-only stdlib census of the vote funnel: sheets/week from gasyoun.github.io git history + returned decisions.json (watcher state + canonical */review/*, worktree-deduped by sheet_id) → weekly table, verdict distribution, completion-vs-sheet-size buckets, note-richness; `--emit json | md` | Hand-counting sheets/returns in ad-hoc shell one-offs per session — or writing another funnel metric from raw git log |
tools/output_caps.py (H3425) | Shared bounded-output helpers (cap_lines/cap_chars_tail/cap_matches/cap_list) — guarantees a single machine-greppable truncation notice, list capped before json.dumps(). Cap table: tools/OUTPUT_CAPS.md | Hand-rolling a [:50] slice + ad-hoc "…truncated" string in a new lister/JSON-summary tool | |
csl-atlas scripts/pd_koshashri_vocable.py + scripts/pd_koshashri_fixture_export.py (H7316, 29-10-2026) | The single KoshaSHRI vocable adapter and deterministic offline fixture exporter: stable entry/sense/citation IDs, 242-text→248-reference preservation, H7296 work lookup, canaries and byte-parity --check. Consumer contract: stage the restricted single-entry response locally, regenerate pd_koshashri_entries.tsv + pd_koshashri_citations.tsv together, run the focused tests; never commit the full response. | Writing a second KoshaSHRI parser or another work registry; collapsing multi-reference citations; guessing unresolved work IDs | |
csl-atlas scripts/pd_locus_fields.py + scripts/lib/pd_locus_parser.py (H7458/C2, 30-10-2026) | Deterministic PD (siglum, locus) materializer over H6949's traversal and C1's continuation inheritance: four synchronized outputs, discourse/page/roman-locus diagnostics, exact reconciliation of every raw marker, pinned source hash and --check. Consumer contract: regenerate all four artifacts together, rerun 14 hand-derived controls, keep unresolved continuations explicit; never rewrite H6949/C1 aggregates. | Building a second PD citation-tail grammar; quoting raw H6949 marker coverage as validated locus coverage; silently dropping unresolved or page-marker residue | |
sanskrit-util | One Python+JS package: IAST ⇄ SLP1 ⇄ Devanāgarī transcode + normalization keys (norm/form_key), parity-tested; v0.3.0 adds the full SLP1-side API. pip: cdsl-sanskrit-util 0.12.0 (19-09-2026, import stays sanskrit_util). Consumer receipt (H8041, 11-11-2026): csl-atlas normalizeLookupQuery exercised end-to-end — vendored src/lib/sanskrit-util.js sha256 B6875AD6…216F == canonical js/index.mjs @ 50a825cd (0.13.0), wrapper 7B267815…5EF5, node --input-type=module exit 0 → rāma→rAma, śiva→Siva, siva→siva, guṇa→guRa, rAma→rAma; registered + available-on-host + exercised — but canonical origin/main already advanced to 669ebc4 (comment-only delta), so byte-match proves current sync only, not future propagation (re-vendor via /cologne-sanskrit-util-sync); full receipt: reports/H8041_REUSE_CONSUMER_RECEIPTS_11-11-2026.md | Re-typing the SLP1 table (62 files) or re-deriving the ś-vs-acute / NFD length-loss traps | |
csl-pywork pipeline | Dict correction + XML-generation toolchain: updateByLine.py, parseheadline.py, diff_to_changes*, make_xml.py, generate_dict.sh | Forking updateByLine.py an 84th time — pull the vendored original | |
| web endpoint template | makotemplates source for getword/servepdf/serveimg PHP endpoints; 40+ dicts regenerate from it | Editing 37 drifted per-dict copies — fix the template, regenerate, hardening propagates | |
kosha/app/render.py | Code-faithful Python port of the PHP SAX display engine (basicdisplay+basicadjust) for MW/PWG/AP90, golden-snapshotted | Re-porting Cologne entry rendering to Python | |
ZettelkastenWiki | Canonical static-site generator for note/FAQ/wiki collections: stdlib core, render hooks, i18n, reusable pytest invariant harness. pip: zettelkastenwiki 0.6.1 (19-09-2026) | Writing site generator #5 (csl-guides stays Docusaurus by design) | |
csl-pyutil | Shared CDSL Python helpers (reviews, crosswalks, sheets). pip: csl-pyutil 0.24.1 (19-09-2026) — was vendored or git+main-pinned before | Vendoring the helpers or git-pinning main | |
ruwritingstyles | Style-review engine (rws CLI + web UI). pip: ruwritingstyles 2.25.3 (19-09-2026); the corpus DB stays local, never published | Checkout-only installs | |
| RU translation kit | Parameterized mw_ru/pwg_ru kit — parallel translate / QA-judge / re-translate / corpus-check stages over one build_src.py | Cloning a third kit — parameterize the existing one by dictionary | |
| CI & hygiene templates | CodeQL, Dependabot (+auto-merge), pre-commit, CoC, CONTRIBUTING, branch protection via the /cologne-* skills | Hand-writing CI per repo — run the skill (PHP repos use Semgrep, not CodeQL) | |
tools/document_check.py (H2797, H4545) | The /document sweep engine: the 26 registration surfaces (8 repo-local + 18 org) as config, grep/judgment modes + kind gating, per-artifact path::kind, collision-proof receipt with provenance (--force to overwrite), --check-doc runbook-parity gate, worktree-aware GH_ROOT; --selftest fixtures | Hand-walking every hub from memory after a ship — or writing a second "is this registered?" lister | |
tools/hub_append_check.py (H8002, 11-11-2026) | The hub-append reflex as code (census D.hub-append-reflex port): given a derived artifact, lists the hub/registry rows it OWES vs which already mention it — FINDINGS / PROJECT_INTERLINKS / REUSE_INDEX / METADOCS_INDEX class — with kind gates + --reason n/a overrides; matcher/kind tables reused from document_check.py (A7), hermetic --selftest, wired into /document + /artifact-propagate default path (claude-config PR #931) | Re-deriving «куда добавить строку» from memory after a ≥10-minute derivation — run this and edit from its OWED appends: line | |
hyperresearch V8 (H4506, certified 11-09-2026, B−) | 16-step tier-adaptive deep-research orchestrator: 14 subagents (fetcher/analyst/critics/synthesizer), SQLite vault with suggested-by provenance + deprecation handling, verbatim-quote citation discipline enforced at 3+ checkpoints. Currently non-executable on this Mac — 0/16 step skills resolve on Claude Code or opencode, CLI not installed, 9/14 subagent files hardcode a Windows-only path; one surviving full-tier run (research/notes/final_report_who-marcis-gasuns-511a3f.md) proves the design works where the CLI is actually present. Full findings: REPORT_hyperresearch-v8-certification_11-09-2026.md | Writing a new deep-research skill/pipeline from scratch — fix this one (2 named gaps) instead |
tools/handoff_trace.py (H4508, 15-09-2026) | Post-hoc handoff-execution trace correlator: joins git log (mint/exec-claim/close/residual commit shapes, all-refs) + the per-box session-ledger.jsonl by handoff id into one lifecycle timeline (--h H####), no drain driver touched. --selftest 8/8 offline; --stale-brief prints the foreman-brief slice (claimed, unclosed, >24h). Stage map + named gaps (gates stage has no durable per-H record; worktree stage is per-box only): docs/OBSERVABILITY_TRACE_MAP_2026-09-15.md | Eyeballing git log + ledger jsonl by hand per "is H#### actually done" question; re-deriving the claim/close commit-message regex per session | |
|---|---|---|---|
Systema FAQ-RAG measurement kit — FaqRagScoreFloor.php + BotFaqRetrievalSmoke.php + FaqEvalSetBuild.php over tests/fixtures/faq_rag_eval.json (H3766, 31-08-2026; census H7162) | The estate's only LIVE student-facing retrieval lane, with its measurement kit: precision-at-threshold score-floor derivation (ruling R3 ≥95 % — honestly reports «недостижимо» where correct/wrong score bands overlap, per-category floors with coverage stated); token-economy smoke proving top-K retrieval keeps the eval-correct section (prompt −64 %, smoke 23/24); eval-set builder from real anonymised support traffic — FINDINGS §635: the eval set measures the CHANNEL (8,219 inbound → 80 labelable), not the retriever. Consumers: student support DM lane category-F live delivery (H3799, release 1.90.42) + cabinet bot_faq_retrieval stage 1 (prod ON, PR Systema#2235: top-3 67→77 %, recall@5 69→83 %, MRR 0.575→0.713). Successor lane: Hermes W1 trio H6181/H6185/H6191 (open, gate deploy:hermes-24-01a10bca-wiring); related pilot: SuflerWeek0Autopilot.php (H5560). Full surface map: retrieval-surfaces census | Re-deriving BM25 tuning + floor calibration + prompt-shrink proof for the next bot lane; trusting a green retrieval eval that never measured coverage; re-discovering the channel-ceiling trap by building another eval set before asking what fraction of traffic is FAQ-answerable at all | |
tools/search_critic.py (H4511, 15-09-2026) | Critic/rerank instrument over tools/recall_index.py: a committed rubric (docs/RUBRIC_H4511_SEARCH_CRITIC_2026.md) labels each recalled hit relevant/partial/noise + one-line why, then computes precision@5 before vs after dropping noise mechanically (grade_query/precision_before_after, --selftest green). Judge is pluggable via a --labels JSON side-file — the mission ran it as a single-model inline judge (this session), not an auto-invoked LLM call (named gap: wiring tools/session_runner.py as an automated judge). Live evidence on 6 real queries: reports/h4511/REPORT.md (mean precision@5 0.4333→0.4805, BM25-only — --strategy hybrid/--vector were found stalling live on this box, a second retrieval-stack defect distinct from the already-known broken context_pack MCP) | Hand-eyeballing recall_index hits for "is this actually useful"; re-deriving a rubric-first grading harness per search-quality question — reuse the rubric + label_hits/precision_before_after API | tools/recall_index.py |
tools/night_lane.py (H4756, 15-09-2026) | Async night lane: JSONL job queue + driver on a free-first price ladder — (a) free local Ollama on the Mac (ollama list probe) → (b) OpenRouter cheap/free → (c) DeepSeek v4-flash off-peak only, price card + peak windows imported from deepseek_bulk.py (§375, never re-invented); kinds class-verifier / test-draft / mech-sweep; a peak-window paid call or --free-only (Mac on battery) holds the row queued rather than running it or dropping it; the spend ledger records $0 free-local rows too, so savings are measurable. Wiring + real probe: docs/NIGHT_LANE_MAC_WIRING.md | Re-deriving the peak-window / price-card policy per script; writing another «run it at night» driver that silently bills peak rates, or one that loses the job it refused |
|---|---|---|
tools/dom_xss_drill_check.py + data/practice/dom-xss-h5106/corpus.json (H5106, 19-09-2026) | DOM XSS source-to-sink review drill distilled from BookIndex PRs #141/#294/#299: 12-case mutation corpus (8 exploitable — data-bearing innerHTML, deny-list URL-sanitizer bypass, boot-catch exception sink, dashboard DOM assembly, seeded helper-indirect canary — + 4 safe near-miss controls), offline stdlib checker with naive (same-line grep) vs taint (structured source→transforms→sink; deny-lists never count as sanitizers) detectors and a 4-run goal loop (--goal-loop: naive 5/8 RED baseline, taint 8/8 GREEN ×3, negative control refuses a mislabeled near-miss with exit 1). Drill + rubric: reports/DOM_XSS_DRILL_H5106_19.09.26.md | Re-inventing a DOM XSS review exercise per security pass; hand-grading reviewer verdicts with no false-positive control — reuse the corpus, the rubric, and the canary pattern (naive detector must provably miss ≥1 case) |
tools/h5108_sensitivity_scan.py + tools/h5108_sensitive_data_drill.py (H5108, 18-09-2026) | Repository-wide sensitive-data scanner (9 classes: working_tree, index_staged, reachable_ref, reflog_only, unreachable_dangling, pseudo_ref, filename_carrier, odb_all_objects, origin_all_objects — full odb walks + by-SHA probes over every repo in a fleet, never ref walks) + the synthetic multi-ref forensic drill (seeds 13 invented canary carriers across branch history, rescue refs, tags, refs/pull/*, reflog-only commits, dangling commits, odb-only objects, fetch-shadow, staged-deleted index, filenames, ORIG_HEAD, second-clone odb; 8-step H4646 cleanup ladder; stop = 3 consecutive full scans, 0 survivors). --selftest ~5 s, synthetic estates only — safe to run, touches no production data. Evidence + coverage matrix: reports/sensitive_data_forensic_drill_2026-09-18.md | Hand-grepping worktrees + git grep HEAD to claim «data removed» — the naive baseline reports CLEAN 0/13 while carriers persist in reflogs, dangling commits, odb and clones (see §776) |
tools/jev_probe.py (H5275, 23-09-2026) | Thin stdlib client for the TypeSafe Jev «System One» API (POST /v1/systemone, {state, model, questions} — NOT an OpenAI chat API; questions = DICT keyed by question-id, live-verified): choice(≤255-map)/noul(0–1, criteria REQUIRED)/score(criteria=REQUIRED 2–10 list) primitives, client-side pre-send validators, --dry-run default key-redacted request print, --run live call with $0.042/1M cost accounting, --selftest 16 offline checks on a live-shaped fixture; question-template registry ad_composite (3 separate questions) + sufler (intent/urgency/reply-confidence); key from ~/.secrets/typesafe.env, never echoed. The SHARED client of the jev pilot batch H5274–H5280 (plan) | Re-deriving the {state,model,questions} shape + score-criteria-is-a-list trap (live 422) per lane; ad-hoc curl with the key in argv; re-testing the 400-messages-shape trap per consumer |
tools/zcode_cli_log_latency_report.py (H7656, 01-11-2026) | Verified-aggregate latency/failure report over ZCode CLI JSONL logs: parsed-JSON walk (not raw-line grep) with range-checked integer statusCode/status, structured reason fields, retry-delay and per-hour TTFT percentiles, gateway/admission/ENOTFOUND counters; exact-allowlist provider redaction (privacy contract: no bodies, tokens, session ids, URLs); 19-check fixture self-test. Born from the Air reconnect diagnosis; read-only over any box's ~/.zcode/cli/log/. | Grep-counting "429" in 100 MB/day JSONL (was 40x overcounted on 01-11); hand-parsing ttft.model.* events per investigation; trusting one healthy point probe as absence-of-fault evidence |
> Download instead of clone: kosha is the org Sanskrit data-hub (06-07-2026) — > the machine-readable manifest > datasets.json > indexes these assets with keying/rights/consumers, and the public tier ships as > data-v* releases (first: > data-v0.1.0, 7 datasets). > Plan: DATA_HUB_ROADMAP.md.
| Asset | What it is | Reuse instead of |
|---|---|---|
| Nīlakaṇṭha licence register | 151 tradition-attested Pāṇini deviations from the *Bhāratabhāvadīpa* over all 24,694 ṭīkā-bearing MBh shlokas; precision 91.5 % hand-checked on all 165 hits, rejects + reasons kept. Join key deviation_term_sa (56 operation names). Combined with the Gītā probe: 178 rows. No pratīka column (measured unrecoverable); agreement_n is 1 by construction, not a quality score | Re-sweeping the vulgate ṭīkā for *ārṣa*/*chāndasa* — and re-scraping sanatana.in blind (FINDINGS §420: the old endpoint 200s with an empty body) |
| corpus_lexicon | 1.09 M word-aligned Sa→Ru rows (per verse-pair, SLP1-keyed, ~190k keys); output JSONL gitignored, builder linked; verified restore (11-11-2026, H8040): gasyoun/pwg-ru-data Git-LFS → 290,543,363 B, sha256 9f3d852f…4072 byte-identical to shadow_assets.jsonl — recipe + receipts in REUSE_MAP §7 | Re-aligning/re-translating — grep deepseek finds nothing, look at the builder; the lexicon is consume-only, NOT rebuildable — restore the mirror, never rebuild |
| SanskritRussian glossary | Ranked Sa→Ru glossary, three layers: surface 190,838 · lemma 40,370 · root 2,021; live search site, 87% token coverage | Deriving another Sa→Ru signal (TM, root-gloss, RU frequencies) |
mw_roots.tsv | MW verbal-root inventory, 2,113 records with explicit verb_type (750 genuineroot + 1363 root), server-regenerated | Re-scanning mw.txt (3 repos did → 750/809/1163) |
mw_etymology.tsv | Headword→root derivation table (Pāṇinian, 10 dicts); distinct from the root inventory | Re-deriving derivations — WhitneyRoots & csl-atlas consume it verbatim |
| VisualDCS DCS ingest | Canonical CoNLL-U → SQLite build + dcs_lemma_summary.json frequency table | Re-parsing CoNLL-U (5 ingest paths). ⚠ UD Tense=Past conflates aorist/perfect |
| DCS akshara/varṇa/ligature frequency | Grapheme-level frequency over the full DCS corpus (4.24M words): akshara 7,347 · varṇa 48 · conjunct 999, corpus-wide + per DCS time-slot, freq- & varṇamālā-ordered CSVs; reproducible script + provenance, legacy .xlsx cross-check 19/20 (feeds Nagari script-teaching order) | Re-counting aksharas/conjuncts from CoNLL-U for teaching or graphemic stats |
| DCS↔CDSL crosswalk | DCS lemma ↔ CDSL headword linkset, 12,946 rows (81.4% linked); reusable LOD linkset | Re-deriving DCS↔CDSL joins |
resources/data/hyphenation_sa/ (H4487) | Sanskrit wordform lists, ~202.5K words each, three transliteration schemes (HK/Continental-cedilla ç=ś/Devanagari — the second is not modern IAST despite the source filename, see README), gzip-landed from yadisk:Sanskrityatina/20_Hyphenation/, own work (MG rights). Not TeX-pattern hyphenation rules — flat wordlists only; no current Systema/karaoke/PWG consumer, see the landing README for the full verdict | Re-fetching or re-typing the same wordlist from yadisk; re-deriving a Sanskrit spellcheck/hyphenation-input corpus from scratch |
| DCS-2026 annotation-layer census | Per-text WordSem/Treebank/IsMantra token counts + only-2026 flag, all 270 CoNLL-U texts; WordSem semantic markup is corpus-wide (219/270), Vedic Treebank (74) + IsMantra (44) are Vedic-selective, and 29/30 only-2026 texts carry zero WordSem | Re-scanning CoNLL-U to learn which texts have semantic/syntax/mantra annotation before filtering |
| union headwords | Cross-dict union headword index, 323k headwords + per-dict provenance | Rebuilding a headword master — SpellCheck tags candidates union=N |
| MW↔Heritage crosswalk | Entry-level MW→Heritage crosswalk: 185,803 MW entries, 25,140 Heritage-covered, 97.6% anchor-resolved | Re-hitting the (bot-walled) INRIA site |
| Zaliznyak grammar index | Compact Zaliznyak-style grammar tokens over all PWG: 98,639 headwords, 342 paradigm tokens (data-v0.4.0 deposit count, H7352 28-10-2026); FAIR data package | Transcribing static grammar tables — paradigms via vidyut |
| kosha frequency layer | SLP1-keyed sidecar, 83,277 rows: whole-corpus counts + per-period vectors + core-vocab coverage (DCS M9) | Re-deriving frequency ordering — LEFT-JOIN it |
PWG <ls> count tables | Every <ls> element in PWG bucketed to its bibliography abbreviation by longest-prefix match: a frozen 2024-09-11 table (ALL 739,056 — the provenance of the scan-index tracker's Citation count) and a current regeneration (ALL 799,500), each with input hashes. ⚠ a work-family rollup, so fold by abbreviation before summing, and divide only by its own snapshot's ALL; the cleaned-*string* counts in sortedcrefs.txt are a different object (MED. 30 vs 12,990) | Re-extracting <ls> counts from pwg.txt, or comparing a count against one from the other family |
| Indische Sprüche | All 7,537 Böhtlingk subhāṣitas as JSONL (public domain) | Re-scraping sayings; for citations reuse existing PWG <ls> links, not a new crosswalk |
| ortho-drift reform maps | 19th-c.→modern spelling maps: de 15,685 / ru 7,709 / fr 254 / en 71 forms (search-normalization asset) | Hand-normalizing archaic gloss text in full-text search |
| do-not-file corpus | 2,297 deliberately non-standard headword spellings across 33 dicts | Re-flagging known-deliberate variants (csl-atlas annotates against it) |
[generated-block: reuse-index-assets — do not hand-edit; regenerated by registry-sync (tools/gen_reuse_index.py)]
_Auto-generated by tools/gen_reuse_index.py from kosha datasets.json + interlinks_edges.tsv. 154 manifest datasets below; edge-registered non-manifest assets follow._
| Asset | Owner | Tier | Release / evidence | Live consumers |
|---|---|---|---|---|
akshara-mt-benchmark-full — akshara.ru MT benchmark FULL corpus (H3597) | kosha (external site, census frozen 27-08-2026) | restricted | not-applicable | 2 |
akshara-mt-benchmark-pilot — akshara.ru MT benchmark pilot corpus (H3455 lane A) | kosha (external site, scraped 24-08-2026) | restricted | not-applicable | 2 |
amarakosa-mw-pwg-sense-crosswalk — Amarakosha varga/synsets x MW/PWG sense-unit inventories (structural split envelope) | kosha | public | not-applicable | 3 |
archive-stopword-sqlite — DCS parallel-passage stop-word run, queryable DB (M9 D2) | VisualDCS | restricted | not-applicable | 0 |
ashtadhyayi-com-data — ashtadhyayi.com site dataset — Aṣṭādhyāyī (sutraani/) + Dhātupāṭha (dhatu/) | data | public | not-applicable | 2 |
bhs-mw-dcs-key2-crosswalk — BHS (Edgerton) × MW/PWG × DCS Buddhist loci key2 crosswalk | SanskritLexicography | public | unreleased | 0 |
bloomfield-rv-citations — Bloomfield 1906 Vedic Concordance — direct Ṛgveda citations (pratīka index) | kosha | public | data-v0.2.0 | 2 |
bloomfield-rv-x-elizarenkova-ru — Bloomfield 1906 RV citations × Elizarenkova Russian RV — citation-to-translation pointer join | kosha | public | unreleased | 2 |
cdsl-headword-ngram-oracle — Pan-CDSL headword character-n-gram oracle (allngramtxt.txt) | CORRECTIONS | public | not-applicable | 3 |
cohort-start-chteniya-pack-freeze — Start chteniya cohort pack freeze (Hitopadesa-0 + subhashita-beginner + sandhi L1-3) | kosha | public | not-applicable | 2 |
corpus-lexicon — Sa→Ru word-aligned corpus lexicon | SanskritLexicography | restricted | not-applicable | 2 |
corpus-sandhi — Corpus sandhi — frequency-ranked rules across 41 texts | kosha | public | data-v0.2.0 | 1 |
correction-events-log — Correction-event log, full trio (all/typed/final) | csl-observatory | public | not-applicable | 3 |
correction-loci — Correction loci — locus-resolved correction records across all csl-corrections change files (v1.0) | csl-corrections | public | not-applicable | 2 |
csl-observatory-a15-gh-ecosystem-inputs — CDSL GitHub maintenance ecosystem: A15 analysis inputs (76-repo snapshot, issue/commit ledgers, 52,498-event correction ledger) | csl-observatory | public | not-applicable | 1 |
cyrillic-proper-noun-slp1 — Cyrillic proper-noun -> SLP1 validated lookup table (H3985, SL GAPS §6) | SanskritLexicography | public | not-applicable | 2 |
dataset-disciplines — Dataset x discipline join — kosha registry through the IndologyScholars meso crosswalk (H4737) | IndologyScholars | public | not-applicable | 2 |
dcs-cdsl-xref — DCS lemma ↔ CDSL headword crosswalk | csl-apidev | public | data-v0.5.0 | 3 |
dcs-compound-dictionary — DCS compound (samāsa) dictionary derivation | VisualDCS | public | not-applicable | 1 |
dcs-full-sqlite — DCS 2026 full corpus database | VisualDCS | restricted | not-applicable | 3 |
dcs-grapheme-frequency — DCS akshara/varṇa/ligature frequency tables | VisualDCS | public | not-applicable | 1 |
dcs-lsc-pilot — Sanskrit lexical semantic change scores (PPMI-cosine, DCS 5-slot chronology) | VisualDCS | public | not-applicable | 0 |
dcs-nominal-class-split — DCS nominal pooled-class split — long-ī and -ant/-at, on an external lexical signal | VisualDCS | public | not-applicable | 1 |
dcs-nominal-class-split-adj — DCS nominal pooled-class split, ADJ half — the -ant surplus G2 could not see | VisualDCS | public | not-applicable | 1 |
dcs-nominal-paradigm-grid — DCS nominal paradigm — case × number grid per declension class | VisualDCS | public | not-applicable | 1 |
dcs-parallel-passages-full — DCS parallel-passage alignments, full-size export (PARA/Polnorazmernye) | VisualDCS | public | not-applicable | 1 |
dcs-reading-pack-hitopadesa-0 — DCS reading pack — Hitopadesa, Prastavika (opening) | kosha | public | data-v0.2.0 | 2 |
dcs-reading-pack-kiratarjuniya-1 — DCS reading pack — Kiratarjuniya 1 (Bharavi) | kosha | public | data-v0.2.0 | 2 |
dcs-reading-pack-nala-1 — DCS reading pack — Nala 1 (Nalopākhyāna, Mahābhārata 3.50) | kosha | public | data-v0.2.0 | 1 |
dcs-reading-pack-nala-2 — DCS reading pack — Nala 2 (Nalopakhyana, Mahabharata 3.51) | kosha | public | data-v0.2.0 | 2 |
dcs-reading-pack-nala-3 — DCS reading pack — Nala 3 (Nalopakhyana, Mahabharata 3.52) | kosha | public | data-v0.2.0 | 2 |
dcs-sintagmatic-appendix6-periods — DCS syntagmatic tables of frequent lexical cores, by historical period (Приложение 6) | VisualDCS | public | not-applicable | 1 |
dcs-sintagmatic-appendix7 — DCS syntagmatic table — all corpus lemmas (Приложение 7) | VisualDCS | public | not-applicable | 2 |
dcs-stem-cooccurrence-full — Sanskrit-stem co-occurrence table (full) | VisualDCS | public | not-applicable | 1 |
dcs-upasarga-prefixus — DCS prefixed-verb decomposition (MG 2014 workbook) — in-form × prefix × dhātu × Palsule № | upasarga-stat (derived-only — xlsx never committed) | public | unreleased | 1 |
dcs-verb-class-prefix-frequency — DCS verb-form frequency by class, with prefixed forms | VisualDCS | public | not-applicable | 2 |
dcs-verb-example-ids — DCS verb-form example-ID lookup table | VisualDCS | intermediate | not-applicable | 1 |
dcs-verb-form-frequency-prelim — DCS verb-form frequency in corpus (preliminary) | VisualDCS | public | not-applicable | 0 |
dcs-verb-roots-by-class — DCS verb roots grouped by class (with attested forms) | VisualDCS | public | not-applicable | 2 |
dcs-verbal-paradigm-attested — DCS verbal paradigm — per-root attested finite + non-finite cells | VisualDCS | public | not-applicable | 1 |
defgen-heritage-second-reference — Definition-generation eval: Heritage (Huet) French glosses as an independent second reference (H2408) | kosha | public | not-applicable | 2 |
dict-corpus-concordance — CDSL headword ↔ DCS corpus concordance (B1, concordance core) | kosha | public | data-v0.2.0 | 2 |
gita-etymology — Bhagavadgita etymology notes | kosha | public | data-v0.2.0 | 1 |
gita-gold-master — Bhagavadgita gold word-by-word analysis (all 18 adhyayas) | kosha | public | data-v0.2.0 | 3 |
gita-gold-vs-dcs-tagging — Gita gold master lemma/POS vs DCS own tagging — divergence map | kosha | public | unreleased | 0 |
gita-inflection-qa — Gita inflection-engine QA — divergence ledger (E1 attested-corpus check) | kosha | public | data-v0.2.0 | 0 |
gita-morphology-gold — Bhagavadgita gold morphology + compound dataset | kosha | public | data-v0.2.0 | 1 |
gita-prose — Bhagavadgita interlinear prose paraphrase (Prose sheet) | kosha | public | unreleased | 2 |
gita-reading-pack — Bhagavadgita reading packs — all 18 adhyayas (gold) | kosha | public | data-v0.2.0 | 1 |
gita-reading-pack-difficulty — Gita reading-pack difficulty -- reduced 3-axis ordering (W2a follow-up) | kosha | public | data-v0.2.0 | 2 |
gita-sandhi — Bhagavadgita sandhi — frequency-ranked rules | kosha | public | data-v0.2.0 | 2 |
grammar-lab-derivation-gaps — Grammar Lab g1 topics x vidyut derivation failures, corpus-frequency weighted (H4795) | kosha | public | not-applicable | 1 |
grammar-lab-g1 — Grammar Lab Wave-1 topic graph (Whitney + Zalizniak root alternation / verbal morphology) | SanskritGrammar | public | not-applicable | 2 |
handoff-lifecycle-gold — Handoff lifecycle gold standard (170 adversarially-verified verdicts) | Uprava | public | not-applicable | 2 |
headword-overlap-matrix — Headword pairwise-overlap matrix (15-dict union) | SanskritLexicography | public | not-applicable | 3 |
headword-overlap-matrix-ext — Headword pairwise-overlap matrix, Tamil-fold extension (18 dicts) | kosha | public | not-applicable | 0 |
heritage-forms-crosswalk-extras — Heritage (INRIA) form-level crosswalk extras (gloss, disagreements, forms-only) | SanskritLexicography | restricted | not-applicable | 2 |
heritage-mirror — Sanskrit Heritage (INRIA) local mirror | SanskritLexicography | restricted | not-applicable | 2 |
hitopadesa-text-concordance — Hitopadeśa per-text word concordance pilot — every word of the text feeding the dictionary (Tamilex corpus-dictionary pattern, H4034) | kosha | public | not-applicable | 2 |
indische-sprueche — Böhtlingk Indische Sprüche subhāṣitas | SanskritLexicography | public | not-applicable | 2 |
kochergina-corpus-concordance — Kochergina 1987 (third-party RU-SA) headword ↔ DCS corpus concordance — read-only sibling of the B1 family (H4748) | kosha | public | not-applicable | 2 |
kompozity-names-compound-splits — DCS proper-name compound splits with frequency vectors (Kompozity names.csv) | VisualDCS | public | not-applicable | 3 |
kosha-db — kosha unified lookup database | kosha | restricted | not-applicable | 2 |
kosha-lemma-frequency — DCS lemma frequency sidecar (whole-corpus + per-period) | kosha | public | data-v0.5.0 | 2 |
kosha-mastery-schedule — kosha combined mastery schedule (five drill families, H3742) | kosha | public | unreleased | 1 |
kosha-raw-sqlite — kosha raw per-dict SQLite inputs (mw, pwg, ap90) | kosha | intermediate | not-applicable | 1 |
kosha-sense-frequency — DCS per-sense frequency sidecar (3-layer: WN synset / MW sense / semdom; attested WordSem + estimated MFS) | kosha | public | unreleased | 4 |
kosha-srs-deck-b1-demo — SRS deck — Rung B1 demo (Nala 1 content vocabulary, frequency-ordered) | kosha | public | data-v0.2.0 | 1 |
les-1990-static-dump — LES-1990 («Лингвистический энциклопедический словарь», Ярцева 1990) static HTML dump | les-1990 | public | not-applicable | 2 |
lift-pilot-cdsl — CDSL LIFT pilot export (fourth serialization, H721) | csl-standards | public | not-applicable | 1 |
ligature-canon-union — Two-tier ligature canon: union of four inventories + coverage curve | VisualDCS | public | not-applicable | 3 |
markup-tag-census — Markup-tag frequency census (44 Cologne v02 dicts) | SanskritLexicography | public | not-applicable | 3 |
mbh-vulgate-critical-presence — MBh Nīlakaṇṭha vulgate × BORI critical-edition presence verdicts (H2845) | csl-atlas | public | not-applicable | 2 |
mdf-pilot-cdsl — CDSL MDF pilot export (50-case neutral-model slice, MW-focused) | csl-standards | public | not-applicable | 2 |
morph-signature-freq — DCS morphological-form frequency table (difficulty-scorer signal) | kosha | public | data-v0.2.0 | 1 |
morphology-attestation-audit — Generated↔attested morphology audit (A3, W1b) | kosha | public | data-v0.2.0 | 1 |
morphology-attestation-audit-inflections — Generated↔attested morphology audit over inflections — the non-circular side (A3, H3782) | kosha | public | unreleased | 2 |
morphology-curriculum — Graded morphology curriculum — corpus-attested paradigms, ordered teaching syllabus | kosha | public | data-v0.2.0 | 2 |
morphology-drills — Morphology drills — fill/match practice items, evidence-linked to a DCS corpus locus | kosha | public | data-v0.2.0 | 3 |
morphology-giveback-candidates — csl-inflect give-back candidates — the defensible subset of A¬G (A3, H3782) | kosha | public | unreleased | 1 |
mw-defgen-eval-sample — MW definition-generation eval — frozen 500-headword sample + attestations + scored baseline outputs (H730) | kosha | public | data-v0.2.0 | 0 |
mw-etymology — Headword→root derivation table (Pāṇinian, 10 dicts) | csl-orig | public | data-v0.5.0 | 2 |
mw-heritage-crosswalk — MW ↔ Heritage (INRIA) entry-level crosswalk | SanskritLexicography | public | data-v0.5.0 | 2 |
mw-roots — MW verbal-root inventory | csl-orig | public | data-v0.5.0 | 3 |
mw-sense-dcs-join — MW senses x DCS citations — full-volume join (per-sense-freq axis) | kosha | public | unreleased | 2 |
nws-halle-lemmas — NWS (Nachtragswörterbuch des Sanskrit, Halle) lemma scrape — 167,991 lemmas | pwg-ru-data | restricted | not-applicable | 3 |
panini-derivation-status — vidyut-prakriya derivation harness over the AG bucket (A4, W2a) | kosha | public | data-v0.3.0 | 4 |
paninian-corpus-concordance — Pāṇinian sūtra ↔ DCS corpus concordance (A4, W2b) | kosha | public | data-v0.3.0 | 3 |
paninian-sutra-coverage-map — Aṣṭādhyāyī sūtra-coverage / dark-class map (A4, W3a) | kosha | public | data-v0.3.0 | 3 |
parallel-passage-concordance — Bloomfield-style parallel-passage concordance across the DCS corpus (B3) | kosha | public | data-v0.2.0 | 2 |
pd-dcs-coverage — Poona Dictionary × DCS corpus coverage crosswalk (H1336) | csl-atlas | public | not-applicable | 2 |
pd-koshashri-citations — KoshaSHRI agni fixture — citation-reference layer (H7316) | csl-atlas | restricted | not-applicable | 2 |
pd-koshashri-entries — KoshaSHRI agni fixture — entry layer (H7316) | csl-atlas | restricted | not-applicable | 2 |
pronoun-corrections — Curated pronoun-paradigm corrections (E1/W4 fix) | kosha | public | data-v0.2.0 | 2 |
pw-pwg-kurzfassung-cut — PW × PWG Kurzfassung cut measure — sense-block diff + pwg_ru cut queue (H4805) | kosha | public | unreleased | 2 |
pwg-de-edition-v1 — PWG DE edition graph — OntoLex-Lemon + TEI Lex-0 sidecars | SanskritLexicography | public | not-applicable | 3 |
pwg-ls-additions-215 — PWG [Cologne Addition] pwgbib entries for the 215 true-unknown Andhrabharati ls strings | csl-observatory | public | not-applicable | 3 |
pwg-ls-counts — PWG <ls> citation counts per bibliography abbreviation (work-family rollup) | csl-observatory | public | not-applicable | 4 |
pwg-mw-ap-sense-coverage — MW/AP90 sense units vs the PWG-family store: matched / unalignable / absent-candidate verdicts (wave 4) | SanskritLexicography | public | not-applicable | 2 |
pwg-print-anchors — PWG entry L → printed volume/column (print-scan anchor table, H3457) | csl-orig | public | not-applicable | 2 |
pwg-ru-data-workingset — PWG→RU nonstop-pipeline working set (TM store, layers, manifests, raws, gate logs) | pwg-ru-data | restricted | not-applicable | 3 |
pwg-ru-mdf-export — PWG->RU translation cards as MDF (bilingual \de RU / \ge EN records) | SanskritLexicography | restricted | not-applicable | 2 |
pwg-scan-index-campaign — PWG literary-source scan-indexing campaign registry (2025-2026) | csl-observatory | public | not-applicable | 5 |
pwg-sense-attestation-window — PWG per-sense attestation window (Ceiling C2 phase 1) | SanskritLexicography | public | unreleased | 3 |
pwg-tm-canonical-v1 — PWG German–Russian translation memory (canonical v1 four-format pack) | SanskritLexicography | public | not-applicable | 2 |
pwg-work-dating-table — Curated per-work dating table for PWG's cited works (Ceiling C2 phase 2) | SanskritLexicography | public | unreleased | 2 |
reading-pack-difficulty — Reading-pack difficulty scores + graded ordering (W2a) | kosha | public | data-v0.2.0 | 3 |
reading-pack-metre — Reading-pack per-verse metre annotation (W3a) | kosha | public | data-v0.2.0 | 2 |
roots-frequency-curriculum — Roots frequency + attestation curriculum (W2b) | kosha | public | data-v0.2.0 | 2 |
ru-gloss-layer — Inline Sanskrit→Russian gloss layer (reading packs, W-RU-a) | kosha | public | unreleased | 2 |
russiantranslation-src-raw-ingests — RussianTranslation raw/working translation-source ingests (all *.jsonl except corpus_lexicon) | SanskritLexicography | intermediate | not-applicable | 1 |
sa-ru-glossary — Ranked Sa→Ru glossary (surface/lemma/root layers) | SanskritRussian | restricted | not-applicable | 2 |
samasa-trainer — Samasa (compound) analysis trainer | kosha | public | data-v0.2.0 | 4 |
samudra-corpus-db — SamudraManthanam Sa↔Ru verse-parallel corpus | SamudraManthanam | restricted | not-applicable | 3 |
samudra-offline-packs — SamudraManthanam offline app packs (base + dict) | SamudraManthanam | restricted | not-applicable | 1 |
sandhi-curriculum — Graded sandhi curriculum — ordered teaching syllabus | kosha | public | data-v0.2.0 | 2 |
sandhi-drills — Sandhi drills — join/split/identify practice items | kosha | public | data-v0.2.0 | 3 |
sandhikosh-tsv — SandhiKosh TSV layer — the only public Sanskrit sandhi benchmark (LREC 2018), 13,929 split junctions | SandhiKosh | restricted | not-applicable | 2 |
sanskrit-upasarga-semantics — Sanskrit root x preverb (upasarga) semantics | kosha | public | data-v0.2.0 | 3 |
sanskritgrammar-whitney-dcs-vidyut-crosswalk — SanskritGrammar Whitney-no x DCS lemma x vidyut dhatu crosswalk | SanskritGrammar | public | unreleased | 0 |
sanskrityatina-reverse-index — Sanskrityatina 04_Reverse coding legend — 187,992-headword reverse-dictionary stock taxonomy (H4475) | SanskritLexicography | public | not-applicable | 0 |
sanskrityatina-reverse-vs-union16 — Sanskrityatina reverse-index full stock (266,703 IAST headwords) joined to the 16-dict union headwords — 91.56% coverage, 22,508-form Russian-school-only residue (H4800) | SanskritLexicography | public | not-applicable | 0 |
sanskrityatina-sch-cases — Sanskrityatina 04_Reverse/Experimental SCH research cases — six-file keyed TSV with L/pc page refs (H4533) | SanskritLexicography | public | not-applicable | 0 |
scherzl-government-dcs-adjudication — Scherzl case-government relations × DCS-treebank adjudication | SanskritGrammar | public | not-applicable | 2 |
semdom-amarakosha-crosswalk — semdom <-> Amarakosha crosswalk (SIL semantic domains <-> AK vargas + synsets, ID pairs) | SanskritLexicography | public | not-applicable | 2 |
sense-alignment-pilot — Cross-dictionary aligned senses, PWG · MW · Apte · MD · ŚKDR · VCP (H3744 slice 1 + H3862 slice 2 + H4751 W2 scaling + H4745 MD extension + H5273 root-vs-nominal gate) | kosha | public | not-applicable | 3 |
sense-corpus-concordance — Per-sense corpus attestation (नागदन्त layer, H1455 wave-1) | kosha | public | not-applicable | 2 |
sense-crossdict-pilot — Pilot cross-dictionary sense view (PWG · MW · Apte) — 500 headwords | kosha | public | not-applicable | 1 |
sense-dating — Per-sense first-attestation era buckets (PWG v1, H4019, additive) | kosha | public | not-applicable | 2 |
sense-portraits — Diachronic sense portraits — per-sense frequency x first-era buckets (H4735 join) | kosha | public | not-applicable | 2 |
shiva-sutras-machine — Śiva Sūtras (Māheśvara Sūtras) + pratyāhāra table, machine-readable | kosha | public | not-applicable | 1 |
stopovye-parallel-passages — DCS parallel-passage alignments, 'stop-word' variant export (PARA/Stopovye) | VisualDCS | public | not-applicable | 0 |
subhashita-reader-pack — Beginner subhāṣita reader pack — graded Indische Sprüche band (W-RU-b + gloss.ru) | kosha | public | not-applicable | 3 |
sundarakanda-two-tier-apparatus — Sundarakāṇḍa two-tier commentary apparatus (Leonov LP volume) | CommentaryStrategies | restricted | not-applicable | 2 |
tamil-fold — csl-santam Tamil/Capeller/MW corpus fold (mwd+cap+otl, Pahlavi separate) | csl-santam | public | not-applicable | 1 |
thematic-vocab-drills — Thematic vocabulary drills — recognition/recall items grouped by Amarakosa theme | kosha | public | not-applicable | 2 |
thematic-vocabulary — Thematic vocabulary axis — corpus vocabulary grouped by Amarakosa varga | kosha | public | not-applicable | 2 |
union-headwords — Cross-dictionary union headword index | SanskritLexicography | public | data-v0.4.0 | 3 |
union-headwords-16dicts — Cross-dictionary union headword index, 16 dicts (now-2026 key1+key2) | SanskritLexicography | public | not-applicable | 3 |
upasarga-atlas — Unified upasarga atlas — prefixed-verb decomposition x Gita semantics x verb-class frequencies | kosha | public | unreleased | 0 |
uttarapada-dict-vs-corpus — MW uttarapada index × DCS Kompozity — dictionary productivity vs corpus attestation | VisualDCS | public | not-applicable | 1 |
varga-series-diachrony — Consonant series (varga) by DCS period + Cramér's V | VisualDCS | public | not-applicable | 1 |
vidyut-chandas-meters — vidyut-chandas metre definitions (vendored) | kosha | public | data-v0.2.0 | 1 |
vocab-curriculum — Frequency-graded vocabulary curriculum — ordered teaching syllabus | kosha | public | data-v0.2.0 | 3 |
vocab-drills — Vocabulary drills — recognition/recall practice items | kosha | public | data-v0.2.0 | 2 |
which-dictionary-routing-benchmark — Which-dictionary routing shared-task benchmark (v1.0) | csl-guides | public | not-applicable | 1 |
wisdomlib-l8-cdsl-xwalk — Wisdomlib L8 English definitions × CDSL headwords crosswalk (definition → headword linkset) | SanskritLexicography | public | not-applicable | 1 |
wisdomlib-sanskrit-layers — Layered Sanskrit corpus extracts of the wisdomlib.org Sanskrit portal (full store, L3-L8, 2026) | sanskrit | public | not-applicable | 2 |
zaliznyak-drills — Zaliznyak declension-class drills — classify a headword or spot the odd one out | kosha | public | not-applicable | 2 |
zaliznyak-grammar-index — Zaliznyak-style compact grammar-token index over all PWG | SanskritLexicography | public | data-v0.4.0 | 3 |
zaliznyak-lectures-transcripts — А.А.Зализняк public-lecture transcript census (Sanskrit/Vedic + Slavic linguistics) | SanskritLexicography | public | not-applicable | 0 |
zaliznyak-oddbank-drills — Zaliznyak -ant/-at oddball drills — split-verdict exception bank | VisualDCS | public | not-applicable | 2 |
zaliznyak-token-distribution — Zaliznyak token distribution over 98,639 headwords (H6053) | kosha | public | not-applicable | 1 |
| AMAR/amar.txt lemma source (24 vargas, 5,590 synsets) -> data/ak_lemma_cdsl_crosswalk.tsv (H4746: first lemma-level AK x CDSL crosswalk; AK absent from CDSL per H3862, sibling B10) | AMAR | edge | SanskritLexicography PR #2226; data/AK_LEMMA_CDSL_CROSSWALK_2026.md | 1 |
| canonical CONTRIBUTING standard bootstrapped org-wide | COLOGNE | edge | /cologne-contributing-all skill | 0 |
| maintainer corrections (Jim Funderburk pattern) | CORRECTIONS | edge | PROJECT_INTERLINKS spine section | 1 |
| vote/sheets/commentarystrategies_typed_link_q41.html | CommentaryStrategies | edge | 258 Q41 loci — typed-link grammar verdict feed for the sanskrit-util linkid library (H3341); awaiting vote, see REVIEW_SHEETS_INDEX.md | 0 |
| vote/sheets/nilakantha_licence_46.html | CommentaryStrategies | edge | 46 licence cards — licence-ruling feed for the Нилакантха register (H006 open); awaiting vote, see REVIEW_SHEETS_INDEX.md | 0 |
| DCS-2026 lemma export -> simple-search | DCS | edge | S1 R5 applied (MG chat vote 27-08-2026); PROJECT_INTERLINKS line ~463 (feed live) | 1 |
| upstream DCS corpus data (CoNLL-U) | DCS | edge | PROJECT_INTERLINKS corpus feeds | 1 |
| *pref_all.* consolidated preface editions in ocr-prefaces.mdx | GRA | edge | PROJECT_INTERLINKS untapped-feeds (wired 26-06) | 2 |
| INDOLOGY-L mailing-list archive atlas (62115 msgs 1990-2026; split from IndologyScholars Indology/ subtree) | IndologyArchiveAtlas | edge | PROJECT_INTERLINKS.md IndologyScholars-split (H460) | 0 |
| curation/meso_discipline_crosswalk.csv meso-code to discipline crosswalk (50 confidence-scored mappings) + curation/disciplines.csv spine | IndologyScholars | edge | H3567 ruling F5; read by IndologyScholars pipeline/disciplines.py into person_discipline; csl-atlas + kosha named as intended consumers, neither reads a discipline mapping as of 26-08-2026 | 0 |
| *pref_all.* consolidated preface editions in ocr-prefaces.mdx | MD | edge | PROJECT_INTERLINKS untapped-feeds (wired 26-06) | 2 |
| full MW inflection tables for form→lemma lookup | MWinflect | edge | kosha/scripts/build_inflections.py ingests sibling MWinflect nominals+verbs pysanskritv2/tables/calc_tables.txt (K1 + K2a H181); probe 15-09-2026: kosha.db inflections 6,916,522 cologne_mwinflect rows incl 67,140 verb rows (pre/ipf/ipv/opt x a/m/p); verbs leg unblocked upstream by sanskrit-lexicon/MWinflect#50 merged 08-07-2026 (opened 05-07, py3 lambda-tuple fix); vidyut is the E1 hybrid overlay (14,199 natva-fix + 17 gap-fill), not the base; memo reports/DECISION_MEMO_A18_MWINFLECT_KOSHA_EDGE_15-09-2026.md | 1 |
| FAQ gold-question eval seed (ors_faq/eval/gold_faq.json + escalation_set.json) — kept alive as the Wave B gold-question source when H3767 parked bot.py/answerer.py; Systema tests/fixtures/faq_rag_eval.json (H3766: eval set 20->100 labeled questions) is its grown descendant feeding the live student FAQ-RAG lane | ORS-FAQ | edge | H7162 retrieval-surfaces census (reports/RETRIEVAL_SURFACES_CENSUS_20-10-2026.md): lineage H3767 (PR ORS-FAQ#295 — eval set kept as Wave B gold-question source) -> H3766 (PR Systema#2235 — eval grown to 100, cabinet bot_faq_retrieval stage 1 ON); channel ceiling FINDINGS 635 | 1 |
| MASKED train dialogue snapshot corpora/train/2026-08-22-masked.jsonl (snapshot 22-08-2026, 2776 dialogues) | ORS-FAQ | edge | H3563 producer-side registration (F4, settled by F3); 2776 = distinct dialogue ids in the producer store, probed 28-08-2026 (2763 dialog_*.txt, files are a strict subset); freeze protocol H3527 corpora/README.md - MASKED SNAPSHOT ONLY, no sample quoted; not yet committed upstream | 0 |
| frozen MASKED eval dialogue snapshot corpora/eval/2026-07-05-masked.jsonl (export 05-07-2026, 2621 dialogues) | ORS-FAQ | edge | H3563 producer-side registration (F4, settled by F3); census of record message-intent-classifier corpora/README.md; PII policy Systema docs/ARCHITECTURE_MESSAGE_INTENT_CLASSIFIER_2026.md - MASKED SNAPSHOT ONLY, raw dialog_*.txt never committed anywhere, no sample quoted; probed 28-08-2026: corpora/ upstream still holds README only | 0 |
| published static FAQ site + bot | ORS-FAQ | edge | ORS-FAQ/CLAUDE.md | 1 |
| source dictionary text for the pwg_ru / pwg_en translation kit | PWG | edge | RussianTranslation kit (5-layer merged source) | 1 |
| voice-conversion audio tooling for karaoke recordings | Retrieval-based-Voice-Conversion-WebUI | edge | ROADMAP_ALL_REPOS_2026H2 P1 grouping; no artifact edge yet | 0 |
| private source texts behind the public style presets | RuWritingStyles-corpus | edge | RuWritingStyles-corpus/README.md | 1 |
| Litpam print-readiness lane: deterministic InDesign/ExtendScript index-building toolkit (Litpam-Indexator/tools/print_ready.py + tools/indesign/) with its gate config and versioned artefact store (gate-report.json, coverage, defect ledger, IDML/PDF proofs) for the two-volume Литературные памятники Ramayana edition | RussianRamayana | edge | H3568 ruling F5, amended by ruling F5b (MG, 29-08-2026): CONSUMER NAMED — the printed two-volume Literaturnye Pamyatniki Ramayana edition itself (Grintser / Nauka), the book this lane exists to make printable. That consumer is OFF-GIT and outside the repo graph, so Gate 2's third test (the consumer contains something that reads it) is repo-shaped and does not apply; the row is ruled real, not measured real, and its retirement condition is DISCHARGED — the ledger verdict edges-registered stands and does not revert at the next census. The 28-08-2026 measurement is not retracted and must not be re-opened as a defect: measured 28-08-2026 across every sibling clone (kosha, SanskritLexicography, CommentaryStrategies, BookIndex, SamudraManthanam, github-spine, csl-atlas, IndologyScholars, SanskritKaraoke, AfanasiyNikitin, Systema-Sanscriticum, ORS-FAQ), zero repo reads any Litpam artefact — the only sibling hits are the unrelated English phrase print-readiness in the dictionary sense and a litpamyatniki.ru URL. Producer side is real and active: H2590 (Book II, merged https://github.com/gasyoun/RussianRamayana/pull/98) plus H2589 (Book I) ship gate-report.json / coverage / defect-ledger artefacts under artifacts/print-readiness/. Book II stands at AUTOMATED_PASS / HUMAN_REVIEW_WAITING, so delivery to the consumer is in flight, not hypothetical. Gate 2's speculative budget is NOT hereby freed: ruling F5a kept the IndologyScholars crosswalk producer-side only because this row held the single slot, and F5b does not re-open F5a — whether the slot is now unspent is a separate human call, and until it is made the answer is no second speculative row anywhere. | 0 |
| Klammerdiagramm SVG generator + samasa right-to-left reading method/rulings | SamasaChakram | edge | PROJECT_INTERLINKS.md SamasaChakram (H1010) | 0 |
| canonical JSONL corpus (web/corpus_builder/jsonl/*.jsonl) → vendored census data/analysis/corpus_truth_reconciliation.json (per-work sha256 pins) | SamudraManthanam | edge | SHARED_CODE s24 vendored-asset drift contract (H4075 fix6, 04-09-2026): scripts/corpus_truth_census.py --check fails loud on count AND sha drift (upstream moved — rebuild and re-commit); tamper proof scripts/corpus_truth_census_tamper_selftest.py (1-byte flip → exit 1, hardlink-safe); blocking wiring .github/workflows/ci.yml:137; R4 path audit 02-10-2026: asset_path repointed to the owner-side feeding asset (old paths were CommentaryStrategies-side census tooling; SamudraManthanam feeding asset = the canonical JSONL corpus (274 files live)), live-probed | 1 |
| corpus.db verse-parallel MBh RU translations → MBh cited-verse RU layer for pwg_ru <ls n="MBH."> citations (5,742 citations / 5,200 joined / 4,386 with RU; join via bori_locus — Samudra files measured to follow BORI critical numbering) | SamudraManthanam | edge | H4727; RussianTranslation/tools/h4727_mbh_samudra_ru_join.py; layer RussianTranslation/data/mbh_cited_verse_ru/h4727_mbh_cited_verse_ru.jsonl; report RussianTranslation/reports/H4727_mbh_cited_verse_ru_join_report.md (SLX PR #2220) | 1 |
| corpus_builder JSONL kochergina headwords → kochergina_corpus_concordance (read-only join into the dict_corpus_concordance family: 20,651 asserted links (exact 19,560 + floor 1,091) over 28,584 SLP1 keys, 53.4% attested; relaxed 3,256 quarantined per the B1 golden-sample ruling; RU gloss text NOT shipped — N10 rights fence) | SamudraManthanam | edge | H4748; kosha data/concordance/KOCHERGINA_JOIN_BUILD_REPORT.md; 25-entry stratified sample 25/25 mechanical PASS (seed 20260915) | 1 |
| gate dicts + verse-aligned corpus_builder JSONL → L0/TMX + name glossaries | SamudraManthanam | edge | RussianTranslation/src/build_src.py + build_l0.py | 1 |
| samskrtam.ru /z/ verb-root deep links (static root→id table) | SamudraManthanam | edge | buhler-sanskrit-book PR #12 | 1 |
| KEWA etymology lane — kewa_pwg_crosswalk.tsv (11,418 rows; OCRed KEWA heading index + form→lemma witnesses joined to PWG key1, ceiling C4 modern-IE lane) | SamudraManthanam,SanskritRussian,csl-orig | edge | H3169; PROJECT_INTERLINKS KEWA lane; RussianTranslation/docs/KEWA_INDEX_NORMALIZATION_AND_PWG_JOIN_25-08-2026.md | 1 |
| DCS text->period diachronic map (PERIOD_MAP over pinned dcs_full.sqlite) | SanskritGrammar | edge | PROJECT_INTERLINKS.md SanskritGrammar-text-period-map (H1000) | 0 |
| declension_cell_coverage (lemma_cell_coverage.csv, 57144 lemmas) derived from pinned dcs_full.sqlite | SanskritGrammar | edge | PROJECT_INTERLINKS.md SanskritGrammar-Sangram-G2 (H1048) | 0 |
| source-archive upstream + docx→mdx/rstTable book pipeline | SanskritGrammar | edge | PROJECT_INTERLINKS SanskritGrammar row; SHARED_CODE §15 | 1 |
| vote/sheets/sanskritgrammar_apte_buhler_viza_p4.html | SanskritGrammar | edge | 114 viza loci — visa-decision feed, apply pass rides H1205 precedent (H3805 cut p4); awaiting vote, see REVIEW_SHEETS_INDEX.md | 0 |
| H4800 dual-run: sanskrityatina reverse stock (reverse-index-full.doc, 266,703 coded headwords) + xlsm hyphenation master list (252,134 tagged words) -> 16-dict union headwords form_key joins (91.56% / 81.37% coverage; 22,508 + 46,985 residue; rework 17-10-2026, sanskrit-util 0.12.0 pinned) | SanskritLexicography | edge | H4800: both TSV joins + builders + provenance committed repo-side, SL PR #2277; primary kosha row sanskrityatina-reverse-vs-union16 | 0 |
| HeadwordLists/union/union_headwords.tsv attestation gate | SanskritLexicography | edge | SanskritSpellCheck PR #13 (union_attestation.py) | 1 |
| RussianTranslation translation memory / TM-TMX lanes (build_l0.py; RU+oral registers) | SanskritLexicography | edge | PROJECT_INTERLINKS.md RU-translation-memory (H215) | 0 |
| ak_lemma_cdsl_crosswalk.tsv — Amarakosha lemma x 15-dict CDSL union key crosswalk, 14,036 (varga,eid,lemma) instances, 96.60%% exact SLP1 coverage (H4746); companion of A58 domain-level semdom-AK | SanskritLexicography | edge | SanskritLexicography PR #2226; data/ak_lemma_cdsl_crosswalk.py --selftest (30-lemma sample PASS) | 0 |
| corpus-aligned RU glosses baked into verses/data/*.json | SanskritLexicography | edge | SanskritKaraoke tools/build_glosses.py | 1 |
| corpus_lexicon.jsonl → 3-layer Sa→Ru glossary | SanskritLexicography | edge | SanskritRussian README + RussianTranslation/src pipeline · CONSUMER RECEIPT (H8041, 11-11-2026): src=SanskritLexicography@b0ca6226 — build_tmx.py sha256 9d1fe416…eaf3, build_corpus_lexicon.py ee72fb50…142f, REUSE_MAP.md B8FB29F7…5c1, corpus_gate.py 6F95739A…bfc4; consumer=RussianTranslation/src/build_tmx.py; command=python build_tmx.py build --in <pwg-ru-data>/corpus/corpus_lexicon.jsonl --out <tmp>/h8041_tmx_sample.tmx --sample 50 && python build_tmx.py validate <tmp>/h8041_tmx_sample.tmx; expected=50 <tu> well-formed TMX 1.4b; observed=exit 0 + validate OK -- 50 <tu>, langs=ru+sa-slp1; input=corpus_lexicon.jsonl 290,543,363 B sha256 9f3d852f…4072 == shadow_assets.jsonl registry; state=REGISTERED + AVAILABLE-ON-HOST + EXERCISED (sample-level; full 1.09M-row build NOT run); limitations=consume-only NOT rebuildable, registry path_by_host.windows (C:\Users\user\…) absent on this host — worktree path differs, specialist JSONLs (grin12 grin3 koch kna fri smirnov kow) absent at src/ so corpus_gate lookup rAma exits 0 with explicit evidence_status=evidence_unavailable; full receipt reports/H8041_REUSE_CONSUMER_RECEIPTS_11-11-2026.md | 1 |
| heritage phase-6 witness cross-validation: heritage_phase6_crossval.md + h3171_results.jsonl + classified disagreement TSV (Heritage sktreader vs pinned local DharmaMitra ByT5, over the H1349 adjudicated glossary sample) | SanskritLexicography | edge | H3171; RussianTranslation/gold/heritage_phase6_crossval.md | 1 |
| union_headwords + mw_roots + mw_etymology in data-v0.1.0 release | SanskritLexicography | edge | kosha data/manifest/datasets.json | 1 |
| vote/sheets/bli_gold_b1_500_v2.html | SanskritLexicography | edge | 500 BLI gold pairs — gold-ratification feed for ROADMAP_ACL_LESSONS B1 eval (H2784 lineage); awaiting vote, see REVIEW_SHEETS_INDEX.md | 0 |
| vote/sheets/h215_gold_full_320.html | SanskritLexicography | edge | G6 MQM gold set, 320 cards — gold-voting wave feed (H1404 lineage); awaiting vote, see REVIEW_SHEETS_INDEX.md | 0 |
| crossword clue glosses from lemma_glossary.jsonl (samskrutam-crossword content pipeline, word selection side is the kosha edge) | SanskritRussian | edge | H720; crossword/CROSSWORD_CONTENT_PIPELINE.md | 1 |
| root-layer RU glosses ↔ Whitney/mw_roots crosswalk | SanskritRussian | edge | PROJECT_INTERLINKS SanskritRussian row (potential downstream) | 0 |
| in-corpus e-text typo reports upstream | SanskritSpellCheck | edge | GRETIL_TEXT_TYPOS.md | 1 |
| nochange/do_not_file_suppress.txt (2,297 deliberate non-standard spellings) | SanskritSpellCheck | edge | csl-atlas PR #188 | 1 |
| ortho-drift census → /tools/ortho-drift page (PH5 ORTHO-CLOCK) | SanskritSpellCheck | edge | H1577 + H3340 verification; csl-atlas PR #308 merged | 1 |
| ortho_drift/*_reform_map.tsv (de 15,685 / ru 7,709) for gloss query normalization (1901-th/-iren + 1918-* subset) | SanskritSpellCheck | edge | H3340; SHARED_CODE §24 contract (check:ortho-edge); csl-guides PR #200 merged 23-08-2026 | 1 |
| Sundarakāṇḍa apparatus (sundara_ch*_commentary_to_add.json, apparatus/ sarga reviewer manifests + gate_ledger.json, valmiki_shlokas/kanda_5_sundarakanda, typed_link_sundara_concordance.*) | Sundara-commentaries | edge | H4849 (16-09-2026): 428 files moved byte-identical to Sundara-commentaries (sha256 parity 428/428); CommentaryStrategies keeps a vendored read-only copy (data/SUNDARA_APPARATUS_VENDORED.md, ~70 scripts/tests have hard path dependencies) — CommentaryStrategies PR #247 open (product-repo PR-only, merge pending) | 0 |
| .92 root crontab installs Hermes lane lines (backup_verify 03:07, gateway_watchdog */10, restore_drill monthly) from the managed scripts/server_guards/cron/root.crontab — hand-added lines are wiped by the next server_guards_apply.sh run | Systema-Sanscriticum | edge | commit 9fa65cf7; reports/HERMES_THREELANE_RECOVERY_INCIDENT_11-09-2026.md | 1 |
| Telegram Track-B harvest driver (telegram-harvest:sync) | Systema-Sanscriticum | edge | Systema PR #286 | 1 |
| prod-DB payments/attendance/referrals → custdev CSVs | Systema-Sanscriticum | edge | Uprava tools/pull_samskrte_db.py; creds pending (H252) | 0 |
| prod-DB read-only SELECT aggregate feed → custdev CSVs (never named/PII rows; distinct writer from pull_samskrte_db.py — see custdev_pii_gate.py) | Systema-Sanscriticum | edge | custdev/CUSTDEV_2026.md §3.4/§3.5; PII-mode split enforced by tools/custdev_pii_gate.py — never merge with the named-CSV upsell adapter | 1 |
| vote/sheets/rws_rcsi_uncertain_628.html | Uprava | edge | 628-row uncertain-catalogue adjudication packset (H4116 wave D3) — evidence-gated feed for the RWS catalogue/Zenodo retag line (H3963); awaiting vote, see REVIEW_SHEETS_INDEX.md | 0 |
| DCS morphology feed for atlas mining | VisualDCS | edge | PROJECT_INTERLINKS corpus feeds | 1 |
| DCS-2026 lemmas.csv → wf.txt refresh + dcs_cdsl_xref.tsv | VisualDCS | edge | csl-apidev simple-search/wf1 + dcs_xref | 1 |
| DCS-data-2026/dcs_full.sqlite Buddhist loci (20 texts, 285,109 tokens) → bhs_dcs_mw_key2_crosswalk.tsv dcs_lemma_id/tokens/loci columns (H4744; 3,171 BHS headwords DCS-attested; register classes pm-excluded/bhs-hybrid/buddhist-classical per-row) | VisualDCS | edge | H4744 report of record HeadwordLists/bhs_dcs_mw_key2_crosswalk_report_15-09-2026.md; SL PR #2225; R4 path audit 02-10-2026: asset_path repointed to the owner-side feeding asset (old path was the SL-side crosswalk output; VisualDCS feeding asset = the pinned DCS sqlite (920,883,200 bytes on disk, > the 500 MB decoy guard)), live-probed | 1 |
| DCS_Sintagmatic.csv (Приложение 7) per-lemma collocate profiles — pwg_collocate_layer.tsv sidecar, 37,183/94,074 k1 covered (top-12 iast:freq) | VisualDCS | edge | H4708 — RussianTranslation/src/pwg_collocate_layer.py --selftest 10/10 + --spot-check 10 (10/10) | 1 |
| VedaWeb 2.0 Rig-Veda accented-text feed (non-derived/vedaweb/, 10551 stanzas joined 0-miss) | VisualDCS | edge | PROJECT_INTERLINKS.md VedaWeb-feed (H096) | 0 |
| akshara/varna/conjunct frequency CSVs (Fonetika regen-2026) | VisualDCS | edge | PROJECT_INTERLINKS Fonetika row (downstream named, wiring unconfirmed) | 0 |
| archive.sqlite parallels → data/crosstext/archive_parallels.json | VisualDCS | edge | CommentaryStrategies crosstext_archive_parallels.py | 1 |
| citation registers | VisualDCS | edge | PROJECT_INTERLINKS corpus feeds | 1 |
| data/frequency/lemma_frequency.tsv (83,277 rows, M9 join) | VisualDCS | edge | kosha build_frequency_layer.py | 1 |
| dcs_ppp_verified.tsv (5,181 corpus-attested PPP forms) | VisualDCS | edge | WhitneyRoots scripts/build_decisions_doc.py | 1 |
| dcs_published_figures | VisualDCS | edge | H2298; VisualDCS/dcs_published_figures.json; SanskritGrammar/data/dcs_published_figures.json + check_claims_consistency.py check 3 | 1 |
| fonetika_regen2026_frequency_csvs | VisualDCS | edge | https://github.com/gasyoun/Uprava/blob/main/PROJECT_INTERLINKS.md | 1 |
| form→lemma maps for glossary rollup | VisualDCS | edge | RussianTranslation/src/build_dcs_maps | 1 |
| frequency-first pwg_ru worklist ordering | VisualDCS | edge | RussianTranslation/src/build_pwg_freq_order.py | 1 |
| indische_sprueche.jsonl (7,537 sayings mirror) | VisualDCS | edge | VisualDCS export_subhashita_jsonl.py | 1 |
| paradigm_nominal (visual/paradigm_nominal.json + sanskrit_nominal_dashboard.html, H1472) | VisualDCS | edge | PROJECT_INTERLINKS.md paradigm_nominal (H1472) | 0 |
| subhashita verse feed (10 curated anuṣṭubhs) | VisualDCS | edge | SanskritKaraoke tools/import_subhashita.py | 1 |
| original transcoder donor code | WhitneyRoots | edge | SHARED_CODE §1 | 1 |
| _memory_site org-memory wiki engine | ZettelkastenWiki | edge | Uprava/_memory_site (Wave-6) | 1 |
| static-site generator engine (extracted from ors_faq/publish.py) | ZettelkastenWiki | edge | ZettelkastenWiki/CLAUDE.md (Wave 2 parity gate) | 1 |
| guard_escape_gate.py vendored copy (hooks/guard_escape_gate.py; per-path pin test_guard_escape_gate_paths.py moved to claude-config 16-09-2026, H4937) | claude-config | edge | H3424 trust-slice escape gate — pin now lives beside the canonical module in claude-config | 1 |
| playbook-invocation ledger (playbook-invocations.jsonl, claude/grok/codex hosts) | claude-config | edge | H3399; consumed by tools/skill_effectiveness.py + tools/llm_task_census.py, read by /playbook-canonical-source | 1 |
| published Cologne workflow skills (Codex counterpart of ~/.claude/commands) | cologne-skills | edge | cologne-skills/README.md | 0 |
| data/pd/README.md | csl-atlas | edge | reusable output: data/pd family README: provenance + builder map for the 15 machine-readable files (hand-maintained, no builder); builder none (documentation); H6719 family register | 0 |
| data/pd/compound_share_by_letter.tsv | csl-atlas | edge | reusable output: MW/GRA dash-truth compound share per letter; builder scripts/compound_share.py; H6719 family register | 0 |
| data/pd/dcs_text_inventory.tsv | csl-atlas | edge | reusable output: DCS 276-text inventory behind the PD x DCS crosswalk; builder scripts/build_dcs_inventory.py; H6719 family register | 0 |
| data/pd/density_fingerprint.tsv | csl-atlas | edge | reusable output: chars/spans/tags per-entry density fingerprint; page /tools/dictionary-density; builder scripts/density_fingerprint.py; H6719 family register | 0 |
| data/pd/entry_size_by_position.tsv | csl-atlas | edge | reusable output: entry-body size by position-in-letter; page /tools/letter-anatomy; builder scripts/letter_anatomy.py; H6719 family register | 0 |
| data/pd/entry_size_by_year.tsv | csl-atlas | edge | reusable output: PWG entry size on real publication year (123,366 entries); page /tools/letter-anatomy; builder scripts/entry_size_chronology.py; H6719 family register | 0 |
| data/pd/letter_anatomy_stats.json | csl-atlas | edge | reusable output: per-letter anatomy stats sidecar incl. SKD/VCP decay-refutation verdicts; page /tools/letter-anatomy; builder scripts/letter_anatomy.py; H6719 family register | 0 |
| data/pd/pd_siglum_raw.tsv | csl-atlas | edge | reusable output: raw PD siglum harvest feeding pd_dcs_crosswalk.py; builder scripts/pd_extract_sigla.py; H6719 family register | 0 |
| data/pd/upasarga_counts.tsv | csl-atlas | edge | reusable output: upasarga profile per letter; page /tools/letter-anatomy; builder scripts/letter_anatomy.py; H6719 family register | 0 |
| audit-trail change files for applied corrections | csl-corrections | edge | PROJECT_INTERLINKS spine section | 1 |
| data/derived/correction_loci.tsv (39,540 locus-resolved records) | csl-corrections | edge | H4178 flip 1 (06-09-2026): consumer live in csl-atlas — scripts/build-correction-feed.mjs reads the sibling TSV read-only, sha-pinned (correction_loci.source.json feedCommit); packet rebuilt green against current feed 61,430 records/16 dicts (human 31,377/bulk 30,053, pc-resolved 39,531), commit 42928c6; consumer shipped #294, rebuilt #331 | 1 |
| Devanagari-tag round-trip diffs back to source | csl-devanagari | edge | csl-devanagari/readme.md (to_devanagari.py / to_slp1.py / diff) | 1 |
| csldoc front-matter scan JPGs (IEG) | csl-doc | edge | prefaces_ieg/README.md | 1 |
| front-matter scans OCR'd into GRA/prefaces/ | csl-doc | edge | /cologne-preface-ocr; GRA prefaces | 2 |
| front-matter scans OCR'd into MD/prefaces/ | csl-doc | edge | /cologne-preface-ocr; MD prefaces | 2 |
| nominal_class_split — mw.json/pwg.json entry-id sets decide -at/-ant lexeme pooling, side-car paradigm_nominal_class_split.json (H3984) | csl-json | edge | split_pooled_nominal_classes.py; nominal_g2_reconciliation_split.md (reconciles: true); SanskritLexicography FINDINGS §630 | 1 |
| each newsletter edition doubles as a blog post | csl-newsletter | edge | csl-newsletter/CLAUDE.md | 1 |
| correction_events_release.csv (50,953 events) | csl-observatory | edge | pending Zenodo DOI re-mint | 0 |
| BHS-unique-key2-18188 + MW/PWG-unique-key2 exports → bhs_dcs_mw_key2_crosswalk.tsv (H4744: first BHS dictionary crosswalk, 18,188 rows; MW gap 74.8% / PWG gap 73.2%) | csl-orig | edge | H4744 report of record HeadwordLists/bhs_dcs_mw_key2_crosswalk_report_15-09-2026.md; SL PR #2225; R4 path audit 02-10-2026: asset_path repointed to the owner-side feeding asset (old path was the SL-side crosswalk output; csl-orig feeding asset = the BHS source v02/bhs/bhs.txt), live-probed | 1 |
| CDSL dictionaries packaged as offline Flutter app | csl-orig | edge | csl-app/README.md | 1 |
| MW-derived CDSL headwords keying paradigm tables | csl-orig | edge | csl-inflect/README.md | 1 |
| MW1899 nominal/verb entries | csl-orig | edge | MWinflect/README.md | 1 |
| SLP1 source text for Devanagari round-trip | csl-orig | edge | csl-devanagari/readme.md steps 2-7 | 1 |
| all dict source texts for the cross-dict headword atlas | csl-orig | edge | PROJECT_INTERLINKS hubs section | 1 |
| canonical CDSL source texts (v02/<dict>/<dict>.txt) | csl-orig | edge | org CLAUDE.md CDSL spine | 0 |
| dict body texts for body-grounded triage | csl-orig | edge | PROJECT_INTERLINKS hubs section | 1 |
| dict data for the search API | csl-orig | edge | SHARED_CODE §5 | 1 |
| etymology_stats → src/data/etymology-oracle.json | csl-orig | edge | csl-atlas PR #178 | 1 |
| inm/pui onomasticon validation of Cyrillic proper-noun keys (446 of 534 keys validated against inm/pui <k1> headwords) | csl-orig | edge | H3985 cyrillic_proper_noun_slp1 table + h3985_cyr_slp1_table.py; validation report RussianTranslation/reports/H3985_cyr_slp1_validation.json | 1 |
| mw_etymology.tsv → crosswalk/mw_derivations.json (620 roots) | csl-orig | edge | WhitneyRoots PR #22 | 1 |
| per-dict XML/SQLite for the PHP web frontend | csl-orig | edge | SHARED_CODE §5 | 1 |
| csl_pyutil/review_sheet.py — render_review_sheet(), the canonical HTML review/voting-sheet emitter (v0.22.0) | csl-pyutil | edge | H3575 verification 26-08-2026: real from csl_pyutil import render_review_sheet in Uprava tools/build_glossary_sense_sheet.py + tools/build_h2582_rec_approval_sheet.py, SanskritLexicography article-comparison/_build_gloss_review_sheets.py, csl-atlas scripts/build-review-sheets.py, SamudraManthanam scripts/build_nkrya_adjudication_sheet.py, CommentaryStrategies scripts/build_licence_register_review_sheet.py, WhitneyRoots scripts/dcs/build_review_sheets.py; /review-sheet skill mandates it ("always"); supersedes the ledger note "no local renderer found" | 0 |
| generate_dict.sh + updateByLine.py build/XML-validation pipeline | csl-pywork | edge | org CLAUDE.md correction workflow | 1 |
| tamil SQLite corpus (mwd+cap+otl, 321,620 entries) fold | csl-santam | edge | H4178 flip 2 (06-09-2026): kosha fold landed — scripts/ingest_tamil_fold.py folds the sibling sqlite/tamil.sqlite (325,838 rows; mwd 166,434/cap 37,413/otl 117,773 = 321,620 EXACT edge match, +4,218 cpd Pahlavi), lexicon-tagged fold with fold_meta provenance (source commit e94ca05 + sha256), committed stats data/tamil/fold_stats.json, manifest row tamil-fold (CC BY-NC-SA 3.0) + data statement; kosha PR #532 merged; cp1252→UTF-8 normalization done at fold time (H1513 normalized only the text export) | 1 |
| five-dictionary sqlite feed for the kosha aligned-sense build | csl-sqlite | edge | H3862 registration (csl-sqlite -> kosha widened 3 to 5 dicts: MW PWG AP90 SKD VCP, read directly by the kosha build script, never loaded into kosha.db); previously recorded only inside the kosha row asset text - now canonical; Sanskrit-cluster spine-completion pass 10-10-2026 | 1 |
| lookup QueryEvent aggregates (k≥5-suppressed, human-ui only) from CDSL frontends' query traffic | csl-usage | edge | H7938/H7939 feed registration (csl-usage docs/FACT_PACK_2026-11-09.md + CONTEXT.md + docs/adr/) | 0 |
| wsd_frame_c1 gold frames (wsd_frame_c1_200.tsv — 200 DCS token occurrences × 48 polysemous PWG lemmas with RU sense menus; DCS token+sentence+chapter via dcs_full.sqlite) | dcs-conllu | edge | H3172; RussianTranslation/src/eval/wsd_frame_c1_200.tsv; WSD_GOLD_SET_ANNOTATION_PROTOCOL_2026.md | 1 |
| GPU morphology/lemma runs | ext:DharmaMitra | edge | csl-atlas #89/#92 | 1 |
| corpustei plaintext (6 sections) for meter QA + typo checking | ext:GRETIL | edge | SanskritSpellCheck PR #24 (local-only, CC BY-NC-SA) | 1 |
| darkone23 mirror → mw_heritage_crosswalk.tsv (185,803 MW entries) | ext:Heritage | edge | HeadwordLists/HERITAGE_MIRROR_INVENTORY.md (H099) | 1 |
| restricted single agni vocable response for H7316 fixture | ext:KoshaSHRI | edge | H7316 PR #653 + reports/PD_KOSHASHRI_FIXTURE_EXPORT_2026.md; public single-entry GET staged locally, all rights reserved, full JSON uncommitted, no bulk feed | 1 |
| DCS×Heritage alignment (130,270 sentences) | ext:Samsaadhanii | edge | approved pilot (GTD Agent-Ready); no LICENSE upstream | 0 |
| Dhātupāṭha verb-generator chart (1,591 roots) → crosswalk/mw_roots.json root-parity join — 562/750 = 74.9 % exact SLP1 (38 fold-witnessed only) | ext:Samsaadhanii | edge | H4741 parity report SanskritLexicography data/samsaadhanii_dhatupatha_mwroots_xwalk/ (PRs #2223+##2224); DeepSeek paired-family verifier loop; VALIDATION-ONLY — dhaatupaatha CC BY-SA 3.0 upstream, no row-level redistribution | 1 |
| Sundarakāṇḍa gold kāraka treebank | ext:Samsaadhanii | edge | A5 pilot 1 (H4738): parity PASS 1,742 anvaya sentences / 38,131 rows — relabeled 16-09-2026 per MG ruling «а» (prior «92» was a sentno chapter-bucket artifact, verifier DISAGREE); VALIDATION-ONLY (no LICENSE upstream); CS PR #246 + relabel PR #249 | 1 |
| morph.cgi live analyses → RussianTranslation glossary-layer witness (form→[{stem,tags}] records; H4740 40-form DCS stratified coverage sample: top-20 frequency + 20 random rank 500-50000; VALIDATION-ONLY — no LICENSE upstream, never overwrites human-reviewed data; cached+throttled+UA-identified) | ext:Samsaadhanii | edge | H4740; RussianTranslation/reports/H4740_scl_morph_coverage.json + H4740_scl_morph_sample.jsonl; builder RussianTranslation/src/scl_morph_glossary.py (selftest PASS) | 1 |
| 4-layer accented Rigveda export → non-derived/vedaweb/ | ext:VedaWeb | edge | H096; host down per SERVER_OUTAGES.md | 0 |
| KEWA heading index digitization — samskrtam.ru/sanskrit-lexicon/KEWA/ static index (9,591 anchors; OCR 2005-2013) upstream of Index/lib/x86_64-win64/Data/KEWA.txt; provenance post samskrtam.ru/kewa-ewa/ | ext:samskrtam-ru | edge | provenance pages fetched 27-08-2026; H3169 source index | 1 |
| ramayana-edition-alignment (scan-backed edition alignment feed) | ext:sanskrit-lexicon-scans | edge | PROJECT_INTERLINKS.md ramayana-edition-alignment | 1 |
| Wisdomlib dictionary glosses x csl-orig MW/AP90 provenance measurement — h4807_provenance_table.tsv (708,875 Sanskrit-dictionary gloss blocks / 255,348 definition pages = L8 index exactly; 90.0% Cologne-declared attribution: MW 25.0% / Apte-1890-as-DDSA 10.0% / other Cologne 55.0% / own-other 10.0%; text-verified borrowing >=32.8% of MW/AP-attributed rows: MW 42.0% / AP90 10.0%; D1 123,362 headword-misses + D2 43,347 no-text-match = defect rows for reader-pack gating) | ext:wisdomlib-sanskrit-layers,ext:csl-orig | edge | H4807 (PR https://github.com/gasyoun/SanskritLexicography/pull/2278): report data/h4807_gloss_provenance/H4807_GLOSS_PROVENANCE_REPORT.md; sample50 50/50 revalidated; builder selftest PASS; csl-orig read-only; BY 4.0 x BY-SA 4.0 => BY-SA derivatives | 1 |
| root CLAUDE.md + SHARED_CODE.md cross-repo spine docs | github-spine | edge | github-spine whitelist .gitignore | 0 |
| cohort_start_chteniya | kosha | edge | H2109→H2110; kosha data/cohort_start_chteniya; Systema resources/data/cohort_start_chteniya via scripts/vendor_cohort_start_chteniya_packs.py | 1 |
| data-v* releases: canonical derived-dataset serving tier | kosha | edge | kosha data/manifest/datasets.json (D-HUB-1..8) | 0 |
| data/concordance/sense_alignment.tsv — H4745 six-dict aligned-sense table (PWG·MW·Apte·MD·ŚKDR·VCP): 67,230 groups / 7,087 aligned / 180 aligned touching MD on the W2 1,487-lemma frame; MD Macdonell added on the English gloss channel (md.zip csl-sqlite, 20,749 entries); PWK recorded NOT reachable (no pwk.zip — pwkvn is the Verzeichnis der Namen, separate dict); consumer = Uprava acceptance-sample strata (G5-gated sense programme) | kosha | edge | kosha PR #587 merged (e98910b6f, 16-09-2026): 31 tests pass; TSV cross-counts independently reproduced 67,230/7,087/180; report numbers match TSV | 1 |
| per-lemma static dictionary cards | kosha | edge | kosha PR #557 merged 14-09-2026 (H4833): 10,725 /w/ pages live, Buhler coverage 75%; consumer wiring = H4843 | 1 |
| sense_alignment_acceptance_sample_w2.tsv — H4751 W2 scaled-wave acceptance sample feed (153 cards stratified on the frozen H3910 strata; review sheet not yet built) | kosha | edge | kosha PR #585 merged (bd758f0f); frozen 120-card artifacts byte-stable; two-pass judge kappa 0.9022 reproduced | 0 |
| vote/sheets/kosha_sense_align_acceptance_w2.html | kosha | edge | 120 acceptance cards — sense-reconciliation W2 acceptance feed (H3785/H3862); awaiting vote, see REVIEW_SHEETS_INDEX.md | 0 |
| shunt delegation core — 350-line read block/redirect + cheap-worker shims (bulk_read/code_write/verify/mcp_shunt, fence, ledger) | opencode-config | edge | MG go 11-09-2026; claude-config consumes the core | 3 |
| pwg_ru_nonstop_lane_store | pwg-ru-data | edge | https://github.com/gasyoun/SanskritLexicography/blob/master/RussianTranslation/docs/ARCHITECTURE_RussianTranslation_pwg_nonstop_multilane.md | 1 |
| rv_hymn_pages_ru_geldner_griffith | rvlinks | edge | https://github.com/gasyoun/Uprava/blob/main/PROJECT_INTERLINKS.md | 0 |
| DCS×Heritage Segmenter sentence alignment (parallel_data_iast.tsv; 130,270 sentences, IAST; VALIDATION-ONLY — upstream has no LICENSE file) | samsaadhanii/datasets | edge | H4739 parity report: VisualDCS derived-data/Samsaadhanii-DCS-SH-alignment/ (G1 row parity 130,270==130,270==expected PASS; 12-row sample hand-checked) | 1 |
| Wordstat keyword rounds feed samskrtam.ru SEO (H5127 night-SEO-growth lane: 17-cluster keyword table, GSC 92d + Yandex suggest; WordStat API blocked on OAuth token -> 0R27 @WAITING) | samskrtamRU | edge | H5127 executed 19-09-2026 (Uprava CHANGELOG 570; FRUIT_LAUNCH_STATUS_19.09.26 check-mark): keyword round delivered for the samskrtam.ru/ORS-FAQ SEO programme; ops repo samskrtamRU holds the WORDSTAT_* creds (gitignored .env, chmod 600) | 1 |
| canonical IAST⇄SLP1⇄Devanagari transcoders (Py+JS) | sanskrit-util | edge | SHARED_CODE §1 | 0 |
| data/derived/headword_candidates.jsonl | telegram-sanskrit-corpus | edge | harvest landed: telegram-sanskrit-corpus PR #19 (H2669) — data/derived/headword_candidates.jsonl, 423 rows on disk (verified 06-09-2026); no consumer wired in SanskritLexicography yet | 0 |
[/generated-block]
| Asset | What it is | Rule |
|---|---|---|
csl-orig | Canonical source text of all ~43 dictionaries (v02/<dict>/<dict>.txt) | Read-only for direct pushes — corrections via the monthly batch PR (/cologne-correction-queue) |
csl-corrections | Change-files / audit trail for the correction workflow | Express corrections as change files applied by scripts, not direct edits |
SamudraManthanam | Parallel Sa–Ru corpus tool (samskrtam.ru) + bundled SLP1-keyed dicts (Kochergina/Kossovich/Knauer/Frish/Smirnov/MW/Apte) + verse-aligned texts | Query it — upstream for pwg_ru gate dicts and mw_ru validation; don't duplicate |
| Hub | What it answers | When |
|---|---|---|
SHARED_CODE.md | Which shared concern comes from where (copy-counts, migration queue) | Before any string helper / correction applier / dict-web endpoint |
PROJECT_INTERLINKS.md | Which repo owns a dataset, who feeds whom, external toolchains | Before deriving an alignment / frequency / crosswalk |
FINDINGS.md (Sanskrit data) | Measured Sanskrit-data gotchas (encoding, coverage, crosswalk caveats) | Before re-measuring a data caveat — append yours in the same pass |
FINDINGS.md (infra) | Platform/network/process gotchas (bot-walls, rate limits, Windows/encoding) | Before re-probing a host or re-hitting a limit (+ SERVER_OUTAGES.md) |
csl-observatory | Org-wide metrics + cross-repo decisions master + correction-event dataset (50,953 events) | Before recomputing error-frequency/typology stats |
csl-atlas | Cross-dictionary headword atlas (consumes every dict + DCS/DharmaMitra morphology) | Before building another cross-dict comparison |
SanskritSpellCheck | Body-grounded headword/typo triage (/cologne-dict-triage) | Before writing a spell/variant checker |
REUSE_MAP.md | Prior answers + rejected approaches for Sa→Ru reuse: why Cyrillic-only sources were NOT reverse-mapped, grin12/grin3 = SPECIALIST corroboration (not correctness authorities), corpus restore recipe (§7) | Before proposing any new Russian source / proper-name extraction — check the prior answer first (H8041 receipt: reports/H8041_REUSE_CONSUMER_RECEIPTS_11-11-2026.md) |
CANONICAL_CUSTOMER_LINKS_2026.md (H4248) | Canonical samskrte.ru customer links: payment/cabinet route patterns, 30-sec live tariff-id snippet (read-only SSH), dated season slice — every student-facing advice must carry a real URL (MG 06-09-2026) | Before composing ANY reply to a student/client about payment, cabinet, schedule |
| Retrieval-surfaces census (H7162, 22-10-2026) | EVERY retrieval/answering surface in the estate — 7 rows (house recall MCP · Systema student FAQ-RAG · ORS-FAQ answerer parked · ask-the-wiki vault QA · OpenClaw seat on prod .92 · Systema Sufler autonomy pilot · NotebookLM MCP), each with repo, state, owner handoff, gate, live check — plus what the 20-10-2026 RAG grill covered (house memory-recall only) vs missed | Before scoping any RAG/retrieval/QA work, minting a "second RAG" handoff, or answering «how many answering surfaces do we run» |
| SANSKRIT_ANALYSIS_ALGORITHMS_SURVEY_2026-11-11.md (H8049, 11-11-2026) | Полный инвентарь алгоритмов анализа/разбора санскрита по уровням + reuse-вердикты: ядро (сандхи ✅ · сегментация 🔧 · морфология ✅ · синтаксис 🏗 · семантика 🏗), справочный и мета-слои, 7 спот-ранов, внешний кросс-референс (DharmaMitra · vidyut · SCL · UD-UFAL · Heritage), отложенный стенограммный поток | Before building or reusing ANY Sanskrit parsing/analysis component — читать карту уровней (отчёт §1) и пробелы (§8): синтаксический парсер = build gap №1 |
| Stack | What it is | Reuse rule |
|---|---|---|
| Samsaadhanii / SCL (Amba Kulkarni, UoHyd) | Morph analyzer, sandhi/segmenter, compound processor, Pāṇinian parser, Amarakośa net, Dhātupāṭha concordance — already consumes our Cologne dicts | Call the live JSON APIs / cross-validate; don't clone the GPL source — reuse map in SAMSAADHANII_INDEX.md |
| Sanskrit Heritage (Gérard Huet, INRIA) | Dictionary + morphology + segmenter (LGPLLR): DICO/ hypertext dict, MW-aligned pages, frequency TSVs, OCaml banks | Pull from the GitHub mirror — ⚠ sanskrit.inria.fr and gitlab.inria.fr are Anubis bot-walled |
| DharmaMitra (Berkeley ecosystem) | AI translation stack (MT, Deep Research w/ references, parallel-passage Explore, segmentation, OCR); already a GPU morphology supplier to csl-atlas | Reuse their MT error taxonomy (wired into pwg_ru/mw_ru judges, crosswalked to MQM); prospective Kosh-API consumer |
| VedaWeb (Cologne / UZH, CC BY 4.0) | Accented Rig-Veda + per-word UZH morphology (Casaretto word-split, lemmatization, Lubotsky padapāṭha), C-SALT-linked | The validation set that unblocks the Vedic accent axis (WhitneyRoots, GRA) — bulk-export once, don't re-hit the API per consumer |
| diglossa.org (beta, Anthrax.js morph analyzer) | "Slow reading" multilingual parallel-text reader with morphological word decomposition; Sanskrit capability surveyed 12-07-2026 | One-off capability survey + verdicts: EXTERNAL_TOOL_SURVEY_DIGLOSSA_SANSKRIT.md — read before re-evaluating or adopting |
_Dr. Mārcis Gasūns_
Плейбук трёх источников: стенограммы занятий → Telegram-чаты групп → Тукан. Использовать перед ответами о нагрузке, домашних заданиях, пропусках и реальном ходе учебы; сайт курса — вспомогательный источник. Не заменять учебные группы публичными каналами или агентскими session transcripts.
Проверенные read-only маршруты (H6682, 13-10): история учебных групп — raw harvest store на LMS .92 (JSONL-каталог, без MTProto-сессии и без ChatExport); переписка Тукана/куратора — telegram_support_messages на .92; обезличенные карточки выводов и покрытие — reports/LEARNING_WORKLOAD_SOURCES_DEID_13-10-2026.md (норма самостоятельной работы численно НЕ установлена — доказанное отсутствие).
Generated by Uprava tools/build_public_site.py from an explicit three-hub allowlist under a deny-rail leak guard (E031 wave 4 / H8046, 11-11-2026). Private paths are withheld by construction, never by review after publication.