Documentation index — NoCap
Evidence state: 2026-07-16 · claim-stack v1 · 3/3 under F1_seal.
Live HTML: https://nocap.think2earn.com
Staging tree: this folder (cf-pages-nocap-agent/).
Hosting (read this first)
Jury-ready staging ships index.html, data/*.csv, and companion .md files. Deploy the staging folder to Cloudflare Pages (exclude WORKSPACE_MAP.local.md and EMAIL_SNIPPET.md). Public strangers recompute from https://nocap.think2earn.com/data/ — see data/README.html.
Audience legend
| Tag | Meaning |
|---|---|
| Public | Stranger-safe; workspace codes only |
| Agent | Local operators; may reference real profiles via local map |
| Both | Shared; public language |
Reading orders
Stranger — ~10 minutes
- CAMPAIGN_OVERVIEW.html — campaign path
- FAIR_CLOCK.html — how speed is measured + non-claims
- GLOSSARY.html — term lookup while reading
- HOW_TO_READ_EVIDENCE.html — what to trust when docs disagree
- data/README.html — public CSVs + 5-minute recompute
- Live site https://nocap.think2earn.com — Verify section + scoreboard
Optional depth: CAMPAIGN_TIMELINE.html, FAILURE_LEDGER.html, RECIPE_CARD.html, NUMBERS_CANON.html.
Agent — deep path
The numbered agent sources below live in the local private repo and are not hosted on this site. Public reviewers can ignore them and follow the stranger path above; every published number is recomputable from the hosted docs + public data/*.csv.
nocap-repo/docs/AGENT_HANDOFF.md— paths, spawn rules, deploy, no GPUs unless asked *(local-only, not hosted)*WORKSPACE_MAP.local.md— private profile ↔ code map *(local-only, never deployed)*nocap-repo/results/claim_stack_v1/MULTI_SEED_CLAIM.md— sealed 3/3 table *(local-only, not hosted)*- HOW_TO_READ_EVIDENCE.html + NUMBERS_CANON.html + EVIDENCE_PATHS.html *(hosted)*
- FAILURE_LEDGER.html + ACCURACY_NOTES.html *(hosted)* + seed
*_SUMMARY.md/ CSVs *(local-only)*
Catalog
Public / stranger-facing (this folder)
| Doc | Audience | Purpose |
|---|---|---|
| GLOSSARY.html | Both | Every important term |
| b32-systems-explained.html | Public | B32×accum16 systems pedagogy (throughput packing; no sealed ΔT%) |
| qkv-explained.html | Public | What Query / Key / Value are in attention (before joint vs split) |
| disjoint-val-explained.html | Public | Official val_disjoint gate · bar ≤ 3.3821 · first-passage (not train CE / EMA) |
| f1-seal-explained.html | Public | AdamW F1_seal denom pedagogy (3.866 h first-passage; not Muon) |
| FAIR_CLOCK.html | Public | Speed metric, denom, under-claiming, non-claims |
| CAMPAIGN_OVERVIEW.html | Public | Campaign path for strangers |
| HOW_TO_READ_EVIDENCE.html | Both | Resolution order + CSV recompute |
| CAMPAIGN_TIMELINE.html | Public | Detailed chronological spine |
| FAILURE_LEDGER.html | Public | Negative results / non-claims table |
| RECIPE_CARD.html | Public | Exact claim-stack recipe + matched pairs |
| NUMBERS_CANON.html | Public | Canonical sealed numbers (codes) |
| ACCOUNT_MAP.html | Public | Workspace roles via codes |
| workspace-codes-explained.html | Public | Clock-W / Claim-W* ops aliases (roles, not speed) |
| CAMPAIGN_RUNS_OVERVIEW.html | Public | Short map of campaign volume runs across workspaces |
| EVIDENCE_PATHS.html | Both | Which files to open |
| data/README.html | Public | Public CSV catalog + recompute |
| FACT_CHECK_CHECKLIST.html | Agent/public QA | Claim → source for HTML |
| ACCURACY_NOTES.html | Agent | Residual conflicts / traps |
| README.html | Agent | Staging / promote / privacy |
Agent / local-only (not hosted on this site)
These live in the local private repo / operator machine only. They are never deployed, so links are intentionally omitted — the paths are shown as plain text.
| Doc | Audience | Purpose |
|---|---|---|
EMAIL_SNIPPET.md | Agent / operator | Draft reply only — NOT sent. BottleCapAI application follow-up; do not deploy to public jury unless operator chooses |
WORKSPACE_MAP.local.md | Agent | Real Modal profiles — never deploy |
nocap-repo/docs/AGENT_HANDOFF.md | Agent | Full agent guide |
nocap-repo/docs/ops/FAIR_CLOCK.md | Agent | Original forensic protocol (partially stale vs multi-seed practice) |
nocap-repo/results/claim_stack_v1/* | Agent | Sealed truth |
nocap-repo/results/f1_seal/* | Agent | Denominator seal |
| books/HF_BOOK_COLLECTION.html | Agent | HF AI/ML foundations collection — TOC/idea workflow (research only; not train data) |
Terminology
There is one glossary: GLOSSARY.html. No separate TERMINOLOGY.md.
Concept explainers
Stranger-facing TensorTonic / process primers. No new sealed numbers on these pages — sealed ΔT% stays on the report / NUMBERS_CANON.
| Doc | Audience | Purpose |
|---|---|---|
| b32-systems-explained.html | Public | B32×accum16 systems path under the claim recipe |
| qkv-explained.html | Public | Query / Key / Value primer (stranger-facing; before split packing) |
| split-qkv-explained.html | Public | Timing-cheap split-QKV vs dual-polar tax; flags |
| split-vs-ortho-explained.html | Public | Split ≠ Ortho — NS×5 kernel vs layout; seal needs both |
| f1-seal-explained.html | Public | AdamW F1_seal · sole public ΔT% denom 3.866 h |
| seeds-explained.html | Public | 3/3 multi-seed seal · each seed’s T / ΔT% |
| not-claimed-explained.html | Public | Hedge defense: F1_seal only / matched ΔT% N/A |
| kill-bank-seal-explained.html | Public | Kill → bank → 3/3 (seal) process jargon |
| workspace-codes-explained.html | Public | Clock-W / Claim-W* · ops aliases, not performance claims |
| TRAINER_DELTA.html · kit/ | Public | Sealed cheap-split trainer + muon opt · PREREG SHA |
| muon-explained.html | Public | Keller Jordan Muon / NS primer (attribution; not Muon-alone ΔT%) |
Entry points: intro.html (Why Muon · Read more) · report.html#insight-kill-bank-seal · GLOSSARY.html.
Improvement rounds
Parallel 6-lane agents improve the evidence/report package without colliding on the same HTML.
| Doc | Purpose |
|---|---|
| PARALLEL_ROUNDS.html | Protocol, lane ownership, conflict matrix, preview/deploy gates |
| reviews/PROMPT_TEMPLATES.html | Copy-paste Task prompts for L1–L6 |
| reviews/rounds/ | Per-round goals + accept/reject logs |
| books/README.html · books/HF_BOOK_COLLECTION.html | L5 literature + HF book collection workflow (local; not deployed) |
Skill nocap-parallel-round | Launch six Task agents with hard file ownership |
Canonical merge targets: report.html, preview-schema.html, muon-explained.html, qkv-explained.html, b32-systems-explained.html, split-qkv-explained.html, split-vs-ortho-explained.html, f1-seal-explained.html, kill-bank-seal-explained.html (+ companion docs). Only L6 edits those pages. Leave cyan index.html until a dedicated promote round.
Site link note
index.html links public data/*.csv, the Verify chart, companion .md paths, and workspace codes. Do not deploy WORKSPACE_MAP.local.md or EMAIL_SNIPPET.md.