Accuracy notes — residual conflicts only

Updated 2026-07-17 against sealed 3/3 claim-stack + CSV recomputes + jury-ready cf-pages-nocap-agent-dev/index.html.
Canonical numbers: NUMBERS_CANON.md. Full claim→source map: FACT_CHECK_CHECKLIST.md.

Resolution order

1. Sealed MULTI_SEED_CLAIM.md / SEED*_SUMMARY.md / f1_seal/SUMMARY.md / F1_BCAROL_SUMMARY.md (Jul 15–16)
2. Public data/*.csv on https://nocap.think2earn.com (same rows as local pulls; see data/README.md)
3. Pulled CSVs under results/claim_stack_v1/seed2029/, seed4242/, results/f1_seal/ (private tree)
4. Public pages (may lag seals)
5. docs/ops/FAIR_CLOCK.md and nocap-repo/README.md (often stale)

Residual conflicts / traps

1. Repo README still pre-dates the seals

nocap-repo/README.md (≈2026-07-12 framing) can still describe official speed as a miss. Trust sealed claim-stack files, not README, for current status.

2. FAIR_CLOCK “same clock account” vs final public denom

docs/ops/FAIR_CLOCK.md was written when cardinal % were expected on Clock-W only. The sealed campaign publishes vs F1_seal = 3.866 h (Clock-W) while candidates ran on Claim-W1 / Claim-W2, with F1_Claim-W2 as a same-workspace cross-check for seed 2029. Task hard rules + MULTI_SEED_CLAIM.md win.

3. Intermediate docs still say 2/2

These are chronology artifacts, not current truth:

• Older FACT_CHECK_CHECKLIST.md / ACCURACY_NOTES.md in this folder (before this refresh)
• SEED2029_SUMMARY.md multi-seed table ends at “2/2 seeds” (written before 4242 sealed)
• SEED1337_SUMMARY.md verdict still says “Not multi-seed”

Current aggregate: MULTI_SEED_CLAIM.md + SEED4242_SUMMARY.md → 3/3.

4. Seed 1337: no local candidate CSV in the tree

results/claim_stack_v1/ has seed2029/*.csv and seed4242/cand_log.csv, but no seed1337 CSV pull. Seed 1337 numbers are taken from SEED1337_SUMMARY.md / MULTI_SEED_CLAIM.md only. If a CSV is later pulled, recompute first-cross and confirm 3.553 h / 3.3741 / 3.3710.

5. Published 3.866 h vs CSV exact hours

F1_seal CSV: 13,918,667.5 / 3.6e6 = 3.866297…. Seal SUMMARY publishes 3.866 h. All page ΔT% use denominator 3.866. This is intentional rounding, not a conflict.

6. Rounding of seed ΔT%

Matches sealed aggregate language.

7. Historical F1 / F2b −2.9%

Historical F1 ~229.1 min is provisional (torch unlogged). F2b quality hit stands; −2.9% vs historical wall is retracted as speed evidence. Dev page states both correctly.

8. claim_muon_wsd path lives outside nocap-repo/results/

Workspace-level results/claim_muon_wsd/log.csv (and board_v3 writeup) support ~3.899 h / ~0.8% slower. Not under the claim_stack_v1 tree.

9. STACK.md “not an under-baseline claim”

claim_ready_stack_v1/STACK.md is correct for 2026-07-14 component bank. Full seals landed later. Use it only for component history.

10. Public CSVs vs private pulls

Jury-ready deploy ships data/*.csv (five files). Strangers should recompute from those URLs — not from private nocap-repo/results/… paths. Seed 1337 remains summary-only on the public site (no data/s1337_*.csv). Workspace codes only in all public text.

11. Live / production vs this refresh

This companion-doc pass did not deploy. After deploy, spot-check that live data/*.csv byte-match staging and that the Verify chart reads the same first-cross rows as data/README.md.

12. “2 / 2 no cross” on the failure ledger

That phrase refers to two paired stock arms (1337/2029), not “2/2 candidate seals.” Do not “fix” it to 3/3.

Explicit non-claims (still true)

• No matched first-cross speedup vs joint stock.
• No published % vs historical F1.
• Claim recipe = candidate stack, not joint stock control.
• Seed 4242 has no stock arm.
• Organizer acceptance of cross-account F1_seal denom is outside local evidence.

Ops (this pass)

• No Modal GPU launches.
• No deploy.
• No production index.html edits.
• Volume ls not required; local CSVs + sealed summaries sufficed.