Skip to content

Research construction: R439 phase-B blocks 506/523 to 24/39 ADDs + certified single-write class boundaries (no new kappa) - #891

Merged
Mister-Meeseeks merged 2 commits into
CrocSwap:mainfrom
navinor:research/agent-04-bit-seed08
Oct 13, 2026
Merged

Mister-Meeseeks merged 2 commits into
CrocSwap:mainfrom
navinor:research/agent-04-bit-seed08

Conversation

@navinor

@navinor navinor commented Oct 13, 2026 •

Copy link
Copy Markdown
Contributor

Finite bit-supplier construction + scoped class-boundary survey; no new kappa. One class search over all ten phase-B blocks of PR 589's tuned punctured R439 word, two shortenings spliced into one verified word, and certified boundaries where the class cannot shorten.

quantity before after
block 506 (B[102:128), frame 506) 26 24
block 523 (B[404:449), frame 523) 45 39
complete exported local ADDs 24,305 24,289
forward positive bound 49,347 49,347 (equal)
composed reverse positive bound 1,113,070 1,090,494
payload 93 bits 93 bits (cap 104)
paid profile / rank mass {1:1323, 2:1342, 3:79, 4:44, 5:102, 6:11, 7:19} / 5,129 unchanged

All 661 register frame paths identical, exact F2 endpoint on all 661 dirty-input columns, inherited H1–H5 conclusions hold via the pinned xcheck_p.py, bills and positive/payload bounds recomputed via the pinned export_records.py. Twelve mutation controls rejected.

Survey (whole-block single-write class of PR #873): class minima are 24 (frame 506), 30 (frame 508 — independently reproducing #873's certified minimum exactly), 39 (frame 523). Seven blocks have no program in the class: frames 507/519/520/522/524 carry order-independent per-row donor-supply certificates (e.g. 507 row 3 needs column 4 three times; every donor whose row fits carries it once), frames 509 and 521 are global order obstructions certified by two independent exhaustive recursions (forward subset DP in search.py, backward Bellman in verify.py, which imports no search code). These are scoped class statements, not impossibility theorems.

Frame 523 vs open PR 857: their best published witness is 40 gates (randomized factorization, reverse prefix grown to 1,228,514, payload 94). The 39-gate witness here is a certified class minimum with a distinct mechanism; its local reverse map is not entrywise dominated but the whole-word reverse prefix drops to 1,090,494 with payload staying 93. Their implementation and savings are not imported.

Concurrent arrivals, disclosed: open PR 880 (frame 507, fixed-factor class, 35→30) and open PR 884 (frame 521, 47→31) landed after this survey's search finished; both are exactly the blocks certified infeasible here, so their shorter programs must use factor-level intermediate donors or multi-write interleavings — independent structural evidence for their mechanisms, not competition for the same claim. Frames 506 and 523 remain uncovered by any other open PR.

Reproduce (standard-library Python 3.11+, ~90 s):

python3 -I -B research/r439-bsection-survey/search.py
python3 -I -B research/r439-bsection-survey/verify.py

Expected: PASS_R439_BSECTION_SURVEY, spliced_frames [506, 523], adds 24305 -> 24289, reverse_positive_bound 1090494, payload_bits 93, 12 controls rejected. search.py regenerates CANDIDATES.json/CERTIFICATE.json byte for byte. Receipt red test (tampered EXPECTED.json rejected) and three development defects caught by these checks are recorded in VALIDATION.md.

Base f567884aeb8851b6e794af1184119a3bfbe1e631 (selected PR 799/PR 589, conditional κ 0.001661634653669567). Only research/r439-bsection-survey/ is added; upstream/ and selected inputs unchanged. OS assignment seed 8, bit-supplier track; no pinned coordination issue existed (only #650 open), so the claim is recorded in CLAIM.md. No native C++/hosted invoice/global assembly/cold replay/Lean/remote CI run here; inherited realization/compiler/hosting/precision/tape/analytic assumptions remain; no new kappa (complex-bound as before).

Credit eumemic (PR 543/PR 555), DreamingOfClouds (PR 545/PR 589), Dylan Crow (record exporter), Zhihao Chen and the retained supplier lineage; class and verification pattern after open #873, fixed-factor framing PR 844 (Gautam Gupta), factorization search PR 857 (zeusinsight); concurrent PR 880 (Krishnendu Purohit) / PR 884 (martin Rv) disclosed above. Prepared with help from opencode. Apache-2.0. No independent human review is claimed.

…DDs, seven class boundaries (no new kappa)

Whole-block single-write rescheduling (PR873's class) over all ten phase-B
blocks of PR589's tuned punctured R439 word. Frames 506 (26->24) and 523
(45->39) are spliced into one verified word: complete exported local ADDs
24,305 -> 24,289, forward positive bound 49,347 unchanged, composed reverse
positive bound 1,113,070 -> 1,090,494, payload 93 bits, paid profile and rank
mass 5,129 unchanged, identical frame paths on all 661 registers, exact F2
endpoint on all 661 columns, inherited H1-H5 conclusions hold. Frame 523's
39-gate witness beats open PR857's 40-gate program by a different, certified
mechanism (minimum re-derived by an independent backward recursion). Frame
508's class minimum is independently reproduced as 30, matching open PR873.

Seven blocks have no program in the class: five carry order-independent
per-row donor-supply certificates (frames 507/519/520/522/524), two are
global obstructions certified only by the exhaustive recursions (509/521).
Scoped class statements, not impossibility theorems; no new kappa (the
selected witness stays complex-bound at 0.001661634653669567).

Adds only research/r439-bsection-survey/. Base f567884; consumed sources
pinned in SOURCES.json. Twelve mutation controls rejected; see VALIDATION.md.
Prepared with help from opencode. Apache-2.0.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants