Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
* -text
2 changes: 2 additions & 0 deletions research/staged544-terminal-extension-budget/.gitignore
Original file line number Diff line number Diff line change
@@ -0,0 +1,2 @@
__pycache__/
*.pyc
Original file line number Diff line number Diff line change
@@ -0,0 +1,8 @@
{
"head": "91f81d131252db704f945891da1ea54e29d46b8b",
"base_url": "https://raw.githubusercontent.com/CrocSwap/integer-mult-bounds/91f81d131252db704f945891da1ea54e29d46b8b/research/staged544-terminal-gap-fill/owner/",
"files": {
"FILL.json.gz": "430233230c9194a52bda171afd21713351fdd02e0054230012109e44047c4c55",
"inputs/TEMPLATE.json.gz": "21297b20f1f2dc76eb9dd8124416068c3cb096bab9df17fa4da19bc68dfe082e"
}
}
8 changes: 8 additions & 0 deletions research/staged544-terminal-extension-budget/FROZEN.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,8 @@
{
"frozen/audit_core.py": "888679ca8502cdcb100420fe5382a052529a494bf76cb332928b86e62612b951",
"frozen/PR3890-PROOF.md": "27a4519a6b75d06f0203c3fc28a60ad95408b726724793e7b8a9774184b6ef9b",
"frozen/PROFILE.json": "64b85c791a331e79ddc18f047b672240c008900b8c004d37cf6d925067373228",
"frozen/WORD-extra.json.gz": "d83c2ec9cecc0ad092d34ea4080b28a4fd7199579356b6d5a8fc6cd53ff93299",
"frozen/WORD-base.json.gz": "08e6c96fbe3dfec96cb8451b3867ef8c1a214109ecd1e5af7fcc28e0f05728d0",
"frozen/SOURCE-CREDITS.md": "d4a13d8b8ed9cdb594f00efb3d1a336e09d934423330f02a1514308a720c278b"
}
37 changes: 37 additions & 0 deletions research/staged544-terminal-extension-budget/LEDGER.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,37 @@
{
"status": "COMPLETE_SCOPED_TARIFF_OBLIGATION_LEDGER",
"no_candidate_emitted": true,
"baseline_local_profile": {
"1": 2482,
"2": 712,
"3": 218,
"4": 109,
"5": 20,
"8": 8
},
"baseline_complete_profile": {
"1": 8041680,
"2": 2306880,
"3": 729648,
"4": 353160,
"5": 69120,
"8": 25920,
"48": 137952
},
"local_terminal_surcharge": "proved lower bound D*(9^s-8^s), every intermediate paid frame fixed; binary terminal rank at most9",
"local_interior_cost": "fixed by theorem; any savings from interior refit are outside its class and must exceed0.006843664113335406486182D before repacking could pay",
"owner_idle_credit": "proved upper bound qD from all841 exact dual inequalities; permits any clock-free fractional reassignment",
"baseline_owner_binding": "all1321920 rows/725760 bins regenerated from immutable external files in full mode; exact aggregate proof checked in default mode",
"actual_new_chronology": "no new assignment constructed; relaxation discards chronology and family restrictions, so rejection also covers their constrained subset",
"fixed_identity_capacity": "proved zero nonnegative extension cone using terminal rank plus actual per-bin capacity",
"dirty_state_and_restoration": "no changed circuit claimed. Frozen words preserve predecessor's91 protected buffers and cleanup; PR3969 exact arbitrary-dirty scalar/permutation identity inherited, not independently replayed here",
"native_phase": "inherited/unverified here, not zero; interior/native recipes fixed in theorem",
"scalar_coefficients": "no scalar coefficient changes; finite4-bit source bound inherited, no uniform precision/tape coefficient derived",
"physical_hosting": "finite capacity certificate fresh, universal physical implementation assumptions inherited/unverified",
"uniform_precision_stock_tape_recovery": "unknown/inherited, not zero",
"all_size_compiler": "unknown/inherited, not zero",
"outer_supplier_release": "unrun; fixed claimed3945 endpoint used only as exact tariff parameter, no root/admission recertification",
"analytic_BIT": "inherited/unverified, not zero",
"scope_limit": "This is a complete local-plus-idle tariff rejection under the explicitly fixed comparison model. It is not a lower bound on every possible multiplication algorithm or on unrelated costs. Other shrinking/extension combinations beyond the17direct refunds, and changed internal frames/templates/entrances remain outside.",
"direct_terminal_refunds": "Exactly17 protected width5->4 direct refunds per provider: savedH1 is canceled by a proved+H1 idle lower bound; all2551patterns checked. Mixed direct refunds plus terminal-only extensions still lose the same positive gap."
}
137 changes: 137 additions & 0 deletions research/staged544-terminal-extension-budget/OWNER-PATTERNS.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,137 @@
{
"derivation": "All1321920 literal FILL rows; run --source-root to reconstruct. All unoccupied bins counted.",
"patterns": [
[
8,
[
0,
0,
0,
1,
0
],
0,
241920
],
[
16,
[
0,
0,
0,
2,
0
],
0,
2233
],
[
16,
[
0,
0,
1,
0,
1
],
0,
159047
],
[
32,
[
0,
0,
0,
0,
3
],
5,
4320
],
[
32,
[
0,
0,
0,
4,
0
],
0,
224
],
[
32,
[
0,
0,
1,
2,
1
],
0,
2719
],
[
32,
[
0,
0,
2,
0,
2
],
0,
4977
],
[
32,
[
0,
1,
0,
1,
2
],
0,
58320
],
[
32,
[
1,
0,
0,
0,
3
],
0,
90720
],
[
48,
[
0,
0,
0,
0,
0
],
48,
137952
],
[
48,
[
0,
0,
0,
0,
5
],
3,
23328
]
]
}
156 changes: 156 additions & 0 deletions research/staged544-terminal-extension-budget/PROOF.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,156 @@
# Terminal-only extension requires a strictly positive interior saving

Chris R / Clear1nsight, substantial OpenAI Codex assistance. Apache-2.0.

## Exact contract and claim

Let k = 1730663236534549309146/10^24, s=1-k, and h(r)=r^s for r>0,
h(0)=0. This is the lower endpoint of the root enclosure claimed in PR3945;
its source moment is sum n_r r(73/r)^k. Comparing sum n_r h(r) is equivalent,
since the common positive factor73^k and the normalizer remain unchanged.
The endpoint is a source pin, not a new root or supplier validation.

Keep either pinned PR3945 word's helper entrances and every paid interior
scalar/native frame unchanged. Enlarge any chosen terminal subspaces by
nonnegative dimensions. They need not be coordinate extensions, uniform
across calls, or equal between providers. Keep the ambient binary dimension9,
helper identities and the bin inventory at T360. Permit an arbitrary new owner
assignment; for the lower bound, drop all time-window and family restrictions,
and even permit fractional packing. Each helper remains whole in one bin and
each unused remainder is paid as one recursive child.

If D>0 is the total number of added helper dimensions over all calls, the new
complete local-plus-idle tariff exceeds the incumbent by more than

0.006843664113335406486182 D.

Thus terminal-only enlargement cannot improve the complete tariff even with
unrestricted repacking in this relaxed model. If an actual implementation adds
other costs, they cannot repair this deficit. If interior repricing introduces
savings, it must supply more than this amount to beat the incumbent; this
necessary allowance is not a sufficient construction. Mixed shrink/enlarge
moves and changed helper entrances are outside the theorem.

## Credited packing dual, newly checked at the changed endpoint

The dual form is from Erwan Secousses / ChaloupeMusic PR3815, Sagar Thakrar
PR3862 and ToikY PR3890. We do not claim to discover it. PR3890's frozen proof
is retained with its original attribution. Its exponent was the earlier
3815 endpoint; here every inequality is checked afresh at the3945 endpoint.
The new result is the terminal-cost implication and binding to the actual
changed3945 inventory and literal allocation.

For widths5..9 put

q=(h(48)-h(3)-5h(5))/20,
pi(w)=-h(5)-(w-5)q,
mu(8)=-pi(8), mu(16)=-2pi(8), mu(32)=-4pi(8), mu(48)=h(48).

For any capacity c and count vector n of widths5..9, with remainder
l=c-sum w*n_w>=0, the dual condition is

h(l) - mu(c) - sum pi(w)*n_w >=0.

The checker enumerates all841 patterns:11 symbolic identities and830 strict
inequalities. Exact rational outward bounds certify every strict inequality
(the minimum exceeds0.007594638501835091171332). It does not use a numerical
optimizer or a floating tolerance. Symbolic zero vectors are checked before
interval evaluation. Bin capacities are8,16,32,48; their counts are
241920,161280,161280,161280. The true local inventories yield widths
5:90720,6:58320,7:171720,8:311040,9:690120 across3240 calls.

All actual incumbent patterns are dual-tight. This is checked both on the
small complete integer pattern certificate and, in the explicit external
replay, by reading all1321920 literal roles and all725760 bins from immutable
source FILL/TEMPLATE data. Therefore the incumbent idle cost equals its dual
lower bound, not merely some unattained estimate.

Weak duality gives, for every new packing, idle_new>=dual_new. Since the
incumbent attains its dual, idle_new-idle_old>=sum delta_n_w*pi(w). Total role
count is fixed. The affine helper prices make this last expression exactly
-qD, even if width changes differ per call. Hence the best possible idle
credit is at most qD. It is not necessary that the changed inventory can
actually attain this bound: a lower bound is enough for rejection.

## Local rank function and the full budget

A terminal frame is a subspace of F_2^9. If its last paid climb had dimension
r and the terminal extends by d while the preceding frame is fixed, the new
charge is h(r+d) instead of h(r). The Grassmann rank identity gives this
increase independently of the extension's orientation. Because0<s<1, h is
concave and its discrete increments decrease. For0<=r<r+d<=9,

h(r+d)-h(r) >= d*(h(9)-h(8)).

This is also checked as nine exact inequalities. Summing all roles gives a
local surcharge at least mD, m=h(9)-h(8). The finite certificate encloses

0.987736182658129815011783 < q < 0.987736182658129815011784,
0.994579846771465221497966 < m < 0.994579846771465221497967,
0.006843664113335406486182 < m-q < 0.006843664113335406486183.

Adding the local and idle inequalities proves the stated complete deficit.
No dirty-state or phase relaxation is used to assert a new working circuit:
this is a necessary-cost rejection of a broader relaxed class, so proving
some candidate semantically valid could only shrink that class.

## Sharper fixed-identity zero-feasibility certificate

Retaining actual owner/time/child/port identities is even more restrictive.
For each role a nonnegative increment d obeys both

d <= 9-dim(terminal),
sum_{roles in bin} d <= capacity - occupied width.

Every occupied bin with positive residual capacity has only width9 roles,
whose terminal dimension is9. There are23328 bins with five width9 roles and
slack3, and4320 bins with three width9 roles and slack5. Every other occupied
bin is saturated. The137952 empty rank48 bins contain no retained role.
Consequently every d is zero. This is a rank-capacity certificate, not a
failed enumeration of frame orientations. It covers call-specific increments.
A proposed role reassignment invalidates this sharper premise, but is still
covered by the more permissive tariff theorem above.

## Dirty state, controls and limits

No candidate word is emitted or admitted. The exact pinned words are retained;
the credited audit core freshly computes their frame invoices and role
entrance/terminal dimensions. Previous PR3969 established relative arbitrary
dirty scalar/permutation identity and the literal owner bridge. Native phases
and universal physical/precision/stock/tape assumptions are not re-audited.
These obligations are unknown/inherited, never zero. The original3945 gain
jointly reprices interior frames and lies outside our rejected class.

Controls include a genuine positive rank1-gap absorption fixture (h(2)-2<0),
noncoordinate subspace rank, omission of local cost falsely creating a gain,
and actual optimized production CLI mutations of dual, capacity, empty-bin
coverage and source bytes. The positive fixture explains why the gate is a
complete-budget test rather than a universal anti-extension assertion.

## Bounded pivot: the17 direct rank-one refunds cannot finance an extension

There is a concrete terminal-only shrinking option, so we priced it rather
than interpreting the one-sided theorem as an obstruction to every change.
Both frozen providers have exactly17 protected width5 helpers whose last
terminal climb has rank1. Their preceding frame is the unique legal terminal
of one less dimension. The other74 protected terminals have zero last climb,
so cannot shrink with the preceding frames fixed. The unprotected helpers
retain FULL terminals under the inherited cleanup contract; adding new private
buffers is outside this scope.

Extend the credited dual to width4 by pi(4)=1-h(5). The supplementary checker
examines all2551 patterns in widths4..9 at the same exact3945 endpoint. It finds
12 symbolic tight and2539 strictly positive inequalities. The original actual
packing still attains the bound. Therefore every width5->4 refund raises the
best possible idle tariff by at least pi(4)-pi(5)=1, exactly canceling its
saved terminal rank1 child. Such a refund can tie, but cannot pay by itself.

For simultaneous direct refunds on R roles and terminal extensions adding D
other dimensions, the idle change is at least R-qD while the local change is
at least -R+mD. The same positive (m-q)D deficit remains. For D=0 the complete
change is nonnegative. This rejects that concrete mixed pivot before any
relocation search. Equal-dimension frame rotations are tariff-neutral here;
changing interior frames, entrances or the protected-buffer set remains open.
No larger terminal search or unpriced new native module is justified by this
pivot. The supplementary code shares the main exact power routine and is not
claimed as an independent numerical implementation.
Loading