Skip to content
Open
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
60 changes: 60 additions & 0 deletions research/p104-delayed-shared-relay/COMPOSITION.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,60 @@
# Composition and ownership accounting

Start from PR5608's authenticated parent construction: the shared587 relay with
the PR5565 commuting cascade. Apply the source-pinned delayed696 construction
before splitting. Every735-input endpoint and chart must agree, with exact
paid change `H4+H7-H5-H6` and unchanged target histogram.

The13 retained PR5608 helper split witnesses keep the same rational vectors,
pivots, charts and rank declarations. Only event cuts move. PR5599's retained
split checker recomputes every full prefix functional and suffix direction
before each application, and all independent endpoints are replayed after each
split. This is essential: an old endpoint proof alone does not authenticate a
new intermediate cut.

Next apply the four exact declarations from PR5601. Three scalar contacts move
to the declared common charts and helper307's endpoint changes from rank5 to4.
The unique-contact adapter rebuilds only affected public MOVE segments. Full
replay verifies the paid delta `-2H2+2H3-2H4+H5` and target delta `H4-H5`.

Finally apply Scratch-net's PR5606 preparation to the retained rank-two prefix573.
At its two actual destination charts, exact current-value identities are

`current573-current572+2current732 = -2 original573`,

`current573-current572-2current710 = -2 original573`.

The verifier derives these identities over all748 source inputs before removing
the573 coordinate. Preparing that role at zero makes each value reconstructible
by coframed COPY/ADD operations. Its two H1 transports disappear, and its H2
productive target is debited. The independent rank-seven successor is retained.
The573 slot is now an owned role, with explicit initialization, reset and final
return. It is not simultaneously credited as an independent input.

The resulting interface has747 independent inputs and nine owned roles in756
slots. There are11 owned lifetimes and peak occupancy two. The exact checker
validates every field, frame, owned lifetime and final row, including arbitrary
initial dirty-owned coordinates and every required zero return. Native evaluation
checks every remaining paid map and source endpoint normalization; the retired
H2 target and its two external normalizers are then explicitly removed, matching
PR5606's declared accounting contract.

Relative to PR5608, the complete local changes are

`Delta paid = -2H1-2H2+2H3-H4-H6+H7`,

`Delta targets = -H2+H4-H5`.

The full caller invokes these changes18 times. The complete bill is66,636 paid
calls and13,030 targets with surplus760. It includes all queries, whole idles,
main adapters, helper normalization, ordinary scalar work and routes. The
owned-lifetime count and peak are actual replay outputs, not inherited literals.

The primary exact interval engine encloses the raw crossing, and an independent
engine checks overlap and positive funding at0.0031029905. Completion, compiler
and all58 outer predicates are checked with the retained power1024 parameters.
The claimed conditional endpoint is0.00310299. The inherited theorem assumptions
and unmaterialized uniform word remain stated in README.md.

Prepared for flyingviper with substantial OpenAI Codex assistance and independent
agents, retaining the predecessor attribution. Apache-2.0.
121 changes: 121 additions & 0 deletions research/p104-delayed-shared-relay/MECHANISM.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,121 @@
# Delaying public preparation through a shared component

This construction continues the verified shared587 relay and148 contraction.
Its source is `contracted-587-relay-WORD.json`, with canonical sorted-key JSON
SHA256 `95078b95bae0c56927f6daaa4430457cc43862c24a9db3b015bb1366e3381055`.
The underlying frozen word is PR5580, head
`6efa248cf456d6101d757e9d09ba29711d18a13a`.

This note proves the new component. The package verifier separately checks the
composed word, complete native/caller/ordinary bill and analytic predicates.

## The changed preparation

Write `P` for the public133 value copied into owned738. At its final line597,
public734 contains `x734-P/2`. All `x` variables below are arbitrary original
inputs; none is assumed zero.

At frame1031, before the relevant producer preparation, let the current public
values699,701,702,703 be `a,b,c,d`. The original preparation forms

* `696 = x696-a+b+c`;
* `700 = x700-b-c`, later updated to `700-d`.

The opposite coefficients of `b+c` permit a useful retiming. Public700 retains
that component after701 and702 have ascended and after `P` has become available.
Consequently696 need not ascend early just to read701 and702 separately.

After the last original zero-frame read of696, add `x700-x734` to696 while
all three fields remain at frame3. Delete its original ascent3-to1031 and
the original three additions from699,701,702. Immediately after734 receives
the owned value at597, perform

1. `MOVE 696,3,597`; `ADD 696,734,+1`;
2. `MOVE 696,597,1031`;
3. `ADD 696,699,-1`; `ADD 696,700,-1`.

The resulting value is exactly the original696 value minus `P/2`:

`x696+x700-x734 + (x734-P/2) - a - (x700-b-c)`.

The changed itinerary uses existing frames. The two new ascent ranks are1
and4, replacing the original rank5.

## Coupling the two later butterfly inputs

At this point, still before the original `700 -= 703` update, define

`S = current696 + current700 + current699 = Z-P/2`,
`Z = x696+x700`.

Add `S` into703 using three ordinary ADDs. Then add the same old `S` into700
using `SCALE 700,2`, `ADD 700,696,+1`, and `ADD 700,699,+1`. The subsequent
unchanged `700 -= 703` update cancels700's entire added quantity.

Thus the later696/703 butterfly receives perturbations `-P/2` and `Z-P/2`
in its two inputs. Its sum delivered to159 changes by `Z-P`. Its complementary
output delivered to222 changes by `-Z`. All other live-input effects cancel.

Delete the remaining owned738 read into159. That read originally added
`P/2`, later scaled by159's factor `-2`, so deleting it adds `P` to the final159
value. Combining the deletion with the changed butterfly leaves only

`out159 += Z`, `out222 -= Z`

on outputs whose final frames have dimension below nine.

Four initial common-zero-frame ADDs cancel these nuisance terms: subtract
both `x696` and `x700` from159, and add both into222. These are distinct
source and destination coordinates. Exact arbitrary-input replay verifies
the cancellation; it does not rely on clean helper inputs.

Owned738 now goes directly from591 to597, a rank7 descent. The intermediate
rank7 frame565 and its scalar contact are removed.

## Full-frame cleanup

Immediately before the original `ADD 133,288,1/2` at full frame127, exactly
five public rows still differ. Let `L=Z-P/2`. Their errors are

| Role | Error |
| --- | --- |
|696|`-P/2`|
|697|`L`|
|698|`-L`|
|700|`L`|
|703|`L`|

At this cut, the actual current rows satisfy

`133=P`, `696=x696-P/2`, `700=x696+2*x700-P/2`,

so `L=(current696+current700)/2`. All six involved rows are at127. First
subtract half of each carrier696 and700 from697 and703; add half of each
to698. Next scale700 by1/2 and subtract696/2, restoring `x700`. Finally add
133/2 to696, restoring `x696`. These nine operations restore every row
before the unchanged suffix. Source identities and operation ordering are
checked over all735 original-input columns.

## Invoice and exact verification

The public preparation replaces `H5` by `H1+H4`. Removing the last owned
contact replaces `H1+H6` by `H7`. The net paid change relative to the
shared587 candidate is therefore

`H4 + H7 - H5 - H6`.

For `H_r=r^t`, `0<t<1`, strict concavity gives
`H7-H6 < H5-H4`, so this change is strictly negative throughout that
range. This comparison is a transport statement, not by itself a new
complete multiplication bound.

The source-pinned `delayed.py` implements the component for either the
published PR5605 word or its commuting-cascade parent. It accepts only their
exact canonical payload hashes and checks the resulting full payload hash.
The standalone component has25187 events and adds18 ADDs and two SCALEs,
with no net MOVE-count change or new register. The full composition is described
in COMPOSITION.md and verified separately on every independent input.

Prepared for flyingviper with substantial OpenAI Codex assistance and independent
agent derivations. Retained source lineage and licenses are credited in README.md.
Apache-2.0.
31 changes: 31 additions & 0 deletions research/p104-delayed-shared-relay/PRICING-CHANGES.diff
Original file line number Diff line number Diff line change
@@ -0,0 +1,31 @@
--- PR5606/pricing.py
+++ delayed-shared-relay/pricing.py
@@ -12,8 +12,8 @@
from fractions import Fraction as Q
import local as L

-KAPPA, TESTED = Q('.00309871'), Q('.0030987101')
-PREVIOUS = Q('.00309839')
+KAPPA, TESTED = Q('.00310299'), Q('.0031029905')
+PREVIOUS = Q('.00310184')


def encode(x):
@@ -62,7 +62,7 @@
second_p[8*(j+1)] += 208
L.need((targets,paid) == (second_t,second_p), 'Independent full invoice')
L.need(max(targets) == 145 and max(paid) == 72, 'Maximum recursive ranks')
- L.need(sum(targets.values()) == 13066 and sum(paid.values()) == 66816, 'Complete field and call counts')
+ L.need(sum(targets.values()) == 13030 and sum(paid.values()) == 66636, 'Complete field and call counts')
L.need(sum(r*n for r,n in targets.items())-sum(r*n for r,n in paid.items()) == 760, 'Complete rank surplus')
return targets,paid,counts

@@ -150,7 +150,7 @@
L.need(wrong_cutoff[0] > 1, 'Old cutoff fails with this helper inventory')
bad=outer_module.ledger(dict(params,bit_saving=Q('.0027')))
L.need(any(x <= 0 for x in bad['strict_checks'].values()), 'Insufficient BIT contract rejected')
- return encode({'kappa':KAPPA,'kappa_decimal':'.00309871','previous_open_pr5599':PREVIOUS,'gain':KAPPA-PREVIOUS,
+ return encode({'kappa':KAPPA,'kappa_decimal':'.00310299','previous_open_pr5608':PREVIOUS,'gain':KAPPA-PREVIOUS,
'relative_gain_percent':100*(KAPPA/PREVIOUS-1),'raw_root_interval':[lower,upper],
'raw_headroom':[lower-KAPPA,upper-KAPPA],
'raw_ceiling':{'interval':[Q(0),cap],'derivative_upper':derivative_upper},'margin_at_previous':E.display(margin(PREVIOUS)),
22 changes: 22 additions & 0 deletions research/p104-delayed-shared-relay/PRICING-EDITS.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,22 @@
[
[
"KAPPA, TESTED = Q('.00309871'), Q('.0030987101')",
"KAPPA, TESTED = Q('.00310299'), Q('.0031029905')"
],
[
"PREVIOUS = Q('.00309839')",
"PREVIOUS = Q('.00310184')"
],
[
"sum(targets.values()) == 13066 and sum(paid.values()) == 66816",
"sum(targets.values()) == 13030 and sum(paid.values()) == 66636"
],
[
"'kappa_decimal':'.00309871'",
"'kappa_decimal':'.00310299'"
],
[
"previous_open_pr5599",
"previous_open_pr5608"
]
]
111 changes: 111 additions & 0 deletions research/p104-delayed-shared-relay/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,111 @@
# Conditional kappa 0.00310299 from delayed shared preparation

This construction raises complete conditional kappa from **0.00310184** in
[PR #5608](https://github.com/CrocSwap/integer-mult-bounds/pull/5608) to
**0.00310299**, a gain of **0.00000115**. Its new mechanism delays the
preparation of public helper696 by reusing a complementary component already
present in helper700. Coupled corrections remove the remaining owned738
delivery into159. Every retained independent helper input is preserved exactly.

The new mechanism has paid delta `H4+H7-H5-H6`, strictly negative for
`Hr=r^t`, `0<t<1`. It is composed with the existing 13 independent helper
splits, Ali Romman's coupled-frame step from
[PR #5601](https://github.com/CrocSwap/integer-mult-bounds/pull/5601), and
Scratch-net's prepared split-helper retirement from
[PR #5606](https://github.com/CrocSwap/integer-mult-bounds/pull/5606).
Those contributions and their costs are explicit, not attributed to this work.

The exact raw crossing is **(0.00310299359333,0.00310299359334)**; the claim
is tested at 0.0031029905. The full bill has **66,636 paid calls, 13,030
independent productive targets and rank surplus760**. The final word has
**747 independent inputs, 539 helpers, nine owned roles, 11 owned lifetimes,
peak owned occupancy two and 25,529 events**. Every added ordinary operation,
owned lifetime, route and target debit enters the complete accounting.

Read [MECHANISM.md](MECHANISM.md) for the new delayed preparation,
[COMPOSITION.md](COMPOSITION.md) for the full construction, and
[RESULT.json](RESULT.json) for the exact native/caller/price receipt.
The selected frontier is unchanged. These are finite exact checks and a
conditional bound calculation, not independent human or proof-kernel acceptance.

## Reproduce

Python3.12 and the standard library suffice:

```sh
python3.12 -B scripts/research_package.py --package research/p104-delayed-shared-relay
```

Normal and optimized modes print `PASS_P104_DELAYED_SHARED_RELAY_KAPPA_00310299`;
the launcher prints `PASS_REGISTERED_RESEARCH_PACKAGE`. Verification authenticates
the official 24-file PR5608 export and its nested unchanged33-file framework
closure before importing code. It reconstructs every stage, checks all13 fresh
prefix/suffix witnesses and every intermediate independent-input identity,
checks all756 final rows and transpose columns, all signed native maps, the
physical caller, complete paid/target bill, two interval engines at the tested
endpoint, the compiler and58 outer predicates. Five controls reject missing
owned initialization/return, missing new cleanup, a falsely retained target
and an uncharged transport. Gzip equality uses complete decompressed payloads.

The checker adaptation from PR5606 changes the explicit interface constants
750/758/749 to748/756/747, inlines the same exception helper and selects MOVE
direction by validated ownership. The verifier compares the remaining checker
body exactly to its pinned source. The retired573 role has no MOVE in the
submitted word. The [pricing diff](PRICING-CHANGES.diff) retains the source's
mathematical tests; only endpoint/comparison/count declarations change.

## Limits and ceiling

The five inherited framework assumptions remain: native all-column implementation,
owned erasure, uniform compiler closure, precision and fixed-tape costs, and
analytic recovery. The inherited helper-endpoint contract remains. The enormous
uniform word and complete rounded multiplier are not materialized.
The prepared-helper step uses the existing owned-erasure contract explicitly;
it does not count the retired prefix as an independent productive target.

The fixed invoice has a strictly decreasing signed moment on[0,0.0034].
Its raw funding ceiling is the interval above, leaving between3.59333e-9 and
3.59334e-9 above the claimed endpoint. This is not an optimum over other words,
inventories or suppliers. The uniform compiler retains logarithm power1024,
tail decay39/40 and the full reserve/outer conditions.

## Sources, independent checks and credit

Base: canonical main `5e0052ca0aa3036a3afa8b0438927377c8d97d53`, priority revision7.
The final upstream refresh reached `d7633ce230196a1d5d061347de7a0fd8be512ed6`;
the guidance, priority catalogue and selected record were unchanged. That update
archived predecessor research; archival intake does not change scientific acceptance.
The assigned scalar-to-moment task is refined to changed public dependencies
and composition with independently introduced target and ownership changes.
Source pins and retention history are in[SOURCES.json](SOURCES.json).

PR5608 is pinned at `5ec3b5b3ee4738f896e823c1dfc04e7ff1789ffb`; PR5601 at
`8f7e870b22885a367b347dc12791da98fd53f83d`; PR5606 at
`cd7d89345c6c5d2114625e5ce912a52ecd5e1d27`. The PR5608 export retains exact
PR5580, PR5565, PR5599 and PR5605 source files and their original notices.
No foreign branch-object cache is required for replay.

Credit Ali Romman for the independent split identity, frame step, source word
and native/caller/price machinery; Pierre-Loic Doulcet/hexapode for the commuting
cascade; Scratch-net for prepared573; Joel Pulikkan/GamingPuzzled for successor
handoffs; DreamingOfClouds for free endpoints and event-level tools; thu-nmrc
for coupled contacts; Gareth Davidson, David Arias, Cinna Davis, Jacob Alan
Sussman, zxv, Cao Yifan and all retained geometry, BIT and analytic predecessors.
The new delayed696 P104 construction, composition and verification were prepared
for flyingviper with substantial OpenAI Codex assistance. Independent agents
derived and checked the current-value identities, ordered cleanup, split
compatibility, prepared ownership and complete invoice. These are separate
agent calculations, not independent human review. Apache-2.0.

## Open next questions

For another owned delivery, identify complementary early public expressions
that persist until its signal is available. First derive an exact current-value
identity and the live-input endpoint effect; only then construct invertible
nuisance corrections and charge its full itinerary.

A separate opportunity is another prepared split prefix: at each of its paid
MOVEs, test whether its current row can be reconstructed from already coframed
public rows modulo that prefix's own independent input. Every reconstruction,
owned lifetime and retired target must be charged. No further prefix candidate
is claimed here. Structured next questions are in[TASK.json](TASK.json).
Loading