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
1 change: 1 addition & 0 deletions research/p104-coupled-observation-audit/.gitattributes
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
retained/** -text whitespace=cr-at-eol
42 changes: 42 additions & 0 deletions research/p104-coupled-observation-audit/CREDITS.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,42 @@
# Credit and exact source

Ali Romman prepared the separate forward audit, rank certificates, proof,
controls and package with substantial OpenAI Codex assistance in research,
implementation, verification and writing. New material is Apache-2.0.
Separate code paths are not independent human or formal-kernel acceptance.

Ali Romman, with substantial OpenAI Codex assistance, supplied the grouped
construction, exact journal, two-receiver example and review request in
[PR #7756](https://github.com/CrocSwap/integer-mult-bounds/pull/7756).
The scientific source is commit `9e051455757d1cc86b822e52e42e1e1c98ead417`.
The [proof](https://github.com/aliromman/integer-mult-bounds/blob/9e051455757d1cc86b822e52e42e1e1c98ead417/research/p104-parallel-bank-punctures/PROOF.md),
[constructor](https://github.com/aliromman/integer-mult-bounds/blob/9e051455757d1cc86b822e52e42e1e1c98ead417/research/p104-parallel-bank-punctures/construction.py),
[edit journal](https://github.com/aliromman/integer-mult-bounds/blob/9e051455757d1cc86b822e52e42e1e1c98ead417/research/p104-parallel-bank-punctures/EDITS.json),
[local source checker](https://github.com/aliromman/integer-mult-bounds/blob/9e051455757d1cc86b822e52e42e1e1c98ead417/research/p104-parallel-bank-punctures/witness.py)
and [source lineage](https://github.com/aliromman/integer-mult-bounds/blob/9e051455757d1cc86b822e52e42e1e1c98ead417/research/p104-parallel-bank-punctures/CREDITS.md)
are retained unchanged with their notices, license and metadata.

The base graph and interface come from Ali Romman's
[PR #7726](https://github.com/CrocSwap/integer-mult-bounds/pull/7726),
scientific commit `3ba9094c54b61f424d2f98cd2c6ca64c01baf9d3`, through the exact
PR7756 export. This audit does not rebuild that earlier export from its
parent source or certify the preceding edits. The preserved
[parent credit](https://github.com/aliromman/integer-mult-bounds/blob/9e051455757d1cc86b822e52e42e1e1c98ead417/research/p104-parallel-bank-punctures/credits/7726-CREDITS.md)
credits Clear1nsight's causal split, relay and physical engine;
DreamingOfClouds's literal splits and toolkit with Claude assistance;
SYLforge's signed word; Rohan Arun's caller lineage; Jerry Chen / CubeJerry's
selected application; and Douglas Colkitt's framework and compiler. All
sixteen retained predecessor notice files remain byte for byte. These
credits follow the source notices and do not infer authorship from the
submitter account.

The new checker authenticates all 31 retained files before import. It uses
the source constructor only to obtain the graph before edit 26. It then
reconstructs the twelve bank rows and both impulses without the source
linear-algebra routines. The unchanged local `coupled_example` routine runs
only after this calculation, as a cross-check. The complete source witness
and its native, caller and price checks were not run.

The source's conditional kappa and all its unresolved assumptions remain
source claims. This package adds local evidence only. It changes no
selected record, shared CI file or other contributor's PR.
147 changes: 147 additions & 0 deletions research/p104-coupled-observation-audit/PROOF.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,147 @@
# Exact audit of the two coupled observations in edit 26

## Result and source boundary

This package checks the local two-receiver claim in PR7756 at scientific
commit `9e051455757d1cc86b822e52e42e1e1c98ead417`. The retained constructor
rebuilds the source graph through edit 25. Its observer supplies the graph
immediately before edit 26. This construction is a dependency, not a newly
verified theorem here. No retained function or global is changed.

There are 1,262 public scalar inputs. Write `x_j` for input `j`, and `b_k`
for bank `k` in this graph. The two source equations are

```text
s0 = x1149 - x1152/2,
s1 = (x1152 - x1151)/2.
```

The result is exact over the rationals. Chart 144 can observe precisely the
line spanned by `s0` in the plane spanned by `s0,s1`. Chart 177 can observe
precisely the line spanned by `s1`. Both observations have the required
response to both paid impulses. Subtracting them restores terminals 130
and 166 when both source equations are zero.

This verifies one claimed scalar implication. It gives no new kappa and
no complete acceptance of PR7756, its earlier edits, its caller or its price.
All five framework assumptions and the selected frontier remain unchanged.

## Complete local banks and explicit observations

The checker enumerates every public input and paid arrival at each chart.
There are six local banks at each chart and no local public entrance.
Their order below is also the column order in the rank certificates.

| Chart | Complete ordered bank list | Observation |
| --- | --- | --- |
| 144 | 101940, 101941, 101949, 101984, 101985, 101996 | `(b101949-b101940)/2 = s0` |
| 177 | 101937, 101938, 101943, 101981, 101982, 101987 | `2*(b101943-b101937) = s1` |

The equalities refer to the unperturbed graph. The new checker propagates
sparse integer numerator vectors forward from the public inputs. Each
vector has one exact positive denominator. It computes all 15,144 public
coefficients of the twelve banks, including zero coefficients. It also
computes both pulse columns at every bank and every terminal on the two
receiver charts. The complete required graph has 904 nodes. Its donor
charts and a separate topological order are checked.

## Certificates that exclude a second source direction

At either chart, let `B` be the matrix whose six columns are the public
coefficient vectors of its banks. Form the 1,262-by-8 matrix

```text
M = [ B0 B1 B2 B3 B4 B5 -s0 -s1 ].
```

A vector in the kernel of `M` gives a local observation and its coefficients
in the source plane. The two complete kernels are

```text
chart 144: span{ (-1/2, 0, 1/2, 0, 0, 0, 1, 0) },
chart 177: span{ (-2, 0, 2, 0, 0, 0, 0, 1) }.
```

Every one of the 1,262 row equations is checked for these vectors. To
certify that each kernel has dimension one, the certificate supplies a
nonzero 7-by-7 minor. All indices below are zero-based column positions
or literal public-input labels.

- Chart 144 uses public rows `2,786,1149,3,4,18,1151` and columns
`0,1,2,3,4,5,7`.
- Chart 177 uses public rows `2,818,1151,3,4,54,1149` and columns
`0,1,2,3,4,5,6`.

Both determinants are exactly `1/512`. The complete matrices are in
`RESULT.json`. Fraction-free determinant evaluation checks the selected
minors separately from rational row reduction. A nonzero minor proves
rank at least seven. The explicit nonzero kernel vector proves rank at
most seven. Thus the two kernel descriptions are complete. In particular,
the six banks at either chart are linearly independent.

The source equations are independent. If only `s0=0` is imposed, a linear
observation that vanishes on all allowed public inputs must belong to
`span(s0)`. Its intersection with chart 177's bank space is zero. That chart
therefore has no nonzero correction of the required type. The same argument
with the charts exchanged applies if only `s1=0` is imposed. This conclusion
concerns these two equations and these banks. It does not exclude a
construction with different constraints, banks, transfers or nonlinear state.
Other lanes in edit 26 are outside this audit.

## Both incoming disturbances

The symbols `u,v` denote the two parallel-basis impulses from the journal.
They are not isolated injections at only the named pivot registers:

```text
u: add u at 101324 and u/2 at 101902; both routes are 117 -> 834.
v: add v at 99605 and v/2 at 100008; both routes are 117 -> 837.
```

Forward propagation of these two additional formal columns gives

```text
observation at 144: s0 - (u+v)/2,
observation at 177: s1 + (-u+v)/2.
```

The errors at terminals 130 and 166 are respectively `-(u+v)/2` and
`(-u+v)/2`. The other two terminals on these charts have zero response to
both impulses. When `s0=s1=0`, subtracting the local observation removes
the entire error at each affected terminal. This holds for arbitrary
rational `u,v` and all constrained old inputs by linearity.

The response matrix is `((-1/2,-1/2),(-1/2,1/2))`, with determinant `-1/2`.
After imposing the source constraints, the two observations recover
`u = -z144-z177` and `v = -z144+z177`. The nonzero determinant does not
provide a physical cross-chart readout; each actual terminal uses only
its own local observation.

## Separate evidence and rejection controls

The new forward propagation and fraction-free rank certificates do not
call the source's entrance expansion, pulse routine or row-space routine.
After the separate calculation, the unchanged source local checker runs
as a cross-check and matches its retained receipt. That check uses reverse
entrance expansion. Both paths share the pinned pre-edit graph constructor.
This is separate implementation evidence, not independent human review.

Four controls flip one coefficient in one parallel impulse. Each produces
a nonzero error after the two prescribed repairs. The resulting residual
vectors at terminals 130 and 166 are retained in `RESULT.json`. Two further
controls flip the nonzero observation coefficient on the third bank; each
breaks the all-input source identity.

A terminal-only control would miss some pulse errors: changing a companion
arrival can leave the terminal response unchanged while changing its local
observation. The controls therefore test the complete repaired expression.
They do not require an irrelevant terminal response to change.

All 31 retained files have Git-blob and SHA-256 pins. The selected base
manifest supplies provenance only; it asserts no unmerged ancestry. Normal
and optimized registered replay check the same certificate.

The complete 32-edit witness, whole native operator, physical catalogue,
mixed caller, interval price gate, selected cold replay, Lean and Docker
were not run. No compiler assumption is discharged. Broader construction
and caller review remain open.
40 changes: 40 additions & 0 deletions research/p104-coupled-observation-audit/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,40 @@
# Two coupled P104 observations: separate exact audit

The two local observations in edit 26 of PR7756 pass a separate exact
check. All six banks at chart 144 observe only the first source direction;
all six at chart 177 observe only the second. Their full two-pulse responses
cancel the errors at terminals 130 and 166. Two explicit 7-by-7 minors,
each with determinant `1/512`, certify the complete intersections.

Read [the proof](PROOF.md), [full bank coefficients and rank certificates](RESULT.json),
[forward expansion and determinant code](audit.py), [verifier](verify.py)
and [source credit](CREDITS.md). No new kappa or whole-supplier acceptance
is claimed. The earlier source graph construction is a pinned dependency.

Use Python 3.12 and the standard library from the repository root:

```sh
python3.12 -B scripts/research_package.py --package research/p104-coupled-observation-audit
```

Both registered commands print `PASS_P104_COUPLED_OBSERVATION_AUDIT`.
The package also works after export, without Git or repository imports.
The source constructor is stopped through its observer before edit 26.
The new checker uses forward integer vectors, rational elimination and
fraction-free determinants. The unchanged source local checker then
provides a reverse-expansion cross-check.

The source has 1,262 public fields. The audit checks twelve complete bank
rows, both pulse columns, and every terminal at the two charts. Four
changed pulse signs and two changed observation signs fail their controls.
Only this local scalar claim is covered. Complete native, caller and price
review, all five compiler assumptions and physical admission remain open.

## Open next questions

- Can a separate implementation verify every grouped return and source
replacement in the exact 32-edit journal? Check one unclaimed group,
including every incoming pulse, before expanding the audit.
- Do all sixteen visits of one mixed root share one physical head after
both charged view changes? This is the separate caller review request
in PR7756; check live coordination before taking it.
Loading
Loading