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
10 changes: 10 additions & 0 deletions research/z4-distinct-source-seed55/NOTICE.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,10 @@
# Source notices

This package is distributed under the repository's Apache-2.0 license.

The following unmodified predecessor snapshots are included for exact replay and attribution:

- `inputs/PR1391/CERTIFICATE.json` and `inputs/PR1391/verify.py`, from `CrocSwap/integer-mult-bounds` commit `6ce8017e4c27c8b220b6c3a27a53bb18f138212a`, PR #1391. SHA256 values are recorded in `SOURCES.json` and checked by `verify.py`.
- `inputs/PR1410-README.md`, from commit `5b1c00b9f2770aabaa0085539405757f7818a14a`, PR #1410. Its SHA256 and Git blob are checked by `verify.py`.

PR #1391 and PR #1410 credit the order-three frame and paid output-sign-frame predecessors, PR #1287 by Ross Richardson and PR #1227 by Sagar Thakrar. This package retains those attributions and does not modify their snapshots.
117 changes: 117 additions & 0 deletions research/z4-distinct-source-seed55/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,117 @@
# Phase-indexed order-four echo across distinct sources

**Scoped recursive-interface construction; no new κ or supplier admission.** This package extends the order-four five-child interface of PR #1391 and the same-source, two-root composition of PR #1410. It composes two five-child calls on independent source banks while preserving an arbitrary dirty output and restoring both sources and workspace.

Research base: `4040a0f46d67b41c36a472a7cd4af791fd4c36cc`. Exact PR #1391 and PR #1410 dependency heads and content pins are recorded in [SOURCES.json](SOURCES.json).

The main construction indexes each child's output reader by the carried workspace phase. If the original child reads through `R_i` and advances workspace by `P`, define

```text
C_(i,t)(D,W) = (D + Q^(-t) R_i W, P W), t in Z/4Z.
```

When `W = P^t w`, the child contributes the original `R_i w`, since `Q^(-t) R_i P^t = R_i`. The typed reader is an eight-row signed selector: it changes which output register each ADD targets while retaining the child's eight ADDs and its 24-coordinate `P` update. This is an explicit change to the child contract.

## Exact boundary recurrence

The pinned predecessor has 24 workspace coordinates. The complete `P^r` gate vectors are

| `r` | ADD | NEG | Total |
|---:|---:|---:|---:|
| 0 | 0 | 0 | 0 |
| 1 | 54 | 6 | 60 |
| 2 | 36 | 12 | 48 |
| 3 | 54 | 6 | 60 |

Let `A(r)` be that vector. With starting workspace phase `t`, the entry charge is `A(t)`. After a child at phase `t`, the workspace has advanced once under `P`; reaching the next phase `s` costs `A((s-t-1) mod 4)`. The exact recurrence is

```text
F_0(t) = A(t)
F_(j+1)(s) = min_t [F_j(t) + A((s-t-1) mod 4)]
g(k) = F_k(0)
```

The verifier compares this recurrence with every path for the two fixed sizes: all 1,024 five-child paths and all 1,048,576 ten-child paths. The exact minima are:

- `g(5) = (54,6)`, total 60, on phases `[0,0,1,2,3,0]`.
- `g(10) = (36,12)`, total 48, on phases `[0,1,0,1,2,3,0,1,2,3,0]`.

Let `T_i=(62,6)` be the complete forward cost of child `i`, including its eight data ADDs and 24-coordinate workspace update. The paired child body and literal inverse contribute `2 sum_i T_i`. For one source, the source copy and reset cost `2d` ADDs; for two sources, the entry copy, two-sided handoff, and exit reset cost `4d` ADDs, where `d=24`. The complete cost recurrences are

```text
E_1(k) = 2 sum_(i=1..k) T_i + (2d,0) + 2 g(k)
E_2(k1,k2) = 2 sum_(i=1..k1+k2) T_i + (4d,0) + 2 g(k1+k2)
```

The inverse uses the same exact phase word in reverse and has the same ADD/NEG vector. These equations include every child, source update, phase change, and restoration charged by this finite interface.

For the two-source call, the ten-child path is at phase 3 after the first five children. At that point the workspace contains `P^3(W+X)`. The handoff applies `P^3(Y-X)`, using two ADDs per each of 24 coordinates, for 48 ADDs. The second five children then read `Y`. The complete word is

```text
W += X
phase-typed 10-child forward body, with W += P^3(Y-X) after child 5
W -= Y
literal inverse of the no-handoff 10-child body
```

The resulting map is

```text
(D,X,Y,W) -> (D + sum_(i=0..4) R_i X + sum_(i=5..9) R_i Y, X, Y, W).
```

All 80 basis columns (8 output, 24 coordinates in each of `X`, `Y`, and `W`) are checked exactly. The two source banks are read-only, and arbitrary initial values in all four banks are allowed.

## Complete invoice

The joint typed invoice is

```text
20*(62,6) + (96,0) + 2*(36,12) = (1408,144)
```

for 1,552 primitives and 2,960 data reads. Two separate calls using the same typed child family cost

```text
2 * [10*(62,6) + (48,0) + 2*(54,6)] = (1552,144)
```

for 1,696 primitives and 3,248 reads. The joint typed call saves **144 primitives and 288 reads**. Its exact prefix row-L1 peak is 21, compared with 16 for the two separate typed calls; this is a coefficient-width cost that could affect fixed-precision admission.

For comparison, the unchanged-reader interface from PR #1410 also composes across two distinct sources when charged with the same 48-ADD handoff. Its exact joint invoice is 1,952 primitives versus 2,016 for two separate original calls, saving 64 primitives and 136 reads. Its prefix row-L1 peaks are 42 joint and 30 separate. This is a matched same-contract construction; it is not included in the typed-call saving above.

## Reproduction and evidence

From the repository root, with CPython 3.12:

```sh
python3.12 scripts/research_package.py --package research/z4-distinct-source-seed55
```

Expected verifier marker:

```text
PASS_Z4_DISTINCT_SOURCE_PHASE_TYPED_ECHO g5=60 g10=48 separate=1696 joint=1552 saved=144 reads_saved=288 paths=1024+1048576 basis=80 controls=5
```

The registered replay checks the base and predecessor pins, all twelve `(phase, reader)` identities, unchanged child ADD/NEG counts, both complete 80-column maps, every phase path at both sizes, exact ADD/NEG/read invoices, prefix row-L1 peaks, and five negative controls. The exact evidence is frozen in [RESULT.json](RESULT.json).

## Scope and limits

The phase recurrence is exact for the stated phase-indexed order-four child alphabet. It is not a lower bound for arbitrary ADD/NEG programs. Output-row reindexing is encoded by the static target of each ADD and costs no additional scalar ADD/NEG instruction in this word model. Address calculation, physical row routing, source hosting, tape travel, fixed-precision admission, recursive assembly into a selected supplier, and analytic transfer remain unproved. The row-L1 increase makes precision a live obligation. The selected result at the pinned base remains κ = `836330098538459/500000000000000000`; this package does not change it or claim a new κ.

## Overlap check

The live refresh found several adjacent connector results. PR #1158 gives an exact readout-price census for a four-source/four-output Hadamard connector. PR #1438 and Cinna Davis's Discussion #1412 study a second arbitrary-dirty root and paid full-row transfers for universal five-source masks. PR #1439 studies a persistent `z = z_initial + Bx` boundary across a paid source mutation `x += Cu`, correcting it by `z += BCu`; PR #1442 proves a weighted query optimum under chronological output-delivery cuts. Here the interface keeps one 24-coordinate dirty workspace and changes between two read-only source banks by the explicit 48-ADD delta `P^t(Y-X)` inside a fixed five-child echo. Its separate evidence is the order-four phase-indexed reader identity, exhaustive five- and ten-child recurrence, and complete arbitrary-dirty two-call map/invoice, not a four-source Hadamard census, two-root transfer bound, general mutable-source connector theorem, or delivery-cut optimum. We make no priority claim for those connector mechanisms and retain their distinct scopes.

## Open next questions

Does a named selected supplier realize all four phase-indexed signed target rows with the charged handoff and no binding routing or precision cost? This matters before the operation saving can alter a multiplication profile. First inspect one retained #1094 child matrix and test `Q^(-t) R_i P^t = R_i` for every phase, then price target routing and the prefix envelope. The blocker is that no physical host, tape schedule, or precision-admitted supplier for this paired interface has been identified.

## Credit and license

This work builds on Cinna Davis's PR #1391 order-four child family and PR #1410 same-source persistent boundary, and retains their attribution. Those results extend Ross Richardson's PR #1287 and Sagar Thakrar's PR #1227. The selected-result context is Rohan Arun's PR #1094 on Jeremy Kothe's #1087 construction. The exact PR #1391 certificate and verifier snapshots are preserved under `inputs/PR1391/`; the direct PR #1410 README snapshot is pinned under `inputs/`. See [NOTICE.md](NOTICE.md) for source heads and snapshot hashes.

The overlap review also credits Cinna Davis's PR #1158 for an adjacent four-source Hadamard connector census, woklerx's PRs #1438, #1439, and #1442, and Cinna Davis's concurrent Discussion #1412 for distinct two-root, mutable-source, and chronological-delivery models.

Prepared by **Cinna Davis** with substantial **OpenAI Codex assistance** in derivation, implementation, exact verification, source pinning, and prose. No independent human review or proof-assistant formalization is claimed. New material is Apache-2.0; the project license is retained.
163 changes: 163 additions & 0 deletions research/z4-distinct-source-seed55/RESULT.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,163 @@
{
"assignment_seed": 55,
"assumptions_and_limits": [
"Output-row reindexing is part of the typed ADD target choice and has no additional scalar ADD/NEG instruction in this word model.",
"Address, tape, source-hosting, precision admission, recursive assembly into a selected supplier and analytic transfer are not charged or proved.",
"The finite state recurrence is exact only for the declared order-four phase-indexed child alphabet; it is not a lower bound for arbitrary words.",
"No new kappa or selected-result change is claimed."
],
"base_commit": "4040a0f46d67b41c36a472a7cd4af791fd4c36cc",
"complete_paid_cost_recurrence": {
"child_forward_ADD_NEG": {
"ADD": 62,
"NEG": 6
},
"d": 24,
"literal_inverse": "counted by the factor 2 on every child and phase boundary; inverse leaves preserve ADD/NEG counts",
"single_source": "2*sum_i(T_i) + (2d,0) + 2*g(k)",
"two_source_split": "2*sum_i(T_i) + (4d,0) + 2*g(k1+k2)",
"wrapper_accounting": "single: W+=X and W-=X; two-source: W+=X, two ADDs per coordinate for P^t(Y-X), and W-=Y"
},
"controls": {
"fixed_reader_midpoint_carry": "PASS",
"missing_final_source_restore": "REJECTED",
"missing_literal_inverse": "REJECTED",
"omitted_source_handoff": "REJECTED",
"typed_child_phase_identity_checks": "PASS for 12 (phase,reader) pairs",
"wrong_handoff_phase": "REJECTED",
"wrong_reader_type": "REJECTED"
},
"dimensions": {
"D": 8,
"W": 24,
"X": 24,
"Y": 24,
"total_basis_columns": 80
},
"handoff": {
"ADDs": 48,
"formula": "W += P^t (Y-X) at the 5|5 boundary, t=phase[5]",
"selected_phase": 3
},
"joint_distinct_sources": {
"ADD_NEG": {
"ADD": 1408,
"NEG": 144
},
"map": "PASS on all 80 basis columns",
"prefix_row_l1_peak": 21,
"primitives": 1552,
"reads": 2960
},
"phase_reader": "C_(i,t): D += Q^(-t) R_i W; W += (P-I)W; type t has same ADD/NEG count as C_i.",
"predecessors": {
"PR1391_certificate_sha256": "2a6ac306ddc6f58514b89f680d8ff3027b4195e2754574f8f04ac1f17785c9d3",
"PR1391_head": "6ce8017e4c27c8b220b6c3a27a53bb18f138212a",
"PR1391_verifier_sha256": "c3d8e76dad9b3cc159affa5e56dbf73612eab0c0ce4a4464f523d4cf47669c05",
"PR1410_head": "5b1c00b9f2770aabaa0085539405757f7818a14a",
"PR1410_readme_sha256": "6dda7d5b83b41c71d73711401edbc146271a546a518ef6b038b0bdbb6578d405"
},
"recurrence": {
"F0(t)": "A(t)",
"Fnext(s)": "min_t F(t)+A((s-t-1) mod 4)",
"g10_ADD_NEG": {
"ADD": 36,
"NEG": 12
},
"g10_exhaustive_paths": 1048576,
"g10_path": [
0,
1,
0,
1,
2,
3,
0,
1,
2,
3,
0
],
"g10_total": 48,
"g5_ADD_NEG": {
"ADD": 54,
"NEG": 6
},
"g5_exhaustive_paths": 1024,
"g5_path": [
0,
0,
1,
2,
3,
0
],
"g5_total": 60,
"terminal_phase": 0,
"workspace_phase_costs_ADD_NEG": [
{
"ADD": 0,
"NEG": 0
},
{
"ADD": 54,
"NEG": 6
},
{
"ADD": 36,
"NEG": 12
},
{
"ADD": 54,
"NEG": 6
}
]
},
"same_reader_control": {
"joint_ADD_NEG": {
"ADD": 1768,
"NEG": 184
},
"joint_prefix_row_l1_peak": 42,
"joint_primitives": 1952,
"joint_reads": 3720,
"map": "PASS on all 80 basis columns",
"phase_path": [
0,
1,
2,
3,
0,
1,
2,
3,
0,
1,
0
],
"saving_primitives": 64,
"separate_ADD_NEG": {
"ADD": 1840,
"NEG": 176
},
"separate_prefix_row_l1_peak": 30,
"separate_primitives": 2016,
"separate_reads": 3856
},
"saving": {
"primitives": 144,
"reads": 288
},
"scope": "Exact phase-indexed typed child family for two distinct source banks and the pinned PR1391 five-child reader sequence; not an admitted selected supplier.",
"separate_calls": {
"ADD_NEG": {
"ADD": 1552,
"NEG": 144
},
"map": "PASS on all 80 basis columns",
"prefix_row_l1_peak": 16,
"primitives": 1696,
"reads": 3248
},
"status": "PASS_Z4_DISTINCT_SOURCE_PHASE_TYPED_ECHO"
}
65 changes: 65 additions & 0 deletions research/z4-distinct-source-seed55/SOURCES.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,65 @@
{
"schema_version": 1,
"base_commit": "4040a0f46d67b41c36a472a7cd4af791fd4c36cc",
"files": [
{
"path": "certificates/selected-result.json",
"usage": "provenance",
"git_blob": "ae6a99f28aa468649eb66a361e764d769363bf11",
"sha256": "10b9e340fd8e354a37a0c4813c90e5de1dd42044b5f6aca4a3fc5a42b685fa32"
},
{
"path": "docs/research/agent-task-catalog.json",
"usage": "provenance",
"git_blob": "c05fafaa25ba95b2d5af86b1800427198e3d20eb",
"sha256": "89b5928bc92c81e2647b6d702a7ad85a6795b8f44e2eb267b2ff3145d8f2b2ce"
},
{
"path": "docs/research/community-batch1099-review.md",
"usage": "provenance",
"git_blob": "6cb6ec061a071a06a2d18cd4dc3e9a2b05d28099",
"sha256": "7e76d2c38e5fce30587835b7251df1430488ae4dd372611ef69a3f03f476b6cb"
},
{
"path": "research/community-frontier1094-audit/SELECTED-SOURCE.json",
"usage": "provenance",
"git_blob": "6a75493790b4eff9c4a98051ef8b01d6aad0d5a8",
"sha256": "79b3de5db3dbcb9829789ba58e9e45758926031e2a7d80667a08429d369be53b"
},
{
"path": "research/community-frontier1094-audit/ASSEMBLY.json",
"usage": "provenance",
"git_blob": "5fd676df060f3b450782fdf5ff5f6beb63a5eeb0",
"sha256": "1122cb3a575f8c788345fc1bf4c108ac7c3250fb81a5988671f5eb04b3ab3045"
}
],
"external_sources": [
{
"repository": "CrocSwap/integer-mult-bounds",
"commit": "6ce8017e4c27c8b220b6c3a27a53bb18f138212a",
"path": "research/z4-paired-frame-restoration-seed56/CERTIFICATE.json",
"local_path": "research/z4-distinct-source-seed55/inputs/PR1391/CERTIFICATE.json",
"git_blob": "0132b117d90f1a7a97c40af3e38503c117604e00",
"sha256": "2a6ac306ddc6f58514b89f680d8ff3027b4195e2754574f8f04ac1f17785c9d3",
"role": "exact predecessor child and recurrence certificate"
},
{
"repository": "CrocSwap/integer-mult-bounds",
"commit": "6ce8017e4c27c8b220b6c3a27a53bb18f138212a",
"path": "research/z4-paired-frame-restoration-seed56/verify.py",
"local_path": "research/z4-distinct-source-seed55/inputs/PR1391/verify.py",
"git_blob": "9d866bbf57a6d1f012a21687bde2e231c90efd14",
"sha256": "c3d8e76dad9b3cc159affa5e56dbf73612eab0c0ce4a4464f523d4cf47669c05",
"role": "pinned signed-ADD child word and basis replay"
},
{
"repository": "CrocSwap/integer-mult-bounds",
"commit": "5b1c00b9f2770aabaa0085539405757f7818a14a",
"path": "research/z4-persistent-pair-seed57/README.md",
"local_path": "research/z4-distinct-source-seed55/inputs/PR1410-README.md",
"git_blob": "001929a89cd64f4bbb0d4194c6bde96f45b7ded2",
"sha256": "6dda7d5b83b41c71d73711401edbc146271a546a518ef6b038b0bdbb6578d405",
"role": "direct same-source predecessor and matched baseline context"
}
]
}
Loading
Loading