Skip to contents

Reproduction proof — msuiter paper lane (PP-1)

Source: vignettes/articles/reproduction-proof.qmd

This document is the one-command reproduction face for every sealed number quoted in vignettes/articles/paper-draft.qmd. It is a verification artifact, not a regeneration claim: everything below re-derives quoted values from committed bytes and committed witnesses, on synthetic data only (G0 master seed 20260928). Regeneration of a campaign from scratch is operator-gated work and is printed (never executed) by --full.

Governing house laws restated where they bind this face:

  • Synthetic-only law — no artifact here touches a real-data cohort; the D3 law keeps COSMIC numerics out of the repo entirely.
  • SIM-provisioning trust law — the R-side scenario faces must run against a build OF THIS TREE, never an ambient stale install. Hence MSUITER_REPRO_LIB.
  • Post-cap witness law — timing rows speak the floor-table epoch f3e894a6933a2ffb (pinned at bench/m7/tests/test_p6_floor_governance.py:156); pre-cap campaigns are quoted with an explicit epoch-mismatch label.
  • Dead-metric law — the yardstick pin verifier and the DOI scanner carry planted-corruption controls that must fire for the face to pass; a silent green is a fail.

0. One command

Provision a library holding a build of the tree (once per tree revision):

R CMD INSTALL --library=<stage-dir> /path/to/this/tree

Then:

MSUITER_REPRO_LIB=<stage-dir> bash inst/repro/reproduce.sh --verify

Attested run of this face, this lane, 2026-10-31 (wall 176 s, rc=0):

FACE m7-tests                 rc=0  ...  PIN  m7-tests   ok  364/364 tests passed.
FACE o5-parity                rc=0       PIN  o5-parity  ok  lines=147 breaches=147
                                          PIN  o5-parity  ok  epoch=f3e894a6933a2ffb
FACE bootstrap-campaign       rc=0       PIN  ok  "ok": true
                                          PIN  ok  claimed 15.36600758177 vs
                                                  recomputed 15.36600758176629
                                          PIN  sealed-ratio   ok  1.536600758177e+01
                                          PIN  sealed-verdict ok MEASURED-false 15.37x
                                          PIN  sealed-secondary ok 14.2209x
FACE yardstick-pins           rc=0
FACE commit-objects           rc=0
PIN  p1-table                 ok  FAIL — 18 offending class(es)
FACE scen-atlas               rc=0  PIN  ok  OVERALL PASS
FACE scen-hrd                 rc=0  PIN  ok  OVERALL PASS
FACE scen-treatment           rc=0  PIN  ok  OVERALL PASS
PIN  seed-master              ok  seed: 20260928 (G0 master
PIN  seeds-bank               ok  4/4 committed Seeds.txt present
REPRODUCE-VERDICT PASS

1. What each face proves, and its committed anchor

Face Command Sealed claim re-derived Anchor (bytes on the branch)
m7-tests python3 -m tests.run_all (in bench/m7) governance/unit surface green bench/m7/tests/*, spoken at :156 epoch pin
o5-parity python3 -m tools.parity_recompute O5 census lines=147 breaches=147, post-cap epoch bench/m7/tools/parity_recompute.py, witnesses under bench/results/p6_regen_20261023-044026/
bootstrap-campaign python3 -m tools.verify_campaign --campaign ../results/bootstrap_vs_sigminer_20261013T175859Z --host <seal> --hw-profile <seal> 15.3660075817662x re-derived from arm witness bytes; MEASURED-false stands; host/hw pins re-checked bench/results/bootstrap_vs_sigminer_20261013T175859Z/{result.md,seal.json}
(pins) sealed-ratio/verdict/secondary byte greps 1.536600758177e+01, MEASURED-false 15.37x, 14.2209x .../result.md:4,5; vignettes/articles/performance.qmd:88
yardstick-pins python3 tools/verify_yardstick_pins.py ratio 1.2323046734094532, iters/seed/threads/k/shape/versions, fixture digest, seconds-pair consistency; planted-corruption control FIRES bench/results/p6_regen_20261016-011928/nmf_yardstick_authoritative.json, bench/m7/adapters/yardstick_cell/matrix.csv
commit-objects git show -s --format=%B greps P10 landing ratio 2.3004 and P-1 MEASURED-FALSE (23.7-32.2x) live in commit-message bytes commits 2296f49, 8213192
p1-table byte grep sealed P-1 gate table FAIL — 18 offending class(es) bench/results/p6_regen_20261016-011928/result.md:6
scen-atlas / scen-hrd / scen-treatment Rscript bench/scenarios/run_{atlas,hrd,treatment}.R CS-1/CS-2/CS-3 self-adjudicating scenarios OVERALL PASS, rc is the law bench/scenarios/* (PR #56 landing 5530af6)
seeds-bank file presence 4/4 campaign Seeds.txt committed listed dirs in reproduce.sh Face 8

The exact sealed-number citations, their citation classes, and their epoch labels are in the draft’s Table 1/Table 2 and Appendix A; this file only fixes how each is re-verified.

2. R-side environment freeze (renv)

inst/repro/renv.lock — Posit renv JSON lockfile: R 4.5.2 stamp, CRAN repository, 111 pinned packages. Provenance (attested, banked in the lane manifest): closure walk over DESCRIPTION metadata from the seeds {S7, rlang, pkgload, testthat, digest, jsonlite, withr, R6, sigminer} over the provisioning library: 118 closure entries -> 111 lockfile pins; the 7 dropped entries are recorded by the generator as missing=['furrr','ggplot2','graphics','grid','rstatix','sigminer','splines'] — graphics/grid/splines ride the base distribution, and furrr/ggplot2/ rstatix/sigminer were local installs without a Repository: field. For a clean-room replay those four resolve from CRAN/Bioconductor under the frozen R 4.5.2 stamp (sigminer 2.3.3 is the ground-truth peer identity, not a build dependency of msuiter itself).

Readback attestation (generator run, banked): LOCK-PARSE ok; R=4.5.2; packages=111 and READBACK 111 R 4.5.2.

Deviation, recorded honestly: the classic Ushey-renv API (renv::write_renv/restore_renv) is no longer the CRAN renv namespace — CRAN renv is Posit’s package, whose successor rrenv is not reachable from this box’s configured mirrors. The freeze therefore uses the Posit API (lockfile_create / lockfile_write / lockfile_read / restore), which serves the same purpose: one lockfile, byte-pinned, readback-attested.

Replay: renv::restore(lockfile = "inst/repro/renv.lock", library = <dir>, prompt = FALSE) (printed, not executed, by --full).

3. Container freeze

Committed Dockerfiles: bench/m7/adapters/Dockerfile.sigprofiler, Dockerfile.sigminer, Dockerfile.nmf_yardstick (build contexts printed by --full). Image identity is attested by the committed docker inspect records in bench/results/docker_m7_PILOT_20261008-200152/image_inspect_*.json — repo digest pins, not floating tags.

4. Seed regime

Master seed 20260928 (G0 synthetic regime) is spoken by the bootstrap seal (result.md, PIN seed: 20260928 (G0 master). Per-campaign seed banks: Seeds.txt committed under the bootstrap dir, both p6_regen dirs, and the docker_m7_degraded dir; Face 8 refuses if any goes missing.

5. What this face deliberately does not claim

  • No real-data result anywhere (G0-PRIMARY gated; the draft prints no “first coverage-measured” claim outside its gate block).
  • No <=2x peer-parity row: the O5 census is quoted AS A CENSUS (147/147 breach vs the compiled yardstick); the parity headline stays behind CAMPAIGN-R exactly as the commit bytes of 2296f49 state it.
  • No signeR comparison (PI-C unverified): peer slots (a)/(b) only.
  • No campaign regeneration: --full prints operator recipes (bootstrap one-key shot, container builds, renv restore) and runs none of them.

6. Attestation bank

Per-face logs of the attested PASS run and of each tooling pass are banked in the lane directory (logs/reproduce-verify-r4.txt plus the per-face /tmp/msuiter-reproduce-* logs at run time); the DOI census run (logs/doi-repin-final3.txt) and lockfile generation log (logs/renv-lock-gen.log) accompany them. The lane manifest maps every file to its sha256 and its one-line claim.