Complete EXP1 v1.1 campaign and publish final evidence

.gitattributes

@@ -0,0 +1,4 @@
+# Preserve the exact generated bytes whose SHA-256 identities are recorded in
+# the EXP1 v1.1 freeze and output manifests, independent of core.autocrlf.
+/ephemeral-sandbox-v1/experiments/analysis/exp1-v11-freeze-record-*.json -text
+/ephemeral-sandbox-v1/experiments/analysis/final-v11-*-tables-*/** -text

.gitignore

@@ -14,6 +14,28 @@
 *~
 .DS_Store
 
+# Python bytecode and tool caches
+__pycache__/
+*.py[cod]
+.pytest_cache/
+.mypy_cache/
+.ruff_cache/
+.coverage
+htmlcov/
+
 # Paper-local benchmark runtime state and Python environment
 /ephemeral-sandbox-v1/.benchmark-state/
 /ephemeral-sandbox-v1/.venv/
+
+# Local EXP1 runtime state and large/reproducible generated artifacts
+/ephemeral-sandbox-v1/tmp/
+/ephemeral-sandbox-v1/experiments/runs/
+/ephemeral-sandbox-v1/experiments/diagnostics/
+/ephemeral-sandbox-v1/experiments/exp1-pre-edit-baseline.json
+
+# Exploratory EXP1 projections and table regenerations. The final-v11 table
+# trees and freeze record are intentionally versioned and are not matched here.
+/ephemeral-sandbox-v1/experiments/analysis/paper-*-expanded*.json
+/ephemeral-sandbox-v1/experiments/analysis/paper-*-expanded*.stderr.txt
+/ephemeral-sandbox-v1/experiments/analysis/pilot-*.json
+/ephemeral-sandbox-v1/experiments/analysis/pilot-*-tables-*/

ephemeral-sandbox-v1/ARTIFACTS.md

@@ -0,0 +1,42 @@
+# Artifact guide
+
+## Scope
+
+This artifact package supports a narrow source-and-treatment claim: Ephemeral
+Sandbox implements private LayerStack workspace sessions and controlled
+filesystem-delta publication; the paper reports a single frozen local treatment
+of startup, public-CLI, and selected resource observations. It is not a
+comparative benchmark or a general coding-agent productivity evaluation.
+
+## Immutable evidence
+
+| Artifact | Location | Integrity anchor | Use |
+| --- | --- | --- | --- |
+| Eligible experiment archive | `experiments/runs/019fb86c-096e-7589-a0a4-a6d6ef5d7f8b/` | content tree `606863f2843a7b19f04e27e2ba5b736d544dd143f56f6d3626611cb29bb44986` | Raw reports, manifests, and run facts. |
+| Frozen table output | `experiments/analysis/final-v11-019fb86c-tables-a/` | output tree `27b53ee5acc049899b4e5821f8d92b14488c7d08ed076ba379af4799c765ad04` | Source for manuscript displays. |
+| Numeric registry | `numeric_evidence.json` | Project-side selector projection of the frozen provenance CSV | Verifies every manuscript number in result tables. |
+| Tagged product source | `plan/source_revalidation.md` | `paper-v1.1-freeze` peeled commit `5c48dae10847fb9e46ba2bea7675bcf2f5a6f4c8` | Source-defined mechanism and interface. |
+| Citation lock | `citation_requests.json`, `citation_lock.json` | Terminal arXiv/Crossref metadata checks | Scholarly bibliography provenance. |
+
+The archive and frozen table output are read-only evidence. The manuscript
+generator reads them but never writes inside either directory.
+
+## Paper-facing outputs
+
+- `main.tex` and `sections/` — buildable manuscript source.
+- `sections/generated_results_tables.tex` — generated LaTeX tables; do not edit
+  it directly.
+- `scripts/generate_latex_results.py` — deterministic table and numeric-registry
+  projection.
+- `scripts/generate_bibliography.py` — bibliography projection from the verified
+  citation lock.
+- `main.pdf` — locally built review PDF.
+- `literature/` — inventory, comparison matrix, and claim-positioning record.
+- `figures/concept-figure-review.md` — final figure QA and explicit waivers.
+
+## Deliberate exclusions
+
+No experiment rerun, new benchmark, external baseline, fault campaign,
+security evaluation, or multi-agent productivity study was performed during
+paper completion. Qualification, smoke, pilot, earlier protocol, and broader
+research-plan material remain excluded from result claims.

ephemeral-sandbox-v1/BUILD.md

@@ -38,7 +38,10 @@ latexmk -pdf -interaction=nonstopmode -halt-on-error main.tex
 
 - PDF: `main.pdf`
 - Primary LaTeX log: `main.log`
-- Skill build-attestation log: `plan/pw0-build-attempt.log`
+- PW0 skill build-attestation log: `plan/pw0-build-attempt.log`
+- PW1 skill build-attestation log: `plan/pw1-build.log`
+- PW2 skill build-attestation log: `plan/pw2-build.log`
+- PW3 skill build-attestation log: `plan/pw3-build.log`
 - Auxiliary files: `main.aux`, `main.bbl`, `main.blg`, `main.fdb_latexmk`, `main.fls`, and `main.out` when produced by the installed toolchain
 
 ## PW0 recorded build
@@ -52,3 +55,65 @@ The skill build recorder executed the declared command successfully on 2026-07-3
 - PDF size and format: 192,209 bytes, four US-letter pages, PDF 1.7.
 
 The log contains no build errors, undefined citations or references, missing files, or overfull boxes. It contains five underfull-box warnings in the provisional source-derived cost table and an empty-bibliography warning. The latter produces an intentionally empty References page while `references.bib` remains comment-only.
+
+## PW1 recorded build
+
+After drafting Sections 2 and 3, the skill build recorder executed the same declared command successfully on 2026-07-30 at `2026-07-30T00:48:19.855325+00:00`.
+
+- Toolchain: Latexmk 4.88 (2026-03-09); pdfTeX 3.141592653-2.6-1.40.29 (TeX Live 2026); BibTeX 0.99e (TeX Live 2026).
+- Exit code: 0.
+- PDF: `main.pdf`, 209,575 bytes and six US-letter pages.
+- PDF SHA-256: `ba4963d3d5f6352e1829946290265671599432e9984e301d5626de7316435327`.
+- Attested input SHA-256: `8a9d0f97487ccf937efd82eb6a5726b2c7c8b5e3e31be1434945826a00731708`.
+- Build log: `plan/pw1-build.log`.
+- Build-log SHA-256: `719696c0aa0efb9fce4796efaab3cce5b27dd17563c4b60d7f4951b539aa7f30`.
+- Attestation: executed.
+
+The parsed LaTeX log contains zero errors, emergency stops, undefined citations, undefined references, missing files, or overfull boxes. The five underfull boxes remain confined to the pre-existing provisional cost table in Section 7, and the comment-only bibliography still produces the expected empty-bibliography warning.
+
+## PW2 recorded build
+
+After drafting Sections 4 and 5, the skill build recorder executed the same declared command successfully on 2026-07-30 at `2026-07-30T01:10:40.550071+00:00`.
+
+- Toolchain: Latexmk 4.88 (2026-03-09); pdfTeX 3.141592653-2.6-1.40.29 (TeX Live 2026); BibTeX 0.99e (TeX Live 2026).
+- Exit code: 0.
+- PDF: `main.pdf`, 221,013 bytes and eight US-letter pages.
+- PDF SHA-256: `801ac91c302ae3ea7d5827d34dd4da09278e8f537409e3426fd4ff30c8ed36e7`.
+- Attested input SHA-256: `36877891dbce1e066589d8a295e436654ccb097970620003cdaf45871f74311b`.
+- Build log: `plan/pw2-build.log`.
+- Build-log SHA-256: `9fc4261ab327472e004f4f62466bf2266218d1b17a66cc5b852ee5dbe7b23265`.
+- Attestation: executed.
+
+The final LaTeX pass contains zero errors, emergency stops, undefined citations, undefined references, missing files, or overfull boxes. The five underfull boxes remain confined to the pre-existing provisional cost table in Section 7, and the comment-only bibliography still produces the expected empty-bibliography warning.
+
+## PW3 recorded build
+
+After completing Sections 6--7 and integrating the four author-supplied
+concept figures, the skill build recorder executed the declared command
+successfully on 2026-07-30 at
+`2026-07-30T03:22:41.491473+00:00`.
+
+- Toolchain: Latexmk 4.88 (2026-03-09); pdfTeX 1.40.29 (TeX Live
+  2026); BibTeX 0.99e (TeX Live 2026).
+- Exit code: 0.
+- PDF: `main.pdf`, 4,485,845 bytes and 14 US-letter pages.
+- PDF SHA-256:
+  `bf893a5ac17e396233232b18d552e77ddddfc2bcba50786d457fd58db71604eb`.
+- Attested input SHA-256:
+  `e31aa83e7db9c45cdf0c3c1f731fe66670fc6641dc20d755704a7673a552e18a`.
+- Build log: `plan/pw3-build.log`.
+- Build-log SHA-256:
+  `39a836a6226fc9b8007dd23fda5deb9e2cd8bf3c453faf17539c9b356871e59f`.
+- Attestation: executed.
+
+The final LaTeX pass contains zero errors, emergency stops, undefined
+citations, undefined references, missing files, or overfull boxes. Seven
+underfull boxes occur in the two narrow Section 7 tables, and the comment-only
+bibliography produces the expected empty-bibliography warning.
+
+All 14 pages were rendered at 144 dpi for visual review, with an additional
+96 dpi grayscale pass. The four figures appear on pages 3, 6, 8, and 10; both
+Section 7 tables remain after their introducing prose. No clipping, overlap,
+broken glyph, or page-number defect was found. The figures remain unchanged
+drafting-stage review assets, and their recorded topology, resolution,
+grayscale-contrast, and style-family disparities remain PW7 work.