PaperBack Recovery Frontier — print-and-scan storage research

Project

PaperBack encodes files as pages of small printed dots and restores them from a scan. Over three weeks we ran controlled sweeps on two office printers — a Canon iR-ADV DX CS840i and an HP LaserJet 4350dtn — scanning every page with an Epson V700 and decoding each attempt as an independent trial. The result is a measured map of which dot densities, printed fill levels, and scan factors actually recover data.

Problem

Can an ordinary letter page on ordinary office hardware hold real data reliably — and at what dot geometry? Below the recovery frontier, pages restore perfectly; above it, they fail. The frontier's exact position depends on printer, fill, and scan resolution, and until now it was unknown.

Audience

Anyone who needs printer-only cold backups: archivalists, air-gapped systems, tinkerers, and anyone who wants a recovery path that needs nothing but a printer, a scanner, and paper.

What AutoClaw did

AutoClaw designed the sweep matrices, ran every print/scan/decode trial through a patched PaperBack control build — an expanded command set plus a JSON control interface for agentic access to the software — analyzed all decode results, produced every chart and report, identified and corrected a capture-labeling error, and rebuilt the binary from the patched sources to confirm reproducibility.

Result

Broad, reliable region below ~80 dpi. Moderate fills restore at scan factors as low as x2.0.
A decisive zone near 100 dpi. Printed fill and scan resolution become the deciding variables.
A narrow frontier at 120 dpi and above. Recovery is possible but printer-specific and fill-hungry.
A working, reproducible toolchain. Patched control binary v8.2, rebuilt from the patched sources at identical size, plus per-block quality telemetry for every trial.

Canon recovery frontier by scan factor. Teal = clean restore; rust = failed; the dark line is a conservative clean edge that never crosses an observed failure.

HP read outcome map across three sweeps — including the original sweep whose raw scans were later lost but whose decode verdicts survive.

Report

The full report — method, outcome maps, recovery heatmaps, the software's own per-block quality views, and print-versus-scan block comparisons:

Read the current report (PDF)

Previous editions, kept for the record:

Figures

Canon outcomes

HP outcomes

Canon frontier slices

HP frontier slices

Canon quality heatmaps

HP quality heatmaps

PaperBack's own quality maps

Print vs. scan, same block

Raw sweeps data

Every campaign is browsable file-by-file — raw scans, print bitmaps, decode logs, quality maps, and notes — at the raw data page; a machine-readable manifest of every file is at raw/index.json. Full-resolution print masters are included as lossless PNG re-encodes of the original BMPs (pixel-identical), so everything fits the 25 MiB per-file hosting limit.

Data

Everything charted here is backed by distributed data files.

canon-outcome-map-final.csv — every Canon candidate: sweep, geometry, floor, verdict
hp-outcome-map-final.csv — every HP candidate
outcome-map-canon-all-20261019.xlsx — Excel workbook with a live, formatted outcome chart
cases-registration.json — print/scan block rectangles, registration shifts, correlation scores
chart-manifest.json — figure inputs and the conservative frontier rule
SHA256SUMS.txt — checksums for report, figures, data, and downloads

Downloads

paperbak-ctl-v8.2-20261017-release.zip — a patched version of Ole Yuschuk's PaperBack with an expanded menu selection and a JSON control interface for agentic access to the software (v8.2 release)
PaperBak-ctl-20261017-v82i-repro.exe — the same v8.2 binary rebuilt from the patched sources (identical size)
paperback-source-bundle.zip — pristine PaperBack 1.10 sources + full patch + build kit + build guide

Checksums for every distributed file: SHA256SUMS.txt. See README-downloads.md for what each artifact is and how to rebuild.