Repository navigation
Releases
0.3.14: improved curved QR recovery
QR 0.3.14 recovers more curved codes with a bounded quadratic fit to alignment measurements. Earlier successful reads keep priority, and format, version and payload validation remain strict. The recovery is available in TypeScript, optional WASM, Rust and Kotlin.
| Accuracy scope | 0.3.13 | 0.3.14 | ZXing-WASM 3.1.7 |
|---|---|---|---|
| Combined synthetic curved images | 718/1,376 | 861/1,376 | 646/1,376 |
| Separately frozen 256-image subset | 65/256 | 118/256 | 77/256 |
| Original photographs | 180/184 | 180/184 | 171/184 |
| Curved sheets | 10/18 | 12/18 | — |
That is 143 additional exact reads in the curved comparison, with all earlier successful results preserved and no wrong QR payloads observed. All 64 negative controls remain rejected. ZXing still reads 14 curved frames QR misses; these dataset results are not a universal accuracy claim.
The twelve sampled new recoveries increase WASM median latency by 0.18–4.27% and TypeScript by 0.66–6.06% on the M4. The packaged WASM is 261,831 bytes, up 8,908 bytes. QR remains faster on the fixed clean/blank frames; random-noise rejection remains about 2.6× slower than ZXing.
Validated across the three ports, packaged TS/WASM parity, browser/worker/session use, 2,944 related photo views and repeated memory use. The specification, install examples and AI-readable documentation are updated.
Benchmark inputs, raw results and limits · Documentation · Choosing QR
Published and publicly verified on npm, JSR, crates.io and Maven Central. Documentation, docs.rs, jsDelivr and unpkg copies were also verified.
Release verification receipt records the exact source revision, seven successful CI/release workflows and six public readback receipts. npm provenance and installed API, JSR native Deno/npm compatibility, signed Maven artefacts and an independent consumer, crate checksums/source, and public documentation/CDN bytes all passed.
QR 0.3.13: selected mesh recovery
QR 0.3.13 improves recovery of curved QR codes using an already sampled alignment mesh.
Some curved QR images produce a usable direct alignment mesh, while the final refined grid still fails validation. Retain the existing mesh from the same invocation that selected the rediscovered header, then try it once under that strict hypothesis only after every earlier path fails. Preserve version, Reed–Solomon and segment checks and earlier successful results.
The packaged TypeScript/WASM reads 653/1,120 evaluated curves, up from 648/1,120 in 0.3.12, compared with ZXing-WASM 3.1.7 at 569/1,120. Five new reads include one in a separate 128-image set frozen before evaluation. Previous full successful results, inputs and search counts remain unchanged across the qualification. Rust and Kotlin implement the same policy; SPEC step 20 and two generated conformance frames describe it. No image sampling or second looks are added. The retained-grid bound is 281,961 module bytes, excluding active buffers and metadata; packaged WASM grows by 404 bytes to 252,923.
Validation: 511 TypeScript tests with required photographs; 123 Kotlin tests; 49 Rust tests plus a doctest under default/all-feature configurations; 84 standalone WASM frames/672 counters, 1,745 option comparisons, 148 encoder vectors, grid/render parity and a stable 1,000-round capacity check. Final packaged parity covers 1,120 curves, 888 regressions including 64 negative controls, 2,944 photo views and 18 sheets. Browser, worker, character-set and session/reset checks pass.
Across two paired M4 runs, the five new reads change WASM median latency by 0.19–0.89% relative to the earlier misses. The two selected-mesh cases that still fail to decode change by -0.24–0.40%; earlier successful curved cases change by -0.45–0.84%. These measurements do not establish a general speed improvement or a worst-case bound.
The paired pure-TypeScript clean-frame median rises from 3.785–3.824 ms to
3.968–4.017 ms (3.77–6.15% per fork); blank-frame medians rise from
2.937–2.953 ms to 3.077–3.080 ms (4.32–4.78%). These measured regressions
are part of the accuracy trade-off. Their cause is not established by these
timings; no general TypeScript speed improvement is claimed.
Prepublication passed: documentation freshness, eight pages’ metadata and landing-page layout at 390/1280 px; isolated npm and Maven consumers; packaged crate library/doctest checks and Rust API documentation with warnings denied; JSR strict dry run, source consumer and actual publisher archive captured on loopback. Public registry consumers remain a post-publication verification step.
Available on npm and JSR as @forgesworn/qr@0.3.13, crates.io as forgesworn-qr 0.3.13, and Maven Central as io.github.forgesworn:qr:0.3.13. Published npm/JSR/Maven consumers, crate source/checksum and docs.rs, thirteen documentation files and six CDN files have been verified.
Complete release verification links the seven successful CI/release workflows and six public verification reports. All seven attached reports were downloaded again and matched against the local evidence.
v0.3.12
Recover eleven more curved QR images by reading a strict header hypothesis from an already sampled refinement grid. Earlier successful results keep priority; version, Reed–Solomon and segment validation remain unchanged.
Across eight evaluated curved-image sets, QR 0.3.12 reads 571/992 against 500/992 for ZXing-WASM 3.1.7. The new 128-image set was frozen before evaluation and is now regression data. These finite results do not establish universal superiority. Photograph accuracy remains 180/184; all 64 negative controls are rejected.
TypeScript, Rust and Kotlin implement the same bounded recovery. The change adds no image sampling, maps or second looks. It may retain one additional grid and perform one additional strict symbol validation after all earlier paths fail.
Across two paired M4 runs, the five measured new reads change WASM median latency by -0.34–0.32% relative to their earlier misses. The three cases with a selected header that still fails to decode change by -0.58–0.01%; earlier successful curved cases change by -0.88–1.46%. These measurements do not establish a general speed improvement or a worst-case bound.
Install from npm (@forgesworn/qr), JSR (@forgesworn/qr), crates.io (forgesworn-qr) or Maven Central (io.github.forgesworn:qr) at version 0.3.12. All four registries are published and verified, including installed npm, native Deno, npm compatibility and independent Gradle consumers. Documentation and CDN files match the release, and the crate archive checksum and sources are verified. See the attached release verification receipt and its six linked public readbacks.
See choosing QR for fit and limitations and performance evidence for exact sources, inputs and measurements.
QR 0.3.11: recover alternate readings of curved codes
QR 0.3.11 recovers two additional difficult curved codes by reading an already sampled grid under one retained alternative orientation and format hypothesis. Earlier successful results keep priority and final validation remains strict.
Across seven measured curved-image sets, QR reads 502/864 against ZXing-WASM 3.1.7's 447/864. The separate 128-image set frozen before this policy's evaluation remains at 62 reads against 57. All previous reads and search counters are preserved in the regression checks. These finite datasets do not establish universal accuracy.
The change is implemented in TypeScript, Rust and Kotlin. It adds no image sampling or secondary attempt; at most two extra strict symbol validations and one saved grid pair are allowed. The limits remain eight second looks, 720 maps and 11,500,000 alignment probes.
WASM median changes for the two new reads range from +0.22% to +0.87% relative to the earlier misses. Existing successful-case median changes range from -0.39% to +0.81%. These measurements do not establish a general speed improvement or the worst-case extra symbol-validation cost.
Install from npm as @forgesworn/qr@0.3.11, JSR as @forgesworn/qr@0.3.11, crates.io as forgesworn-qr@0.3.11, or Maven Central as io.github.forgesworn:qr:0.3.11.
Accuracy, timing evidence and reproduction
All four registry publications, installed consumers, documentation and CDN bytes are verified. Release verification receipt.
QR 0.3.10: recover conflicting-version curved codes
Recover two more curved images by retaining one map with an exact version word and a weak contradictory peer for strict late refinement. Earlier successful results and metadata keep priority; final decoding checks remain unchanged.
| Dataset | 0.3.9 | 0.3.10 | ZXing-WASM 3.1.7 |
|---|---|---|---|
| Earlier distortions,128 images | 106 | 108 | 106 |
| Newly frozen set,128 images | 67 | 67 | 53 |
| Curved sheets,18 images | 9 | 10 | Not measured |
All earlier successful results remain unchanged in the evaluated sets, with no wrong QR payload observed. QR leads on all six measured distortion sets; this does not establish universal superiority or accuracy on every camera.
All 506 TypeScript tests with required photographs, Rust tests and Kotlin tests pass. The 79-frame conformance pack preserves all 77 previous entries. Packaged WASM grows by 2,813 bytes to 240,027 bytes. Maximum recovery work becomes 8 attempts and 11,500,000 alignment probes, with the 720-map cap unchanged.
See performance results and limitations, reproduction evidence and the distribution guide.
Published and verified on npm, JSR, crates.io and Maven Central. Documentation and both CDNs match the released files. See the attached release verification receipt and its six public readbacks for workflow, source and consumer evidence. The npm consumer check passed on retry after a registry propagation race; the package was uploaded once.
The two new recoveries cost 19.99–23.21% more in WASM than the earlier failed attempts. Five qualifying images that still fail cost 16.76–29.45% more. Earlier curved successes have median changes from -1.09% to +0.75%. The additional recovery is an explicit accuracy/time trade-off.
QR 0.3.9: late refinement for curved maps
Recover curved maps that previously received ordinary refinement without the late alignment pass. One strict retry runs only after all earlier paths fail; existing successful results and metadata retain priority.
| Dataset | 0.3.8 | 0.3.9 | ZXing-WASM 3.1.7 |
|---|---|---|---|
| Earlier 128 distortions | 103 | 106 | 106 |
| Previous 128-image holdout | 62 | 63 | 52 |
| Earlier separate 128-image holdout | 71 | 72 | 67 |
| New frozen 128-image set | 79 | 80 | 70 |
| Curved sheets, 18 images | 6 | 9 | Not measured |
The 96-image distortion set stays at 48 correct reads against ZXing’s 42. No earlier successful result changes, and no wrong QR payload is observed in these finite datasets. This does not establish universal superiority over ZXing.
TypeScript, Rust, Kotlin and packaged WASM remain aligned. All 498 TypeScript tests and the 77-frame conformance pack pass; all 74 previous conformance entries remain unchanged. The packaged WASM is 237,214 bytes (+2,423 bytes). The maximum becomes seven refinement attempts, trading extra work on some difficult failures for accuracy. Newly recovered reads take 13–18% longer in WASM than the previous failed reads; routine WASM medians are essentially unchanged. TypeScript clean-frame p95 rises by 0.89–1.09 ms in the paired runs.
See performance results and limitations, reproduction evidence and the distribution guide.
Published and independently verified on npm, JSR, crates.io and Maven Central. Documentation and both CDN copies match the release bytes. All seven workflow runs passed; npm provenance, JSR native/compatibility consumers and signed Maven artefacts with a separate consumer were checked.
Release verification receipt links the six public readbacks.
QR 0.3.8
Recover curved QR codes previously excluded by conflicting format headers. One candidate receives a final strict retry after all earlier paths fail; existing successful results keep priority.
| Distortion set | 0.3.7 | 0.3.8 | ZXing-WASM 3.1.7 |
|---|---|---|---|
| Evaluated 96 images | 44 | 48 | 42 |
| Earlier 128 images | 89 | 103 | 106 |
| Previous 128-image holdout | 53 | 62 | 52 |
| Separately frozen 128 images | 58 | 71 | 67 |
Curved sheets improve from four to six reads out of 18. Earlier complete results are preserved across the evaluated sets, with no incorrect QR payloads found. This is an accuracy/time trade-off: the maximum rises to six refinement attempts. These finite datasets do not establish a universal advantage over ZXing.
TypeScript, Rust, Kotlin and packaged WASM remain aligned. All 495 TypeScript tests and the expanded 74-frame conformance pack pass; packaged WASM matches complete TypeScript results across the 888-case regression, 480 distortions, 2,944 transformed photos and 18 sheets. The WASM asset grows by 250 bytes to 234,791 bytes.
See performance results and limitations, reproduction evidence and the distribution guide.
Version 0.3.8 is published and verified on npm, JSR, crates.io and Maven Central. The documentation site and CDN copies match the release. Seven JSON verification receipts are attached below, covering workflow results, public artefact checks and the applicable consumer checks.
v0.3.7 — bounded curvature recovery
Recover additional curved QR codes by allowing one final strict refinement after all earlier recovery paths fail. The candidate must already be retained with exact version evidence and must not have been attempted at full size. Previous successful results and metadata keep priority.
| Accuracy set | 0.3.6 | 0.3.7 | ZXing-WASM 3.1.7 |
|---|---|---|---|
| Evaluated distortions | 37/96 | 44/96 | 42/96 |
| Earlier distortions | 81/128 | 89/128 | 106/128 |
| Separate set frozen before evaluation | 48/128 | 53/128 | 52/128 |
| Earlier curved sheets | 3/18 | 4/18 | Not run here |
| Transformed photographs | 2,872/2,944 | 2,872/2,944 | Not run here |
No previous read or metadata is lost and no wrong QR payload appears in these checks. The separate set is now regression data. ZXing remains substantially ahead on the earlier 128 distortions; these results do not establish a universal accuracy lead.
TypeScript, Rust, Kotlin and packaged WASM agree. Three new shared fixtures cover normal, inverted and mirrored fifth-attempt recovery. The maximum rises explicitly to five refinement attempts and 7,187,500 alignment probes; the original four-attempt order and map cap remain unchanged.
Two paired M4 runs keep main-noise WASM medians within 0.5% and clean within 0.4%. Four newly recovered curved examples cost 13.4–21.2% more than their earlier misses; the tested persistent curved miss costs 19.7–21.0% more. This release trades additional work on difficult frames for measured accuracy.
See the performance report, source-bound evidence and reproduction commands and installation guide.
Publication is verified across npm, JSR, crates.io, Maven Central, the documentation site and both CDNs. The release verification receipt links all seven successful workflows and six public readbacks. npm verification passed on its second attempt after the public archive became available; the package was uploaded once.
Embedded Rust QR encoder 0.1.0
First release of forgesworn-qr-embedded, a standalone dependency-free no_std QR encoder.
- QR versions 1–8 and all four error-correction levels.
- Raw-byte and single-mode text APIs with 602 bytes of caller-owned encoding buffers.
- Optional
optimaltext segmentation with an explicit 980-byte planner. - Packed module access and documented buffer wiping on ordinary returns and symbol drop.
- Shared specification, standard tables and conformance checks with ForgeSworn QR, including all 82 applicable ZXing matrices and decode round trips.
- Rust 1.87 support, zero-allocation checks, isolated package-consumer checks and embedded target builds.
Install with cargo add forgesworn-qr-embedded@0.1.0; enable optimal only when mixed-mode planning is wanted.
This crate has its own version lifecycle. Existing npm, JSR, Kotlin and full Rust packages remain on their existing versions. Heartwood adoption is opt-in; on-board scan, latency and task high-water measurements remain pending.
QR 0.3.6 — deferred mesh recovery
Recover curved QR codes whose alignment mesh was readable before timing and edge corrections displaced it. Saved meshes are checked only after all previous decoder paths fail, preserving existing successful results and metadata.
| Accuracy set | 0.3.5 | 0.3.6 |
|---|---|---|
| Evaluated distortions | 30/96 | 37/96 |
| Newly frozen distortions | 78/128 | 81/128 |
| Earlier curved sheets | 2/18 | 3/18 |
| Transformed photographs | 2,872/2,944 | 2,872/2,944 |
No lost reads or incorrect QR payloads occur in these sets. ZXing reads 42/96 and 106/128 respectively, so significant curvature limitations remain. The new 128-case set is now regression data.
TypeScript, Rust, Kotlin and packaged WASM agree, with shared fixtures covering reduced coordinates and inverted polarity. At most four retained grids add bounded sampling, memory and final validation work; map, alignment-search and second-look limits stay unchanged.
Two paired M4 runs show 3.0–3.8% extra WASM time on the tested persistent curved miss and 1.3–2.6% on an existing success after failed meshes. Clean and main-noise WASM medians vary within 1.0%; this release prioritises accuracy rather than claiming a general speed improvement.
See the performance report, source-bound evidence and reproduction commands and installation guide.
Public npm, JSR, Maven Central and crates.io packages, installed consumers, documentation and CDN bytes have been verified against this release. See the machine-readable verification receipt and its linked per-channel receipts.