chore(master): release 1.0.0

[ooples]

🤖 I have created a release beep boop

1.0.0 (2026-09-02)

⚠ BREAKING CHANGES

• optimizers: NewtonMethodOptimizer.Step throws NotSupportedException on a training context that cannot re-evaluate the loss (TapeStepContext. SupportsReevaluation false), and LevenbergMarquardtOptimizer.Step always throws. Both previously fell through to the base class's plain theta -= lr*g, so callers who selected a second-order method silently received gradient descent. Use Optimize() with an explicit dataset, or choose Adam/SGD for a first-order method or LBFGS for a quasi-Newton one.

Features

• adamw: let a recipe exempt parameters from decoupled weight decay (ab46661)
• analyzers: make an optimizer owning its own generator a build error (94c3c74)
• autodiff: give ComputationNode a public backward pass (24af063)
• ci: carry unchanged shards forward instead of re-instrumenting them (8f548e6)
• ci: coverage-derived shard selection, tooling and proofs (e64ba20)
• ci: coverage-derived test selection to replace the unconditional 115-shard run (3a8b668)
• ci: emit a per-shard test-impact digest artifact (80d8bdf)
• ci: let the nightly map cover heavy shards too (38917d8)
• ci: measure whether selection would have missed a failing shard (4f58102)
• ci: ship selection in shadow mode, and widen the nightly window (19415ab)
• ci: validate only the delta on master pushes (6833575)
• ci: wire shard selection into the test matrix (282fa30)
• clone: automated, provably-correct cloning for options and layers (b295ba3)
• generators: report the layers the scaffold cannot generate (AIDN046) (198a8b9)
• layers: ask the following norm whether a bias is redundant (940818c)
• layers: make the convolution bias conditional rather than unconditional (c8950e6)
• optimization: LP/QP/assignment solvers + fix solver-shaped defects (9d8303b)
• params: persist input-sized fitted state without counting it (efdde6c)
• perf: judge the census against a series, not a frozen point (2e2b236)
• persist the graph a graph layer was given (7c58aa0)
• serialization: one opt-in contract for layer construction state (7594d78)

Bug Fixes

• address hidden review findings (205e59b)
• address PR 2053 review findings (7a86f01)
• address the codeql review findings from pr 2052 (2d46e91)
• address the review findings on this branch (01a39d8)
• address the second review round, and unbreak the net471 build (d6dc75b)
• align audio transformer optimizer recipes (0a6f65b)
• align remaining paper training invariants (3dc2afd)
• align risk and speech training recipes (6f425b1)
• audio: give silerovad's frame tensors their samples, and dispose them (3246fec)
• audio: make the mel front-end mandatory and replace fabricated spectrograms (#2027) (bc5111f)
• audio: replay GraFPrint dynamic graph during compiled training (56c8e4a)
• audio: run voxlingua's own front end and output contract (284cf3f)
• audio: size sed batch norm from the projection width (3eb7fad)
• audio: stop preprocessing twice in named-layer activations (05dd99d)
• audio: train ACE-Step at the learning rate it documents (207caa8)
• audio: train the event detector on its paper's objective (e940c2c)
• autodiff: stop an identity function adding its gradient to itself (6564dfb)
• benchmarks: migrate the last shim call site, which broke the CI build (683ed0d)
• bound abandoned clone attempts so the sweep stops killing its own host (f3cd8ef)
• builder: consult the compiled plan before the defaulted inference optimizations (9a98ac7)
• causal: honour maxepochs as a maximum and name the lagrangian policy (410ccd3)
• causal: match the reference schedules for castle and amortizedcd (93deefc)
• causal: restore CASTLE raw norm thresholding (b245e49)
• centralize clone topology restoration (feab62b)
• centralize scheduler checkpoint recipes (2998ace)
• ci: address hidden review findings (e279c9e)
• ci: address PR 2029 review findings (917b0e9)
• ci: basename infra matching, atomic digest writes, merged carry ranges (29a3844)
• ci: close PR 2004 test regressions (ad31da6)
• ci: close PR 2026 acceptance regressions (c1c2bc2)
• ci: close PR 2029 review gaps (f9c55e8)
• ci: cover generated row-parallel tests (e3089c3)
• ci: cover generated vector model tests (0544944)
• ci: declare the repvit-sam peak working set the same commit already moved (dabaa89)
• ci: declare the third repvit-sam memory metric the same commit moves (3f00880)
• ci: deleted files, script injection, stale audit map, missing timeout (dc49ce6)
• ci: emit flat range pairs from the digest as well (7a17a42)
• ci: enforce impact selection and exact-tree reuse (b9ded76)
• ci: escalate on any selection failure, and validate a manual run id (55c2e86)
• ci: escalate when the map cannot be read, not just when it is absent (2442815)
• ci: evaluate merge-queue reuse candidates first (b8d19e7)
• ci: fail safely across coverage-map lifecycle (721cf66)
• ci: flat range pairs and precedence, both found only by running the real path (65e9440)
• ci: group the vi/vi alternation before excluding video (ef6bf31)
• ci: harden the impact tooling against the adversarial review's findings (cdd8f74)
• ci: keep draft events from cancelling ready validation (2ae1923)
• ci: keep regression safeguards net471-compatible (c5879a6)
• ci: keep the regression gate and shard-coverage check honest under selection (46cb6ba)
• ci: let master reuse merge-queue validation, not only pr runs (8ae7834)
• ci: make impact selection and exact-tree reuse effective (7f0807c)
• ci: preserve certified regression baselines (0e4b5e2)
• ci: re-key the repvit-sam leases to the baseline the census now uses (e873544)
• ci: resolve final acceptance regressions (a5b12d7)
• ci: resolve final regression shards (604f391)
• ci: resolve PR 2010 regression set (61536d7)
• ci: restore compatibility and baseline validation (2c81989)
• ci: select against the map's coordinates, not the PR's (49580f7)
• ci: split always-run names defensively and count all runnable shards (50737ae)
• ci: stop collecting coverage on Integration C - Core (1ecd634)
• ci: stop master duplicating pr validation, and video running twice (af9058f)
• ci: stop master duplicating pr validation, and video running twice (228221e)
• ci: stop splitting always-run shard names on commas (5745d55)
• ci: tolerate the map workflow not existing yet (561361b)
• ci: unblock PR validation (3466865)
• ci: verbatim directory creation and schema-gated map reads in the carry tool (f59422b)
• clone: close generated lifecycle regressions with CI proof (4a1d022)
• clone: complete automated clone acceptance criteria (1a85906)
• clone: restore generated auxiliary state faithfully (6114cf5)
• clone: restore generated continuation state (3d6649e)
• close PR 2004 clone and serialization regressions (bf0d326)
• cmaes: the step-size rule compared against the wrong reference length (4303e99)
• complete clone state and RepViT invariants (80bcb89)
• complete review follow-up invariants (37773ed)
• consolidate Tacotron2 paper training path (12e0e96)
• conv: apply the activation on the training path and stop grad mode choosing the kernel (2739294)
• conv: preserve fused kernel parity by platform (39985c9)
• conv: preserve platform-fast exact kernels (f653eae)
• daggnn: stop orienting edges on the random seed and assert the recovered dag (02a7676)
• dcn: normalize dcnv3 modulation across sample points and repair the ctor contract (c3a5ca1)
• declare cloze attention's projection widths so the count sees them (f0b21ab)
• deps: align Entity Framework Core package train (35adb27)
• diffusion: declare PointE image generator conditional (1428c34)
• diffusion: keep the conditioner in the chunk stream it was never registered in (4b3a7fc)
• diffusion: stop an absent conditioner taking the whole parameter layout offline (67f87a5)
• diffusion: stop an empty clone from materializing a foundation model (acbd6aa)
• diffusion: stream the component registry instead of a hand-listed parameter surface (2941c5c)
• diffusion: stream the component registry instead of a hand-listed surface (acff35d)
• drop the null-forgiving operator from the conv stem fast path (bc7727d)
• embedding: reject impossible deferred parameter payloads (f761fbc)
• fix generic dictionary optimizer reset traversal (be886db)
• fix post-2051 CI regressions without weakening invariants (9e2d503)
• fix remaining paper-faithful training invariants (bf812ff)
• fix TOTO zero-warmup and DCGRU shape validation (b2b0a58)
• generator: register a component slot declared through a plain interface (f1e045b)
• generators: eliminate repeated semantic scans (e11b664)
• generators: emit clone factories for layers that forward ctor args to a base (be4d880)
• generators: give 15 more layers a working clone factory (4aa6a2a)
• generators: let a declared width reach children held in a collection (d73bbc2)
• generators: let a layer with buffers infer its deferred input width (7cacdb5)
• generators: make [TrainableParameter] on a non-partial class an error (ea88c1b)
• generators: route the value kinds LayerStateBag already supported (7c4cc65)
• generators: scope the axis walk per axis and treat a signed literal as a literal (e3f98a8)
• give a bce-with-logits head a target it is defined for (bdfe9ae)
• gradcheck: flush weight caches on perturbation, drop SVTR's second walk (3e354a3)
• graph: mark the graph tensors input-sized, and stop training one (79bd50b)
• graph: prevent recursive training and arena-backed weights (29ec59c)
• inference: refresh cached weights after parameter restore (bc41891)
• inference: register PagedCachedMultiHeadAttention's projection weights (d0bca53)
• inference: register the cached attention layers' projection weights (fef2aa5)
• initialize first compiled training lifecycle safely (78749d2)
• init: make parallel chunk arithmetic overflow-safe (9f0aa9a)
• init: make seeded parallel weight initialization machine-independent (59e1b18)
• init: make seeded parallel weight initialization machine-independent (f63ba64)
• init: stop a requested initialization seed from being silently discarded (bdaa795)
• layerhelper: remove dead controllerinputsize store in the ntm factory (c93f7ae)
• layers: count the children a composite holds in a collection (2b278c3)
• layers: declare child widths for three more composites (fd9e33c)
• layers: do not chain children while a declaration is still being built (a721700)
• layers: keep spectral normalization's own weights and iteration vectors (b2f4f11)
• layers: let a partial declaration coexist with chained sizing (7cf5279)
• layers: make the declared sub-layer width reach the parameter surfaces (0d2747d)
• layers: materialize before enumerating a layer's trainable value slots (933cc33)
• layers: materialize separable pointwise kernels (3bf0490)
• layers: restore the autodiff tape through MLA attention and RDB activation (0dab560)
• layers: scaffold six more layers and fix what that exposed (5256577)
• layers: seed spectral iteration so sigma means something before training (2c5931f)
• layers: size the temporal module and reconnect the generator's input (022f97b)
• layers: use the implicit-broadcast divide and correct a wrong attribution (8cd7097)
• lstm: re-infer resolved input width from parameter payload (567f865)
• lstmvae: implement the elbo backprop and stop forecasting from an anomaly score (625f6e4)
• make the bulk clone-buffer copy view-safe (2c86812)
• mamba: align Step/Predict on token ids, and stop dropping deferred layers (9aa1a80)
• mbpa: return the k nearest in order and stop refusing batched multi-component output (d06dd55)
• mbpa: return the k nearest nearest-first, and stop asking for a refused pack (1a9b1e8)
• n1: address follow-up review edge cases (301a6ad)
• n1: address quantum and MbPA review findings (b326303)
• n1: stabilise cutie memory readout and repair non-finite comparisons (71cde8d)
• nmf: start from the data's singular vectors instead of an unseeded draw (ee42954)
• nmf: stop the convergence tolerance from gating the random restarts (1721275)
• nn: carry both halves of the dnc controller state the same way (a0bbf8f)
• nn: give the dnc the recurrent controller its paper specifies (3853a20)
• nn: keep RecurrentGemma off the fused compiled training plan (91e3cbb)
• ocr: give abcnet the optimizer its published rate belongs to (4ba640e)
• ocr: give svtr tps its released localization margin (a06df29)
• optimizers: adopt options through one method, and stop deserialize clobbering collaborators (3b56c41)
• optimizers: honour options.seed on the model-based path (7342cbf)
• optimizers: honour options.seed, delete the duplicated options copy, and gate both (6bb6b06)
• optimizers: make every fused optimizer spec describe what its Step() actually does (#2009) (264780d)
• optimizers: rebuild the proximal operator when its options move (1051272)
• optimizers: resolve optimizerbase by symbol, and assert collaborators by identity (0cb4238)
• parameters: gate value-surface materialization on the declaration (cace89f)
• parameters: gate value-surface materialization on the declaration (c35f3cb)
• parameters: guard declared shape axes at their roots, not their result (396366a)
• parameters: invalidate stale generated component snapshots (b9bdfd6)
• parameters: let AIDN082 see the chunked surface, and delete the 13 it was missing (f27d2eb)
• parameters: materialize at the value boundary, and hold it to the declaration (4a4b37e)
• params: append the new slot role, and classify link prediction's cache (2d7ef8f)
• patchgan: drop the bias on blocks whose batchnorm already shifts (6cba715)
• patchgan: restore checkpoints written before the bias mode existed (e7b34cf)
• perf: gr00tn1 grew a parameter slot it should always have had (80d8f19)
• performance: cache echo state weight layouts (f898c2f)
• perf: stop a baseline drawn at a series low reporting a timing regression (5b4add8)
• persist eight learned tensors that three SSM layers trained but never saved (5fac16e)
• persist paper learning rate schedules (3096997)
• persist QuantumLayer's rotation angles, its only learned parameter (d443514)
• persist RWKV's ten learned tensors; [AutoParameters] never registered them (1e44104)
• pointnet: restore the gradient the neighbourhood max-pool was discarding (ef68085)
• post-2052 clone, serialization and CI-reuse failures (969f198)
• preserve overlapping parameter ownership (bee5df8)
• preserve scheduler checkpoint cadence (d81bd80)
• preserve scheduler recipes and ONNX sessions (b7f6e4b)
• preserve strict paper training contracts (2d039ba)
• pretrained: address review — internal Inner + skip (not pass) fixture tests (a1c1962)
• pretrained: address second review pass (f20203a)
• pretrained: permute q/k to interleaved RoPE layout for HF safetensors decoders (e055374)
• pretrained: permute q/k to interleaved RoPE layout for HF safetensors decoders (c6de27e)
• prioritize incomplete COW coverage diagnostics (15dd50b)
• quality: close tensor ownership findings (4c9eb61)
• quantum: encode a valid normalized state instead of sqrt of raw features (f7c9a07)
• quantum: preserve signed inputs as finite amplitudes (dac183c)
• rbm: dispose contrastive-divergence tensor input (c5c5879)
• rbm: reject invalid Gibbs step counts (7b96066)
• recurrentgemma: restore fused compiled training (26db80a)
• recurrentgemma: train off the fused compiled step, as Griffin and Hawk already do (a36bb82)
• recurrentgemma: validate optimizer options and refuse silent embedding scaling (5b532a2)
• register the state-space core on seven SSM layers so training survives a save (979377c)
• regularization: a shrink factor that could invert the sign (52710c6)
• resolve PR 2054 regression failures (fb8f5f1)
• resolve remaining PR 2054 model regressions (909a85f)
• restore blocks whose input declares a free axis, and green the clean build (2b2383d)
• restore DAC recipe and DCRNN persistence (2a10561)
• restore generated clone and parameter lifecycle (20da139)
• restore Hamiltonian derivative training objective (52d46eb)
• restore MedCLIP pretraining recipe (c66dcb5)
• restore net471 invariant test compatibility (ef54439)
• restore net471 option copy compatibility (7e5feb7)
• restore paper DCRNN recurrent diffusion (cb9ecf5)
• restore paper optimizer and FiNS training behavior (9b35c1e)
• restore paper-faithful Deep KWS architecture (e1b634d)
• restore remaining paper training invariants (5ed6c77)
• restore Squeezeformer extended Noam schedule (5796503)
• review: address generator and tensor lifecycle findings (9fec033)
• rg-lru: restore finite fused training for Griffin, Hawk, and RecurrentGemma (2d60bbe)
• rl: train the deep q-network on its published rmsprop settings (422a92a)
• robotics: build helix's optimizer from its own training options (61f0518)
• segmentation: train repvitsam at its family's rate, not the generic default (3cbd616)
• serialization: carry a component's configuration, not just its type (66fec3a)
• serialization: let a restored expression call the framework's own maths (e60651b)
• serialization: write the resolved input shape for every layout node (726e2ec)
• som: Predict passthrough; plus generator symbol-retention hygiene (no perf win) (#2036) (b7087b3)
• ssm: construction-sized parameter surfaces, RWKV7 token contract, D protection (af0e1c2)
• stabilize remaining non-clone training invariants (f80e4e6)
• state: rebind derived views after parameter restore (3496f8b)
• stop allocating full-size scratch activation caches during conv init (2214792)
• stop the GPU quantum prescale returning NaN for every all-zero row (08b2b31)
• stop the LMI search reporting success on an estimate its own grader rejects (65659e0)
• stop two composites counting their children twice (53e1439)
• stop two more composites counting their children twice (4e1e933)
• streaming: honor overridden parameter count during auto-detection (5c0a7ba)
• tests: align MGTSD architecture with forecast options (18e72a7)
• tests: align RecurrentGemma policy guard (4d021a6)
• tests: ask the model whether its own objective can tell the targets apart (21060e1)
• tests: bound codec TTS clone sweep options (dbdb3b8)
• tests: bound exhaustive layer clone sweep (3edc9c1)
• tests: bound Nemotron clone sweep options (90d82a7)
• tests: bound the gradcheck localization sweep by the real clock (5c59046)
• tests: build model inputs from the effective shape, not the declared one (#2025) (a087259)
• tests: classify every clone attempt and keep coverage off the timing shards (68551cb)
• tests: compare SeACo shape dimensions portably (c7a2529)
• tests: compare tensor shape portably (62bf7e4)
• tests: drop duplicate provideslearnableshift left by the rebase (1064ae2)
• tests: fail closed when the configured loss cannot be measured (252fd20)
• tests: give the training-error invariant enough steps to be true at paper learning rates (733b17e)
• tests: implement provideslearnableshift exactly once per ilayer double (796c79a)
• tests: implement provideslearnableshift on the two hand-rolled ilayer mocks (8669d1d)
• tests: keep the RecurrentGemma regression asserts net471-compatible (822a244)
• tests: let Dessurt clear the AdamW transient (8583b8a)
• tests: measure the loss a model actually descends, not mse by default (8e7ee54)
• tests: measure the train-vs-test invariant on the claim it makes (5e5c8dc)
• tests: migrate the last shim call sites, which broke the test build (ea3ca39)
• tests: mirror the scalar target in the space the loss compares (b13f8de)
• tests: prevent clone sweep model accumulation (f5a331d)
• tests: re-measure with more steps when the target effect is under-resolved (357c284)
• tests: read the optimizer options as field OR property after the refactor (c0eef9a)
• tests: repair two harness defects that failed before asserting (#2035) (1f3bc96)
• tests: support legacy echo state allocation checks (e021019)
• tests: support net471 finite checks (0a9cc96)
• tests: support net471 finite gradient assertion (63ed6ec)
• tests: use a finite check that exists on net471 (a31a900)
• tests: well-pose the memorization target for born-rule heads (d5fc54b)
• timegan: restore recurrent sequence architecture (5b6cf58)
• training: address PR 2029 review findings (6ec49d2)
• training: address transitive optimizer reset review (a100390)
• training: align strict invariants with paper objectives (52a9874)
• training: align strict invariants with paper objectives (bd7f232)
• training: close PR 2029 regression gaps (a6b99d0)
• training: close reset and diagnostic review gaps (96d5124)
• training: complete paper-faithful objective fixes (d6e45d1)
• training: eliminate target and fused-path regressions (b4295cc)
• training: flatten gradients in the canonical parameter order (2cc61c9)
• training: honor partial-freeze selection in fused compiled path (012adcd)
• training: preserve partial-freeze invariants in fused plans (51f04de)
• training: record RG-LRU in compiled plans (69896e7)
• training: repair PR 2032 regressions (ff62eed)
• training: reset optimizers a model's own base class owns, not just the base one (d4e584d)
• training: reset optimizers owned through collections and sub-networks (bfffab9)
• training: reset the fused plan's optimizer state, not just the outer optimizer (e6f12e4)
• training: reset the optimizer state a training step actually uses (edec97a)
• training: resolve cross-target review regressions (df22056)
• training: restore fused RG-LRU family parity (6558574)
• training: validate target-dependent gradients generically (b79cc2a)
• tts: measure dittotts memorization past its first adam update (92def7e)
• tts: train OpenVoiceV2 at the learning rate its options declare (27fc29b)
• tts: train the vits family on its published adamw recipe (2d1a248)
• vad: put the audio into the frame tensor the model scores (9aa2b9a)

Performance

• bulk-copy clone buffers and copy tensors before resolving shape (cb548a7)
• fill conv kernels in chunks instead of a full-size staging array (e23dfeb)
• internimage: use dcnv3 linear offset/mask projections as the paper specifies (f76e12c)
• layers: stop re-copying parameters once per level of nesting on write (9cd287e)
• own activation-cache lifetime in the base class, PyTorch-style (1c9ff26)
• resolve a clone's shapes without materialising the activation (dc22d62)
• serialization: work out a component's configuration once per type (96a120f)
• stop the VAE stack pinning every activation for the whole forward (d67773f)
• tests: stop the loss-domain guard allocating per element (e33cf39)
• training: skip unused fused selection walk (0a16ea8)

Reverts

• drop my duplicate provideslearnableshift on the two ilayer mocks (feda8d3)
• remove invariant-weakening fixture changes (fe7440a)

Refactoring

• clone: enforce generated model lifecycle (de55b75)
• complete the half-declared weight groups (69709af)
• declare fused weight caches and recurrent sequence state (528e808)
• declare layer gradients, caches and optimizer moments (1903705)
• engine: drop the broadcast compatibility shim (78e50fe)
• extract the duplicated nndsvd entry in nmf initialization (08a8032)
• layers: normalize spectral weights on the tape, not by mutation (8aeb3d3)
• let the generator own BayesianDenseLayer's weights (598fc8a)
• let the generator own DuelingCombinationLayer's weights (f5cffb6)
• let the generator own NoisyDenseLayer's weights (e86c8fd)
• let the generator own three more layers' parameters (5de90fc)
• optimizers: delete the cached typed options copy, 42 updateoptions overrides to 6 (c687c6f)
• optimizers: move seeding wholly into the base instead of per optimizer (fc11d5e)

Build System

• stop packing six projects nothing publishes or consumes (d8dab69)
• tests/AiDotNet.Tests succeeds on net10.0. (647fb7e)
• tests: target net8.0, which the library ships but nothing compiled (bde2742)

Documentation

• attach the sed constructor documentation to its own constructor (fa4adc5)
• ci: record the map builder's measured behaviour at full scale (0400c8f)

---

This PR was generated with Release Please. See documentation.

[vercel]

Deployment failed with the following error:

Resource is limited - try again in 24 hours (more than 100, code: "api-deployments-free-per-day").

Learn More: https://vercel.com/franklins-projects-02a0b5a0?upgradeToPro=build-rate-limit

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

Walkthrough

The release manifest advances the root package from version 0.231.0 to 1.0.0, and the changelog adds the dated release entry with breaking changes, features, bug fixes, and performance updates.

Changes

Release metadata

Layer / File(s)	Summary
Version and changelog entry .release-please-manifest.json, CHANGELOG.md	The root package version changes to 1.0.0, and a categorized changelog section documents API removals and release updates.

Estimated code review effort: 1 (Trivial) | ~2 minutes

Possibly related PRs

• ooples/AiDotNet#1884: Makes corresponding release-version and changelog updates for the adjacent release.

Poem

One point oh now takes the stage,
New features bloom upon the page.
Breaking changes mark the tune,
Fixes and speed arrive with June.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely identifies the primary change: creating the 1.0.0 release through Release Please.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch release-please--branches--master--components--AiDotNet

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
aidotnet_website	Ready	Preview	Sep 2, 2026 5:44pm UTC
aidotnet-playground-api	Ready	Preview	Sep 2, 2026 5:44pm UTC

[Copilot]

🟡 Changes recommended

The changelog’s new 1.0.0 section is placed above ## [Unreleased], conflicting with the stated Keep a Changelog structure and likely to cause ongoing formatting drift.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 1
• Review effort level: Lite

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.