audiobits 0.1.2 → 0.1.4

package/CHANGELOG.md

@@ -1,5 +1,18 @@
 # audiobits
 
+## 0.1.4
+
+### Patch Changes
+
+- 5340645: Simplify the package README to a tested quick start and direct detailed guidance to the website docs. Clarify installed-version authority in the shipped integration Skill.
+
+## 0.1.3
+
+### Patch Changes
+
+- 7aa39bd: Align the package README with published capabilities and add the public website
+  and relevant npm keywords to package metadata. Runtime behavior is unchanged.
+
 ## 0.1.2
 
 ### Patch Changes

package/README.md

@@ -1,21 +1,22 @@
 # AudioBits
 
-AudioBits is available on npm as `audiobits`. It includes confirmation, impact, thruster, tactile click, gentle rejection,
-glass notification, whoosh and power-up; schema-1 recipes; play/live controls; seeded variation; bounded
-voices; bus gain, mute and routing; native output taps; and explicit lifecycle APIs.
-Prerelease verification target: current Chromium. Other browsers and operating
-systems are unverified and deliberately deferred. Maintainer listening acceptance for the
-unchanged sound definitions is recorded separately.
+Browser-native procedural sound for games and interactive applications. Play
+bundled sounds or build editable JSON recipes without downloaded audio assets
+or runtime dependencies.
+
+[Live gallery](https://audiobits.joacod.com/sounds) ·
+[Documentation](https://audiobits.joacod.com/docs)
 
 ## Quick start
 
+Install in a browser application with a bundler:
+
 ```sh
 npm install audiobits
 ```
 
-Use a browser bundler. Create the engine and sound once; run the final two lines
-inside a user gesture handler, such as a button click. Call `audio.start()`
-directly in that handler and await it before playback.
+Create the engine and sound once. Construction is silent; call `play()` directly
+from a user gesture, such as a button click. It awaits activation before playback.
 
 ```ts
 import { createAudio } from "audiobits";
@@ -24,305 +25,41 @@ import { confirmation } from "audiobits/recipes";
 const audio = createAudio();
 const sound = audio.sound(confirmation);
 
-await audio.start();
-sound.play();
-```
-
-Surface activation failures and retry from a fresh gesture. Use
-`audio.stopAll({ tails: "cut" })` to stop playback and `await audio.dispose()`
-when the host is removed. See the production lifecycle example below for
-hiding and navigation.
-
-## Production lifecycle
-
-For an SPA, invalidate pending activation when hiding or unmounting; suspend on
-hide and dispose on navigation. Returning requires a fresh Play gesture.
-
-```ts
-import { createAudio } from "audiobits";
-import { confirmation } from "audiobits/recipes";
-const audio = createAudio();
-const sound = audio.sound(confirmation);
-let request = 0;
 export async function play() {
-  const token = ++request;
   await audio.start();
-  if (token === request && !document.hidden) sound.play({ seed: 42 });
+  sound.play();
 }
 export function stop() {
-  request++;
   audio.stopAll({ tails: "cut" });
 }
-function hide() {
-  if (document.hidden) {
-    stop();
-    void audio.suspend().catch(console.error);
-  }
-}
-document.addEventListener("visibilitychange", hide);
 export async function dispose() {
-  stop();
-  document.removeEventListener("visibilitychange", hide);
   await audio.dispose();
 }
 ```
 
-`createAudio()` and `audio.sound()` create no context. A failed start reports
-`AudioBitsError` with a retryable `start-failed` code. A blocked resume times out
-after two wall-clock seconds; the context stays owned for gesture retry. Play
-before activation fails with `not-ready`; no input is queued.
-
-The archive exports `audiobits/schema.json` and `audiobits/capabilities.json`.
-The latter identifies package version, shipped primitives, initial Chromium gate and verification browser matrix;
-the schema is structural, while `validateRecipe` enforces additional semantic
-limits. Agent guidance is included in [the AudioBits Skill](skill/SKILL.md).
-
-## Supported data
-
-`defineSound(recipe)` accepts typed authored data, preserves recipe structure, literal parameter names and modes, validates and returns a deeply frozen `Recipe` snapshot.
-Runtime validation remains authoritative for exact schema validity.
-For external JSON, `validateRecipe(unknown)` returns `{ ok: true, recipe }` or
-`{ ok: false, issues }`; every issue has `code`, `path`, and `message`.
-Paths use JSON bracket notation, such as `$["layers"][0]["id"]`.
-`AudioBitsError` carries `code` and `issues`; errors during validation use
-`invalid-recipe`. The input must be plain JSON data without accessors or cycles.
-Unknown fields and versions are rejected. Validation is browser-independent.
-
-Schema version 1 supports:
-
-- One-shot gates in `(0, 60]` seconds or sustained playback until Stop, with
-  1–16 uniquely identified layers. Sustained recipes forbid `duration`.
-- Sine, triangle, sawtooth, and square oscillators; numeric Hz or linear/
-  exponential frequency automation beginning at zero and ending by gate close
-  (at most 60 seconds of onset automation for sustained voices). White noise
-  uses bounded generated buffers; there is no audio-file fetch.
-- Layer gain from -60 to 0 dB and fixed ADSR envelopes. Attack/decay are 0–10 s,
-  sustain is 0–1, release is 0.005–10 s. Effective attack is at least 0.002 s;
-  effective attack plus decay must fit within the gate. A zero decay ramps
-  directly to sustain rather than introducing an instantaneous peak.
-- Lowpass/highpass/bandpass filters with frequency 20–20000 Hz and numeric
-  Q 0.1–20. At most eight filters and 128 frequency points across the recipe.
-- Source/filter frequencies accept numeric values, mappings, variation, or
-  automation. Layer gain accepts numeric values, mappings, or variation;
-  gain automation and variable envelope times are not supported.
-- Up to 16 named parameters with `min < max`, an in-range `default`, and `mode`
-  of `play` or `live`. Live parameters require `smoothing` in 0.005–1 seconds.
-  Names begin with an ASCII letter and contain letters, digits or underscores,
-  up to 64 characters. Declaration numbers are bounded to -60000–60000.
-
-Depth is limited to 16, visited values to 10000, and diagnostics to 100.
-Source/filter frequencies must be below the owning context's Nyquist frequency
-at playback, including all mapping/variation extrema. Other effects,
-expression strings, sequencing, and continuous random modulation are rejected. These are the capabilities shipped in 0.1.0.
-The generated schema is available as `recipeSchema` or `audiobits/schema.json`.
-
-## Runtime ownership
-
-`createAudio({ maxVoices, maxVoicesPerSound, masterGainDb })` defaults to
-32 active voices per engine, eight per sound, and -12 dB master gain. Limits
-must be integers from 1–128; master gain must be -60–0 dB. `setMuted(boolean)`
-uses a separate mute setting and a 5 ms gain ramp.
-
-`sound.play({ at, gainDb, pan, parameters, seed, bus })` creates fresh sources synchronously. `at` is
-absolute audio-context time, obtained by the host's own scheduling logic;
-available after startup as `audio.native.context.currentTime`. Omit it for
-immediate playback. Past timestamps are rejected. Playback gain is -60–0 dB
-(default 0), pan is -1–1 (default 0). Future voices reserve capacity.
-
-`voice.stop()` cancels before onset or releases from the current envelope value.
-Repeated stops cannot extend lifetime. `voice.ended` resolves after owned nodes
-are disconnected; `voice.state` is `active`, `stopping`, `retiring`, or `ended`.
-Stopping voices keep their capacity reservation until cleanup. Oldest-voice
-stealing uses a 5 ms output fade. There is at most one retiring voice beyond
-engine/per-sound limits; another steal finalizes the previous retiree first.
-`audio.counts` distinguishes reserved active/stopping voices from retirees.
-
-Gate duration excludes release. Filters receive a bounded 50 ms tail allowance;
-output fades to zero over the final 5 ms. Layer filters precede their envelopes.
-`stopAll()` releases managed voices according to their recipe envelopes.
-`stopAll({ tails: "cut" })` uses a 5 ms source/output fade. `sound.dispose()` immediately finalizes
-that sound's voices and noise buffers. `audio.suspend()` invalidates pending starts and finalizes all voices before suspending.
-Native suspension/interruption also finalizes voices when its state event arrives.
-`audio.dispose()` invalidates pending startup, finalizes voices immediately,
-disconnects master output, and closes the context once, even while suspended.
-Disposal is terminal and idempotent; no audio-clock progress is needed.
-
-`audio.state` reports `idle`, `starting`, `running`, `suspended`, `interrupted`,
-`closed`, or `disposed`. `subscribe(listener)` returns an unsubscribe function.
-Concurrent starts share one promise. A host controls visibility/navigation policy;
-the engine does not replay sounds or automatically resume them.
-
-Recipes contain no native nodes, callbacks, framework imports, or runtime
-dependencies. Only `audiobits/recipes` imports curated sound data.
-Default headroom checks do not guarantee safe peaks for arbitrary recipes,
-filter resonance, gains, or concurrency.
-
-## Dynamic controls and replay
-
-```ts
-import { createAudio } from "audiobits";
-import { impact, thruster } from "audiobits/recipes";
-
-const audio = createAudio();
-const hit = audio.sound(impact);
-const engine = audio.sound(thruster);
-
-// In a user gesture, await activation before creating voices.
-await audio.start();
-hit.play({ parameters: { intensity: 0.8 }, seed: 42 });
-const voice = engine.play({ parameters: { throttle: 0.2 }, seed: 42 });
-voice.set({ throttle: 1 }); // Update this voice without retriggering.
-console.log(voice.seed, voice.parameters);
-voice.stop();
-await voice.ended;
-await audio.dispose();
-```
-
-Omitted parameters use recipe defaults. Invalid names, non-finite values,
-out-of-range controls, and invalid seeds fail before graph allocation or voice
-stealing. Updates validate every supplied key before changing any parameter;
-play-only controls cannot be updated. `voice.parameters` is a frozen snapshot of
-requested values, not a readback of the current smoothed native parameters.
-`voice.set()` after Stop, retirement, completion, or disposal throws
-`ended-voice`; other control errors use `invalid-control`.
-
-Mappings use `{ control, range: [low, high], scale }`, where scale is `linear`
-or `exponential`. Exponential mapping endpoints must be positive. Mapping ranges
-may descend; random ranges `{ random: [low, high] }` must be ordered. All extrema
-must satisfy the target range. Automation points can reference only play-only
-controls. Direct live mappings support source frequency, layer gain, and filter
-frequency. Native targets ramp linearly over the parameter's smoothing time,
-from their current value on every retarget. For gain, that ramp is in linear
-amplitude after dB conversion; mapping scale describes the control-to-target
-conversion, not the ramp curve. Envelopes remain fixed and independent of live
-level controls, so release remains continuous.
-
-Seeds are unsigned 32-bit integers, including zero. Without one, playback chooses
-a seed and exposes it on `voice.seed`. xorshift32 uses shifts 13/17/5 and divides
-its unsigned state by 2^32; seed zero maps internally to `0x6d2b79f5`. Resolution
-traverses root effects, then layers in array order. Each layer resolves gain,
-layer filters, source frequency, and a noise seed, in that order; automation
-points follow time order. Each variation consumes one draw, and each noise layer
-consumes one draw for its own xorshift32 stream. Object property insertion order
-does not change this traversal. This algorithm reproduces choices and
-noise for the same normalized recipe, controls, seed and sample rate; it does not
-promise identical oscillator/filter samples across browsers or sample rates.
-
-## Noise ownership and bounds
-
-Each noise layer owns one mono, one-second buffer at context sample rate. The
-last 20 ms crossfades into the first 20 ms; looping resumes after that prefix,
-so the wrap follows an ordinary adjacent sample pair. The loop period is about
-0.98 seconds. Both finite and sustained noise loop this bounded resource.
-This treatment has signal evidence and maintainer listening acceptance;
-individual listening observations were not recorded. There is no shared noise cache or rendered-output cache.
-
-Noise supports integer sample rates from 8000 through 192000 Hz. Unsupported
-rates fail with `noise-rate` before voice allocation. A layer retains at most
-768000 sample bytes (192000 at 48 kHz); generation temporarily uses one extra
-buffer of that size. A recipe has at most 16 layers, and existing voice limits
-bound concurrent buffers. The thruster owns ten nodes and one buffer per voice;
-with its default eight-voice limit plus one retiree, sample storage is at most
-1728000 bytes at 48 kHz. The engine master owns two additional nodes for independent gain and mute. These are
-owned sample-storage bounds, not measurements of all browser memory.
-
-Control updates allocate no new audio nodes or buffers. Finishing a voice clears
-its buffer-source references and disconnects owned nodes, including cancellation
-before onset, stealing, sound disposal, and engine teardown. Browser-internal
-reclamation timing remains outside the library's control.
-
-## Buses and routing
-
-After `await audio.start()`, `audio.master` is the root bus. `audio.bus(name,
-parent = audio.master)` creates a named bus or reuses the live bus with that
-name; use `setParent()` to move an existing bus. Names contain 1–64 characters
-and cannot be blank. At most 32 buses, including master, may be live. Routes
-form a single-parent tree; invalid, disposed, foreign-engine, and cyclic
-parents fail before changing the previous connection. Reassigning the same
-parent does not create another route. `sound.play({ bus })` defaults to master
-and rejects foreign/disposed buses before voice allocation or stealing.
-
-`bus.setGainDb(value, rampSeconds = 0.005)` accepts -60–0 dB and 0–10 seconds.
-It holds the current native value before ramping. `bus.setMuted(boolean)` uses
-an independent 5 ms stage, preserving volume automation. `audio.setMuted()`
-controls master mute, including a setting made before startup.
-
-```ts
-import { createAudio } from "audiobits";
-import { confirmation } from "audiobits/recipes";
-const audio = createAudio();
-const sound = audio.sound(confirmation);
-// In a gesture handler, after activating this engine:
-await audio.start();
-const effects = audio.bus("effects");
-effects.setGainDb(-6, 0.1);
-const voice = sound.play({ bus: effects });
-voice.stop(); // Release this voice.
-audio.stopAll({ tails: "cut" }); // Fade managed voices over 5 ms.
-effects.dispose(); // Finalize routed voices and descendant buses.
-await audio.dispose();
-```
-
-Each bus owns two gain nodes. Suspension and disposal finalize voices
-without waiting for audio-clock progress.
-
-Non-master bus disposal is idempotent and stops routed voices recursively.
-A later lookup of that name creates a fresh bus. `master.dispose()` fails;
-dispose the engine to release master. Engine disposal removes all owned
-output and closes its context, even while suspended or starting.
-
-## Native analyser and caller ownership
-
-`audio.native` is available after startup and exposes the owned `context`,
-master `output`, and `connect(node)` tap helper. The helper rejects foreign
-contexts, duplicate taps, the output itself, and the context destination.
-It returns an idempotent detach function. Master already connects to the
-speaker destination. Leave an analyser's output unconnected to avoid a second
-audible path. The vanilla example reads peak levels with a host animation
-frame, cancels that frame, detaches the tap, and disconnects its analyser
-before engine teardown.
-
-```ts
-import { createAudio } from "audiobits";
-const audio = createAudio();
-// Execute in a gesture handler.
-await audio.start();
-const native = audio.native;
-const analyser = native.context.createAnalyser();
-const detach = native.connect(analyser);
-const samples = new Float32Array(analyser.fftSize);
-analyser.getFloatTimeDomainData(samples);
-// Host teardown:
-detach();
-analyser.disconnect();
-await audio.dispose();
-```
+Surface rejected activation, for example `void play().catch(console.error)`, and
+retry from a fresh gesture. Bind `stop()` to Stop and await `dispose()` when the
+host is removed. For pending activation, page hiding, and navigation, follow the
+[lifecycle guide](https://audiobits.joacod.com/docs/lifecycle).
 
-Caller-created nodes, sources, connections, and animation listeners are
-caller-owned. Stop unmanaged sources explicitly: `stopAll()` cannot stop them.
-Direct native graph operations remain the caller's responsibility and cannot
-be serialized as recipes. Closing the engine context invalidates these native
-nodes; retained native handles do not transfer ownership back to the engine.
+## Sounds and controls
 
-The site and vanilla demo invalidate pending Play actions, cut managed tails,
-and suspend on hide. Returning does not resume; a fresh gesture starts audio.
-Late activation after suspension is cancelled and cannot replay an old request.
-The engine also clears owned voices/effects on native interruption; automatic
-native recovery is suspended until a fresh `start()` request. Physical OS
-interruption still requires manual evidence. Step 04 passed maintainer manual
-verification on 2026-10-06. Browser/device details and individual observations
-were not supplied; automated checks remain separate from that acceptance.
+Eight bundled sounds are available from `audiobits/recipes`: confirmation, impact,
+thruster, tactile click, gentle rejection, glass notification, whoosh, and power-up.
+Use [custom JSON recipes](https://audiobits.joacod.com/docs/recipes) with
+`defineSound()` for typed authoring or `validateRecipe()` for external data.
 
-Buses provide gain, mute, and routing. Shared effects are deferred until
-multiple real sound requirements justify an API.
+[Parameters and seeds](https://audiobits.joacod.com/docs/parameters) vary playback
+and update live controls. [Buses](https://audiobits.joacod.com/docs/buses) provide
+gain, mute, and routing; [native taps](https://audiobits.joacod.com/docs/native)
+connect caller-owned analysers and nodes.
 
-## Curated sounds
+The [API reference](https://audiobits.joacod.com/docs/api) explains public options
+and limits. Exported types, `audiobits/schema.json`, and
+`audiobits/capabilities.json` describe the installed package; semantic validation
+still requires `validateRecipe()`. Integration guidance also ships in the
+[AudioBits Skill](skill/SKILL.md).
 
-The eight exports live in `audiobits/recipes`: `confirmation`, `impact`, `thruster`,
-`tactileClick`, `gentleRejection`, `glassNotification`, `whoosh`, and `powerUp`.
-Intensity controls impact, click and power-up; brightness controls glass; size
-controls whoosh; throttle is live on thruster. Other sounds have no parameters.
-New sound definitions use only schema-1 sources, filters and envelopes. Their
-signal and resource properties are tested; the current eight curated sounds have
-passed the maintainer listening gate in Chrome. Output devices and detailed
-listening coverage were not specified.
+Current Chromium is the only automated browser verification target. Other
+browsers and operating systems are unverified. Automated checks do not establish
+subjective listening quality or physical-device compatibility.

package/dist/capabilities.json

@@ -1,5 +1,5 @@
 {
-  "packageVersion": "0.1.2",
+  "packageVersion": "0.1.4",
   "schemaVersion": 1,
   "status": "stable",
   "recipe": {

package/package.json

@@ -1,8 +1,17 @@
 {
   "name": "audiobits",
-  "version": "0.1.2",
+  "version": "0.1.4",
   "description": "Procedural browser audio: versioned recipes, live controls, managed voices and buses.",
   "license": "MIT",
+  "homepage": "https://audiobits.joacod.com/",
+  "keywords": [
+    "web-audio",
+    "procedural-audio",
+    "sound-effects",
+    "sfx",
+    "game-audio",
+    "typescript"
+  ],
   "repository": {
     "type": "git",
     "url": "https://github.com/joacod/audiobits.git",

package/skill/SKILL.md

@@ -8,8 +8,9 @@ description: Integrate AudioBits procedural sounds into a browser application, i
 This guide ships with the `audiobits` npm package. Confirm the installed
 package version and read `audiobits/capabilities.json` and
 `audiobits/schema.json` before choosing primitives. Package metadata describes
-shipped behavior, not registry availability. Prerelease verification target: current Chromium. Other browsers and operating
-systems are unverified and deliberately deferred.
+shipped behavior, not registry availability. Current Chromium is the only
+automated browser verification target. Other browsers and operating systems are
+unverified and deliberately deferred.
 
 Install with `npm install audiobits` and use a browser bundler. Follow the
 [package quick start](../README.md#quick-start) to play the bundled `confirmation`
@@ -18,8 +19,13 @@ sounds without browser activation, then invoke `start()` synchronously inside
 a user gesture and await it before `play()`. Surface failures and allow a fresh
 gesture retry. Never queue playback across Stop, hiding, or disposal.
 
-Use the [package quick start and API reference](../README.md) for activation,
-validation, ownership, limits, seeded replay, and native interop. Read the
+Use the installed package's exported types, schema, capabilities, and actual
+version to determine supported options and limits. The current website may
+postdate an older installed package. Consult the [API guide](https://audiobits.joacod.com/docs/api),
+[lifecycle](https://audiobits.joacod.com/docs/lifecycle),
+[seeded replay](https://audiobits.joacod.com/docs/parameters), and
+[native interop](https://audiobits.joacod.com/docs/native) for human-readable guidance,
+checking it against the installed contracts. Read the
 [controlled sound example](references/controls.md) for live thruster and bus setup.
 
 Keep recipe data JSON-only. Run `validateRecipe` for semantic validation;

package/skill/references/controls.md

@@ -1,8 +1,10 @@
 # Controlled sound and routing
 
 Call `play()` directly from a gesture. Surface rejected promises, bind Stop,
-and call `dispose()` when the host unmounts. See the package reference for
-bus limits and caller-owned native nodes.
+and call `dispose()` when the host unmounts. See the
+[bus guide](https://audiobits.joacod.com/docs/buses) and
+[native ownership guide](https://audiobits.joacod.com/docs/native), checking options
+against the installed package's exported types and capabilities.
 
 ```ts
 import { createAudio } from "audiobits";