@torrent-tv/proxy 2.87.4 → 2.87.6

package/services/media/container/Container.js

@@ -1,603 +1,603 @@
-/**
- * @file Base Container — abstract per RFC 9559 / ISO 14496-12.
- *
- * A Container knows how to read its own format's track table and index.
- * Concrete containers (MatroskaContainer, Mp4Container, AviContainer) implement
- * spec-specific parsing. All byte access goes through `readRange(start,end)` so
- * the class works over torrent piece windows.
- *
- * Spec refs:
- *  - Matroska RFC 9559 §5: EBML, Segment, SeekHead, Tracks, Cues, Clusters
- *  - MP4 ISO/IEC 14496-12 §8: ftyp, moov, trak, tkhd, mdhd, hdlr, elng, stbl
- *  - AVI RIFF §: LIST hdrl, idx1
- */
-
-/**
- * What one file declares about itself. Every field is either a value the
- * container states or `null`, which means the container does not state it —
- * a defined absence, not "nobody has looked yet".
- *
- * @typedef {object} ContainerMediaInfo
- * @property {string} format - "matroska" | "mp4" | "avi" | "unknown".
- * @property {number | null} durationSeconds
- * @property {number | null} startTimeSeconds - Where this file's own timeline
- *   begins. Two files of one release need not agree on it, and the difference
- *   is what keeps a soundtrack shipped separately aligned with its picture.
- */
-
-export class Container {
-  /**
-   * Two ways of reading, because two readers of one file want different things
-   * and the file is opened once.
-   *
-   * `readRange` fetches what is missing: the head and the track table are
-   * kilobytes, they are needed before anything can be offered, and the codec
-   * probe has already pulled the head of every file that plays. `readHeld`
-   * reads only what is already downloaded and asks the swarm for nothing, which
-   * is what the cue walk needs — turning subtitles on must not pull bytes the
-   * viewer is not waiting for. `isHeld` says whether a range can be read that
-   * way at all, so the walk can leave a cluster for next time instead of
-   * blocking on it.
-   *
-   * They are given per container rather than per call because a container is
-   * cached per file and both readers want the same parsed head. Where the
-   * caller supplies only `readRange`, the held reader is that one and every
-   * range counts as held — which is the right answer for a local file, and for
-   * a torrent it is the caller's job to say otherwise.
-   *
-   * The torrent is NOT passed in and must not be: `readHeld` and `isHeld` are
-   * the only two facts about it this layer needs, and reducing it to two
-   * functions is what keeps the container ignorant of where its bytes live.
-   *
-   * @param {object} params
-   * @param {(start:number,end:number)=>Promise<Buffer|null>} params.readRange
-   * @param {number} params.fileSize
-   * @param {string} [params.label]
-   * @param {(start:number,end:number)=>Promise<Buffer|null>} [params.readHeld]
-   * @param {(start:number,end:number)=>boolean} [params.isHeld]
-   */
-  constructor({ readRange, fileSize, label = "", readHeld, isHeld }) {
-    this.readRange = readRange;
-    this.fileSize = fileSize;
-    this.label = label;
-    this.readHeld = typeof readHeld === "function" ? readHeld : readRange;
-    this.isHeld = typeof isHeld === "function" ? isHeld : () => true;
-  }
-
-
-  /**
-   * Whether one of ffmpeg's banner streams and one of this container's tracks
-   * can be the same track.
-   *
-   * @param {{ language?: string, title?: string }} banner
-   * @param {{ language?: string, name?: string }} container
-   * @returns {boolean}
-   */
-  static pairingHolds(banner, container) {
-    return pairingHolds(banner, container);
-  }
-
-  /**
-   * ffmpeg's description of a file's subtitle streams, corrected by what the
-   * container itself declares.
-   *
-   * The banner cannot express the difference between "no track is marked" and
-   * "every track is marked", because Matroska's `FlagDefault` defaults to 1 and
-   * ffmpeg has already applied that default by the time it prints. The
-   * container can be asked, and this is where the two readings are lined up —
-   * by position, checked pair by pair rather than assumed.
-   *
-   * @param {object[]} bannerTracks
-   * @param {object[]} declared - This container's own subtitle tracks, in its order.
-   * @returns {object[]}
-   */
-    /**
-   * The picture's facts, from the two readings that state them.
-   *
-   * Audio and subtitles have had this since the flags were first read from the
-   * file; video never did. Every figure the encode is planned from — the size,
-   * the frame rate, whether it is HDR, how many bits a sample carries — was
-   * taken from ffmpeg's `-i` banner alone, and the `VideoTrack` the container
-   * declares was read and then used for nothing but a line in the log.
-   *
-   * Which reading wins is decided per field by what each one IS, and the rule
-   * is the one `readMediaInfo` already states: a fact the container DECLARES is
-   * read from the container; a fact only the media has is measured from the
-   * media.
-   *
-   * - the coded size and the frame rate are the BANNER's. Both are declared by
-   *   the container as well, but what the encoder receives is what the decoder
-   *   produced, and the ladder and the scale filter have to be sized to that. A
-   *   container that declares something else is mis-declaring, and using its
-   *   numbers would size the encode to a picture that never arrives;
-   * - the bit depth and the HDR signalling are the CONTAINER's where it states
-   *   them. They are not properties of the decoded frames at all — they are the
-   *   file saying how its samples are to be read — and ffmpeg prints them only
-   *   as a side effect of naming a pixel format. HDR is not compared for
-   *   disagreement: both sides give a boolean, and a boolean cannot say "I did
-   *   not look";
-   * - the display size is the container's alone; the banner has no such field.
-   *
-   * Where only one side states a field, that side answers whatever the rule
-   * would have preferred. Where both state it and they DISAGREE, the
-   * disagreement is reported: it is a fact about the file, and until now
-   * nothing could see it.
-   *
-   * @param {{ width?: number|null, height?: number|null, fps?: number|null, isHdr?: boolean, bitDepth?: number|null }} banner
-   * @param {object | null} declared - The container's own `VideoTrack`.
-   * @returns {{ width: number|null, height: number|null, fps: number|null, isHdr: boolean, bitDepth: number|null, displayWidth: number|null, displayHeight: number|null, disagreements: string[] }}
-   */
-  static mergeVideoFacts(banner, declared) {
-    const number = (value) => (Number.isFinite(value) && value > 0 ? Number(value) : null);
-    const fromBanner = {
-      width: number(banner?.width),
-      height: number(banner?.height),
-      fps: number(banner?.fps),
-      isHdr: banner?.isHdr === true,
-      bitDepth: number(banner?.bitDepth)
-    };
-    if (!declared) {
-      return { ...fromBanner, displayWidth: null, displayHeight: null, disagreements: [] };
-    }
-    const fromContainer = {
-      width: number(declared.width),
-      height: number(declared.height),
-      fps: number(declared.fps),
-      isHdr: declared.isHdr === true,
-      bitDepth: number(declared.bitDepth)
-    };
-    const disagreements = [];
-    const note = (field, mine, theirs) => {
-      if (mine !== null && theirs !== null && mine !== theirs) {
-        disagreements.push(`${field} ${theirs} in the container against ${mine} in the probe`);
-      }
-    };
-    note("width", fromBanner.width, fromContainer.width);
-    note("height", fromBanner.height, fromContainer.height);
-    if (fromBanner.bitDepth !== null && fromContainer.bitDepth !== null && fromBanner.bitDepth !== fromContainer.bitDepth) {
-      disagreements.push(
-        `bit depth ${fromContainer.bitDepth} in the container against ${fromBanner.bitDepth} in the probe`
-      );
-    }
-    // HDR is deliberately NOT compared. Both sides give it as a boolean, and a
-    // boolean cannot say "I did not look": a container with no Colour element
-    // and one that states SDR are the same `false`, as are a probe that printed
-    // no colour metadata and one that printed BT.709. Reporting that as a
-    // disagreement would report it on almost every file. Either side saying yes
-    // is taken as yes, which is the safe direction — the cost of tone mapping a
-    // picture that did not need it is smaller than showing a washed-out one.
-    return {
-      width: fromBanner.width ?? fromContainer.width,
-      height: fromBanner.height ?? fromContainer.height,
-      fps: fromBanner.fps ?? fromContainer.fps,
-      // The container declares these; the probe only reflects them.
-      bitDepth: fromContainer.bitDepth ?? fromBanner.bitDepth,
-      isHdr: fromContainer.isHdr || fromBanner.isHdr,
-      displayWidth: number(declared.displayWidth),
-      displayHeight: number(declared.displayHeight),
-      disagreements
-    };
-  }
-
-  /**
-   * Line ffmpeg's banner up with what the container declares, and say whether the
-   * two are describing the same thing in the same order.
-   *
-   * ffmpeg numbers a file's streams `0:a:N` / `0:s:N` over every stream of that
-   * kind, in the order the container declares them; the container reading is a
-   * list in that same order. So position is the correspondence — but a position
-   * match that is merely assumed is worth nothing, and it is CHECKED: each pair
-   * has to agree on language or on title. One pair that agrees on neither, or a
-   * length that differs, means the two readings are not about the same thing, and
-   * then the container reading is not used AT ALL. A wrong flag is worse than a
-   * missing one, because `0:a:N` is what the encoder is given.
-   *
-   * One alignment for every media kind, because it is one rule. It was written
-   * twice — once for subtitles, once for audio — down to a `pairingHolds` that
-   * was byte-for-byte the same function under two names.
-   *
-   * @param {object[]} bannerTracks
-   * @param {object[]} declared
-   * @param {string} noun - "subtitle" or "audio"; only for the reason text.
-   * @returns {{ aligned: boolean, reason: string, banner: object[], container: object[] }}
-   */
-  static alignWithBanner(bannerTracks, declared, noun) {
-    const banner = Array.isArray(bannerTracks) ? bannerTracks : [];
-    const container = Array.isArray(declared) ? declared : [];
-    if (banner.length === 0) {
-      return { aligned: false, reason: `the probe found no ${noun} stream`, banner, container };
-    }
-    if (container.length === 0) {
-      return { aligned: false, reason: `the container declares no ${noun} track`, banner, container };
-    }
-    if (container.length !== banner.length) {
-      return {
-        aligned: false,
-        reason: `the container declares ${container.length} ${noun} tracks and the probe found ${banner.length}`,
-        banner,
-        container
-      };
-    }
-    for (const [order, track] of banner.entries()) {
-      if (!Container.pairingHolds(track, container[order])) {
-        return {
-          aligned: false,
-          reason:
-            `${noun} ${order} is "${normalise(track?.title) || "-"}"/${normalise(track?.language) || "-"} ` +
-            `in the probe and "${normalise(container[order]?.name) || "-"}"/` +
-            `${normalise(container[order]?.language) || "-"} in the container`,
-          banner,
-          container
-        };
-      }
-    }
-    return { aligned: true, reason: "", banner, container };
-  }
-
-  /**
-   * What the banner says, with the flags only the container knows added — or the
-   * banner alone where the two could not be lined up.
-   *
-   * `take` says which of the container's fields this kind of track wants, and is
-   * the only part that differs between them.
-   *
-   * @param {object[]} bannerTracks
-   * @param {object[]} declared
-   * @param {string} noun
-   * @param {(containerTrack: object, bannerTrack: object) => object} take
-   * @param {object} absent - The same fields as `take` returns, for a track whose
-   *   container reading could not be used. Not "the container says no" — the
-   *   container has not been heard from.
-   * @returns {{ tracks: object[], aligned: boolean, reason: string }}
-   */
-  static mergeDeclaredFlags(bannerTracks, declared, noun, take, absent) {
-    const { aligned, reason, banner, container } = Container.alignWithBanner(bannerTracks, declared, noun);
-    if (!aligned) {
-      return {
-        tracks: banner.map((track) => ({ ...track, ...absent })),
-        aligned,
-        reason
-      };
-    }
-    return {
-      tracks: banner.map((track, order) => ({ ...track, ...take(container[order], track) })),
-      aligned: true,
-      reason: ""
-    };
-  }
-
-  static mergeSubtitleFlags(bannerTracks, declared) {
-    return Container.mergeDeclaredFlags(
-      bannerTracks,
-      declared,
-      "subtitle",
-      (track) => ({
-        // Matroska applies `eng` when Language is absent, and ffmpeg reports
-        // that default as if the file had stated it. Preserve only an explicit
-        // container value; otherwise let cue-text detection decide later.
-        language: track.declaresLanguage === true
-          ? (track.languageBcp47 || track.language || "")
-          : "",
-        declaresLanguage: track.declaresLanguage === true,
-        isDefault: track.isDefault === true,
-        declaresDefault: track.declaresDefault === true,
-        // Read from the file rather than guessed from the track's name. Both
-        // are stated by the container itself (RFC 9559 §5.1.4.1) and neither
-        // reaches ffmpeg's `-i` banner, which is where every other field here
-        // comes from.
-        isForced: track.isForced === true,
-        isHearingImpaired: track.isHearingImpaired === true,
-        // FlagEnabled, so the browser can leave an unusable track out of the
-        // menu. It stays in the list and keeps its number: ffmpeg creates a
-        // stream for it either way.
-        isEnabled: track.isEnabled !== false,
-        // The RFC 5646 tag where the file writes one, kept BESIDE the code
-        // rather than replacing it: this list is aligned against ffmpeg's
-        // banner, which prints the three-letter form.
-        languageBcp47: typeof track.languageBcp47 === "string" ? track.languageBcp47 : ""
-      }),
-      {
-        declaresDefault: false,
-        isForced: false,
-        isHearingImpaired: false,
-        isEnabled: true,
-        languageBcp47: ""
-      }
-    );
-  }
-
-  /**
-   * The same for audio, with the flags RFC 9559 §5.1.4.1 defines for it.
-   *
-   * `FlagOriginal`, `FlagCommentary` and `FlagVisualImpaired` do not appear in
-   * the banner at all, so without this the audio menu cannot tell a director's
-   * commentary from the film.
-   *
-   * @param {object[]} bannerTracks
-   * @param {object[]} declared
-   * @returns {{ tracks: object[], aligned: boolean, reason: string }}
-   */
-  static mergeAudioFlags(bannerTracks, declared) {
-    return Container.mergeDeclaredFlags(
-      bannerTracks,
-      declared,
-      "audio",
-      (track, banner) => ({
-        isOriginal: track.isOriginal === true,
-        isCommentary: track.isCommentary === true,
-        isVisualImpaired: track.isVisualImpaired === true,
-        isEnabled: track.isEnabled !== false,
-        isDefault: track.isDefault === true,
-        declaresDefault: track.declaresDefault === true,
-        languageBcp47: typeof track.languageBcp47 === "string" ? track.languageBcp47 : "",
-        channels: Number.isFinite(track.channels) ? track.channels : null,
-        title:
-          typeof banner?.title === "string" && banner.title.length > 0
-            ? banner.title
-            : (typeof track.name === "string" ? track.name : "")
-      }),
-      {
-        declaresDefault: false,
-        isOriginal: false,
-        isCommentary: false,
-        isVisualImpaired: false,
-        isEnabled: true,
-        languageBcp47: "",
-        channels: null
-      }
-    );
-  }
-
-
-  /** @returns {string} Human name: "matroska" | "mp4" | "avi" | "unknown" */
-  get formatName() {
-    return "unknown";
-  }
-
-  /** Whether `head` (first bytes) looks like this container. */
-  static detect(_head) {
-    return false;
-  }
-
-  /**
-   * All tracks declared by the container, in container order.
-   * Includes disabled tracks (isEnabled=false) to preserve declaredIndex alignment with ffmpeg.
-   * @returns {Promise<import("../tracks/index.js").ContainerTrack[]>}
-   */
-  async readTracks() {
-    throw new Error("readTracks not implemented");
-  }
-
-
-  /**
-   * This container's subtitle tracks and where their cues are, in ONE shape
-   * whichever container answers.
-   *
-   * `tracks` carries a `clusterPositions` list for a container that stores cues
-   * in clusters and a `samples` list for one that states each cue's own byte
-   * range; a caller reads neither, and asks {@link Container#readHeldCues}
-   * instead. `declared` is what the container says about its subtitle tracks in
-   * its own order, and empty means the container said nothing — a real answer,
-   * not a missing one.
-   *
-   * @returns {Promise<{ tracks: object[], declared: object[], secondsPerTick: number, segmentDataOffset: number } | null>}
-   *   Null where this container declares no subtitles at all.
-   */
-  async readSubtitlePlan() {
-    return null;
-  }
-
-  /**
-   * The cues this container can read RIGHT NOW for one track, without fetching.
-   *
-   * Every container answers this, and each reads what its own specification
-   * says: Matroska walks the clusters its Cues table names, an MP4 reads the
-   * samples its table states. The caller therefore chooses a container once —
-   * from the bytes — and never again. It used to choose twice, once by file
-   * extension for the container and once by whether a track carried a sample
-   * list for the reading, and two choices that must agree and are made from
-   * different evidence are a disagreement waiting to happen.
-   *
-   * `progress` is what has already been read, kept by the caller because it
-   * belongs to the file rather than to one pass: `walked` holds cluster
-   * positions, `harvested` holds sample offsets per track. Each container adds
-   * to the one it uses.
-   *
-   * @param {object} _plan - This file's subtitle plan.
-   * @param {object} _track - The track asked about.
-   * @param {{ walked: Set<number>, harvested: Map<number, Set<number>> }} _progress
-   * @returns {Promise<{ found: Map<number, object[]>, covered: number, indexed: number }>}
-   *   `found` is track number to the cues found in THIS pass — Matroska fills
-   *   every track from one walk, so it is a map and not a list.
-   */
-  async readHeldCues(_plan, _track, _progress) {
-    return { found: new Map(), covered: 0, indexed: 0 };
-  }
-
-  /**
-   * Keyframe times for the video track, ascending seconds. Null when index
-   * absent (MPEG-TS, fragmented MP4, truncated).
-   *
-   * **It reads, and does not remember.** It used to keep the answer, and that
-   * made the table the one fact in this proxy stored in two places: here, and
-   * in the file's `KeyframeTable` on the main thread. Traced 2026-09-15, this
-   * copy had no reader: nothing ever asked a container twice for its keyframes.
-   * The parse itself has since moved to the thread the sessions are on, so
-   * there is no longer a second thread for a second copy to sit in either.
-   *
-   * The remembering lives where the fact is USED, which is also the only place
-   * that can do the rest of the job: bound how long a viewer waits for it
-   * while letting the read go on, join a second asker to a read already
-   * running, and keep the answer reachable by a session created before it
-   * landed. A container cannot do any of those — it does not know there is a
-   * viewer — so a copy here was never the whole answer, only a second one.
-   *
-   * @returns {Promise<{times:number[],tolerance:number}|null>}
-   */
-  async readKeyframeIndex() {
-    return this.parseKeyframeIndex();
-  }
-
-  /**
-   * How THIS format states where its keyframes are. Overridden by every
-   * container that has such a table; the default is the honest answer for one
-   * that does not.
-   *
-   * @returns {Promise<{times:number[],tolerance:number}|null>}
-   */
-  async parseKeyframeIndex() {
-    return null;
-  }
-
-  /**
-   * What this file DECLARES about itself as a whole, as distinct from what its
-   * individual tracks declare.
-   *
-   * The rule this method exists to hold: a fact the container declares is read
-   * from the container; a fact only the media itself has is measured from the
-   * media. Both halves used to be asked of ffmpeg, so the same header was read
-   * twice — measured 2026-09-03, this layer read one `.mka` header in 8 ms while
-   * a second ffmpeg read the same header over HTTP for 8121 ms, in the same
-   * second, on the same file.
-   *
-   * `null` is not "unknown, ask someone else". It means the container does not
-   * declare the field, which is a final answer about the container and the point
-   * at which a caller may go to the media — see `docs/container-architecture.md`.
-   *
-   * @returns {Promise<import("./Container.js").ContainerMediaInfo>}
-   */
-  async readMediaInfo() {
-    if (!this.mediaInfo) {
-      this.mediaInfo = {
-        format: this.formatName,
-        durationSeconds: null,
-        startTimeSeconds: null
-      };
-    }
-    return this.mediaInfo;
-  }
-
-  /**
-   * Subtitle-specific: where cues live (Matroska cluster positions or MP4 sample ranges).
-   * Returned via track objects' clusterPositions/samples, so base has no extra method — tracks carry it.
-   */
-
-  /**
-   * The TEXT FIELD of one subtitle cue, taken out of this container's framing.
-   *
-   * How a cue's bytes are wrapped is stated by the container's own
-   * specification, so each subclass answers for itself: Matroska reorders an ASS
-   * dialogue row, drops its two timing fields and prepends a read order
-   * (`matroska.org/technical/subtitles.html`); an MP4 prefixes a `tx3g` sample
-   * with its length (ISO/IEC 14496-12 §12.6); a subtitle FILE states its own
-   * field order in `[Events]`. None of that is a fact about the subtitle format,
-   * and the format's own markup — `{\pos(…)}`, `\N` — is not a fact about the
-   * container. The second half is `TextSubtitleTrack`; this is the
-   * first, and the two are applied in that order.
-   *
-   * Static because de-framing reads no instance state: a caller that has bytes
-   * and knows the format needs no container built over the whole file. The
-   * instance form below exists so a caller that DOES hold a container gets the
-   * right answer without naming the subclass.
-   *
-   * @param {Buffer} _payload - The cue's bytes as the container stores them.
-   * @param {string} _codecId - CodecID / sample entry type / file extension.
-   * @returns {string} The text field, markup still in place.
-   */
-  static cueTextOf(_payload, _codecId) {
-    throw new Error("cueTextOf not implemented");
-  }
-
-  /**
-   * @param {Buffer} payload
-   * @param {string} codecId
-   * @returns {string}
-   */
-  cueTextOf(payload, codecId) {
-    return /** @type {typeof Container} */ (this.constructor).cueTextOf(payload, codecId);
-  }
-}
-
-// ---------------------------------------------------------------------------
-// Lining ffmpeg's banner up with what a container declares. Here because the
-// correction is about what a CONTAINER states and the banner cannot.
-// ---------------------------------------------------------------------------
-/**
- * Which subtitle track the FILE says to show, read from the file rather than
- * from ffmpeg's description of it.
- *
- * Why this exists. The browser decides which subtitle track to turn on from
- * `isDefault`, and until now that came from ffmpeg's `-i` banner, which prints
- * `(default)`. In Matroska `FlagDefault` DEFAULTS TO 1 and ffmpeg has already
- * applied that default by the time it prints — so a file whose muxer wrote the
- * flag on no track arrives looking exactly like one that wrote it on every
- * track: everything marked. The banner cannot tell the two apart, and the
- * difference is the whole question, because one of them means "show this one"
- * and the other means "the file has no opinion".
- *
- * The container itself can be asked, and the EBML reader already walks the
- * Tracks element for subtitle extraction. What it now also records is whether
- * the element was WRITTEN, which is the fact the banner destroys.
- *
- * The awkward part is lining the two readings up. ffmpeg numbers its subtitle
- * streams `0:s:0`, `0:s:1`, … over EVERY subtitle stream, picture-based ones
- * included, in the order the container declares them; the container reading is
- * a list in that same order. So position is the correspondence — but a position
- * match that is merely assumed is worth nothing, so it is CHECKED: each pair
- * has to agree on language or on title. One pair that agrees on neither, or a
- * length that differs, means the two readings are not describing the same
- * thing in the same order, and then the container reading is not used at all.
- */
-
-/**
- * Language codes that carry no information, and so cannot confirm a pairing.
- *
- * ffmpeg prints `und` for a stream with no language; Matroska's own default
- * for `Language` is `eng`, which is why an absent element cannot be read as a
- * statement either — but `eng` is also a real answer, so it is not listed here
- * and is compared like any other.
- */
-const EMPTY_LANGUAGES = new Set(["", "und", "unknown"]);
-
-/**
- * @param {unknown} value
- * @returns {string}
- */
-function normalise(value) {
-  return typeof value === "string" ? value.trim().toLowerCase() : "";
-}
-
-/**
- * Whether one banner stream and one container track can be the same track.
- *
- * Agreement on either the language or the name is enough; both being empty is
- * not agreement, because two tracks that say nothing about themselves say
- * nothing about their pairing either.
- *
- * @param {{ language?: string, title?: string }} banner
- * @param {{ language?: string, name?: string }} container
- * @returns {boolean}
- */
-function pairingHolds(banner, container) {
-  const bannerLanguage = normalise(banner?.language);
-  const containerLanguage = normalise(container?.language);
-  if (
-    !EMPTY_LANGUAGES.has(bannerLanguage) &&
-    !EMPTY_LANGUAGES.has(containerLanguage) &&
-    bannerLanguage === containerLanguage
-  ) {
-    return true;
-  }
-  const bannerTitle = normalise(banner?.title);
-  const containerName = normalise(container?.name);
-  if (bannerTitle.length > 0 && bannerTitle === containerName) {
-    return true;
-  }
-  // Nothing to compare on either side. Not a disagreement — a file may name
-  // neither — so it does not break the alignment; it simply adds no support.
-  return (
-    (EMPTY_LANGUAGES.has(bannerLanguage) || EMPTY_LANGUAGES.has(containerLanguage)) &&
-    (bannerTitle.length === 0 || containerName.length === 0)
-  );
-}
-
+/**
+ * @file Base Container — abstract per RFC 9559 / ISO 14496-12.
+ *
+ * A Container knows how to read its own format's track table and index.
+ * Concrete containers (MatroskaContainer, Mp4Container, AviContainer) implement
+ * spec-specific parsing. All byte access goes through `readRange(start,end)` so
+ * the class works over torrent piece windows.
+ *
+ * Spec refs:
+ *  - Matroska RFC 9559 §5: EBML, Segment, SeekHead, Tracks, Cues, Clusters
+ *  - MP4 ISO/IEC 14496-12 §8: ftyp, moov, trak, tkhd, mdhd, hdlr, elng, stbl
+ *  - AVI RIFF §: LIST hdrl, idx1
+ */
+
+/**
+ * What one file declares about itself. Every field is either a value the
+ * container states or `null`, which means the container does not state it —
+ * a defined absence, not "nobody has looked yet".
+ *
+ * @typedef {object} ContainerMediaInfo
+ * @property {string} format - "matroska" | "mp4" | "avi" | "unknown".
+ * @property {number | null} durationSeconds
+ * @property {number | null} startTimeSeconds - Where this file's own timeline
+ *   begins. Two files of one release need not agree on it, and the difference
+ *   is what keeps a soundtrack shipped separately aligned with its picture.
+ */
+
+export class Container {
+  /**
+   * Two ways of reading, because two readers of one file want different things
+   * and the file is opened once.
+   *
+   * `readRange` fetches what is missing: the head and the track table are
+   * kilobytes, they are needed before anything can be offered, and the codec
+   * probe has already pulled the head of every file that plays. `readHeld`
+   * reads only what is already downloaded and asks the swarm for nothing, which
+   * is what the cue walk needs — turning subtitles on must not pull bytes the
+   * viewer is not waiting for. `isHeld` says whether a range can be read that
+   * way at all, so the walk can leave a cluster for next time instead of
+   * blocking on it.
+   *
+   * They are given per container rather than per call because a container is
+   * cached per file and both readers want the same parsed head. Where the
+   * caller supplies only `readRange`, the held reader is that one and every
+   * range counts as held — which is the right answer for a local file, and for
+   * a torrent it is the caller's job to say otherwise.
+   *
+   * The torrent is NOT passed in and must not be: `readHeld` and `isHeld` are
+   * the only two facts about it this layer needs, and reducing it to two
+   * functions is what keeps the container ignorant of where its bytes live.
+   *
+   * @param {object} params
+   * @param {(start:number,end:number)=>Promise<Buffer|null>} params.readRange
+   * @param {number} params.fileSize
+   * @param {string} [params.label]
+   * @param {(start:number,end:number)=>Promise<Buffer|null>} [params.readHeld]
+   * @param {(start:number,end:number)=>boolean} [params.isHeld]
+   */
+  constructor({ readRange, fileSize, label = "", readHeld, isHeld }) {
+    this.readRange = readRange;
+    this.fileSize = fileSize;
+    this.label = label;
+    this.readHeld = typeof readHeld === "function" ? readHeld : readRange;
+    this.isHeld = typeof isHeld === "function" ? isHeld : () => true;
+  }
+
+
+  /**
+   * Whether one of ffmpeg's banner streams and one of this container's tracks
+   * can be the same track.
+   *
+   * @param {{ language?: string, title?: string }} banner
+   * @param {{ language?: string, name?: string }} container
+   * @returns {boolean}
+   */
+  static pairingHolds(banner, container) {
+    return pairingHolds(banner, container);
+  }
+
+  /**
+   * ffmpeg's description of a file's subtitle streams, corrected by what the
+   * container itself declares.
+   *
+   * The banner cannot express the difference between "no track is marked" and
+   * "every track is marked", because Matroska's `FlagDefault` defaults to 1 and
+   * ffmpeg has already applied that default by the time it prints. The
+   * container can be asked, and this is where the two readings are lined up —
+   * by position, checked pair by pair rather than assumed.
+   *
+   * @param {object[]} bannerTracks
+   * @param {object[]} declared - This container's own subtitle tracks, in its order.
+   * @returns {object[]}
+   */
+    /**
+   * The picture's facts, from the two readings that state them.
+   *
+   * Audio and subtitles have had this since the flags were first read from the
+   * file; video never did. Every figure the encode is planned from — the size,
+   * the frame rate, whether it is HDR, how many bits a sample carries — was
+   * taken from ffmpeg's `-i` banner alone, and the `VideoTrack` the container
+   * declares was read and then used for nothing but a line in the log.
+   *
+   * Which reading wins is decided per field by what each one IS, and the rule
+   * is the one `readMediaInfo` already states: a fact the container DECLARES is
+   * read from the container; a fact only the media has is measured from the
+   * media.
+   *
+   * - the coded size and the frame rate are the BANNER's. Both are declared by
+   *   the container as well, but what the encoder receives is what the decoder
+   *   produced, and the ladder and the scale filter have to be sized to that. A
+   *   container that declares something else is mis-declaring, and using its
+   *   numbers would size the encode to a picture that never arrives;
+   * - the bit depth and the HDR signalling are the CONTAINER's where it states
+   *   them. They are not properties of the decoded frames at all — they are the
+   *   file saying how its samples are to be read — and ffmpeg prints them only
+   *   as a side effect of naming a pixel format. HDR is not compared for
+   *   disagreement: both sides give a boolean, and a boolean cannot say "I did
+   *   not look";
+   * - the display size is the container's alone; the banner has no such field.
+   *
+   * Where only one side states a field, that side answers whatever the rule
+   * would have preferred. Where both state it and they DISAGREE, the
+   * disagreement is reported: it is a fact about the file, and until now
+   * nothing could see it.
+   *
+   * @param {{ width?: number|null, height?: number|null, fps?: number|null, isHdr?: boolean, bitDepth?: number|null }} banner
+   * @param {object | null} declared - The container's own `VideoTrack`.
+   * @returns {{ width: number|null, height: number|null, fps: number|null, isHdr: boolean, bitDepth: number|null, displayWidth: number|null, displayHeight: number|null, disagreements: string[] }}
+   */
+  static mergeVideoFacts(banner, declared) {
+    const number = (value) => (Number.isFinite(value) && value > 0 ? Number(value) : null);
+    const fromBanner = {
+      width: number(banner?.width),
+      height: number(banner?.height),
+      fps: number(banner?.fps),
+      isHdr: banner?.isHdr === true,
+      bitDepth: number(banner?.bitDepth)
+    };
+    if (!declared) {
+      return { ...fromBanner, displayWidth: null, displayHeight: null, disagreements: [] };
+    }
+    const fromContainer = {
+      width: number(declared.width),
+      height: number(declared.height),
+      fps: number(declared.fps),
+      isHdr: declared.isHdr === true,
+      bitDepth: number(declared.bitDepth)
+    };
+    const disagreements = [];
+    const note = (field, mine, theirs) => {
+      if (mine !== null && theirs !== null && mine !== theirs) {
+        disagreements.push(`${field} ${theirs} in the container against ${mine} in the probe`);
+      }
+    };
+    note("width", fromBanner.width, fromContainer.width);
+    note("height", fromBanner.height, fromContainer.height);
+    if (fromBanner.bitDepth !== null && fromContainer.bitDepth !== null && fromBanner.bitDepth !== fromContainer.bitDepth) {
+      disagreements.push(
+        `bit depth ${fromContainer.bitDepth} in the container against ${fromBanner.bitDepth} in the probe`
+      );
+    }
+    // HDR is deliberately NOT compared. Both sides give it as a boolean, and a
+    // boolean cannot say "I did not look": a container with no Colour element
+    // and one that states SDR are the same `false`, as are a probe that printed
+    // no colour metadata and one that printed BT.709. Reporting that as a
+    // disagreement would report it on almost every file. Either side saying yes
+    // is taken as yes, which is the safe direction — the cost of tone mapping a
+    // picture that did not need it is smaller than showing a washed-out one.
+    return {
+      width: fromBanner.width ?? fromContainer.width,
+      height: fromBanner.height ?? fromContainer.height,
+      fps: fromBanner.fps ?? fromContainer.fps,
+      // The container declares these; the probe only reflects them.
+      bitDepth: fromContainer.bitDepth ?? fromBanner.bitDepth,
+      isHdr: fromContainer.isHdr || fromBanner.isHdr,
+      displayWidth: number(declared.displayWidth),
+      displayHeight: number(declared.displayHeight),
+      disagreements
+    };
+  }
+
+  /**
+   * Line ffmpeg's banner up with what the container declares, and say whether the
+   * two are describing the same thing in the same order.
+   *
+   * ffmpeg numbers a file's streams `0:a:N` / `0:s:N` over every stream of that
+   * kind, in the order the container declares them; the container reading is a
+   * list in that same order. So position is the correspondence — but a position
+   * match that is merely assumed is worth nothing, and it is CHECKED: each pair
+   * has to agree on language or on title. One pair that agrees on neither, or a
+   * length that differs, means the two readings are not about the same thing, and
+   * then the container reading is not used AT ALL. A wrong flag is worse than a
+   * missing one, because `0:a:N` is what the encoder is given.
+   *
+   * One alignment for every media kind, because it is one rule. It was written
+   * twice — once for subtitles, once for audio — down to a `pairingHolds` that
+   * was byte-for-byte the same function under two names.
+   *
+   * @param {object[]} bannerTracks
+   * @param {object[]} declared
+   * @param {string} noun - "subtitle" or "audio"; only for the reason text.
+   * @returns {{ aligned: boolean, reason: string, banner: object[], container: object[] }}
+   */
+  static alignWithBanner(bannerTracks, declared, noun) {
+    const banner = Array.isArray(bannerTracks) ? bannerTracks : [];
+    const container = Array.isArray(declared) ? declared : [];
+    if (banner.length === 0) {
+      return { aligned: false, reason: `the probe found no ${noun} stream`, banner, container };
+    }
+    if (container.length === 0) {
+      return { aligned: false, reason: `the container declares no ${noun} track`, banner, container };
+    }
+    if (container.length !== banner.length) {
+      return {
+        aligned: false,
+        reason: `the container declares ${container.length} ${noun} tracks and the probe found ${banner.length}`,
+        banner,
+        container
+      };
+    }
+    for (const [order, track] of banner.entries()) {
+      if (!Container.pairingHolds(track, container[order])) {
+        return {
+          aligned: false,
+          reason:
+            `${noun} ${order} is "${normalise(track?.title) || "-"}"/${normalise(track?.language) || "-"} ` +
+            `in the probe and "${normalise(container[order]?.name) || "-"}"/` +
+            `${normalise(container[order]?.language) || "-"} in the container`,
+          banner,
+          container
+        };
+      }
+    }
+    return { aligned: true, reason: "", banner, container };
+  }
+
+  /**
+   * What the banner says, with the flags only the container knows added — or the
+   * banner alone where the two could not be lined up.
+   *
+   * `take` says which of the container's fields this kind of track wants, and is
+   * the only part that differs between them.
+   *
+   * @param {object[]} bannerTracks
+   * @param {object[]} declared
+   * @param {string} noun
+   * @param {(containerTrack: object, bannerTrack: object) => object} take
+   * @param {object} absent - The same fields as `take` returns, for a track whose
+   *   container reading could not be used. Not "the container says no" — the
+   *   container has not been heard from.
+   * @returns {{ tracks: object[], aligned: boolean, reason: string }}
+   */
+  static mergeDeclaredFlags(bannerTracks, declared, noun, take, absent) {
+    const { aligned, reason, banner, container } = Container.alignWithBanner(bannerTracks, declared, noun);
+    if (!aligned) {
+      return {
+        tracks: banner.map((track) => ({ ...track, ...absent })),
+        aligned,
+        reason
+      };
+    }
+    return {
+      tracks: banner.map((track, order) => ({ ...track, ...take(container[order], track) })),
+      aligned: true,
+      reason: ""
+    };
+  }
+
+  static mergeSubtitleFlags(bannerTracks, declared) {
+    return Container.mergeDeclaredFlags(
+      bannerTracks,
+      declared,
+      "subtitle",
+      (track) => ({
+        // Matroska applies `eng` when Language is absent, and ffmpeg reports
+        // that default as if the file had stated it. Preserve only an explicit
+        // container value; otherwise let cue-text detection decide later.
+        language: track.declaresLanguage === true
+          ? (track.languageBcp47 || track.language || "")
+          : "",
+        declaresLanguage: track.declaresLanguage === true,
+        isDefault: track.isDefault === true,
+        declaresDefault: track.declaresDefault === true,
+        // Read from the file rather than guessed from the track's name. Both
+        // are stated by the container itself (RFC 9559 §5.1.4.1) and neither
+        // reaches ffmpeg's `-i` banner, which is where every other field here
+        // comes from.
+        isForced: track.isForced === true,
+        isHearingImpaired: track.isHearingImpaired === true,
+        // FlagEnabled, so the browser can leave an unusable track out of the
+        // menu. It stays in the list and keeps its number: ffmpeg creates a
+        // stream for it either way.
+        isEnabled: track.isEnabled !== false,
+        // The RFC 5646 tag where the file writes one, kept BESIDE the code
+        // rather than replacing it: this list is aligned against ffmpeg's
+        // banner, which prints the three-letter form.
+        languageBcp47: typeof track.languageBcp47 === "string" ? track.languageBcp47 : ""
+      }),
+      {
+        declaresDefault: false,
+        isForced: false,
+        isHearingImpaired: false,
+        isEnabled: true,
+        languageBcp47: ""
+      }
+    );
+  }
+
+  /**
+   * The same for audio, with the flags RFC 9559 §5.1.4.1 defines for it.
+   *
+   * `FlagOriginal`, `FlagCommentary` and `FlagVisualImpaired` do not appear in
+   * the banner at all, so without this the audio menu cannot tell a director's
+   * commentary from the film.
+   *
+   * @param {object[]} bannerTracks
+   * @param {object[]} declared
+   * @returns {{ tracks: object[], aligned: boolean, reason: string }}
+   */
+  static mergeAudioFlags(bannerTracks, declared) {
+    return Container.mergeDeclaredFlags(
+      bannerTracks,
+      declared,
+      "audio",
+      (track, banner) => ({
+        isOriginal: track.isOriginal === true,
+        isCommentary: track.isCommentary === true,
+        isVisualImpaired: track.isVisualImpaired === true,
+        isEnabled: track.isEnabled !== false,
+        isDefault: track.isDefault === true,
+        declaresDefault: track.declaresDefault === true,
+        languageBcp47: typeof track.languageBcp47 === "string" ? track.languageBcp47 : "",
+        channels: Number.isFinite(track.channels) ? track.channels : null,
+        title:
+          typeof banner?.title === "string" && banner.title.length > 0
+            ? banner.title
+            : (typeof track.name === "string" ? track.name : "")
+      }),
+      {
+        declaresDefault: false,
+        isOriginal: false,
+        isCommentary: false,
+        isVisualImpaired: false,
+        isEnabled: true,
+        languageBcp47: "",
+        channels: null
+      }
+    );
+  }
+
+
+  /** @returns {string} Human name: "matroska" | "mp4" | "avi" | "unknown" */
+  get formatName() {
+    return "unknown";
+  }
+
+  /** Whether `head` (first bytes) looks like this container. */
+  static detect(_head) {
+    return false;
+  }
+
+  /**
+   * All tracks declared by the container, in container order.
+   * Includes disabled tracks (isEnabled=false) to preserve declaredIndex alignment with ffmpeg.
+   * @returns {Promise<import("../tracks/index.js").ContainerTrack[]>}
+   */
+  async readTracks() {
+    throw new Error("readTracks not implemented");
+  }
+
+
+  /**
+   * This container's subtitle tracks and where their cues are, in ONE shape
+   * whichever container answers.
+   *
+   * `tracks` carries a `clusterPositions` list for a container that stores cues
+   * in clusters and a `samples` list for one that states each cue's own byte
+   * range; a caller reads neither, and asks {@link Container#readHeldCues}
+   * instead. `declared` is what the container says about its subtitle tracks in
+   * its own order, and empty means the container said nothing — a real answer,
+   * not a missing one.
+   *
+   * @returns {Promise<{ tracks: object[], declared: object[], secondsPerTick: number, segmentDataOffset: number } | null>}
+   *   Null where this container declares no subtitles at all.
+   */
+  async readSubtitlePlan() {
+    return null;
+  }
+
+  /**
+   * The cues this container can read RIGHT NOW for one track, without fetching.
+   *
+   * Every container answers this, and each reads what its own specification
+   * says: Matroska walks the clusters its Cues table names, an MP4 reads the
+   * samples its table states. The caller therefore chooses a container once —
+   * from the bytes — and never again. It used to choose twice, once by file
+   * extension for the container and once by whether a track carried a sample
+   * list for the reading, and two choices that must agree and are made from
+   * different evidence are a disagreement waiting to happen.
+   *
+   * `progress` is what has already been read, kept by the caller because it
+   * belongs to the file rather than to one pass: `walked` holds cluster
+   * positions, `harvested` holds sample offsets per track. Each container adds
+   * to the one it uses.
+   *
+   * @param {object} _plan - This file's subtitle plan.
+   * @param {object} _track - The track asked about.
+   * @param {{ walked: Set<number>, harvested: Map<number, Set<number>> }} _progress
+   * @returns {Promise<{ found: Map<number, object[]>, covered: number, indexed: number }>}
+   *   `found` is track number to the cues found in THIS pass — Matroska fills
+   *   every track from one walk, so it is a map and not a list.
+   */
+  async readHeldCues(_plan, _track, _progress) {
+    return { found: new Map(), covered: 0, indexed: 0 };
+  }
+
+  /**
+   * Keyframe times for the video track, ascending seconds. Null when index
+   * absent (MPEG-TS, fragmented MP4, truncated).
+   *
+   * **It reads, and does not remember.** It used to keep the answer, and that
+   * made the table the one fact in this proxy stored in two places: here, and
+   * in the file's `KeyframeTable` on the main thread. Traced 2026-09-15, this
+   * copy had no reader: nothing ever asked a container twice for its keyframes.
+   * The parse itself has since moved to the thread the sessions are on, so
+   * there is no longer a second thread for a second copy to sit in either.
+   *
+   * The remembering lives where the fact is USED, which is also the only place
+   * that can do the rest of the job: bound how long a viewer waits for it
+   * while letting the read go on, join a second asker to a read already
+   * running, and keep the answer reachable by a session created before it
+   * landed. A container cannot do any of those — it does not know there is a
+   * viewer — so a copy here was never the whole answer, only a second one.
+   *
+   * @returns {Promise<{times:number[],tolerance:number}|null>}
+   */
+  async readKeyframeIndex() {
+    return this.parseKeyframeIndex();
+  }
+
+  /**
+   * How THIS format states where its keyframes are. Overridden by every
+   * container that has such a table; the default is the honest answer for one
+   * that does not.
+   *
+   * @returns {Promise<{times:number[],tolerance:number}|null>}
+   */
+  async parseKeyframeIndex() {
+    return null;
+  }
+
+  /**
+   * What this file DECLARES about itself as a whole, as distinct from what its
+   * individual tracks declare.
+   *
+   * The rule this method exists to hold: a fact the container declares is read
+   * from the container; a fact only the media itself has is measured from the
+   * media. Both halves used to be asked of ffmpeg, so the same header was read
+   * twice — measured 2026-09-03, this layer read one `.mka` header in 8 ms while
+   * a second ffmpeg read the same header over HTTP for 8121 ms, in the same
+   * second, on the same file.
+   *
+   * `null` is not "unknown, ask someone else". It means the container does not
+   * declare the field, which is a final answer about the container and the point
+   * at which a caller may go to the media — see `docs/container-architecture.md`.
+   *
+   * @returns {Promise<import("./Container.js").ContainerMediaInfo>}
+   */
+  async readMediaInfo() {
+    if (!this.mediaInfo) {
+      this.mediaInfo = {
+        format: this.formatName,
+        durationSeconds: null,
+        startTimeSeconds: null
+      };
+    }
+    return this.mediaInfo;
+  }
+
+  /**
+   * Subtitle-specific: where cues live (Matroska cluster positions or MP4 sample ranges).
+   * Returned via track objects' clusterPositions/samples, so base has no extra method — tracks carry it.
+   */
+
+  /**
+   * The TEXT FIELD of one subtitle cue, taken out of this container's framing.
+   *
+   * How a cue's bytes are wrapped is stated by the container's own
+   * specification, so each subclass answers for itself: Matroska reorders an ASS
+   * dialogue row, drops its two timing fields and prepends a read order
+   * (`matroska.org/technical/subtitles.html`); an MP4 prefixes a `tx3g` sample
+   * with its length (ISO/IEC 14496-12 §12.6); a subtitle FILE states its own
+   * field order in `[Events]`. None of that is a fact about the subtitle format,
+   * and the format's own markup — `{\pos(…)}`, `\N` — is not a fact about the
+   * container. The second half is `TextSubtitleTrack`; this is the
+   * first, and the two are applied in that order.
+   *
+   * Static because de-framing reads no instance state: a caller that has bytes
+   * and knows the format needs no container built over the whole file. The
+   * instance form below exists so a caller that DOES hold a container gets the
+   * right answer without naming the subclass.
+   *
+   * @param {Buffer} _payload - The cue's bytes as the container stores them.
+   * @param {string} _codecId - CodecID / sample entry type / file extension.
+   * @returns {string} The text field, markup still in place.
+   */
+  static cueTextOf(_payload, _codecId) {
+    throw new Error("cueTextOf not implemented");
+  }
+
+  /**
+   * @param {Buffer} payload
+   * @param {string} codecId
+   * @returns {string}
+   */
+  cueTextOf(payload, codecId) {
+    return /** @type {typeof Container} */ (this.constructor).cueTextOf(payload, codecId);
+  }
+}
+
+// ---------------------------------------------------------------------------
+// Lining ffmpeg's banner up with what a container declares. Here because the
+// correction is about what a CONTAINER states and the banner cannot.
+// ---------------------------------------------------------------------------
+/**
+ * Which subtitle track the FILE says to show, read from the file rather than
+ * from ffmpeg's description of it.
+ *
+ * Why this exists. The browser decides which subtitle track to turn on from
+ * `isDefault`, and until now that came from ffmpeg's `-i` banner, which prints
+ * `(default)`. In Matroska `FlagDefault` DEFAULTS TO 1 and ffmpeg has already
+ * applied that default by the time it prints — so a file whose muxer wrote the
+ * flag on no track arrives looking exactly like one that wrote it on every
+ * track: everything marked. The banner cannot tell the two apart, and the
+ * difference is the whole question, because one of them means "show this one"
+ * and the other means "the file has no opinion".
+ *
+ * The container itself can be asked, and the EBML reader already walks the
+ * Tracks element for subtitle extraction. What it now also records is whether
+ * the element was WRITTEN, which is the fact the banner destroys.
+ *
+ * The awkward part is lining the two readings up. ffmpeg numbers its subtitle
+ * streams `0:s:0`, `0:s:1`, … over EVERY subtitle stream, picture-based ones
+ * included, in the order the container declares them; the container reading is
+ * a list in that same order. So position is the correspondence — but a position
+ * match that is merely assumed is worth nothing, so it is CHECKED: each pair
+ * has to agree on language or on title. One pair that agrees on neither, or a
+ * length that differs, means the two readings are not describing the same
+ * thing in the same order, and then the container reading is not used at all.
+ */
+
+/**
+ * Language codes that carry no information, and so cannot confirm a pairing.
+ *
+ * ffmpeg prints `und` for a stream with no language; Matroska's own default
+ * for `Language` is `eng`, which is why an absent element cannot be read as a
+ * statement either — but `eng` is also a real answer, so it is not listed here
+ * and is compared like any other.
+ */
+const EMPTY_LANGUAGES = new Set(["", "und", "unknown"]);
+
+/**
+ * @param {unknown} value
+ * @returns {string}
+ */
+function normalise(value) {
+  return typeof value === "string" ? value.trim().toLowerCase() : "";
+}
+
+/**
+ * Whether one banner stream and one container track can be the same track.
+ *
+ * Agreement on either the language or the name is enough; both being empty is
+ * not agreement, because two tracks that say nothing about themselves say
+ * nothing about their pairing either.
+ *
+ * @param {{ language?: string, title?: string }} banner
+ * @param {{ language?: string, name?: string }} container
+ * @returns {boolean}
+ */
+function pairingHolds(banner, container) {
+  const bannerLanguage = normalise(banner?.language);
+  const containerLanguage = normalise(container?.language);
+  if (
+    !EMPTY_LANGUAGES.has(bannerLanguage) &&
+    !EMPTY_LANGUAGES.has(containerLanguage) &&
+    bannerLanguage === containerLanguage
+  ) {
+    return true;
+  }
+  const bannerTitle = normalise(banner?.title);
+  const containerName = normalise(container?.name);
+  if (bannerTitle.length > 0 && bannerTitle === containerName) {
+    return true;
+  }
+  // Nothing to compare on either side. Not a disagreement — a file may name
+  // neither — so it does not break the alignment; it simply adds no support.
+  return (
+    (EMPTY_LANGUAGES.has(bannerLanguage) || EMPTY_LANGUAGES.has(containerLanguage)) &&
+    (bannerTitle.length === 0 || containerName.length === 0)
+  );
+}
+