// Build-time gates for the telehealth video feature. Deliberately its OWN module, not a constant in
// `videoContract.ts`: the contract file is the frozen wire shape transcribed from the server design,
// and a local ship switch has no business living in it. EVERY entry point imports from here, so the
// feature is turned on in exactly one place instead of several that can disagree - as of Task 13 that
// is: `EncounterVideoPane` (start a visit), `StartCaseConferenceButton` (start a conference),
// `AvailableMeetingBanner` (JOIN one - the patient's only way in) and the late-joiner
// `listAvailable()` lookup in `VideoService`.
//
// THE LOOKUP IS GATED TOO, and it is the one that is easy to miss: it is not an affordance, it is
// an authenticated `GET /R4/video-session` that fires on EVERY app mount for EVERY signed-in user.
// Ungated it would put a request for an unshipped feature on the critical path of every session,
// against a route that exists only on staging.

/**
 * **ON since 2026-08-02**, flipped deliberately after the first real two-party browser call against
 * deployed staging. Everything below is the history of why it was closed and what it cost to open -
 * keep it, because the reasons are the argument for the next flag someone is tempted to skip.
 *
 * **What that call actually verified, in a browser:** two-way audio (the silent-call defect is closed),
 * a rendering self-view, camera-off NOT reading as "waiting", the surface handover on navigation, and
 * leave/rejoin. It also found FIVE defects a green unit suite could not have caught - which is the
 * whole case for having made the flag conditional on a browser rather than on the tests.
 *
 * **TWO CONDITIONS REMAIN UNSEEN, recorded on as untested rather than passed:**
 * 1. **TRUE two-way video** - the test machine has ONE webcam, so two browsers could never both hold
 * it and one side always had `cameraAvailable: false`. Needs a second camera on live staging.
 * 2. **The DOCKED encounter pane** - the primary clinical surface - has still never run in a browser.
 * It mounts only inside the encounter workspace, reached from the clinician queue.
 * **CORRECTION 2026-08-02:** this used to say the queue was "blocked by new Appointments not
 * reaching the HealthLake search index". That was wrong, and it was a bug in my test fixture
 * rather than in the product - there is no indexing problem. The org appointment search scopes on
 * `actor=PractitionerRole/<id>`, and the fixture omitted the PractitionerRole participant, so it
 * was never *matchable* rather than merely not-yet-indexed. A fixture that includes the role is
 * reachable immediately. Left in place because a wrong diagnosis quoted in a comment outlives the
 * conversation that produced it.
 *
 * **This is not a release.** The only accounts are Spencer, Toncy and Ed, and video must be finished
 * before anything ships - which is why ON is safe now and would not have been on a live product.
 *
 * **Still asserted, in the other direction.** `EncounterVideoPane.gate.test.tsx` pins this as `true`
 * via `importActual`; both `.gate` test files now MOCK the flag closed so "a closed gate really closes
 * the entry point" stays covered whatever this ships as. Changing it in either direction is deliberate.
 *
 * history
 * OFF because THE BROKER DOES NOT EXIST YET.
 *
 * (Lane A) owns `POST/GET /R4/video-session*`, and as of 2026-08-01 `aws lambda
 * list-functions` returns **zero** video functions on HIS-Dev - the handlers are its T6 and are not
 * deployed. Meanwhile this client is fully wired: Task 9 mounted a real "Start video" button in the
 * encounter workspace and Task 11 pointed the app at the LIVE transport. So without this flag a
 * plugin build ships a practitioner a button that 404s, and the previous state (the in-memory stub)
 * was worse - it fed FABRICATED `Meeting`/`Attendee` objects straight into the real Chime SDK.
 *
 * The control stays VISIBLE and disabled with honest copy rather than hidden. Spencer's product call
 * on: a visible "coming soon" beats a feature that silently isn't there, because a missing
 * button reads as "this product has no telehealth" while a disabled one reads as "not yet".
 *
 * **THE BROKER HALF OF THIS CONDITION IS NOW MET** - Task 12's integration checkpoint ran green
 * against deployed staging on 2026-08-01: all seven routes answered, a real Chime meeting was minted
 * and torn down, and every response was a **superset** of what this client expects (zero shape
 * incompatibilities). So the transport is no longer the thing holding the flag closed.
 *
 * **BOTH TASK 9 REVIEW DEFECTS ARE NOW FIXED** (2026-08-01), so neither is holding this closed any
 * longer:
 * 1. ~~"Waiting for the patient to join..." is WRONG whenever the patient has no video.~~ It is now
 * derived from zero remote **ATTENDEES** (`realtimeSubscribeToAttendeeIdPresence`, tracked in
 * `ChimeCallSession`), not zero remote tiles - so an audio-only patient, and one who turns their
 * camera off mid-visit, both read as PRESENT. `remoteTiles` remains, for layout only.
 * 2. ~~A docked call has NO surface once the practitioner navigates off its encounter.~~
 * `EncounterVideoPane` now registers itself as the docked surface while it is showing the call,
 * and `CallOverlay` renders the FULL floating surface when a live call has no such surface
 * mounted. The one-grid rule is unchanged: while the pane is up, the overlay is silent.
 * Task 13 replaced the minimal Leave-only affordance this used to describe, because the same
 * branch turned out to serve a PATIENT in an `encounter#` visit - who has no encounter workspace
 * at all, so a Leave-only stub was their entire view of the visit.
 *
 * **STILL DO NOT FLIP - the browser call is still owed.** Nothing in the video client has run in a
 * **browser** yet: no real camera, no real microphone, no real two-party call. Everything below the
 * transport is mock-verified, which is precisely how the silent-call defect survived a clean review
 * once already - and the presence fix above is mock-verified in exactly the same way.
 *
 * FLIP TO `true` WHEN: one real two-party call has been observed on a deployed build - hear the
 * other person, see your own tile, watch the camera light go out on Leave, and (the two fixes above)
 * confirm an audio-only participant does NOT read as "waiting" and that leaving the encounter mid-call
 * still leaves a way out.
 *
 * ** IS DONE - the prerequisite that used to sit here is CLEARED** (Spencer picked option (a)
 * on 2026-08-02: the client names a `teamId`). `findTeamForPatient` and its tiebreaker-free GSI query
 * are gone from the invite path; the practitioner picks which of the patient's teams to pull in, and the
 * server pins the named team to the encounter's subject before notifying anybody - refusing "not this
 * patient's team" byte-identically to "no such team" so the caller-supplied id is not an existence
 * oracle. The server half is COMMITTED, NOT DEPLOYED: `git push origin dev` is what deploys it, so
 * confirm the invite route is live before relying on this in a smoke.
 *
 * **THAT TWO-PARTY CALL IS NOW ACTUALLY REACHABLE.** Until Task 13 nothing called `joinAvailable()`,
 * so there was no way for a second person to get INTO a meeting from the UI - the condition on this
 * flag could not have been met even in principle. The banner is what closes that.
 *
 * ## HOW TO ACTUALLY RUN THAT CALL - this flag is a SHIP gate, not a TEST gate
 *
 * Read on its own, the condition above looks circular: the observation needs a call, and the flag
 * shuts every door a call could come through (claude-4 read it exactly that way on 2026-08-02, and the
 * omission was mine). It is not circular - the wording just never said how, so it says it now.
 *
 * **The smoke runs on a TEMPORARY, UNCOMMITTED flip.** Set this to `true`, have Spencer run the
 * human-gated build + `deploy:local`, run the call, then decide:
 * - it works commit the flip **together with** the two gate tests below, and put the browser
 * evidence in the commit message. That is the whole point of the gate tests: the flip cannot be
 * quiet.
 * - it does not restore the one file (from a COPY - never `git checkout --` in this repo,
 * ) and nothing was ever claimed.
 * A plugin build compiles source and runs no tests, so a red suite never blocks the build - but it
 * WILL red, deliberately: `EncounterVideoPane.gate.test.tsx` asserts the OFF default and
 * `AvailableMeetingBanner.gate.test.tsx` does the same for the JOIN half (3 tests total). That is the
 * tripwire working, not a break.
 * **Say so on the session board before flipping.** An uncommitted flag flip is invisible to every
 * other session, and this board has been bitten by state that lived in exactly one working tree.
 *
 * This is already the codebase's idiom, not a new workflow: `EncounterVideoPane.test.tsx` and
 * `AvailableMeetingBanner.test.tsx` mock this module ON to describe the behaviour, while their `.gate`
 * counterparts render against the real OFF value. Exercise on, ship off.
 */
export const TELEHEALTH_VIDEO_ENABLED = true;