// Known Issues

What we know about

Real list of things we are tracking. Updated when we find a bug, fix one, or narrow a feature description to match what actually ships. Better to read this than to discover it mid-match.

We publish this list because pretending bugs do not exist is the worst version of a product page. If you hit something not on this list, please email support@altrosstudios.com and we will add it. Refunds are self-serve in the account portal within 14 days or 2 hours of play, no support ticket required.

First-launch disclosure modal originally said "your local coach"

fixed in next release
Scope: Test channel v1.7.0-rc1 · Updated 2026-05-19

The disclosure modal shipped in v1.7.0-rc1 contains the phrase "your own coach personalize to you ... That part is local to your machine." This understates how the coach actually runs: the LLM call is cloud for Pro and cloud-default for Base. Only the personalization layer (preferences in your SQLite DB) is local.

Workaround. Read the privacy policy at altrosstudios.games/gamedemon/privacy — the policy text is precise about which pieces are local and which are cloud. The next rc will rewrite the modal body to match.

Coach calls accumulate on the HUD during play

fixed in next release
Scope: Test channel v1.7.0-rc1 · Updated 2026-05-19

A pre-CQ3 tuning request bumped the HUD call hold from 12s to 5 minutes so calls would remain readable across rounds. In practice the stack of up to 8 calls all sit at full opacity for 5 minutes and cover the screen mid-round.

Workaround. In Settings → Live Coaching, switch the overlay to Compact mode (renders one call at a time). The next rc reverts the hold to 15s + 3s fade and caps the stack at 3.

Post-match auto-route off the Live screen can interrupt rating

fixed in next release
Scope: Test channel v1.7.0-rc1 · Updated 2026-05-19

When a match ends, Game Demon auto-routes from the Live screen (IV) to the Coach Chat screen (III) after 60 seconds. The v1.7 chip selector has more steps than the legacy thumbs UI, so 60 seconds is not enough to rate the last few calls.

Workaround. Stay actively on the Live screen and rate calls before navigating yourself. The next rc keeps you on Live if you are already there when the timer fires, and bumps the off-screen timer to 120s.

Data request form opens an email instead of self-serve

open
Scope: Website (both channels) · Updated 2026-05-19

The form at altrosstudios.games/privacy/data-request is live but submits via a pre-filled email to support@altrosstudios.com. The self-serve Lambda backing the form is built and waiting in the repo; it will be deployed alongside the next release.

Workaround. Use the form as-is. Send the pre-filled email. We fulfill within 7 days. EU and UK residents have the GDPR 30-day window; we hold ourselves to 7 regardless.

Preference aggregator runs on demand, not on a schedule

open
Scope: Test channel v1.7.0-rc1 · Updated 2026-05-19

The v1.7 changelog originally described the preference aggregator as "daily." It is actually on-demand: clicking "Refresh from recent feedback" in Settings → My Coach runs it. There is no background scheduler yet.

Workaround. Click Refresh in Settings → My Coach after giving a batch of feedback. The scheduled tick will land in the next release.

Map callout data is educated-guess pending an accuracy pass

investigating
Scope: Both channels (CQ3 v1.6+) · Updated 2026-05-19

All seven active-duty maps ship with structured callout data so the coach can ground tactical advice. The bounds and common-hold descriptions are reasonable defaults, not pro-VOD-validated. Some calls may name the wrong area or describe a hold that no longer matches current meta.

Workaround. Flag "Wrong call" with a freeform note when you see a callout you do not recognize for that area. Flagged callouts go to the top of the next accuracy-review batch.

Coach occasionally says you have no armor when you have kevlar + helmet

open
Scope: Both channels · Updated 2026-05-19

The coach occasionally reprimands the player for buying no armor in rounds where they actually purchased kevlar AND helmet. Root cause is on the GSI to GameContext translation path: kevlar and helmet are independent purchases, but some advisor code paths still read armor as a single boolean. The v1.1 canonical-state work covered the prompt-build layer; a few legacy reprimand templates still take the shortcut.

Workaround. Flag the miscall with "Wrong call" in the chip selector. Those flagged rows go straight to the next coach-quality batch. Fix is queued for v1.7.0-rc2.

Tilt detection softens tone but does not suppress cues

open
Scope: Both channels (v1.4.x+) · Updated 2026-05-19

When you lose 3+ rounds in a row the coach detects "tilted" state and softens the delivery (Demon less mocking, Veteran warmer, etc.). The coach does NOT reduce call frequency while tilted — same cadence, gentler tone.

Workaround. Mute Game Demon between rounds if you need silence. A future release will add an opt-in "low-fire-rate while tilted" mode that reduces cadence too.

Savant and Analyst personas overlap more than they should

investigating
Scope: Both channels · Updated 2026-05-19

During CQ2 development the Savant persona drifted toward Analyst phrasing on enough cue types that the two voices are less distinct than designed. Internal audit estimates roughly 80% of the Savant training data needs a rewrite.

Workaround. Use Demon, Veteran, or Analyst if voice distinctness matters. Savant rewrite is on the V2 roadmap.

Backend entitlement refresh can 401 every 15 minutes

open
Scope: Stable channel (v1.0.5) · Updated 2026-05-19

The periodic entitlement refresh in the background uses a Steam authentication ticket. Steam tickets go stale every ~15 minutes; when that happens the next refresh hits 401 and the local entitlement cache stops updating until you restart Game Demon.

Workaround. Coaching still works (the live-coach path reads cached entitlement). If you see "Pro features unavailable" mid-session, restart the app. The persistent refresh path lands in v1.7 stable.

Several v1.7-rc1 fixes are in the branch but not in the binary yet

fixed in next release
Scope: Test channel v1.7.0-rc1 · Updated 2026-05-19

After the v1.7.0-rc1 build was packaged and uploaded, three small fixes landed on the branch: disclosure-modal copy correction, post-match auto-route gate, HUD pile-up fix. These will all be in v1.7.0-rc2.

Workaround. Use the workarounds listed for each item above. v1.7.0-rc2 will be rebuilt and pushed to the test channel shortly.