DLSS 5 Autopilot 2.0.6

Updating from 2.0.x with update now downloads this build, checks its
SHA-256 against the SHA256SUMS.txt published with the release, and offers
a one-click restart. Your settings, library, hidden games and profiles carry
over.

Added

• Frame generation files you downloaded yourself (#370). On the
  optiscaler route of a DirectX 12 game, the frame generation files row
  takes a version.dll you downloaded (dlssg_for_sm86 with its
  dlssg_sm86.ini, or DLSS Enabler) through add your own.... The next
  install places it beside the game and records it, OptiScaler never takes
  the name those files need, and uninstall takes the
  files out and puts back anything that was there before. The tool never
  downloads these files. The dlssg_for_sm86 version.dll was checked
  against a real download; DLSS Enabler's was not. Neither has been run in
  a game here (no RTX 20/30 card), so it is marked experimental.
• Use the NVIDIA card on a laptop (#427). On a machine with an NVIDIA
  card and a second GPU, the install sets Windows' graphics preference to
  High performance for the game - and, for a 32-bit game on the feeder
  route, for its 64-bit helper - unless you already chose one. Uninstall
  puts it back, unless you changed it since. The row is use the NVIDIA
  card (Windows setting); it only shows where there are two GPUs. Not
  tried on a laptop here.
• add a game (#441) in the library's header: pick a folder the scan did
  not find. From 2.0.6 on the folder is remembered, so a full scan and
  later updates keep it; a game added by hand in an earlier version needs
  adding once more.
• What worked for games like this one. A shared result now says whether
  the game ships DLSS, FSR or XeSS of its own, and the compatibility list is
  counted per kind of game and per driver as well as per game. The route
  the tool recommends before an install changes only when both routes
  have 20 or more results from games of that kind and their ranges do
  not overlap; after a route fails, the next one it names and the
  autopilot's order use games of that kind from 5 results. Driver advice
  compares drivers within one route. Results carry the kind of game from
  2.0.6 on, so the table by kind of game starts empty.
• A compatibility page (#304), rebuilt from the shared results.

Fixed

• A failed first install is taken back out (#438). A download that
  broke after the first files went in left them in the game folder, and the
  game then ran on half a route (#438: DXVK's d3d9.dll on its own). A
  fresh install that fails on a download or a full drive now removes what
  it wrote and puts the game's own files back (a locked file still leaves
  them recorded for uninstall - close the game and install again). An install over one of ours keeps what is there and
  records the new files for uninstall. When the install was a switch
  from another route, that route was taken out first; the message after a
  failed download, a full drive or a locked file says so, and installing
  it again brings it back.
• Connection failures say what happened (#434, #438). Windows blocking
  the socket (a firewall or security suite), a connection cut before or
  during a download, a
  name lookup that failed and a server that did not answer each get their
  own answer, with the address, instead of "install again".
• The game's own crash was blamed on "an overlay" (#480). Windows
  sometimes records the faulting module as a full path, and the game's own
  exe given that way was called "neither the game nor ours". Modules are
  now compared by file name.
• "Nothing here recorded why" above the Windows event that recorded it
  (#460). When you say the game closed itself and Windows recorded a fault,
  the headline says so.
• "Did the game start?" (#460, #461). Your answer now also replaces
  verdicts that ask you to switch something on in the game, not only the
  ones that point at the overlay.
• Life is Strange and other 32-bit Unreal Engine 3 games (#479) were
  set up as DirectX 11 because their exe names d3d11.dll. They draw with
  DirectX 9 unless their own Engine.ini turns D3D11 on, and are now read
  that way. A report where ReShade finds a Direct3D 9 device on a D3D11/12
  install says so in the headline.
• The extras app picked instead of the game (#497, Halo). Bonus apps
  beside a game (digital extras, artbook, soundtrack) no longer win a tie
  with the game's own exe.
• The 64-bit helper "went away" (#482). When the feed stops waiting for
  its helper and the helper's own log shows it finished starting later, the
  report says the feeder's wait ran out, and no longer asks you to read a
  log it already carries. Frames fed by the 32-bit feed count as drawn.
• ReShade's own d3d9.dll in front of DXVK (#348). On a DirectX 9
  install the report names a ReShade d3d9.dll that Windows finds before
  DXVK's, and says whether this install wrote it.
• An older bridge add-on beside another route (#439).
  dlss5-dx11-bridge.addon64 was cleaned up only by the bridge route. Every
  other ReShade route now moves it aside when it installs, uninstall puts
  it back, and the report says to install again instead of 'move it out'.
• The previous build stayed as dlss5-autopilot.old.exe (#450). It is
  now removed on the second start of the new version, once that version has
  shown it starts.
• Sharing a result the logs cannot see (#437). When the report says
  neural rendering was not switched on, or that the game made no DLSS
  call, share the result asks what you saw instead of filing it as a
  failure.
• The tool's own frame generation files read as the game's (#458). A
  nvngx_dlssg.dll this tool had just installed promoted a DirectX 11 game
  to DirectX 12. Files an install of ours wrote are no longer evidence about
  the game.
• Reports keep the log lines that decide the verdict (#458), instead of
  only the last lines of each log.
• The multi-frame generation package: the preview and --check now
  refuse it on a card it does not support, as the install already did,
  and show its warnings.