Files
frameledger/docs/HANDOFF.md
T
poli0981andClaude Opus 5.5 24a87159f4 fix: the owner's eight sessions of 2026-10-20 - a x2.52 state, the driver's DLSS word lost at exit, Start directly losing the game Steam starts, two summary windows, display sizes, a bare DLSS line; the Steam note; the two load-flaky tests (beta.27 PR-1) (#330)
The owner ran the installed 0.1.0-beta.26 (sessions 64-71) and asked for the results to be checked,
the two flaky tests fixed, bugs fixed, and a release if there were bugs. Read from a backup copy of
the ledger and the logs (docs/spike-notes/23-owner-runs-2026-10-20.md).

- FgWindow.IsNearWholeMultiplier: a window is a state's only within 10 % of a whole multiplier of two
  or more (FgSlices.Slice.IsActive); FgMixRecord.Parse reads stored rows by the same rule. Silent Hill
  2 under Dynamic MFG listed x2.52 for 12.8 s - windows where the multiplier stepped; it now reads
  x2.0, x3.0, none and 7.36 % not classified.
- NgxDriverState / CaptureSession: the driver is not asked after the game has gone, and CREATED |
  EVALUATE stand once a reading carried them. The last NGX reading, taken after the target exited,
  replaced every in-game one: none of the 14 answered sessions kept the two bits, so DLSS
  (driver-reported) and the NVIDIA App's mode (D63) never showed on an NGX-direct title.
- CaptureOrchestrator: a launch says it ended on its own (Recorded) with the pids it recorded
  (Targets), so a process turned away while it ran - the game Steam starts after Lies of P hands
  itself over - is recorded when it ends; the launch's own process never is.
- SessionSummaryOpener: one window per session; opening it again brings it forward.
- DisplayTally: the sizes are the mode's last sample that saw the window (a game leaving exclusive
  fullscreen closes it first; five sessions stored none).
- DlssText.Line: no line that says only "DLSS".
- Tests: the guard's scan set includes the test host above the harness, which under a full gate
  releases modules while it starts its other classes - ModuleScanFailed (reproduced 1 run in 12 of
  the assembly with coverage; 15 of 15 green after), ProcessUnreadable on 2026-10-19.
  ShmDrainIntegrationTests.InjectAsync asks again on those two reasons only (4 times at most, 250 ms
  apart); the pause case waits for READY; every injection suite waits for the harness's flushed
  "presenting" line instead of 800 ms (tests/Shared/HarnessReady.cs).
- Docs: 03_METRICS (the mix, the driver-reported rung, display sizes), 04_CAPTURE, 06_DATA_MODEL,
  08_UI, 14_TESTING, 18_GPU_VENDOR_APIS, 19_SAFETY (direct starts counted by the crash safety -
  measured, the policy unchanged), LIMITATIONS, guide (troubleshooting, results), spike-notes section
  23, HANDOFF items 20 and 21, CHANGELOG.

Then, on the owner's word about Steam games (Play > Start directly says a Steam game may hand itself to Steam - the notice, the tooltip and Settings (en/vi/ja); why a note rather than a fixed Steam rule (beta.27 PR-1)):

The owner, 2026-10-20: Steam games hand themselves to Steam; a note in the UI, or a fixed setting
that Steam's games always go through the Steam client - the choice left open. The note:

- GamePlayer: PlayResult.SteamGame - the game the agent starts (Start directly, or an App running as
  administrator, or a Steam record with no link) is one Steam's own records name; the notice then
  reads Play_Started_Steam_Format / Play_Elevated_Steam_Format. Store records are read once a press.
- Strings: Play_ToolTip_Direct and Settings_PlayMode_Body say a Steam game may hand itself to Steam
  and the game Steam then starts is recorded; Settings adds that a few will not start outside Steam
  at all. en/vi/ja, ja marked review; Strings.Designer regenerated.
- Why not the fixed rule: three of the five Steam games started directly on 2026-10-20 ran and were
  measured from their first frame, and a Vulkan game is measured only when FrameLedger starts it
  (Everhex Demo) - recorded in 08_UI.
- Docs: 08_UI, guide/08-settings, LIMITATIONS, CHANGELOG.
- Tests: GamePlayerTests (Steam started directly and elevated say it; a loose or GOG game does not).

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-20 10:31:47 +07:00

33 KiB
Raw Blame History

HANDOFF — read this first, then stop reading it

The one file a new session opens to pick up P0 work (corrected 2026-10-04). It carries sequencing, decisions and traps. It deliberately carries no status.

The rule this file lives under

This file must never restate what another document already states. Four documents disagreeing about the gate's composition is what §S23-4 and §S23-5 were raised for, and rules-validate now fails a fifth copy in the shipped data. A handoff that summarises status becomes the next stale copy within a day — that has already happened three times here (#45's note, then #46–#52; 15_ROADMAP's status block; spike-notes §8 correcting a stale claim and going stale in the same file).

So: status lives where it lives. Read it there.

Question File
What is unresolved and what does it block? docs/20_OPEN_QUESTIONS.md — §S24 is the index; the ✅/◐/❓/🔴/🅓 markers are load-bearing
What was measured, on what machine? docs/spike-notes.md
Which P0 item is where? docs/15_ROADMAP.md
What landed? CHANGELOG.md
What does the software promise users? README.md, legal/DISCLAIMER.md — both carry accuracy blocks

And verify a status claim against the code before planning on it. Four times now those files have asserted in the present tense something that had been closed, or was never true. The 2026-08-05 session found the ledger stale across seven PRs and two documents contradicting themselves inside one section.


How this file is split (2026-10-19, beta.24, owner decision D66)

Everything this file held is still here or one link away, moved verbatim: what to do next and the open owner-only items stay below; the history of every train, the passed owner-only items and the traps moved to docs/handoff/. Code and documents cite the old sections by name — each name resolves here:

Cited as Now in
§Where P0 actually stands, queue items 1–6 (§2b, "item 3" …) handoff/history/p0-queue-1-6.md
queue item 7 (7a–7c), §Separable, §M5 handoff/history/p0-queue-7-m5.md
§P2 (D1–D10) handoff/history/p2.md
§P3 (D11–D17) handoff/history/p3.md
§P4 handoff/history/p4.md
§beta.3 handoff/history/0.1.0-beta.3.md
§beta.4 handoff/history/0.1.0-beta.4.md
§beta.5 handoff/history/0.1.0-beta.5.md
§beta.6 handoff/history/0.1.0-beta.6.md
§beta.7 handoff/history/0.1.0-beta.7.md
§beta.8 handoff/history/0.1.0-beta.8.md
§beta.9 handoff/history/0.1.0-beta.9.md
§beta.10 handoff/history/0.1.0-beta.10.md
§beta.11 handoff/history/0.1.0-beta.11.md
§beta.12 handoff/history/0.1.0-beta.12.md
§beta.13 handoff/history/0.1.0-beta.13.md
§beta.14 handoff/history/0.1.0-beta.14.md
§beta.15 handoff/history/0.1.0-beta.15.md
§beta.16 handoff/history/0.1.0-beta.16.md
§beta.17 handoff/history/0.1.0-beta.17.md
§beta.18 handoff/history/0.1.0-beta.18.md
§beta.19 handoff/history/0.1.0-beta.19.md
§beta.20 handoff/history/0.1.0-beta.20.md
§beta.21 handoff/history/0.1.0-beta.21.md
§beta.22 handoff/history/0.1.0-beta.22.md
§beta.23 handoff/history/0.1.0-beta.23.md
§beta.24 handoff/history/0.1.0-beta.24.md
§beta.25 handoff/history/0.1.0-beta.25.md
§Owner-only, an open item §Owner-only below
§Owner-only, a passed item (3, 12, 14–19) handoff/owner-only-passed.md
§Traps, §Two corrections carried forward handoff/traps.md
Dnn §Decisions index below

START HERE — 2026-10-19, 0.1.0-beta.26: a ×1336 frame-generation state, fixed

Status is docs/changelog/0.1.0-beta.26.md; this is what no other file carries. The owner ran the installed beta.25 the evening it shipped and sent screenshots: "Bạn có thể thử kiểm tra kết quả (fix nếu cần), cập nhật docs, làm chip và trigger lại sonarcloud thử nhé." Silent Hill 2's summary listed a ×1336.0 frame-generation state; the owner pointed at the cause ("có khoảng thời gian đầu game đang biên dịch shader") and then: "Bạn có thể cân nhắc (nếu thực sự cần) bản beta26 fix lỗi nhé." It was needed: every beta.25 session of a title that compiles its shaders under a loaded frame-generation runtime can store such a row, and the summary and the DLSS line print it. #327 (the fix and the owner's runs), #328 (the SonarCloud chip — the owner marked the last six issues the same day, and main's quality gate passes) and this release.

No new owner decision. The fix makes the window rule say what 03_METRICS already promised ("windows with too few frames to read") about APPLICATION frames, not presents, and bounds the ratio at ×8 (FgWindow.GenerationCeiling) — no shipping frame generation passes ×6; the bound is raised only with a title measured above it. A stored mix is read by the same rule, so no session is rewritten and none keeps the row.

Measured on the owner's ledger (a read-only backup copy, decoded with the App's own loader in a scratch test): the ×1336.0 row was Silent Hill 2's first fifteen seconds — 1,369, 4,877 and 1,303 presents with 1, 0 and 1 application frames per five seconds, then a 13,348 ms frame: the game compiling its shaders at start (the owner). The only stored value above ×8 in 60 sessions. Lies of P read no FSR mode with ffxQuery hooked (bit 11 in hooks_installed_mask): item 21's first branch (spike-notes §22).

Traps met on the way:

  • A row written by a released build is read by every later one. Changing what makes a state changes nothing on a stored session unless its reader applies the rule too (FgMixRecord.Parse); the row would keep the old meaning.
  • A session's timeline is one scratch test away: SessionSeriesLoader over a ledger copy prints presents, generated and application frames per five seconds (a PagesLoadTests partial, never committed). Decoding the blobs by hand is not needed.
  • SonarCloud's PR analysis lists only issues new in the PR: whether a change clears a finding on main shows only in main's next analysis.

Owner-only: items 20 and 21 annotated (spike-notes §22); nothing added.

Owner-only — no PR can close these

The items that passed — 3, 12, 14–19 — moved verbatim to handoff/owner-only-passed.md on 2026-10-19 (D66); the numbers are kept, and a number is never reused.

  1. §S23-2 — branch protection. Rules / validate is not a required status check on main, so the gate that makes the anti-cheat blocklist un-removable is advisory. Second-order problem: rules-publish.yml filters on four paths for both push and pull_request, so requiring the context as-is blocks every PR touching none of them, forever. Land the skip-shim (drop paths:, short-circuit inside the job) first; the setting is the owner's.

  2. Real-title verification runs for the upscaler/FG/RT work, against each game's own settings menu. Live as of 2026-09-03: the slGetNewFrameToken producer's table — Cyberpunk 2077 off / ×2 / ×4 and Hell Is Us DLSS + FG ×4, one launch each, reading presents/tokens off the FG counts: line against §S31's rows P1–P4. Run twice, 2026-09-03 (P4) and 2026-09-04 (P1); done. What is still owed here is the quality preset on a title that chains sl::DLSSOptions — Alan Wake 2 is the candidate — and the DLSS-G identity on a title where kFeatureDLSS_G is evaluated, if one exists.

  3. §H7's confirmation against the overlays actually resident on the dev machine. The compare-and-restore path is proven against a fixture (ctest fl_unhook_inline); §H7 still says the six real overlays are "worth doing once the Overlay has real hooks", and it has them since P1. One capture with each overlay active; the result is a spike-notes row.

  4. Re-ratify §S18 blocker 3 and §S27's basis once P2's PR-C/PR-F make the Agent the shipped injecting entry point (§P2, decision D4). The PRs restate both entries; whether the restatement stands is the owner's.

  5. D33's first real use (beta.9). Done by 2026-09-27, read from a copy of the owner's ledger: three normal hooked sessions of GIRLS' FRONTLINE 2 under the exception (NetEase Yidun) on 2026-09-26 and 09-27, the longest 3 h 9 min; the ja review below is still owed. (The item as written:) Turn the option on; GIRLS' FRONTLINE 2 should be listed eligible (NetEase Yidun, NEP2.dll, three sessions); make the exception, turn hooking on and play one session: hooked, marked Hooked · exception, not stopped by the 30 s re-scan or by the Overlay when NEP2.dll loads. Aniimo should read not eligible with its NEPKernel.sys; the option off makes the next session Tier 2 with the exception kept. And the ja text of Safety_Exception_*, which ships as English until a reviewer signs.

  6. beta.10 on real hardware. (a) The display mode: one DXGI game taken through exclusive full-screen → borderless → windowed in one session — the summary's shares, the Sessions column, the Trend's three share metrics — with a note of what the game's own menu called each mode. (b) The admin mode: turn it on → the disclosure → the UAC prompt → Settings says Running elevated; decline it → the administrator prompt was declined; sign out and in with the logon task → the prompt waits on the taskbar and one Agent results; start a game from FrameLedger while elevated → Task Manager's Elevated column says No; a game that runs as administrator → hooked; the CPU temperature with PawnIO (closes §M10 if it reads). (c) The Unreal Engine version on each Unreal game's page. (d) The Legal Gate shows Privacy 2.5 and Disclaimer 2.8 once. (e) The ja text of Safety_AdminMode_*, English until reviewed. (f) docs/LIST_GAME_TESTED.md — reviewed and approved by the owner before it was committed (D35, 2026-09-27); keep it current as games are run.

  7. beta.11 on real hardware. (a) GIRLS' FRONTLINE 2: the option on → Eligible → Make exception… (the disclosure ac-exception-dialog/2, with its Trial paragraph) → hooking on → two sessions: the card says trial: 0 of 2, then 1 of 2, then Exception in force since … alone. (b) The library at your window's width, and the list; Refresh after installing a game, and with the USB drive unplugged (Not installed, the drive named). (c) Delete all sessions on a game's page and in Settings ▸ Data. (d) Sessions at a glance and the Trend with a second unit. (e) The Legal Gate shows the Disclaimer 2.9 and the EULA 1.4 once. (f) The ja text of Safety_Exception_Trial (English until reviewed) and the App strings marked review.

  8. beta.12 on real hardware. (a) The game's memory: one hooked game and one that is not (hooking off, or a blocked game), each beside Task Manager's Details tab at the same moment — the summary's VRAM (this game) against Dedicated GPU memory, RAM (this game) against Memory (private working set); the Dashboard's live line; the Sessions columns, the Trend's memory metrics and the Sensors charts. (b) The Legal Gate shows the EULA 1.5, the Disclaimer 2.10 and the Privacy Policy 2.6 once, rendered, with the GPL as your licence, needing no acceptance. (c) Help ▸ User guide, Help ▸ About (its three tabs; Open the licences folder in the installed copy), and View on GitHub opening the v0.1.0-beta.12 tag. (d) The ja strings marked review (the memory labels, the About window, the first-start summary). (e) docs/LIST_GAME_TESTED.md is as of beta.10: refresh it from a copy of the ledger, with the games' memory figures, and approve it before it is committed (D35).

  9. beta.13 on real hardware. (a) The first delta: update an installed 0.1.0-beta.12 through Help ▸ Check for updates — the download is a fraction of the full ~107 MB, and the App comes back as beta.13; if it downloaded the full package instead, say so (Velopack falls back silently). (b) On this PC: with an Easy Anti-Cheat service running for any game, start a hook-enabled game — the notice says the anti-cheat is running on this PC, not in the game. (c) Two graphics cards, if a machine with them is at hand: the Sensors chart's GPU figures are the game's card. (d) The ja review of the new safety strings, with the others still marked for it.

  10. beta.14 on real hardware. (a) The update from beta.13: choose Beta in Settings ▸ Updates, Help ▸ Check for updates, then Restart to update with no game running; Settings then reads Automatic (Beta). The velopack: lines in ui-*.log say what was downloaded — a delta of about 2 MB; say so if it was the full package. (b) Explorer ▸ Properties ▸ Details of FrameLedger.exe and FrameLedger.Agent.exe name FrameLedger and the copyright; Task Manager lists FrameLedger Agent. (c) The wheel on every page, the dialogs and the summary window; over a chart the wheel scrolls the page and Ctrl+wheel zooms; the Logs page follows the newest line. (d) Hide scroll bars on and off, without a restart. (e) The import list in the dark theme, and Compare's tick boxes. (f) A crash: an Unreal game that crashes reads crashed (… CrashReportClient.exe); The Witcher 3 closed normally no longer reads crashed. (g) The new measurements on the RTX 5080: the VSync share against the game's own VSync setting, the render scale against the DLSS mode, the clocks, fan and limits. (h) The ja strings marked review. (i) The Privacy Policy 2.7 wording — before PR-5 and PR-7 merge.

  11. beta.17 on real hardware. (a) A theme switch: on Settings, Dark → Light → Dark, and again with a session's summary window open — every description and caption follows at once. (b) Hide scroll bars on: the Dashboard, Games, a game's page, Logs, About, Help ▸ User guide and a drop-down list draw no bar, and the wheel still scrolls; off, the bars come back without a restart. (c) Help ▸ User guide lists twelve pages and Limitations; a link to a section opens that section. (d) The wiki, once the release is out: the Wiki tab shows Home, the sidebar's two halves and a footer naming v0.1.0-beta.17 — then tick Restrict editing to collaborators only (Settings ▸ General ▸ Features ▸ Wikis), or anyone can change a page until the next release overwrites it. (e) The ja strings marked review.

  12. beta.24 on real hardware — decision tables written before the runs. (a) The mix (D62): a DLSS 4 title (Cyberpunk 2077 or Hell Is Us) with the NVIDIA App's Frame Generation Mode on Use 3D app setting, one session of about two minutes each at ×2, ×3 and ×4 chosen in the game's menu, plus its menus → the summary's Frame generation in this session lists ×2.0 / ×3.0 / ×4.0 with shares near the time spent, each with its own Native → Displayed, and the line Set in the game — no NVIDIA App frame-generation override. Then the MFG title with the App's Fixed ×6 → a ×6.0 row and Set in the NVIDIA App: Fixed ×6. Then Dynamic, Target FPS = Max refresh rate, Multiplier = Up to 6x → several rows or Not classified, and Set in the NVIDIA App: Dynamic, up to ×6, target: max refresh rate. Record the stored driver_profile values for the Dynamic run: NGX_DLSSG_DYNAMIC_MULTI_FRAME_COUNT_MAX 5 and the target 0x01000000 confirm the decode; anything else corrects NvidiaOverrides.FrameGenerationContext. (b) The quality mode (D63) on Expedition 33: the App's Super Resolution mode Quality, then DLAA, Ultra Performance, Performance, one short session each → the header reads DLSS (driver-reported) Quality · NVIDIA App and so on. Stored performanceMode: Ultra Performance 3 (already measured on Silent Hill 2) and DLAA 5 confirm NGX's numbering; DLAA 4 would mean the report follows the profile's numbering above 3 — then UpscalerNames.FromDriverReport changes, and the row goes into spike-notes. (c) The libraries (D64): the same sessions' DLSS line names the super-resolution library and from the graphics driver when the override's copy is loaded; preset L → DLSS 4.5 (preset L). With Model Presets on Use 3D app setting, the game's own library and no generation beyond what was observed. Record the names the override loads under %ProgramData%\NVIDIA\NGX (the summary's Loaded: line) — they correct 05_DETECTION §Shipped is not loaded. (d) DLSS 5 on the owner's DLSS 5 title: DLSS 5 runtime loaded. If it is not shown, record the module names it loads (Loaded: line); nvngx_dlssnr.dll is then corrected in CensusNames and 05_DETECTION. (e) SonarCloud (D67): bulk-mark per 13_CI_CD §SonarCloud; optionally its Ignore issues on multiple criteria rows; close PR #315. (f) The Legal Gate shows Disclaimer 2.11 and Privacy Policy 2.12 once. (g) The ja strings marked review. (2026-10-19, the owner's first runs — spike-notes §22: (a) the Dynamic context line reads the App's own words, and the mix found ×2.0 / ×3.0 / ×6.1 — plus a ×1336.0 row that was a defect, fixed in beta.26; (c) the override loads 160_E658700.bin 310.9.16.0 and 1B0_E658703.dll 2.14.16.0 from the NGX folder; (e) done — every issue marked; the last six, triaged in 13_CI_CD §SonarCloud (beta.26), the same day, and main's quality gate passes. The rest is not run yet.) (2026-10-20, the owner's runs of beta.26 — spike-notes §23: (a) done — ×2, ×3 and ×4 chosen in Cyberpunk 2077's menu read ×1.9 (ten seconds), ×3.0 and ×4.0, each with its own rates, under "Set in the game — no NVIDIA App frame-generation override"; Fixed ×6 on Expedition 33 read ×6.0 under "Set in the NVIDIA App: Fixed ×6"; Dynamic on Silent Hill 2 read ×2.0 and ×3.0, plus a ×2.52 row that was a defect, fixed in beta.27. The Witcher 3 under Fixed ×6 presented ×2.0 throughout: the rows are the frames, the line is the setting. (b) could not pass before beta.27: every stored driver word had lost CREATED | EVALUATE, read after the game had closed, so an NGX-direct title under the override never read its mode — run it on beta.27. (c) The Witcher 3 under the override loaded 160_E658700.bin 310.9.16.0 and 1B0_E658700.dll 2.14.16.0; Silent Hill 2 and Expedition 33 loaded 1B0_E658703.dll.)

  13. beta.25 on real hardware — FSR's quality mode (D65), decision table written before the run. Lies of P (FSR through AMD's FidelityFX libraries) launched by FrameLedger, not attached, with the game's own FSR at Quality, then a second session at Performance → the summary's header reads FSR 3 Quality, then FSR 3 Performance (or FSR Quality where the SDK 2.x libraries name no version), and the CSV export's # quality_mode: says quality (game). Then: no mode on both → the title sizes its frames without asking, or asked before the hook was in: record the session's render → output resolution and the Loaded: line in spike-notes, and D65 stays built and silent for that title. A wrong mode (the Quality session reading Balanced) → the answers this title receives are not the vendor's documented ratios: record the stored upscaler_quality and the render size in spike-notes, and the match is withdrawn until it is explained. With the game's dynamic resolution on (if it offers one) → no mode. (2026-10-19, spike-notes §22: the Quality session read no mode with the query hooked — the first branch; 1707×961 is the engine's own rounding of 66.67 %, not the size answer, so the title asked before the hook was in or never asked. D65 stays built and silent for Lies of P. The Performance session is not run.) (2026-10-20, §23: the Performance session, started through Steam, read "FSR 3 · 1280×720 → 2560×1440" and no mode — the first branch on both. 1280×720 is exactly the vendor's Performance size, so a query the hook met would have matched: the title did not ask while the hook was in. It cannot be launched by FrameLedger as this item asks — started directly it hands itself to Steam within seconds, and before beta.27 the game Steam started was not recorded at all.) Answered 2026-08-05, do not re-ask: remove gameguard and keep guard (approved over a red Rules gate, with the reasoning recorded in the merge commit); vendor NVAPI headers and amd64/nvapi64.lib; Windows 11 is the measurement scope only — SupportedOSPlatformVersion stays at Win 10 22H2 and the D3DKMT two-OS requirement is deferred with a written rationale.


Decisions index

Where each owner decision is recorded — the file and its section, and a label to find it by. The decision itself is the text there, not this table.

D Recorded in About
D1 handoff/history/p2.md FrameLedger.Application gains a reference to FrameLedger.Shared (A); Domain still
D2 handoff/history/p2.md never assigned — the numbering skips D2 and D3 (§P2 says so)
D3 handoff/history/p2.md never assigned — see D2
D4 handoff/history/p2.md the Agent ships --console consent grant
D5 handoff/history/p2.md FileGameConsentStore is retired in B; the CaptureHost writes a build-tree ledger.db
D6 handoff/history/p2.md --data-dir exists only under --console (F; 12_BUILD §Debugging), so --serve can never
D7 handoff/history/p2.md the FR-2.4 kill switch is the FOURTH input of HookedCaptureGate, not an upstream
D8 handoff/history/p2.md FR-11 is NOT a consent precondition in P2
D9 handoff/history/p2.md P2 never clears hook_blocked_reason
D10 handoff/history/p2.md the pipe (07_IPC) is P3, not P2; --serve is watcher-driven and logs
D11 handoff/history/p3.md the pipe's transport
D12 handoff/history/p3.md commands join as one more producer
D13 handoff/history/p3.md SessionProgress is a third ICaptureObserver
D14 handoff/history/p3.md Safety_ lives in FrameLedger.Shared, once
D15 handoff/history/p3.md the App opens ledger.db directly, under 06_DATA_MODEL §Writer ownership: it writes games'
D16 handoff/history/p3.md the settings registry (closes §G's row, written into 06_DATA_MODEL in PR-3): ui.language
D17 handoff/history/p3.md FL_MOCK, if the owner wants it, is --serve --mock under --data-dir only (D6's shape), so a
D18 handoff/history/0.1.0-beta.3.md the guard has one override, and its shape is fixed
D19 handoff/history/0.1.0-beta.3.md the DLSS preset derived from the render scale is STILL not shown
D20 handoff/history/0.1.0-beta.3.md the bug bundle does not get the machine's hardware
D21 handoff/history/0.1.0-beta.4.md a process an anti-cheat driver protects is TargetUnreadable, and nothing offers a way past it
D22 handoff/history/0.1.0-beta.5.md the per-game guard override (D18) is withdrawn; a finding about the game turns its hooking off
D23 handoff/history/0.1.0-beta.5.md Tier 2 is a held session: duration, telemetry and the reason
D24 handoff/history/0.1.0-beta.5.md a row follows its executable to another drive letter when exactly one other root holds it
D25 handoff/history/0.1.0-beta.6.md a file name is not an identity
D26 handoff/history/0.1.0-beta.6.md the Agent merges two entries for one executable under two drive letters
D27 handoff/history/0.1.0-beta.6.md other bytes at the entry's path on another letter: the entry follows them with hooking off
D28 handoff/history/0.1.0-beta.6.md HHW is measured through an entry pointed at swiftshader\Game.exe
D29 handoff/history/0.1.0-beta.7.md a per-entry Recording switch (games.record_sessions) and the import's known-tools list
D30 handoff/history/0.1.0-beta.8.md strict anti-cheat rules: any .sys in the install tree, and per-title lists
D31 handoff/history/0.1.0-beta.8.md the store half's matrix
D32 handoff/history/0.1.0-beta.8.md games blocked by anti-cheat: hide the hooking part, and scan the library ahead
D33 handoff/history/0.1.0-beta.9.md the user-mode anti-cheat exception (CLAUDE.md rule 2, amended)
D34 handoff/history/0.1.0-beta.10.md the admin mode asks Windows at EVERY Agent start
D35 handoff/history/0.1.0-beta.10.md the tested-games list comes from a COPY of the owner's ledger (scratchpad, Python sqlit…
D36 handoff/history/0.1.0-beta.10.md legal text changes where it would become false (Privacy 2.5, Disclaimer 2.8; EULA stays…
D37 handoff/history/0.1.0-beta.10.md each PR merges when its CI is green, then the release PR and the GPG-signed tag
D38 handoff/history/0.1.0-beta.11.md the user-mode exception's evidence moved under the grant, as a trial
D39 handoff/history/0.1.0-beta.11.md charts across a game's sessions: both — Sessions at a glance on the Sessions tab, and A…
D40 handoff/history/0.1.0-beta.11.md Delete all sessions: both places — the game's page and Settings ▸ Data
D41 handoff/history/0.1.0-beta.11.md Refresh does all three: reload the library, offer the games the stores installed since,…
D42 handoff/history/0.1.0-beta.11.md the library switches between a grid that fills the window and a list (ui.library_view)
D43 handoff/history/0.1.0-beta.12.md the game's memory is measured from OUTSIDE the game, in both tiers
D44 handoff/history/0.1.0-beta.12.md a user guide in English, guide/, rendered in the App (Help ▸ User guide), linked from t…
D45 handoff/history/0.1.0-beta.12.md the legal split
D46 handoff/history/0.1.0-beta.12.md documentation is corrected in place (struck, dated) with an index, docs/README.md; hist…
D47 handoff/history/0.1.0-beta.14.md the update channel follows the running copy
D48 handoff/history/0.1.0-beta.14.md the guard's narrow fix, both halves
D49 handoff/history/0.1.0-beta.14.md all four metric groups
D50 handoff/history/0.1.0-beta.14.md the managed binaries' identity
D51 handoff/history/0.1.0-beta.15.md the scan set cuts a parent/child link only when it is proven wrong: both creation times…
D52 handoff/history/0.1.0-beta.15.md the App's --data-dir is a viewer over a copy: no Agent is started or contacted (its com…
D53 handoff/history/0.1.0-beta.17.md the wiki's source is the repository
D54 handoff/history/0.1.0-beta.17.md the wiki's End-user half is the user guide, one source: it grows past D44's eight pages…
D55 handoff/history/0.1.0-beta.19.md a game you uninstalled leaves the library by itself when it has no session (amends D41'…
D56 handoff/history/0.1.0-beta.19.md an NW.js runtime is the import's game: an executable beside nw.dll + package.json ranks…
D57 handoff/history/0.1.0-beta.21.md the four itch "launchers" are malware: the documentation is corrected, the import is no…
D58 handoff/history/0.1.0-beta.21.md Wuthering Waves is on the title list (Steam 3513350, Anti-Cheat Expert), beside the fix…
D59 handoff/history/0.1.0-beta.21.md Play, in two modes switched in Settings: "1: press Play → the platform opens → it runs;…
D60 handoff/history/0.1.0-beta.23.md a game's artwork is what is already on the PC
D61 handoff/history/0.1.0-beta.23.md the band takes the artwork's colour as far as every text on it reads: primary and secon…
D62 handoff/history/0.1.0-beta.24.md the mix: every frame-generation state a session held, with its share, never averaged
D63 handoff/history/0.1.0-beta.24.md the NVIDIA App's override is the quality mode where the driver reports it applied
D64 handoff/history/0.1.0-beta.24.md DLSS labels are evidence-bound: versions and folder class, DLSS 4/4.5 only when shown, 5 only loaded
D65 handoff/history/0.1.0-beta.25.md FSR's quality mode through ffxQuery — deferred from beta.24, built in beta.25; none at a dynamic resolution
D66 handoff/history/0.1.0-beta.24.md the long documents are split; every original path stays as the index of its parts
D67 handoff/history/0.1.0-beta.24.md SonarCloud stays on Automatic Analysis with .sonarcloud.properties; the owner marks in its UI

Keeping this file honest

It goes stale the moment it starts describing state. If you find yourself writing "X is done" here, that belongs in CHANGELOG.md or §S24 — put it there and write a pointer. The only things that belong here are: what to do next and in what order, why that order, decisions that live in no other file, and traps.

Since 2026-10-19 (beta.24, D66) this file is split, and the split is kept up by the release: at each release the previous START HERE section moves, verbatim, to docs/handoff/history/<version>.md and gets its row in §How this file is split; an owner-only item the owner reports passed moves to handoff/owner-only-passed.md with its number; a new decision gets its row in §Decisions index. tools/docs-split-check.ps1 holds the history index and its files to each other.