# The Reserve — Answer Matrix & Audio Cross-Check **Status:** working audit, 2026-09-24. Zach's ask: "make sure we have a valid answer matrix and a list of all tags needed... if anything doesn't make sense, we need to verify that and then demand it gets fixed by Brett, or propose a solution." **Sources cross-checked:** - `Blackwood_Reserve_v6_TRIPLE_AUDITED_MASTER.pdf` (in this folder) — Brett's "Triple-Audited Master," version 6, 19 September 2026. Itself an audit document that already flags its own unresolved fields — see below. - The real audio files Zach copied off his laptop (`temp barrel game/`), now uploaded to R2 and indexed in `audio-manifest.csv` (machine-readable twin of this doc — same rows, CSV columns instead of prose, meant for a script/agent to consume directly rather than parse markdown). - `../GAME-INFO.md` and `../tech-build/hardware-decisions.md` — the already-locked game design (player flow, RFID item types, station behavior). **This doc does not re-decide anything GAME-INFO.md has already locked** — where the PDF conflicts with a locked design decision, that's flagged as a finding for Brett, not silently resolved either way. **Format note (why CSV + this markdown, not one or the other):** this pairing is the real, common pattern for prop/continuity tracking in themed entertainment and immersive/escape-style shows — one canonical table (here, `audio-manifest.csv`) that a script or agent can consume directly, plus a human-readable narrative view (this file) built from the same data, not a second independent source of truth. Same principle as this repo's own `GAME-INFO.md`/firmware split: one real answer, multiple readable forms. --- ## 1. Bottle tag list (the master answer key) 24 physical wine bottle NFC tags — 8 barrels × 3 bottles each. This is **taken directly from PDF section 1** ("Current bottle assignment master," Brett's current-as-of-9/19 data) — not reconstructed, not inferred. | B | Game | Bottle 1 (winery / # / symbol / year) | Bottle 2 | Bottle 3 | |---|------|------|------|------| | 1 | The Toast That Never Happened | Vale & Company / 026 / silver stag / 1874 | Montgomery Vineyard / 084 / tower / 1881 | Holloway Brothers / 147 / crossed keys / 1869 | | 2 | The Guest Who Was Never There | Bellamy Cellars / 061 / black swan / 1865 | Mourning Hill / 072 / crescent moon + stars / 1867 | Kingsley Court / 201 / crown / 1884 | | 3 | The Warning That Came Too Late | Northgate Cellars / 205 / compass rose / 1876 | Graybridge / 187 / stone arched bridge / 1882 | Southwick & Son / 119 / anchor / 1877 | | 4 | The Last Bake | Fenwick & Daughters / 034 / three wheat sheaves / 1868 | Alder & Finch / 057 / perched bird + Hearth Blend / 1873 | Cinderfall / 052 / flame / 1864 | | 5 | The Missing Painting | Blackthorn Estate / 013 / thorned rose / 1877 | Easton & Wray / 133 / balance scales / 1870 | Peregrine Vale / 176 / falcon + Winter Press / 1880 | | 6 | The Applause That Never Came | Oakhaven / 064 / two acorns + oak leaves / 1866 | Briar & Bell / 111 / bell + briar / 1872 | Saint Orison / 093 / sunburst / 1858 | | 7 | The Empty Chair | Wrenford House / 156 / raised-wing bird + Autumn Claret / 1879 | Larkspur Lane / 091 / bird + Cellar Red / 1878 | Redgrave Vineyard / 128 / rampant lion / 1883 | | 8 | The Conservatory | Foxmere / 068 / fox head / 1871 | Ashcombe Estate / 019 / oak tree / 1871 | Thistle Row / 142 / purple thistle / 1875 | ## 2. Drawing Room / Library tag list Per GAME-INFO.md's already-locked design: 24 Drawing Room object tags (3 per game), one per game token also valid at that game's Drawing Room station. Library items are physical books only — **no electronics/tags** (GAME-INFO.md, locked). | B | Drawing Room marker | 3 objects (storyline variants) | Correct-answer object (PDF's labeling — see Finding F6) | Library book | |---|------|------|------|------| | 1 | Writing desk beneath Blackwood family portrait (no separate emblem) | Engagement announcement; Clara's note; Edmund's opened message | Edmund's opened message | Manor Guest Register | | 2 | **UNVERIFIED** | **UNVERIFIED** | **UNVERIFIED** | **UNVERIFIED** | | 3 | Brass compass rose | Mud-stained riding glove; brass storm lantern; next-morning stable receipt | stable receipt | The Coachman's Guide to Roads and Posting Inns | | 4 | **UNVERIFIED** | **UNVERIFIED** | **UNVERIFIED** | Large brown book, gold wheat sheaf spine | | 5 | Three small gilt picture-frame markers | Mud-stained handkerchief; upright brass candlestick; framed shipping receipt | framed shipping receipt | Large red book, gold picture-frame spine | | 6 | Silver music-note markers (locked 8/8) | Broken violin string; loose sheet music; scorched velvet curtain tassel | scorched velvet curtain tassel | Blue book, silver violin spine | | 7 | Brass leaf | Silver spectacles; small medicine case; signed carriage receipt (10:40 PM) | carriage receipt | *Final Journeys*, tall black book, brass carriage wheel | | 8 | **UNVERIFIED** (older brass-leaf marker retired, reassigned to B7) | Pruning shears; small copper watering can; cork stopper | cork stopper | Ivory/bone gardening book, green fern spine | --- ## 3. Findings for Brett ### Already admitted by Brett's own audit (PDF, not our finding — but repeating it here since it's the highest-priority real gap) **F1. Barrel 2 is entirely unresolved** — marker, Drawing Room objects, and Library identifier are all `UNVERIFIED` in the PDF itself. Yet real audio **already exists** for B2's Drawing Room correct/incorrect stage (see `audio-manifest.csv` rows for B02 `drawing_room`) — meaning the spoken clue was recorded before the physical props/marker it refers to were locked. Needs Brett to close this before B2 can be built. **F2. Barrel 4's Drawing Room marker/objects are unresolved** — same pattern as F1: audio already exists ("Last Bake - Drawing Room - Correct/NOT Correct") referencing a physical marker that isn't locked yet. **F3. Barrel 8's current marker is unresolved** — the PDF says the older brass-leaf marker was reassigned to B7, leaving B8 without a confirmed current marker, even though B8's Drawing Room objects/correct-answer (cork stopper) and bottle assignment are otherwise locked. ### New findings from cross-checking the actual audio files against the PDF/folder names **F4. Two files have the wrong barrel number baked into their filename** (not just a typo — this is the code a human/script would key off of): - `Guest Who Wasnt/Guest who was never there- 1st bottle go to drawing room - correct - RES_B03_C01.mp3` — story and folder are clearly B2 ("Guest Who Was Never There"), but the code says `RES_B03`. Should be `RES_B02_C01`. - `Applause that never came/Appluase that never - Drawing Room - correct - RES_B02_C02.mp3` — story/folder are B6, code says `RES_B02` (also misspells "Applause"). Should be `RES_B06_C02`. **F5. One file is explicitly self-flagged bad by whoever produced it:** `The Conservatory/The Conservatory - Both bottles CORRECT - RES_B08_C02nope.mp3` — the literal `nope` in the filename. Needs a real re-record/replacement before B8 ships. **F6. Possible conflict with the already-locked "storyline variant" design.** GAME-INFO.md (2026-09-10 refinement, locked) says the 3 Drawing Room objects per game are **storyline variants** — "any of the 3 is a valid correct pairing with the token, not 1 correct + 2 deliberately wrong." The PDF (section 2) instead labels exactly ONE object per barrel as "CORRECT" (e.g. B1: "CORRECT: Edmund's opened message" out of 3 listed objects), which reads like the older pick-1-of-3 model. **This may not actually be a conflict** — the PDF's "CORRECT: X" could just be naming one representative variant for documentation, with the real station logic (per GAME-INFO.md) still accepting any of the 3. But it reads ambiguously as written, and the actual Drawing Room audio files only have a binary `correct`/`NOT correct` pair per barrel (not one clip per each of the 3 variants) — **worth an explicit one-line confirmation from Brett**: does "correct" audio play for ANY of that game's 3 valid objects, or only for the one specific object the PDF names? This determines whether 1 audio pair per barrel is enough or whether 3 variant-specific pairs are needed. **F7. Clip-code scheme doesn't uniquely identify a stage.** Within most folders, `C02` is reused for two entirely different stages — "Drawing Room correct/incorrect" AND "Both bottles correct/incorrect" both carry the code `C02` in the same folder (see e.g. Empty Chair: 4 different audio files, all labeled C02). The human-readable prefix is what actually disambiguates them today, not the C0X code. **Not a functional bug** as long as whoever loads these onto a DFPlayer SD card keys off the description, not the code — but worth standardizing before that happens, to avoid a real mix-up. See proposed fix in §5. **F8. Barrel 1 doesn't follow the rest of the set's clip-numbering.** Every other barrel uses C01 (bottle 1) → C02 (Drawing Room + both-bottles). Barrel 1 additionally has an unrelated `C01`-coded "barrel clue" file, and uses `C03` (not `C02`) for its "both bottles" stage, plus an explicit `UNUSED The toast - RES_B01_C03.mp3` leftover file. Functionally the 6 required stage-outcome clips ARE all present for B1, but the numbering is inconsistent with every other barrel — flag for cleanup, not a missing- content issue. **F10. Filename hygiene** (cosmetic, batch-fixable, listed for completeness): 6 files have doubled `.mp3.mp3` extensions (Empty Chair, Conservatory ×2, Applause, Missing Painting) — copy-paste artifacts, harmless to playback but worth cleaning before these go on an SD card. Full list in `audio-manifest.csv`'s `flags` column. ### Findings from independent content verification (Whisper transcription — real speech-to-text, not just filenames) Zach's ask: verify what the audio actually SAYS, not just what its filename claims. Ran Whisper locally against all 51 clips (free, open-source, no external service) — full transcripts are the `transcript` column in `audio-manifest.csv`. This surfaced bugs filename inspection alone couldn't have caught — several bigger than the naming issues above: **F11. Two files are self-announcing, literal placeholders — highest priority, unambiguous, no judgment call needed:** - `Guest Who Wasnt/Guest who wasnt - Drawing Room - NOT correct - RES_B02_C02.mp3` — the clip itself says: *"We do not have the finalized drawing room clue for this memory yet. This audio is a placeholder and must be replaced before installation."* - `The last bake/The Last Bake - 1st bottle go to drawing room - correct - RES_B04_C01.mp3` — the clip itself says: *"...the exact object clue for this memory has not yet been finalized. This is a production placeholder and must be replaced before installation."* These directly confirm F1/F2 (B2 and B4 Drawing Room content unresolved) — not our inference, the audio production itself flagged it. **These two cannot ship as-is; there is zero ambiguity here.** **F12. A "correct"-labeled file actually contains the "NOT correct" script, verbatim.** `Missing Painting/Missing Painting - Drawing Room - correct - RES_B05_C02.mp3` and `Missing Painting/Missing Painting - Drawing Room - NOT correct - RES_B05_C02.mp3` have **identical spoken content** — both say *"That is not correct. The painting had been transferred before it disappeared..."* A player who gets B5's Drawing Room stage RIGHT would currently hear a failure message. Real functional bug, not a labeling nit — needs the correct-outcome clip re-recorded. **F13. B1's "both bottles correct" clip sends the player to the wrong book.** It says: *"Go to the library and find the coachman's guide to roads and posting-ins."* That's **B3's** Library book (see §2/PDF section 2). B1's real Library book is the **Manor Guest Register**. Filename/code (`RES_B01_C03`) is correctly B1 — this is a content error, likely a copy- paste from B3's script when B1's was generated, not a filing mistake. **F14. Two files in the "Guest Who Wasnt" (B2) folder contain OTHER barrels' actual spoken content, not just a wrong code in the filename:** - `Guest who was never there- 1st bottle go to drawing room - correct - RES_B03_C01.mp3` — filename reads as B2's story, but the words are B3's Drawing Room briefing almost verbatim ("...only one of the three objects can prove when the messenger reached Blackwood Manor..." — matches PDF's B3 script and B3's glove/lantern/stable-receipt objects exactly). This is not B2 content at all. - `barrel clue missing.mp3` — despite living in the B2 folder, this is **word-for-word identical** to `The Toast/The toast - barrel clue RES_B01_C01.mp3` (B1's opening clue, silver stag's head bottle). Real B1 content, misplaced. Combined with F1 and the F11 placeholder, this means **Barrel 2 currently has no genuine Reserve-station audio of its own for its opening/Bottle-1 stage** — every file in that slot is borrowed/misplaced content from other barrels. Only B2's "Both bottles correct/incorrect" and "Drawing Room correct" clips (rows 15-17, 19 in the manifest) actually contain B2-specific content (black swan bottle, crescent-moon-and-stars bottle) — the rest of the folder is filler. B2 needs real content produced, not just a marker/prop decision (F1) — the audio gap is bigger than the PDF's own admission suggested. **F15. One line worth Brett's eyes, not asserted as a bug:** the "NOT correct" B1 Drawing Room clip references *"Klara's answer"* leaving the room with Edmund. The PDF's own supersession ledger says: *"Do not use the erroneous Clara-place-card reconstruction. Correct object = Edmund's opened message."* This may be an intentional distractor line (the "incorrect" clip can reasonably reference the WRONG theory to steer players away from it) — or it may be a leftover fragment of the retracted Clara reconstruction that should have been rewritten along with everything else B1's supersession note touched. Flagging for a human call, not marking as confirmed-wrong. --- ## 4. Complete list of tags needed (summary count) - **24 wine bottle tags** (8 barrels × 3 bottles) — see §1, all 24 identities defined. - **24 Drawing Room object tags** (8 barrels × 3 storyline variants) — 18 defined (barrels 1, 3, 5, 6, 7, 8), **6 undefined** (barrels 2 and 4, 3 objects each — blocked on F1/F2). - **8 game token tags** (GAME 1–8) — identities are the barrel numbers themselves; no separate physical/label data needed beyond "which barrel." - **Library:** physical books only, no tags (GAME-INFO.md, locked). **Barrel 2 is behind Barrel 4 in real readiness, not just tied with it.** Both have unresolved Drawing Room props (F1/F2), but per F14, B2 is also currently missing genuine Reserve-station audio for its own opening/Bottle-1 stage (every file in that slot is misplaced content from B1 or B3) — B4's audio, by contrast, is otherwise complete apart from its one self-announced placeholder (F11). B2 needs both a design decision (marker/objects/Library) AND new audio production; B4 only needs the design decision plus one re-record. ## 5. Proposed process: getting tags coded in a row Real, minimal process — matches how the rest of this repo's device/tag work already gets done (see `platform/tools/reserve_*` firmware, which all read tag UIDs into a lookup table, never hardcode logic per physical tag): 1. **One spreadsheet row per physical tag** (bottles + objects + tokens = 56 rows total once B2/B4 are resolved). Columns: `tag_type` (bottle/object/ token), `barrel`, `position` (bottle 1/2/3, or which of the 3 object variants), `label_text` (what's printed/written on the prop, from §1/§2 above), `uid` (blank until physically written), `status` (planned/written/verified-on-reader). 2. **Write + verify in the same pass, not two passes**: for each row, write the tag, then immediately read it back on the SAME station firmware that'll use it in the show (not a generic phone NFC app) — catches a read-fails-at-the-actual-reader problem immediately instead of during tech rehearsal. Fill in `uid` and flip `status` to verified only after that real read-back succeeds. 3. **This spreadsheet becomes the lookup-table source**, not a parallel document — the firmware's answer tables (per GAME-INFO.md: "answers in lookup/config tables so content can change without rewriting code") should be generated FROM this sheet, not hand-typed separately, so the two can't drift. 4. Keep `audio-manifest.csv`/this doc and the tag spreadsheet cross- referenced by barrel number — that's the join key throughout.