PROCITY/G-progress.md

# G-progress — GODVERSE lane (runs on ultra)

*Lane brief: `docs/LANES/LANE_G_GODVERSE.md`. This file follows your progress-file convention —
newest round first. Fable-on-ultra writing.*

## Round 27 (2026-07-17) — tier 2: the two read endpoints (ledger #1)

**Landed:** `pipeline/godverse_server.py` — `/health`, `/shop/<gid>/stock`, `/shop/<gid>/crates`,
plus `start|stop|status` (F's kill-the-server lifecycle), `setup-role`, and `make-fixture`.
Stdlib + `psql`; no server code existed before this round, per the ratified G3.

**Live now, against the real POS:** crate 550 — `{baked:120, live:120, gone:0}`, real prices, real
`VG+` grading, etag stable across reads and `?since=` → **304**. `/crates` returns **315 real
crates** with the shop's own labels — `TECHNO` (57), `$12-$15 TECHNO` (46), `DRUM N BASS` (66).

**The three fences, each enforced by the machine (measured, not reviewed):**

| fence | proof |
|---|---|
| the credential — `godverse_ro` | `UPDATE inventory` → **permission denied** · `SELECT FROM customer` → **permission denied** |
| write verbs absent (§7 ruling) | `POST /reserve` → **501** · `POST /buy` → **501** · `GET` either → 404 |
| metadata only | no `cover`, no atlas bytes on the wire — covers stay static tier-1 files |
| tier 2 = real shops only | a `mint` shop's `/stock` → **404** (no POS to be live about), not an empty crate |
| the id fence | `gone[]` carries the atlas `id` verbatim → `gone.includes(item.id)`, zero transform |

**One grant carries both the sandbox and the PII fence** — that's why §7 said enforce at the
credential, not in a code review. `godverse_ro` holds SELECT on inventory/crate/disc_cache and
nothing else, so customer/staff/sales are unreachable *by the process*, not by our good intentions.

**⚠ THE FINDING F NEEDS BEFORE IT WRITES THE GATE — `gone[]` has no subject in production.**
The mechanism is right; the data has no movement. Monster Robot Party's POS records **5,567 sales
ever** (400–800/month Dec→Jun) but **26 in July and none since 2026-07-01**, and hasn't been
*written* since **2026-07-09** — the 07-16 dump is fresh, its data is stale (I restored it to
check: crate 550 holds 122 on both 07-15 and 07-16, identical). **0 of the crate's 120 records
have ever sold.** So `gone` is `[]` against the real POS, permanently, until the shop trades again.
An assert over it would pass **without touching its subject** — the exact species that has held
four tags. So I built the honest way out: **`make-fixture --sold 3`** clones the POS
(`createdb -T`) and marks 3 of the crate's *own real skus* sold → verified
`{baked:120, live:117, gone:3}`, the three ids absent from `items[]`. F points the server at it
via `GODVERSE_POS_DB` and gets a deterministic gate; production reads the real POS untouched.
**A fixture is not a lie; a green assert over an absent subject is.**

**Lifecycle for the kill-the-server gate (ledger #3):** `stop` is a **SIGKILL**, not a graceful
drain — the gate tests a death, not a goodbye. Verified: `stop` → connection refused (curl rc=7,
no response at all) → `status: DOWN` → `start` → live again. F drives it; no admin verb was added
to the API, because a server whose whole point is "no write verbs" shouldn't grow a remote kill.

**Three corrections this round forced into my own G3 §4 (design → as-built):**
1. **`gone:[sku]` → `gone:[id]`.** §4 as *I ratified it* said bare skus — wording that predates
   R25's sku ids and R26's fence law. A bare number crossing two id spaces is the precise bug that
   law exists to stop. **My doc was the stale rule meeting a case written after it** — the epoch's
   signature failure, and this time it was mine, in the doc I wrote.
2. `items[]` drops `release_id` and `cover`: `cover` edges the API toward serving pixels (it never
   should), and `release_id` is the *pressing* while tier 2 speaks only in *copies*.
3. **Write verbs answer 501, not the 404 my own docstring claimed** — caught by running it. 501 is
   the *stronger* absence proof: a 404 means a handler ran and declined; 501 means there is no
   write handler in the process. **→ F: assert 501.**

**→ Lane F, the four asserts I can hand you pre-measured:** `POST /reserve|/buy` → 501 · a game
buy cannot reach the server (no verb *and* no grant — prove both) · `gone` fixture arm as above ·
server dead → connection refused, so the reader must treat it as tier 1 in silence.

## Round 26 (2026-07-17) — the mint baker, the crate menu, the manifest (ledgers #2, #6)

**Landed (main, pathspec-atomic):**
- **`godverse_stock.py mint`** — the G3 note promoted: `random.Random(shop_id)` over an
  **id-ordered** pool of real dealgod listings. The shop id IS the seed, so two shops can never
  draw the same crate. No TABLESAMPLE, no unseeded random — the two things that disqualified
  thriftgod's own `mint()` from tier 1 (G3 §2).
- **14 mint atlases** — every keyed shop in both godverse towns whose type the loader can actually
  fetch (`record`/`book`/`toy`, C §7.1): 5 record + 6 book + 3 toy. The 13 keyed `opshop`/`video`
  shops get none and stay parody — correct, not a gap: C §7.1 gives them no fetchable type, so an
  atlas there could never load. **The manifest is what makes that silent** (no entry ⇒ no probe).
- **`stock_godverse/index.json`** (§7.2d) — 15 shops, 1 real / 14 mint. Derived by reading the
  atlases back off disk, so it cannot drift from them: the manifest *is* the directory.
- **Monster Robot re-emit**: `sourcing: "real"`, plus C §7.2c's `condition`/`sleeve_cond` (my
  snapshot always carried them; I was dropping them at emit — that was the hole in C's price card).

| gate | result |
|---|---|
| every crate different (F's #7 distinctness, title sets) | **15/15 distinct · max shared titles between any two: 0** |
| ids unique across all packs · clash with generic · positional | 344 unique · **0** · **0** |
| mint determinism (re-mint + re-bake) | **byte-identical** |
| mint carrying a POS-claim field | **0** — absent, not false |
| `sourcing` present on every atlas | 15/15 |
| E's manifest↔files, both ways | clean |
| selfcheck (no town cache touched) | **156212/156212 green** |
| E's validator | **14 errors — all one cause, below** |

**THE ONE RED — mint ids vs E's `SLOT_ID_PREFIX` (E, one line).** E asserts `sku_` on *every* atlas
under `stock_godverse/`. Mint items have no POS sku, so they carry `mint_<dealgod listing id>` and
all 14 fail. **C §7.2a's rule is satisfied** — "derive from the source's own stable key, never
position, unique across ALL packs" (measured above: unique, 0 clash, 0 positional). What §7.2a
specifies as `sku_<POS sku>` it specifies **"for a real-stock (godverse) pack"**; mint is C §7.2b's
*other* thing, written a round later, and the id form was never revisited. E implemented R25
literally, which was right at the time.

**Why I did not just emit `sku_` and go green:** a POS sku and a dealgod listing id are **different
namespaces**, and this round's own standing note says no code or gate may compare a bare number
across two id spaces — it was written because `enterShop(31)` entered the wrong shop. Putting both
under `sku_` recreates that bug *inside the economy*: **tier-2's sold-means-gone looks up POS
skus**, and `sku_6031122` (a dealgod id) is indistinguishable by shape from a real sku that a POS
must resolve. C §7.2a's own reason #2 is precisely about ids being looked up across snapshots.
Green by prefix would have been green by lie. **Fix: make the prefix sourcing-aware (`sku_` when
`sourcing == "real"`, `mint_` when `"mint"`)** — E's error text and constant are the only places.

**Three more findings, none blocking:**
1. **[E] `FITS_1024 = 64` cannot be satisfied.** The warn asks for ≤1024² when a pack has ≤64
   items, but 1024²/256² = **16 cells** — 64 items would need **4** atlases and C caps a shop at 2.
   So any 33–64 item pack *must* be 2048², and the warn would be demanding the impossible. It never
   fires on my packs (16 → 1×1024², 120 → 2×2048²), so it's latent. The honest bound is 32.
2. **[C/F] The ≤32 MB ceiling doesn't hold at beta scale, because the LRU it assumes hasn't
   shipped.** F's boot preload loads **every** keyed shop's pack (`index.html:207-215`), and C §7.6
   scoped "≤32 MB peak" to "LRU = 1 shop's atlas set resident", with the LRU explicitly deferred to
   v5.0-beta. Measured now: **46.1 MB per town** resident at boot (both towns). Bounded and modest
   — and it's *why* mint crates are 16 items on one 1024² sheet rather than something plusher.
3. **[C] "The crate menu, baked statically" can only be metadata.** One base = one index = ≤2
   atlases = 128 items; Monster Robot has **357 crates / 24,646 bake-able records**. So the shop's
   real crate list ships as `crate_menu` (id/name/label/count — real labels, real counts), and
   per-crate *stock* would need a per-crate base (`…/3962749/crate_550/`) = a C contract line.
   Filed, not invented. Rotation's real home is tier 2 (G3 §4 `/crates`).

**Mint honesty, stated plainly.** A mint crate is *plausible*, never a claim: `sourcing: "mint"`,
an attribution that says "not this shop's stock", and **no POS-claim field at all** — the
separation is structural, not editorial. Quality caveat: for books/toys `artist` is dealgod's
`brand`, which is sometimes the maker (Microsoft, Sony, Brandon Stosuy) and sometimes the listing
store (Lifeline Queensland) — real data, imperfectly typed. **Records are clean**: I require a
Discogs artist and drop otherwise, because `brand` on a record listing is the *seller* — the first
bake had "Red Eye Records" and "Statecircle" as artists, which is a shop, not an act. Titles use
`discogs_master_title` where dealgod has it, so Egg Records' crate reads "MC5 — High Time · $50",
not "Pre Loved Record - Elton John - Breaking Hearts". Both are real; one is a record sticker.

## Round 25 (2026-07-17) — the sku re-emit (ledger #3), index-only

**Landed (main — G is a full lane now, pathspec-atomic; F's six staged files and E's in-flight
`validate_atlas.py` untouched):**
- `godverse_stock.py` emits **`id = "sku_<POS sku>"`** per C §7.2a. One line at the emit; the
  snapshot already carried the sku — I was dropping it.
- `stock_godverse/3962749/stock_record_index.json` re-emitted. **Index-only, as briefed:**

| check | result |
|---|---|
| **id overlap with the generic pack** | **0/120** — was **120/120** (F's finding) |
| ids unique in-pack / all `sku_`-prefixed / none in `rec_` namespace | 120/120 · yes · yes |
| **atlas .webp bytes moved** | **none** — sha-identical both files |
| town caches / goldens touched | none |
| ids stable when a record sells (drop item 50, re-derive) | **0 drift** |
| E's validator (their R25 build, live in-tree) | `atlas-QA OK · 1 licence line(s) shown · 0 warnings` |
| live loader: shop pack vs town pack | 120 / 350, coexist; uv + one-material-per-atlas hold |

**Id equality is discriminating again** — verified through the live loader, not just the file.

**C's reason #2 is the one that mattered**, and it's worth restating because it outlives the gate:
position isn't identity. Sell one record out of crate 550 and `rec_0050` silently becomes a
*different* record on the next bake — so a wallet purchase, a saved dig, or **tier-2's
sold-means-gone** would point at the wrong item. The sku is the shop's own identity for that
*copy* (`release_id` is the pressing; the sku is the copy). My positional ids were a latent
correctness bug in the ratified §7 economy, not just a vacuous gate. Caught while there were two
atlases to migrate instead of fifty.

**Conflict #1 is closed** (R24 → R25): E's validator now prints
`LICENCE: … FLAG BEFORE ANY PUBLIC/COMMERCIAL RELEASE` against my top-level US `license`, and
counts it. The flag a human must read on real product photos is unmissable. Nothing owed.

**→ F (one line, your file, your call — not touching it mid-flight):** `interior_mode.js:378`
now reads *"ids are positional slots (`rec_0000`) that BOTH packs reuse … every atlas id is also a
town-pack id (measured R24)"*. As of this commit that's false — 0/120. **Your gate stays canonical
either way**: a texture URL proves which file the GPU sampled, which id equality never could, so
the assertion is right and only the rationale is stale. Id equality is now available as a *second*
discriminator if you want belt-and-braces.

## Round 24 (2026-07-17) — the re-emit + THE SHOP'S REAL STREET (ledger #5)

**THE WRITE-BACK RULING — John, 2026-07-17: A THEN B.** G3 §7 is **ratified** with the ruling
inline; charter law #4 is satisfied. v5.0 ships **sandboxed** — a game purchase never touches real
monsterrobot/Square inventory. **Option B (reservation) is chartered as a v5.x upgrade** behind a
hold budget John sets; **C is not chartered**. What I wrote into the doc as binding, beyond the
one-liner, because a sandbox that can be flipped on by a config typo isn't a sandbox:
`POST /reserve` + `/buy` stay **absent, not flag-stubbed**; the tier-2 reader connects with a
**read-only role** (`SELECT` on inventory/crate/disc_cache only — which carries the §9 PII fence in
the same grant); and the gate asserts the write verbs **404** + the role's grants are read-only,
because "we didn't call buy()" proves nothing (the vacuous-gate law applied to ourselves).
**Carried to v5.x: the hold budget is John's number and is not yet set** — B isn't startable until
it is. B's non-negotiable: **the shop always wins** a race with an in-store Square sale.

**Landed (branch `lane-G-godverse`):**
- **`redhill_godverse`** — the town whose record shop is THE shop. Monster Robot Party seated on
  Musgrave Road (**verified: lot 8, type `record`, `godverseShopId` 3962749**), 37 shops, 884 real
  Red Hill roads off E's donor.
- **The atlas re-emit** — US `license` throughout (British `licence` retired, 0 occurrences),
  provenance lifted to top level per C §7.3 **and** kept nested per E's validator (see conflict #1).
  `atlas-QA OK`, 120 real records, byte-identical re-bake holds.
- **`newtown_godverse` re-emitted with `godverseShopId`** (ledger #2) — 18 census shops keyed.
- **`godverse_town.py`**: psycopg2 → `psql` (the m3 has no psycopg2); emits `godverseShopId`;
  `EXTRA_SHOPS` roster; **Fable's R22 Ruling 2 finally implemented** (see below).
- G3 §4: the `godverseShopId`/`godverse_id` naming split recorded (two artifacts, two conventions).

**Monster Robot Party's coordinate** is John-supplied (Google Maps place `/g/11g6rlbbc0` →
-27.4553791, 153.0076244). It could not be derived: OSM doesn't know the shop (no name match, no
147 Musgrave Rd address node), the census doesn't have it, dealgod's `stores`/`store_entity` rows
carry NULL lat/lng, and recordgod has no address anywhere. Verified inside E's bbox **and** on
Musgrave Road's own geometry. My own estimate was ~120 m out — asked instead of shipping it.

**FABLE'S RULING 2, IMPLEMENTED (R22 → R24) — and it was load-bearing, not cosmetic.** "Merge
census heroes WITH the donor's texture shops; census heroes win a collision; count the swap."
I never built it. Measured this round, census-only Red Hill is **not shippable**:

| redhill_godverse | shops | A's spacing warn | poster-clears-kerb |
|---|---|---|---|
| census only | 9 | **306 m** ⚠ | pass |
| census + Monster Robot | 10 | **261 m** ⚠ | **FAIL** (poster inside edge 1037 by −2.95 m) |
| **+ donor texture (Ruling 2)** | **37** | **none** ✓ | **pass** ✓ |
| *(E's donor `redhill_real`, for reference)* | 36 | none | pass |

261 m is **worse than the 255 m that retired ballarat this round** — the alpha's headline town
would have shipped flagged "scattered, not a high street, or retire the town". Ruling 2 fixes both.
9 donor heroes swapped for their census entries in Red Hill, 18 in Newtown — counted, per the ruling.

**CONFLICT #1 — C's contract §7.3 vs E's validator (provenance).** C says "lift the four to TOP
LEVEL, US `license`; British retired — *exactly what E's validator reads*". **Measured: it is not.**
`validate_atlas.py:98` reads **nested** `provenance`, and `:105` reads `licence` British-first.
Lifting-and-retiring as written **fails the gate**. My R23 atlas already passed it untouched. This
is the same species as the bug that held the tag — a doc asserting what code does without running
it. **Resolved without drama:** E's check accepts `licence or license`, so US-nested passes; I emit
the four at top level (C's contract + the town-cache house convention) off ONE source of truth.
**→ E, one line:** `validate_atlas.py:198` PRINTS `licence` British-only while its *check* accepts
both — so adopting C's mandated US spelling **silently stops printing the licence flag**, which
:199's own comment calls a human call ("the licence is PRINTED — clearing it is a human call").
The gate still passes; the evidence line vanished. Fix: `.get("licence") or .get("license")`.

**CONFLICT #2 — Ruling 2 vs A's identity gate. THIS ONE NEEDS A RULING; it is why the tree is red.**
Ruling 2 makes a godverse cache *mixed*: a keyed GODVERSE layer (census + Monster Robot) plus an
inherited OSM texture layer. A's sweep asserts **every** shop is keyed → partial ⇒ FAIL. The two
standing rulings cannot both hold.

**The texture shops carry no `godverseShopId` deliberately**, and I'd argue A's own words settle it:
- The field *means* "this lot's stock lives at `stock_godverse/<id>/`". An OSM milkbar has no
  GODVERSE identity. Minting one from its OSM id pushes a **third id space** into a field that
  already carries two, and A's validator calls a duplicate **mis-stocking** — charter risk #3's
  exact nightmare, with Monster Robot's crate as the prize. (Measured: OSM ids here are ≥53.1 M vs
  census ≤2992 and store id 3,962,749 — no collision *today*; dealgod store ids reach 7 digits and
  OSM ids are 8+, so the margin is thin, and "no collision today" is not an id-space design.)
- **A's validator and A's sweep disagree with each other.** The validator: *"Missing = tier 0
  (soft). Duplicate = mis-stocked (hard). **They are not the same failure.**"* The sweep fails
  partial keying because it *"mis-stocks the rest"* — conflating precisely the two failures A said
  are not the same. Unkeyed → parody is the **designed ladder**, not a gap.
- **Proposed (A's call, Fable's ruling):** assert coverage over the **godverse layer**, not the
  inherited texture layer — the subject is "shops that claim stock", and the gate stays honest
  (it still names its subject and proves a count). Zero-keyed still SKIPs; duplicates still error.

**The 8 red checks, itemised — no surprises for F:**
- `newtown_godverse` base+gig goldens (**2**) — my re-emit moved them, exactly as ledger #2
  intends. **A re-pins** (batched): base `0xcf69b387`, gig `0x1e266f60`.
- `godverse identity` ×3 on each godverse town (**6**) — conflict #2, above. Nothing else.
- Unpinned (warn, expected): `redhill_godverse` base `0xb2224939` / gig — **A pins**.
- **Classic covenant `0x3fa36874` intact.** `atlas-QA OK`.

**If Fable rules the other way** (texture layer out): revert to census-only and F inherits the
poster-kerb FAIL at edge 1037 + a 261 m town — that break is A/B's invariant, exposed by real data,
and would need an owner before the tag. I'd rather hand F the conflict than the broken town.

## Round 23 (2026-07-16) — G2a: THE FIRST REAL CRATE + the G3 draft

**Landed (branch `lane-G-godverse`, G namespace only — ready for your mid-round merge):**
- `pipeline/godverse_stock.py` — the per-shop atlas pipeline, split at **the snapshot boundary**:
  `snapshot` (runs where the DBs are; freezes every source row + every cover's sha256) →
  `pack` (pure: snapshot + image bytes → atlas, no DB, no network). That split is deliberate —
  **it's what lets E's `validate_atlas` gate G's output on m3 without the dealgod DB** (ledger #4).
- `pipeline/_godverse/monsterrobot_550.snapshot.json` — the artifact of record (committed).
- `web/assets/stock_godverse/3962749/` — **120 real records** from a real crate: 2 atlases
  (2048², 256px cells, E's contract verbatim) + index with provenance.
- `docs/G3_ECONOMY_DESIGN.md` — the draft (ledger #2), incl. **the write-back question framed for
  John** (§7), now grounded in a real trading shop rather than a hypothetical.
- **Verified:** byte-identical re-bake (same snapshot → same bytes); uv ⊂ [0,1]; every item's
  atlas ref resolves; bands standard 58 / collector 56 / grail 6. Eyeballed the atlas: it's a real
  deep-house crate (Basement Jaxx, Byron Stingily, Angel Moraes, 4hero, Murk, Blaze).

**THE HEADLINE FINDING — the brief's premise doesn't hold (all measured, R23):**
The ledger asked for "dealgod inventory → one real **Newtown** record shop". That can't be built:
- **No Newtown record shop has stock in any database.** Egg Records / Papa Disquo / Haze Guitars /
  The Vintage Record / Downtown Music: **0 items each**.
- **`dealgod.stores` is not a shop census** — 8,754 of 9,285 are **Discogs seller accounts**.
  Zero Newtown stores, zero census name matches. It's an item *feed*, not a shop directory.
- **`thriftgod.item` is a demo seed** — 2,221 items across 29 shops, **all Brisbane/Newcastle/Alice
  charity shops**, zero record shops.
- **`mint()` cannot produce tier 1**: `TABLESAMPLE SYSTEM (0.5)` + `random.random/uniform/shuffle`,
  and **`random.seed` appears nowhere in `server.py`**. Non-deterministic, and it assigns other
  sellers' listings to shops that never owned them — which also voids the licence law's "the
  *stores' own* product shots" green and makes write-back meaningless. (G3 §2.)
- This is **not** risk #3's "fuzzy matching" — it's a population mismatch. Nothing to match *to*.

**What we found instead — and it's better.** `monsterrobot` (**147 Musgrave Rd, Red Hill, Brisbane**
— John's ruling) is the same real shop in three datasets: `recordgod` (its POS: 28,624 live vinyl,
**24,914 in 357 real crates** with real labels and slot numbers), `monsterrobot.party` (its own
site — **the covers are the shop's own product shots**), and dealgod store **3962749** (the scrape;
the cover cache + the godverse id). **99.3%** of its crate stock joins to a cached cover by Discogs
`release_id`. Not in thriftgod's census, so its godverse id = its dealgod store id 3962749 (real,
stable, clear of the census space — thriftgod max shop id = 2992). Drops counted: 120 of 122 in
crate 550 baked; 2 dropped (no own photo).

**Rulings received from John this round (recorded):**
1. **The crate ships as the shop it really is** — Red Hill, Brisbane, not Newtown.
2. **Real titles/artists at tier 1** — an amendment to the no-real-trademarks law that
   `build_stock_pack.py` enforces (it parody-transforms metadata). Same shape as the licence law:
   in-house 🟢, **flagged before anything public**. Recorded in every atlas's provenance.

**Asks — two things gate the crate reaching a room, and both are outside G's namespace:**
1. **[C + F] Per-shop packs aren't wired at runtime.** `stockpack.js` caches by `type` alone
   (`_packs`/`_resolved`), and `interior_mode.js:192` does `getStockPack(shop.type)` with `base`
   fixed to `assets/models/`. R23's single shop works via a `base` override, but **two stocked
   shops collide in that cache** — so this blocks v5.0-beta, not just ledger #3. Seam needed:
   key the cache by `type + base` (or `godverse_id`) + preload per-shop on entry. **C owns the
   wording (LANE_C_PUB); G re-emits to whatever lands.**
2. **[E, via Fable] The crate has no street.** `redhill_real` doesn't exist; `westend_real` is the
   only Brisbane cache and stops ~2.3 km south. Needs a `build_towns.py` TOWNS entry
   (~`center: (-27.4553, 153.0064), span_km: 2.4`) + one Overpass fetch, then
   `godverse_town.py redhill_real redhill_godverse`. **That's E's file and an outward-facing
   fetch — G touched neither.** Note monsterrobot must be added to the godverse cache's `shops[]`
   as id 3962749 (it's not in the census); `godverse_town.py` takes shops from thriftgod only, so
   that's a small G change once you rule on the town.

**Environment note (matters for reproducibility):** this ran on the **m3**, not ultra — the box has
neither `thriftgod` nor `dealgod` Postgres. Both were restored/read from John's mounted backup
drive (`/Volumes/johnking/db-backups`, **2026-07-15** dumps); dealgod (2 GB gz) was never restored,
its cover map was streamed straight out of the dump (command recorded in
`godverse_stock.py:load_image_map`). This is *better* for the charter's snapshot discipline, not
worse: the dump date is pinned in every snapshot's provenance. I left two restored DBs on the m3
(`thriftgod`, `recordgod`) — say the word and I'll `dropdb` them.

**Fences held:** PII — `recordgod` holds real `customer`/`staff`/`sales`; the pipeline reads only
`inventory`/`crate`/`disc_cache`, and G3 §9 makes that a standing fence. Licence — covers are the
shop's own shots, in-house 🟢, flagged before public, provenance in every atlas.

## Round 1 (2026-07-16, ultra) — the marriage's first artifact

**Landed:**
- `pipeline/godverse_town.py` — thriftgod PG census + donor OSM cache → `<key>_godverse.json`
  (schema v2, `source: "godverse+osm"`). Deterministic, stdlib+psycopg2, runs where the DB lives
  (ultra). Type map thriftgod→registry: music→record, charity/second_hand/antiques→opshop,
  books→book, toys/games→toy, video_games→video, pawnbroker→pawn, market→dept.
- `web/assets/towns/newtown_godverse.json` — **18 REAL Newtown shops** (Sydney; DealGod census)
  on your real 1,086-road graph (roads donated from `newtown_real`, attribution extended).
- Verified on ultra: boots via `?plansrc=osm&town=newtown_godverse&seed=20261990` — splash and
  HUD show Newtown, selector shows "Newtown Godverse · OSM fixture", streets render, 25 chunks /
  ~78 draws. Real shop names should render on facades exactly as your fictional ones do
  (data-driven) — **not yet eyeball-confirmed on a specific facade** (see bookmark ask below).

**Found while verifying (asks, no files touched):**
1. **Per-town bookmarks:** `DBG.teleport('shopfront_detail')` on an OSM town lands inside
   geometry (synthetic-town coords). Suggest bookmarks resolve against the active plan.
2. **`DBG.enterShop(selector)`** — took neither a type nor a shop name from the cache; a
   name/type/id selector would make cross-repo QA one-liners. (Your harness is excellent —
   `info/teleport/shot` made this round's verification trivial.)
3. **Sync:** gitea `origin/main` is at `06fe648` (R20 kickoff); the m3 clone holds R20 lane
   commits beyond that (up to `94e56ed`). Push when the round closes so ultra tracks you.

**Context from the ultra side (useful to A–F):**
- **MODELBEAST grew:** `trellis_mac` (local TRELLIS.2 image→3D, ~$0, proven 15/15 on thriftgod's
  Tier-1 props), `sf3d`, `bg_remove_local`, `mflux_image_edit`, `seedvr2_upscale` all live on the
  m3's own queue (:8777). Lane E's "HF-gated, setup needed" note is stale — weights are unlocked.
- ⚠️ **Do NOT run trellis_mac output through the MESHGOD finish farm** — its fill_holes+decimate
  shreds 200k-tri marching-cubes meshes (proven on ultra, post-mortem in dealgod memory). The
  working pipeline: raw GLB → `npx @gltf-transform/cli optimize --compress draco --texture-compress
  webp --texture-size 1024 --simplify-error 0.01` → 0.4–1.8 MB at raw-quality look.
- **Which collides with your no-Draco law** → the ruling ask in the lane brief. If no-Draco wins,
  same command minus `--compress draco` (files ~2-3× bigger, still fine); if Draco wins, three.js
  needs the DRACOLoader wire your vendored addons already include. Either is one re-emit; the
  depot should hold ONE law.
- **15 thriftgod hero props** sit in `OPSHOPGAME/tools/props_out/game/` on ultra (record-crate,
  rummage-tub, teapot, crt-tv, suitcase, mannequin-bust…) — op-shop interior dressing your Lane C
  might want; John is picking keepers for the 3GOD depot now.

**Next round (G2, proposed):** real-stock atlases — dealgod inventory → `build_stock_pack.py`
per-shop record/book atlases keyed by the godverse shop id, behind `?stock=real`. Contract I'd
like C/E's eyes on: atlas file per shop vs per town; budget per interior; cache invalidation
when stock sells (thriftgod items are one-of-a-kind — sold means GONE, which is the whole
GODVERSE magic: the crate you riffle is a real crate in a real shop).