v3.36.12 — Copper as a first-class metal + JM Bullion restore + Silver:Copper ratio

[3.36.12] - 2026-08-17

Added — STRK-347: Poller dashboard shows both FBP price-source + buy-link URLs for FBP-backed vendors

• Dual URL editor for FBP-backed vendors (STRK-347): the home-poller provider
  editor (/providers) now surfaces, per coin, both the provider_coins.fbp_url
  (the FindBullionPrices page the price is scraped from) and the vendor's own
  buy-link URL, each editable and clearly labeled price-source vs buy-link, in both
  the By-Coin and By-Vendor views. FBP-backed vendors (currently jmbullion) carry
  an "FBP-backed" badge, driven by a FBP_VENDOR_IDS set that mirrors the runtime
  FBP vendor module. Editing the FBP URL writes through a dedicated
  updateCoinFbpUrl (via POST /providers/coin-fbp-url) that touches only
  fbp_url — never the coin's identity columns — and getProvidersByVendor now
  surfaces fbp_url so the By-Vendor view can render it. Operator-facing only; no
  change to the app or its data.

---

[3.36.11] - 2026-08-17

Removed — STRK-346: Remove unused FBP slug resolver + provider_coins.fbp_match column

• Dropped the dormant FindBullionPrices slug resolver (STRK-346): STRK-334
  shipped an on-demand year-preferring slug resolver (resolve-fbp-slugs.js) and
  a provider_coins.fbp_match keyword-hint column to auto-populate each coin's
  fbp_url, but go-live chose direct hand-assignment instead — the resolver's
  positive-token matching can't disambiguate FBP's multi-variant products (single
  coin vs tube vs monster-box vs fractional). The resolver never ran in the hot
  path, so it and its fbp_match threading are removed as dead code. No runtime
  behavior change: the JM Bullion vendor module still reads coin.fbp_url
  directly. The inert fbp_match column is left in existing databases (libSQL
  DROP COLUMN needs a table rebuild) but is no longer created or referenced.

---

[3.36.10] - 2026-08-16

Added — STRK-334: Restore JM Bullion market pricing via FindBullionPrices gap-fill

• FindBullionPrices.com JSON-LD gap-fill poller vendor (STRK-334): JM
  Bullion is blocked at the source by its Webscale/reCAPTCHA challenge, so its
  market prices had gone stale. A new poller vendor module scrapes
  FindBullionPrices.com's product pages — which expose clean JSON-LD ItemList
  data with no bot challenge — and uses those listings to gap-fill JM Bullion
  pricing back onto the market table. A year-preferring slug resolver picks the
  correct year-specific FBP product URL (older slugs 302-redirect), and a new
  provider_coins.fbp_match column records the resolved FBP mapping per coin.
• Market footer FindBullionPrices attribution (STRK-334): the market view
  footer now carries a FindBullionPrices.com attribution link crediting the
  gap-fill data source.
• Re-enabling JM Bullion as a live market vendor is a post-deploy step once the
  gap-fill feed is publishing.

[3.36.9] - 2026-08-16

Added — STRK-345: v2 day-file backfill tool for new metals

• devops/pollers/shared/backfill-v2-day-files.js (STRK-345): backfills a
  metal's data/v2/spot/{iso}/YYYY/MM/DD.json archive from year-file daily
  history. The v2 per-day archive begins at each metal's poller go-live —
  copper's first day file is 2026-08-15 while the legacy metals' floor is
  2026-03-25 — so the client's hourly-resolution pulls 404 across a new
  metal's pre-go-live window (the STRK-302 epic's pre-seed intent covered the
  daily layer but never the day-file archive). sqld cannot source the gap
  (spot_prices reaches back only to 2026-02), so the tool reads the
  published year files and emits one honest single-sample OHLCA entry per day
  (n: 1, noon-UTC t) in the standard v2 envelope. Skip-existing by
  default so it can never clobber a live hourly file; --dry-run and
  --overwrite supported. Rollout checklist for future metals added to
  .context/data-pipelines.md. 12 unit cases.

[3.36.8] - 2026-08-16

Fixed — STRK-344: Per-metal seed merge for existing profiles

• loadSeedSpotHistory merges seed rows per metal (STRK-344): the seeder
  previously hydrated the 180-day seed window only for empty profiles and
  skipped existing ones entirely (behind a global migration_seedHistoryMerge
  read), so a newly added metal (copper) could never obtain its 8–180-day
  chart history — every range ≤180d reads only spotHistory. A new pure
  helper (_seedEntriesToMerge) now runs the merge per metal: any metal
  lacking a history row older than 30 days receives its seed rows inside the
  180-day window, skipping days that already have live rows (live wins).
  Idempotent every boot; the flag write is retained purely as the
  seed-pass settle sentinel Playwright waits on. Companion to v3.36.7's
  per-metal backfill fix — same global-check-vs-per-metal-need root cause,
  one layer down. Server-side day-file parity is tracked separately
  (STRK-345).

[3.36.7] - 2026-08-16

Fixed — STRK-343: Copper sparkline flat on existing profiles

• Per-metal backfill freshness check (STRK-343): backfillStaktrakrHourly's
  fresh-load-vs-incremental decision was global across metals — any recent
  api-hourly row (always true on an existing profile via the legacy four)
  forced the 24 h window, so copper never received its 7-day deep backfill and
  its sparkline flat-lined on every ≤180d range. The window decision is now a
  pure helper (_staktrakrBackfillHoursBack) that requires EVERY metal in
  METALS to have both a recent hourly row (≤24 h) AND deep coverage (a row
  ≥6 days old, proof its 7-day pull already landed) before taking the 24 h
  fast path — recency alone would strand profiles that kept syncing after the
  new metal shipped. Timestamps are compared in the UTC frame (stored rows
  are UTC-naive), so the decision is offset-independent. Any uncovered metal
  extends the pull to 7 days and self-heals on the next sync; deriving the
  tracked set from METALS means the next metal addition inherits the
  behavior with no new code.

[3.36.6] - 2026-08-15

Added — STRK-341: The Silver:Copper ratio pair — Copper Phase 4

• Ag:Cu joins RATIO_PAIRS (STRK-341): one pair, not the cross-product, per the locked
  decision — Silver:Copper is the comparison a stacker holding both actually makes.
  Silver numerator over copper denominator (xag/xcu), copper accent, one decimal
  (the pair runs ~157–165, an order of magnitude above every other pair, so 1dp keeps it
  legible without false precision). The ratios panel, the /ratios/ page selector, and
  the panel modal pick the pair up automatically from the registry.
• The Ag:Cu chip rides the copper spot card (STRK-341): chips live on their pair's
  denominator card by convention, so the chip's visibility structurally follows copper's
  Settings › Metal Order opt-in — no copper figure ever surfaces on a copper-less
  dashboard, with no extra config plumbing. resolveChipContent gains a per-card
  numerator override (default gold; silver for this pair) and the chip tooltip explains
  the ratio in plain English.
• Monthly/daily boundary verified (STRK-341): pre-2013 copper history is monthly
  (World Bank Pink Sheet), so the Ag:Cu series is sparse there by construction —
  buildRatioSeries emits points only where both metals printed on a shared trading
  date and never fabricates values, now pinned by a unit test.
• Four existing pairs unchanged (STRK-341): pinned by regression assertions; the
  four-pair-era count assertions across the unit and Playwright suites were updated to
  the five-pair spec.

Changed — STRK-341: dashboard alignment refinements from live review

• Copper card per-ozt moved to a hover title (STRK-341): the visible $0.41 /ozt
  second line made the copper card's change/timestamp rows sit lower than the other
  four cards. The stored per-troy-ounce figure now lives in the price's hover tooltip
  (alongside the shift-click edit hint), restoring row alignment.
• Six-card totals rows are uniform two-line (STRK-341): the v3.36.5 per-row
  conditional wrap fixed truncation but let long-value cards grow taller than their
  neighbors, misaligning row separators across the six cards. Every row now stacks
  uniformly — label left on the first line, value right-aligned on the second — so all
  six cards stay height-aligned and $123,456.78 still never truncates. Five-card
  layouts (copper disabled) remain untouched.

---

[3.36.5] - 2026-08-15

Added — STRK-306: Copper dashboard cards, Metal Order entry, and the six-card layout

• Copper spot card and copper totals card (STRK-306): full parity with the other four
  metals — live price, sparkline, trend-period chip and select, sync icon, timestamp, and
  the eight-row totals card (items, weight, avg cost, purchased, melt, retail, gain/loss,
  realized, all fed by the existing METALS-driven loops). Both cards ship hidden: copper
  is opt-in via Settings › Appearance › Metal Order per the epic decision, and
  getMetalOrderConfig() appends the new row as disabled to stored configs, so existing
  users see no change until they enable it.
• Copper displays $/lb, stores $/ozt (STRK-306): the spot card shows the
  industry-standard dollars-per-pound figure ($6.02) with the stored per-troy-ounce value
  ($0.41) on a secondary line — a raw $0.41 next to gold at ~$4,376 reads as broken.
  formatSpotCardValue() is the single display-conversion seam (exact 7000/480 grains
  ratio); every stored, synced, charted, and computed value stays per troy ounce, and the
  shift-click inline editor accepts the price in the displayed $/lb unit and converts back.
• Six-card summary layout (STRK-306): with copper enabled, applyMetalOrder() stamps
  visible-card counts (data-cards) on both containers and new CSS tiers key on them —
  at ≥1350px totals rows keep the inline label-left/value-right layout and a value only
  wraps to its own right-aligned line when it genuinely doesn't fit (so $123,456.78
  never truncates and $0.00 rows never waste a second line; review round on the live
  preview replaced the prototype's uniform stacking). The five spot cards sit five
  across at ≥960px with the user's first-sorted metal taking the full-width top slot
  below that (no orphan card). With copper disabled the stamps read 4/5, no new rule
  matches, and the dashboard is pixel-identical to v3.36.4 — verified by byte-identical
  screenshot comparison against baseline.
• All Metals leads by default (STRK-306 review decision): the default Metal Order now
  puts the All Metals card first at every width for new users. Stored configs are never
  reordered — existing customized orders keep working, including on mobile.
• Derived metal lists (STRK-306): the trend-control wiring (TREND_METALS), the
  Realized-row visibility toggle, and the provider-info metal list now derive from METALS
  instead of hardcoding four names, so they can never silently miss a future metal.

---

[3.36.4] - 2026-08-15

Added — STRK-303: Copper in the user-selectable spot providers — Part 2

• Copper joins the canonical provider map (STRK-303): SPOT_PROVIDER_METAL_SYMBOLS
  gains copper: "XCU", wiring copper through the Metals-API and MetalPriceAPI latest
  and timeseries paths (XCU is per troy ounce on both — probe-verified 2026-08-15; the
  unconditional 1/rate inversion is unchanged, and correct even though copper's bare
  rate ~2.42 is the one wired rate ≥ 1), the CUSTOM provider's {METAL} substitution,
  and all three wired-metal selection gates (sync, history pull, cost preview).
• gold-api copper via COMEX HG with pound→troy conversion (STRK-303): the Gold API
  provider gains a /price/HG endpoint and a metal-aware parseResponse that divides
  copper's per-pound quote by exactly 7000/480 (≈14.5833) troy ounces per pound. The
  ~9% COMEX-futures-over-LME-spot premium is accepted by decision — a user who selects
  gold-api gets that provider's consistent view of copper.
• metals.dev copper is batch-latest only, unit-asserted (STRK-303): copper rides the
  /v1/latest batch call (which requests &unit=toz) and is accepted only when the
  response echoes unit: "toz" — industrial metals default to metric tonnes on this
  API, a 32,150× error. A non-toz echo rejects the whole payload; the per-metal
  /metal/spot endpoint (no unit parameter) deliberately gains no copper entry, and
  the /timeseries parser skips copper for the same reason. Copper history comes from
  Metals-API/MetalPriceAPI instead.
• Copper checkbox in "Metals to track" (STRK-303): every provider panel's metal
  selection grid now surfaces copper (config-truthy since v3.36.2), so users can opt
  out of the fifth metal's API calls. History-pull cost previews and the confirm
  dialog count copper automatically via the wired-metal gates.

---

[3.36.3] - 2026-08-15

Fixed — STRK-342: Spot provider registry hygiene

• Unknown metals no longer resolve to palladium (STRK-342): the METALS_API and
  METAL_PRICE_API symbol resolvers used a ternary chain whose else-branch was XPD, so any
  metal outside the four wired ones was silently priced as palladium (~$1,316/ozt for
  copper's true ~$0.41). Palladium is now matched explicitly and unmapped metals return
  null — a provider miss, never a wrong price. This unblocks STRK-303 Part 2.
• CUSTOM provider no longer substitutes undefined into user endpoints (STRK-342):
  with copper defaulted into config.metals since v3.36.2, the CUSTOM fetch loop and the
  Metals-API/MetalPriceAPI batch-history URL builders interpolated undefined into live
  request URLs. Unmapped metals are now skipped before URL construction.
• One canonical symbol map (STRK-342): the four-metal metal↔symbol map was redeclared
  eight times across constants.js, api.js, and spotLookup.js (and had already
  drifted — spotLookup gained copper in v3.36.2 while the rest did not). A single
  SPOT_PROVIDER_METAL_SYMBOLS map in constants.js now feeds every provider path, with
  unit tests scanning for inline-copy drift.
• "Metals to track" checkboxes actually work now (STRK-342): the per-provider metal
  selection in Settings → Spot Price rendered every box checked and persisted nothing —
  the listener targeted markup retired with the v2 settings shell, and was itself never
  invoked. Checkboxes now hydrate from the stored selection and persist changes
  per provider via a delegated handler.

---

[3.36.2] - 2026-08-15

Added — STRK-305: Copper as a first-class inventory metal — Phase 2

• Copper is selectable in the add/edit form (STRK-305): items save, reload, edit,
  bulk-edit, import, export, filter, search, and autocomplete like any other metal.
  Copper's spot resolution works end to end — the v2 feed's copper price is now ingested
  (xcu joined the frontend metal map), historical lookups reach the 59 years of seeded
  copper history, and melt/premium values compute from real copper spot. The dashboard
  spot card, summary card, and Metal Order entry are deliberately not added — that is
  Phase 3 (STRK-306) — and copper items total correctly with the cards absent.
• New avoirdupois-ounce weight unit (avdp) (STRK-305): copper bullion is sold in
  avoirdupois ounces, and entering a "1 oz copper round" as troy ounces would overstate
  its metal content — and melt value — by ~9.7%. The new unit converts on entry
  (1 avdp oz = 0.9114583 ozt) and displays back in avdp, while storage stays troy ounces
  like every other metric lens. Picking Copper auto-defaults the unit to avdp (only from
  plain "oz" — an explicit unit choice is never overridden). The unit code is avdp,
  not cu — cu is the constitutional-silver unit, and a collision would have
  silently corrupted junk-silver valuations.
• Copper theming (STRK-305): a --copper token in all four themes (tuned per
  background like its metal siblings), metal-accent card class, chip, and thumb
  placeholder colors.

Added — STRK-304: Copper history backfill — Phase 1b

• Copper now has a full price history back to 1968 (STRK-304): the offline seed
  bundle and year history files carry copper from January 1968 to today — 546 monthly
  points (1968–mid-2013) sourced from the World Bank Pink Sheet's LME grade-A cathode
  series, then 3,602 daily points from the price API's copper feed, which begins
  2013-07-23. Pre-2013 points are deliberately stored monthly-sparse rather than
  forward-filled — the source is a monthly average, and inventing daily points would
  fabricate precision it never had. Provenance is recorded per row (LME-WB vs
  StakTrakr) so the two eras stay distinguishable.
• The two eras are calibrated onto one quote basis (STRK-304): across the 157
  months where both sources overlap, the World Bank series runs a systematic ~9%
  above the API's copper quote (median ratio 1.0944, stdev 0.026) — a benchmark
  difference, not noise. Pre-2013 values are scaled by 0.9138 onto the API basis,
  which closes the seam at the 2013 splice to 0.5% and prevents a permanent false
  step in any long-range chart.
• The bundle's copper hold-back gate is open (STRK-304): v3.36.0 deliberately kept
  copper out of the shipped bundle until history existed. The bundle now includes all
  five metals, and sub-dollar prices keep four decimal places in the bundle as well,
  matching the poller's precision rules. Existing metals' bundle series are
  byte-identical to the previous release.

Fixed — STRK-304: Seed tooling path resolution and writer parity

• The seed updater resolves the repo data directory correctly again (STRK-304): the
  script computed the data path relative to its own location with a depth that predates
  the shared-poller migration, silently pointing at a nonexistent directory. It now
  resolves four levels up and refuses to run if the target does not exist.
• Both year-file writers now emit the same compact format (STRK-304): the seed
  updater wrote year files single-line while the bundle gap-fill pretty-printed them, so
  whichever tool ran last reformatted the whole file. Both now write the compact format
  the repo predominantly uses; the four recent year files collapse to single-line in
  this release as the one-time migration (content verified unchanged row for row).

---

[3.36.0] - 2026-08-14

Added — STRK-303: Copper (Cu) spot feed — Phase 1a

• Copper now flows through the spot pipeline end to end (STRK-303): the poller fetches
  copper (XCU) alongside gold, silver, platinum and palladium, writes it to the database,
  and publishes it through the v2 API. Nothing changes in the app yet — copper has no
  spot card, no summary card, and cannot be selected on an item. This release only starts
  collecting the data so a chart has history to draw when the interface work lands
  (STRK-305, STRK-306).
• The tracked metal set is now defined in one place (STRK-303): new
  devops/pollers/shared/spot-metals.js owns the metal list, and the API query string,
  database key set, JSON entry order, exporter ISO map and manifest array are all derived
  from it. The set had been duplicated across five files in two different casings with no
  link between the copies.

Fixed — STRK-303: Spot price derivation and sanity bounds

• Spot prices are no longer inferred from their own magnitude (STRK-303): the poller
  guessed whether the feed had returned a price or its reciprocal by testing whether the
  value was above 1. That only ever worked because every tracked metal was worth far more
  than $1 per ounce, which put the two candidates orders of magnitude apart. The response
  carries the direct price in a separate field all along — the code's own comment said so —
  so the guess has been replaced with reading that field.
• Sanity bounds are now per metal (STRK-303): a single $5–$50,000 range cannot cover
  metals four orders of magnitude apart. Because a rejected price throws and the poller
  treats that as a fatal error, an out-of-range value would have ended the entire run
  rather than dropping one metal.
• Sub-dollar metals keep more precision (STRK-303): prices below $1 are now stored to
  four decimal places instead of two. Two decimals is ample at $4,000 an ounce but
  quantises a 40-cent metal by roughly 0.6% on every reading, and that error is permanent
  once written to history.
• Historical backfill understands that copper starts partway through (STRK-303):
  archived files written before copper was tracked are complete with four metals, so the
  completeness check only requires copper from a configured cutover onward. The all-or-
  nothing gate is retained after that point — it is what makes a missed poll visible.

---

[3.35.101] - 2026-08-13

Fixed — STRK-338: Market coin names could silently stop caching

• Market manifest cache now survives compression (STRK-338): The cached list of coins
  the Market block tracks — the canonical names, weights, and metals behind every row —
  was written by one module through the app's compressing storage helpers but read back by
  another with those helpers bypassed. Values under 4,096 characters are stored verbatim,
  so the two halves agreed and nothing broke; past that size the reader could no longer
  parse what the writer had stored, silently discarded the cache, and fell back to a short
  built-in coin list until the next successful sync. The cached roster measured about
  57% of that limit and grows with every coin added, so this was on track to start
  happening on its own. Both sides now use the same helpers, and the duplicated restore
  logic that let them drift apart has been collapsed into one place (STRK-338).

---