One GET /api/stats call, aggregated in a single pass over the library.
Completion funnel, breakdowns by system, genre, decade, condition and
rating, a backlog that links into the filtered library, and a value card.
Form was picked before colour, and most of the page is not a chart: single
numbers are stat tiles, the breakdowns are bar lists with the value printed
per row, which is also the table view.
The colour work, in order:
* one hue for the breakdown bars — identity is on the axis labels, so
colour has nothing to encode, and a darker-where-bigger ramp would just
double-encode bar length
* an ordinal ramp for the funnel, since owned/played/finished are ordered
stages rather than peers
* validated with the dataviz validator against this app's real card
surfaces rather than a reference one, which caught that the documented
ordinal light-end measures 1.91:1 here and fails the 2:1 floor; the ramp
starts a step darker
* status colour used once, on the stale-valuation notice, with an icon and
text so it never carries meaning alone
Two bugs found by rendering it and looking, which the validator cannot see:
* the ratings card was showing condition data under a ratings heading — a
chart whose title did not describe its contents. Fixed by adding a real
rating distribution rather than relabelling the card.
* dark mode rendered light cards on a dark page. Copying the reference
pattern's `color-scheme` onto the container overrode how every
descendant resolved light-dark(), and the `:root`-prefixed media
override never matched at all, because Angular's emulated encapsulation
scopes selectors in component styles. Both replaced by light-dark()
values that inherit the app's own scheme.
The value card reports coverage, age and source beside the total, and flags
valuations older than 90 days, because a bare figure mixes fresh with stale
and silently omits everything unpriced.
148 backend tests.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Two changes aimed at the first real run, since the integration cannot be
exercised here without a subscription.
Refresh now reports which product each game matched — name, console and the
source's id — next to the prices, and dry-run surfaces it before anything is
written. This is the failure that would otherwise go unnoticed: a lookup for
the DS "Chrono Trigger" resolving to the SNES original returns entirely
plausible numbers for the wrong game, and nothing in a bare price would say
so.
The matched id is then stored on the game, and later refreshes look it up
directly instead of repeating the title search. Cheaper, and stable — a
search that drifts to a different edition next month cannot silently
re-price something that was already matched correctly.
IPriceProvider takes an optional sourceId so this stays provider-agnostic.
eBay ignores it, having no stable per-product identifier in Browse.
139 backend tests.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
An eBay production keyset needs account verification, which leaves pricing
blocked on someone else's review queue. These are two routes that are not.
CSV price guide, POST /api/prices/import: no account, works immediately.
Column names are matched by alias, so a PriceCharting bulk export
(product-name / console-name / loose-price) and a hand-kept
title,system,loose,cib,new sheet both parse, along with currency symbols,
thousands separators and blank cells. Rows match on title + system, so the
same game on two consoles is priced separately, and rows for games not in
the library are reported rather than silently added.
PriceCharting adapter: paid, but access is immediate with no review, and it
quotes the same three tiers this app stores, so no inference is needed.
Their API docs are not reachable without an account, so the parser follows
the widely-used convention — integer pennies under hyphenated keys — and is
tolerant enough that a naming difference degrades to "no price" instead of
throwing. One method to adjust if it differs.
Providers are now a registry rather than a single service. /api/prices/status
lists each one with what it is configured for, what its numbers actually
mean, and how to enable it; refresh takes an optional provider name and
falls back to the first configured one. With none configured it answers 503
pointing at the CSV route.
Checked against the real library: a five-row guide in PriceCharting's own
column names priced four games and reported the fifth as not owned, with
each effective value following that copy's condition. Demo figures were
cleared afterwards.
135 backend tests.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Researched the options first. PriceCharting is the standard for retro prices
but requires a paid subscription for both its API and its bulk download.
eBay's sold-price data sits behind the Marketplace Insights API, which is a
limited release closed to new developers. The free game-price APIs cover
current digital storefronts, not physical retro copies. So there is no free
route to sold prices, and this uses eBay Browse — active listings, which are
asking prices, labelled as such rather than presented as valuations.
Schema now holds three prices per game (loose, CIB, new), with marketValue
as whichever tier matches that copy's condition. Changing a condition
re-prices from the stored tiers with no further lookup, and the dashboard
can later show both actual value and what a collection would be worth
complete.
The judgement lives in classification and aggregation, both pure and both
tested without credentials:
* listings are sorted into tiers from their titles, and accessories,
reproductions and multi-game lots are discarded — a "box only" listing
at $45 counted as a copy would halve the loose estimate for a $130 cart
* the discard qualifier is required. The first version matched a bare
"box", which threw out "complete in box" and "with box and manual",
i.e. most of the CIB tier, while keeping exactly the listings the
filter existed to remove. A test asserting on tiers rather than counts
caught it.
* median with an interquartile trim, since one optimist asking 50x moves
a mean and not a median
* sample counts travel with the estimate, because a tier drawn from two
listings warrants less confidence than one drawn from thirty
Credentials are optional: with none set, /api/prices/status reports
configured=false and refresh answers 503 with instructions, while the rest
of the app is unaffected.
101 backend tests.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Rating, notes, condition, region, purchase price and date, plus a market
value carrying the timestamp and source that make it interpretable.
Condition is load-bearing rather than cosmetic: price feeds quote per
condition, so it selects which quoted price applies to a copy. Market value
records when it was captured and where it came from — a collection total is
only as good as its staleness — and an edit to an unrelated field leaves
that timestamp alone, so a stale price cannot start looking freshly checked.
Two storage decisions worth naming:
* Money is stored as integer minor units. SQLite has no decimal type and
EF Core maps decimal to TEXT, which compares lexically: "9.00" sorts
above "10.00" and SUM is unavailable. A value converter keeps decimals
in C# while ordering and totalling work. A test pins the ordering.
* Enums serialise as names. The default is ordinals, which meant the API
rejected the browser's {"condition":"Cib"} with a 400 while the C# tests
passed, because they round-tripped ints and never spoke the client's
dialect. The tests now share the API's serializer options.
Also fixes a data-loss bug in the Python tools. Both built their PUT body
from a hardcoded list of field names, so any column added to the model was
omitted and therefore nulled. Adding collector fields meant the next art or
enrichment run would have erased every rating, note, condition, price and
valuation in the library. Payloads are now built by excluding the handful of
server-owned fields, so new columns carry through by default.
The migration was rehearsed against a copy of the live database before being
applied: 105 rows, descriptions and developers intact.
67 backend tests, 8 frontend.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The only backup was the Docker volume. Export writes the caller's whole
library as JSON or CSV; import reads either back, into the same account or
a different one.
Rows are matched on title + system rather than id, so a file is portable
between accounts and instances, and the same game on three consoles stays
three entries. Merge adds and updates but deletes nothing. Replace wipes
first, and is gated behind an explicit confirm dialog in the UI. dryRun
reports what would happen and writes nothing.
CSV is hand-rolled rather than pulling a dependency, but handles the parts
that actually bite: quoted fields containing commas, escaped quotes,
embedded newlines and CRLF endings. That is not hypothetical here — 101 of
the 105 descriptions contain newlines, and two titles contain accents, so a
naive split-on-comma would corrupt most of the library. Exports carry a BOM
so Excel reads them as UTF-8.
Verified against the real library, not just fixtures: 105 games exported to
CSV, imported into a scratch account and re-exported compare identical
field for field.
15 new tests cover round-trip fidelity, merge vs replace, dry run,
per-user isolation on the destructive path, malformed input, and the
awkward-quoting case. 51 backend tests total.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The API had no tests. These run against the real application through
WebApplicationFactory — same pipeline, same Identity configuration, same JWT
validation — with only the SQLite file, upload folder and signing key
swapped, so a passing test says something about what ships.
The ownership tests pin the rule the old PHP API got wrong: a second user
sees an empty library, gets 404 (not 403, which would confirm the id exists)
when reading, updating or deleting someone else's game, and cannot reassign
ownership by putting ownerId or userId in the request body.
Also covered: the password policy, that login is indistinguishable between a
wrong password and an absent user, that the availability endpoint leaks no
row data, that a token signed with an untrusted key is refused, that the
sort parameter is allow-listed rather than interpolated, and that uploads
must decode as an image regardless of extension or content type.
36 tests, ~1s. Program is now declared public partial so the test host can
reach it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The 2018 stack (Angular 5.2 / CLI 1.7, PHP, MySQL) had not been touched since
July 2018. Rebuilt rather than upgraded in place: the frontend was 17 major
versions behind, and of ~16,700 lines of PHP only ~150 were application logic —
the rest was four near-identical vendored copies of php-crud-api plus
class.upload.php.
Backend — ASP.NET Core 10, EF Core, SQLite
* ASP.NET Core Identity (PBKDF2) + JWT bearer auth
* Clean REST API replacing php-crud-api's filter[]/transform query syntax
* Box art uploads re-encoded to WebP via SkiaSharp
* Imports the 105 games recovered from the 2018 dump on first run
Frontend — Angular 22, zoneless, signals, Material 22
* Standalone components, lazy routes, functional guards and interceptor
* Vitest replaces Karma/Jasmine; fonts and icons bundled, no CDN calls
* No provideAnimations: @angular/animations is deprecated in v22 and
Material no longer depends on it (pinned by a test)
Docker
* Multi-stage builds for both services, non-root at runtime
* nginx serves the SPA and reverse-proxies the API, so everything is
same-origin; one volume holds the database, uploads and DP keys
Security issues in the old code, not carried across:
* Two endpoints exposed unauthenticated CRUD over every table
* The client chose whose rows to read (filter[]=userId,eq,N); ownership now
comes from the JWT subject server-side
* Login was hardcoded to a single username
* crypt() with one global salt, silently truncating passwords to 8 chars
* JWT secret was the literal string "testing", tokens never expired
* Token travelled in the query string rather than a header
* Uploads were anonymous with the path built from the client filename
* Access-Control-Allow-Origin: *
The live MySQL password committed in 2018 remains in git history and must be
rotated independently of this change.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
ckochandClaude Opus 5