Next P1 slice from UX_AUDIT_2026-10-09.md:
- Marketplace hub tabs now live in the URL (?tab=...) so refresh, share and
back/forward keep the tab; unknown values fall back to Browse.
- ListingOffersPanel shows an error state on a relay outage instead of
returning null (a failure no longer reads as no offers).
- Nav customizer Reset reseeds from DEFAULT_TAB_IDS (4) instead of a
hardcoded 3, matching MIDDLE_SLOTS.
New marketUi.offers.loadFailed (en/es).
P0 fixes from UX_AUDIT_2026-10-09.md:
- useMarketplaceItem now filters admin-hidden listings and skips
showcase-only (not-for-sale) items, so a direct /listing/... URL cannot
surface removed or non-sale content.
- Listing detail hides Make an offer for logged-out users and for
fixed-price listings (Buy is the single primary CTA there).
Borrowed from the Plebeian market review: a /marketplace-guidelines page
(peer-to-peer, buyer-beware, no checkout, prohibited items, soft
moderation), linked from the marketplace hub; new marketTerms en/es keys.
Records the remaining borrowables in the audit doc.
Full app audit (UX_AUDIT_2026-10-09.md), heavy on the marketplace flow
sequence. Headline verified findings: useMarketplaceItem does not filter
admin-hidden or exclude showcase-only listings (a hidden/not-for-sale
listing is reachable by URL); the buyer CTAs compete on fixed-price
listings and Make an offer is shown to logged-out users; the grid's Mark
as sold skips MarkSoldDialog so no trade attestation is recorded; the hub
tabs are not URL-addressable; nav.marketplace and nav.trades are
duplicate labels; and the nav customizer Reset hardcodes 3 tabs against
MIDDLE_SLOTS 4. Includes a recommended fix order. Audit only, no code
changes.
Capture the recommendation in AGENTS.md (and mirror it in TRUST_SCORE.md):
do not build the cvm-server per-observer GrapeRank scorer yet. The full
confidence model is provider-side and consumed via NIP-85; the client has
a local confidence gate; the blockers are the per-observer signing key and
8-hop compute plus lifecycle/deploy/monitoring, with ~zero users making
the ROI negative. A house-perspective scorer is the cheaper path; revisit
when there are users, no trustworthy provider, or skate-specific signals
to bake in.
A lone friend-of-a-friend no longer reads as trust. useViewerGraph now
counts distinct follows per friend-of-a-friend (fofCounts), and
deriveLocalScore returns verified only at FOF_MIN_RATERS (2) or more.
Provider assertions are unaffected; the full GrapeRank confidence model
stays provider-side. Docs updated.
Fifth slice of TRUST_SCORE.md. useNetworkAlerts reads the viewer's NIP-85
assertions for the accounts they follow and surfaces those trusted people
reported or muted. New /alerts page lists them with Ignore / Unfollow /
Mute / Report (reusing useFollowActions, useMuteUser, ReportDialog) —
inform, never auto-act. Settings -> Trust links to it with a live count;
ignored accounts are device-local. /alerts added to the smoke list.
Completes the safety loop. useMuteList reads the viewer's kind-10000 mute
list plus optimistic local mutes (lib/localMutes.ts); useMuteUser
republishes the list with the target added, records it locally and
invalidates content queries. useAdminModeration carries a new
mutedPubkeys set, and the shared filterAdminHidden/filterAdminHiddenMixed
drop muted authors — so every list hook filters them with no per-hook
changes (admin tooling with includeHidden still sees them). The flagged
notice gains a Mute action.
MarketplaceItemCard ran useTrustScore(item.pubkey) AND rendered
TrustBadge for the same author. It only needed the >=60 bit, which the
shared viewer graph (via useTrustBucket.localScore) already provides, so
the extra heavy per-card query is gone. Behaviour preserved.
TrustBadge called useTrustScore per card, which fetched the viewer's
follow list plus up to 100 friends' follow lists for EVERY card. The
graph does not depend on the target, so useViewerGraph computes it once
per viewer (one shared React Query) and deriveLocalScore returns the same
100/90/60/0 ladder (the NIP-05 30 tier classified as unknown anyway).
TrustBadge now costs no per-card network work on top of the batched
assertion read. Behaviour preserved; logged-out users still use the house
graph. Also records the ANALYZE=1 bundle findings in AGENTS.md.
Settings -> Trust is now functional, not read-only. useSetTrustProvider
publishes a kind-10040 designation for a custom provider (npub or 64-hex +
wss relay, the five 30382:<type> rows) and can clear it to fall back to the
community provider. The card validates input, prompts the signer, and
invalidates the provider + assertion queries so scores refresh. Docs
updated.
- Settings -> Trust section (?section=trust, TrustSettingsCard): a
show/hide-marks toggle (useTrustDisplay; Flagged is never hidden) and the
active provider + relay (useTrustProvider), linked to /trust.
- TrustBadge respects the display preference.
- Name AGPL-3.0 on the License page by updating its existing copy
(licenseLabel + body) in en/es.
- Settings + card i18n keys; /settings?section=trust added to the smoke
list. Docs updated.
Replace the license file with the unedited AGPL-3.0 text (copied in full
from the teambtcmap/btcmap-android LICENSE, byte-identical). No project
header and no 'or later' grant, so SPDX is AGPL-3.0-only. AGENTS.md records
the rationale and the section-13 source-offer implication.
Add the canonical AGPL-3.0 text at /LICENSE (with the project copyright
notice on top) and set package.json license to AGPL-3.0-or-later. Chosen
because the trust work reviews AGPL reference implementations and we
reimplement rather than copy; AGPL section 13 source-offer is covered by
the public repo + the License page's source link. Recorded in AGENTS.md.
Legibility pass: a trust verdict has to be explainable and sourced.
- TrustBadge tooltips now attribute the claim (from your network / the
community provider / the provider you chose) and link to /trust.
- New /trust page: the three states, why absence is not distrust, who
computes it, presence-is-not-endorsement, and point-of-view. Public,
linked from the badge; added to the smoke-test route list.
- New trustUi.* keys and four walletUi.trust attribution keys (en/es).
Second pass of TRUST_SCORE.md.
- useTrustAssertions now batches: every TrustBadge registers its subject into
a shared per-viewer union and ONE query fetches the union (no per-card
query), with a debounced refetch when a list grows. Same hook API.
- A direct follow (or self) is no longer shown as Verified - only a
friend-of-friend or a provider assertion counts; Flagged is never
suppressed.
- useUserReports reads public NIP-56 reports for an account, and
TrustFlaggedNotice shows a flagged counterparty with the reporters grouped
by the viewer's network plus Report / Dismiss. Wired into the listing
seller card and the coach detail page.
- New walletUi.trust.* keys (en/es).
First slice of TRUST_SCORE.md.
- lib/nip85.ts: parse kind-10040 provider designations, resolve a provider
with a house fallback, parse kind-30382 assertions (rank/followers/
reporters/muters/hops; absent hops means no path, absence is not
distrust), and author-check the provider.
- hooks/useTrustAssertions.ts: batch read from the resolved provider,
signature-verified via onlyVerified + the author check.
- lib/trustLadder.ts: three buckets and thresholds (flagged = trusted
reporters >= 2; verified rank >= 2).
- hooks/useTrustBucket.ts: prefers a NIP-85 assertion (verify + flag),
falls back to the local WoT ladder (verify only).
- TrustBadge.tsx: now renders Verified / Unknown / Flagged with glyph,
colour and tooltip; Unknown hidden by default, Flagged always shown.
Reaches the map/tour/event cards, marketplace card, listing detail and
coach detail through the shared component.
Open follow-ups (in TRUST_SCORE.md): batch the bucket at list level, a
'who reported' breakdown, a provider-choice setting, alerts, confidence
scoring.
Ideas-only review of a personalized web-of-trust approach: a spec-correct
NIP-85 consumer (kind-10040 provider designation, house fallback, author +
signature verification, the rank/followers/reporters/muters/hops tags, and
the absence rules), a Verified/Unknown/Flagged display with a viewer mode,
counterparty alerts, confidence-gated scoring, presets and negative
signals. Notes what we already have, concrete file touch points, license
and privacy caveats, and a recommended first slice. Links it from the
parked identity/conflict section.
Typing in the map search pill now opens a suggestions dropdown built
from the items that already match the active filters:
- Spots: title matches; selecting one flies to it and opens its sheet.
- Places: distinct city/region/country tokens split out of each item's
location; selecting one sets the query to that token (e.g. spain ->
Spain).
buildSearchSuggestions lives in lib (pure, testable) and
MapSearchSuggestions.tsx is presentational; MapPage owns the list and
keyboard nav (ArrowUp/Down, Enter, Escape). Adds mapPage.searchSuggestPlaces
(en/es only; other locales fall back to English).
A NIP-09 kind-5 delete is only honoured by relays and was never applied
locally, so on refresh the persisted React Query cache and the Dexie spot
cache re-rendered a just-deleted spot for a few seconds until fresh data
(or the admin hidden-label query) filtered it out.
Add src/lib/localDeletions.ts: a device-local tombstone keyed by coord
and event id, written optimistically by useDeleteEntity on the owner path
(rolled back if signing/publishing fails). The shared filterAdminHidden/
filterAdminHiddenMixed now also drop tombstoned items, so every list hook
filters them from first render with no per-hook changes. useSkateSpot
returns null for a tombstoned spot, and cacheSpots/cacheSpot refuse and
evict tombstoned rows. Prune by tombstone only (callers use limits 8-2000,
so pruning by 'not in fresh result' would wipe valid cached spots).
The language list now has a search input in both the header dropdown
(pinned above the scrolling list) and the inline list used in the mobile
nav sheet. Filtering matches the English name, native name, or locale
code, and shows a localized empty state. Adds nav.searchLanguages and
nav.noLanguages to en/es (other locales fall back to English).
The shared DropdownMenuContent used overflow-hidden with no max-height,
so any menu taller than the viewport (e.g. the language picker with 22
locales) was clipped and could not scroll to its last items. Apply
Radix's available-height CSS var as max-height and switch to
overflow-y-auto (canonical shadcn pattern).
