feat(agent): 'Vad din agent vet' page rendering the ledger context (P2) (#935)

* feat(agent): 'Vad din agent vet' page rendering the ledger context (P2)

The human-facing surface for the openwiki ledger-context: a read-only page
that renders the exact payload the AI agent reads (Accounted://ledger/context
+ the briefing digest) as a legible profile of how this company books.

- Route app/(dashboard)/agent-knowledge (server component) calls the shared
  buildLedgerContext(supabase, companyId) directly: one payload, two
  renderers, no new API or data path.
- Sections mirror the payload 1:1: coverage/freshness strip, counterparty
  patterns (monochrome confidence bars + seen/agree evidence), supplier
  patterns, explicit rules shown as authoritative instructions distinct from
  observed patterns, account usage, VAT profile, conventions.
- Nav entry in the Analys group (icon Brain), ungated so it doubles as an
  upsell; flip requiredCapability to paywall.
- Design per .claude/rules/design.md (PageHeader, Card, Table, Badge,
  AccountNumber BAS tooltips); sv + en strings (agentKnowledge namespace).
  VAT/BAS labels stay Swedish in both locales per i18n rules.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat(agent): deep entity-resolved analysis + radial graph for the knowledge page

Reworks the 'Vad din agent vet' page from tables into a radial-hub graph
driven by a new full-history deep analysis, per founder feedback.

- fix(rpc): median_booking_lag_days now measures real posting promptness via
  committed_at, not entry_date (which the bank flow sets to the transaction
  date, giving a ~0 tautology: 151/152 on prod). migration 20260708120000.
- feat(rpc): get_ledger_deep_context (migration 20260708130000): full-history,
  deterministic, read-side. Merges counterparties by normalize_counterparty_key
  (e.g. Claude = 14 bookings across 12 name variants, weekly, 9 710 kr, always
  5420), mines booked verifikat for SEK spend (coalesce amount_sek), detects
  recurrence cadence, dominant account + share, plus supplier entities. Storno
  excluded, corrections kept; 19xx/26xx excluded from the dominant contra.
- LedgerGraph: radial SVG (company center, accounts inner ring, payees outer
  ring), hover/focus reveals variants + spend + cadence + account. Keyboard
  focusable nodes with per-node accessible names + a screen-reader data table.
- Page fetches the deep context alongside the light context; coverage strip
  gains tracked-payee / recurring / tracked-spend stats. sv + en strings.
- 14 pg tests (light + deep) green; both RPCs applied to prod + version-matched.

Reviewed by an adversarial multi-lens pass (accounting/SQL, frontend/a11y,
prod-fact verification); all four verified findings fixed (SEK currency,
storno-lag guard, keyboard a11y, spacing tokens).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat(agent): gentle mount fade-in for the radial map (reduced-motion safe)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(agent): render the page for a rules-only company (empty-state edge case)

isEmpty ignored explicit_rules, so a company with configured mapping rules
but no posted transactions hit the 'hasn't learned anything' empty state and
lost its rules section. Rules are independent of bookings. (CodeRabbit)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat(agent): show Kompetens (skills) + Fakta (memory) on the knowledge page

The 'Vad din agent vet' page now shows the full picture of what the agent
knows: alongside the booking map, a compact read-only view of its Kompetens
(the Swedish accounting/tax knowledge atoms it ships with, grouped by
tier as chips with active/dormant state) and the Fakta it remembers (top
learned facts with kind + source), each linking to /settings/assistant for
full management. Server-rendered via a new buildAgentCompetence() that
mirrors GET /api/agent/skills + /api/agent/memory. Also renders in the
no-bookings case so a new company still sees its agent's competence.
sv + en strings.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* refactor(agent): restructure knowledge page - hero graph + tabbed detail

Declutters the page per feedback: the booking map is the always-visible hero,
and the supporting detail (Kompetens · Minne · Regler & profil) moves into
tabs so only one view shows at a time instead of a long card stack. Split
AgentCompetenceSections into standalone CompetenceCard + FactsCard for the
tabs; removed the top stat row on request. sv + en.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat(agent): Reconciliation Aurora rewrite of the ledger knowledge graph

Full rewrite of LedgerGraph: node area = sqrt(spend), colour = cadence,
shape = supplier/counterparty, confidence = depth-of-field; on-mount
descriptor-collapse animation with xN badge; cadence pulse veins;
deterministic seeded layout; framer-motion only (no new deps); keyboard
navigation, reduced-motion and sr-only support. Build-verified; 3-lens
adversarial review findings fixed.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(agent): sample-size-honest confidence in the ledger knowledge graph

dominant_account_share was raw cnt/total, so a counterparty with a single
booking rendered as '100% säkerhet': fake certainty by construction (the
data_quality_master Item-C / P3 finding). New migration replaces
get_ledger_deep_context with a Laplace-smoothed share (cnt+1)/(total+2)
(1/1 -> 0.67, 3/3 -> 0.80) and exposes the raw evidence as
dominant_account_count / dominant_account_total. The detail card now shows
'Bokförd hit i k av n fall' under the confidence bar; the existing focus
buckets, stroke widths and percent labels inherit the honest value
unchanged. pg-real test updated to guard the n=1 case.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

DECISIONS.md

@@ -46,9 +46,12 @@ One line per decision: `[YYYY-MM-DD] <decision>: <why>`. Appended by agents and
 [2026-07-07] Reconciled ledger-context prereqs INTO dev_docs/bank_transaction_ai_normalization.md (§14): plan is the strategic superset; ledger-context RPC gets interim normalizeCounterpartyName() now, re-keys to entity_id at Phase 2/Layer F. Closed 4 gaps: RPC in Layer F substrate list, supplier-side digest patterns, storno/correction exclusion (§13+§14), pending_operations audit+FK for agent-suggestion attribution.
 [2026-07-08] Ledger-context prereq trifecta folded into the P1 branch pre-merge (normalize_counterparty_key SQL mirror of normalizeCounterpartyName + supplier_patterns CTE + storno filter + evidence{seen,agree,share,last_booked} format) instead of follow-up PRs: shipping first then fixing would break the payload shape consumers had just learned. Storno filter deliberately asymmetric: account_usage excludes source_type='storno' only; counterparty CTE has NO source_type filter because correctEntry() relinks transactions.journal_entry_id to the correction (the join self-heals) and excluding 'correction' would drop exactly the human-corrected booking. Faithful-mirror discipline: bare "KORT " prefix is NOT stripped (TS doesn't either); hardening the prefix list must change the TS+SQL pair together (pg test pins this). Payload caps trimmed 20/20 -> 15/15 + supplier 10 to hold the 12 KB budget with evidence objects.
 [2026-07-08] Ledger-context dominant-contra VAT bug, found by the switch-on check (calling gnubok_get_agent_briefing on real prod data, not synthetic tests): counterparty patterns for foreign SaaS (Google/ngrok/Supabase) showed dominant_account 2614 (reverse-charge output VAT) instead of 5420 (software expense). Cause: the dominant_account CTE excluded only 19xx, so on a reverse-charge booking (expense + 2645 + 2614 + 1930) the three non-bank accounts tie and the account_number ASC tiebreak picks the low VAT number 2614. Fix (migration 20260708110000): also exclude 26xx (always moms in BAS, never characterizes a counterparty); 23xx/24xx/25xx/27xx stay eligible so loan/tax counterparties (e.g. ALMI) still surface their real account. supplier_patterns unaffected (aggregates supplier_invoice_items.account_number = expense only). Regression pg test asserts 5420 over 2614; verified it fails on the old function.
+[2026-07-08] Built the P2 "Vad din agent vet" page (/agent-knowledge, Analys nav group, icon Brain): the human render of the same ledger-context payload the AI agent reads. RSC calling buildLedgerContext(supabase, companyId) directly (one payload, two renderers, no new API/data path), sections mirror the payload 1:1 (coverage strip, counterparty patterns with monochrome confidence bars + seen/agree evidence, supplier patterns, explicit rules distinguished as instructions, account usage, VAT profile, conventions). Chose a STANDALONE page over a 3rd tab in /settings/assistant: faithful to the plan's page spec + its sales-demo intent (Fortnox shows your data, this shows your patterns). UNGATED (shows the user their own derived booking stats, no live AI compute, doubles as an upsell) rather than CAPABILITY.ai-gated: set requiredCapability on the nav item to paywall. VAT treatment labels + BAS account names stay Swedish in both locales (regulatory domain terms, per .claude/rules/i18n.md). Verified: build compiles the route (client/server boundary resolved), lint+tsc clean on new files, i18n sv/en parity (54 keys). Data correctness already proven live on prod via the briefing + RPC.
+[2026-07-08] Deep entity-resolved ledger analysis + "Vad din agent vet" graph. (1) Fixed median_booking_lag_days: was median(entry_date - transaction date) = ~0 tautology (bank flow dates the voucher to the transaction; verified 151/152 entries entry_date==txn date on prod); switched to committed_at (real posting lag, median 90 for Arcim). migration 20260708120000, applied to prod. (2) New get_ledger_deep_context RPC (migration 20260708130000): full-history, deterministic, read-side deep analysis merging counterparties by normalize_counterparty_key (delivers "Claude = 14 bookings across 12 name variants, weekly, 9710 kr, always 5420"), mining booked verifikat for spend (sum abs bank amount), recurrence cadence (median gap of distinct dates), dominant account (19xx+26xx excluded) + share, supplier entities from supplier_invoices. Chose read-side compute-on-read (no new tables/backfill job/embeddings) as the pragmatic v1 over the full persistent counterparty_entity substrate (bank_transaction_ai_normalization.md) which stays deferred: name-key merge alone delivers the wow; cross-name merges needing the bank-account hard key or fuzzy/embeddings are the deferred layer. (3) Page reworked from tables to a radial-hub SVG graph (company center, accounts inner ring, payees outer ring, hover reveals variants/spend/cadence/account+share). Both RPCs applied to prod + version-matched; light get_ledger_usage_stats stays the agent digest, deep RPC feeds the page. Then restructured the page (hero graph + Kompetens/Minne/Regler tabs) and added a read-only Kompetens (atoms) + Fakta (memory) overview mirroring /settings/assistant.
 [2026-07-08] Bedrock prod outage + Docker build failure both root-caused to dependabot #884 (a1fad319, 2026-07-06) bumping @anthropic-ai/bedrock-sdk 0.29.1->0.32.0. Runtime: 0.32.0 streaming returns an empty event stream ("request ended without sending any chunks", no HTTP status) - proven NOT a creds/region issue (prod diagnostic logged AKIA key + eu-west-1). Two prior sessions mis-diagnosed it as an AWS_* env collision and shipped/reverted #937 (BEDROCK_AWS_* rename) with no effect. "Works locally, fails on prod/CI" because local node_modules was stale at 0.29.1 while prod/Docker build fresh from the lockfile (0.32.0). Fix: pin back to ^0.29.1 + regenerate lockfile. FOLLOW-UP: add a dependabot ignore/exact-pin so it does not re-bump to 0.32.x and re-break both.
 [2026-07-08] One reconciliation PR adopts 3 prod-orphaned migrations (20260707113729 enrichment + 20260708120000/130000 ledger-stats RPCs) plus their pg-tests/fixtures onto main, instead of waiting on #927+#935 to merge: prod ledger was 3 versions ahead of the repo, leaving the default Supabase branch MIGRATIONS_FAILED and blocking every preview branch from being created. SQL committed byte-identical under the exact apply-time versions -> no-op on prod (idempotent), clean on fresh replays, and a no-op on #927/#935's next rebase. Carries #935's DB layer only (migrations + pg-tests + fixtures), not its UI/lib/i18n. Root anti-pattern: all three applied to prod via MCP apply_migration without committing the file (CLAUDE.md "never leave the remote DB ahead of the repo").
 [2026-07-08] Pinned @anthropic-ai/bedrock-sdk to exact 0.29.1 (dependabot #884 auto-bumped it to 0.32.0, which broke Bedrock streaming in prod: empty stream / "request ended without sending any chunks"). Guarded three ways against accidental re-bump: exact pin in package.json, dependabot ignore, and a pinned-dep check in scripts/checks/no-new-antipatterns.mjs (check:guards). Unpin only once 0.32.x streaming is verified against Bedrock.
+[2026-07-09] "Vad din agent vet" hero reworked from the flat radial-hub map into "Reconciliation Aurora" (LedgerGraph.tsx), a cinematic dark SVG panel, on the founder's explicit "maximum wow" directive (a design judge-panel of 4 divergent concepts picked it over circle-packing / flow-pulses / living-constellation). Encodings: node area = √spend, colour = cadence (weekly/monthly/irregular, the one semantic axis), shape = supplier vs counterparty, confidence = optical depth-of-field (crisp = sure, blurred = unsure). Signature on-mount act: the top ~5 name-merged payees' raw bank descriptors scatter in and collapse into one node with a counting-up ×N badge. Two deliberate calls: (a) a self-contained ALWAYS-DARK panel that bends design.md's flat/achromatic-surface rule (justified: it is a demo hero, self-contained, premium-not-neon, and DOF/glow only read on dark); (b) ZERO new deps: rejected d3-hierarchy (circle-pack) and d3-force (constellation) because a fixed 3-level tree is ~40 lines of deterministic polar trig and force physics would reshuffle the layout every load, breaking the repeatable live demo; framer-motion@12 (already installed) drives the entrance/merge, CSS keyframes drive idle pulse/breathe (single clock, no 40 JS loops), and confidence buckets into 3-4 shared SVG blur filters attached only after each node's entrance spring settles (avoids per-frame feGaussianBlur re-raster). Adversarial 3-lens review run; 3 low findings fixed (StrictMode CountUp freeze, filter-on-animating-subtree, center-out inverse permutation).
 [2026-07-09] Invoice online payment ships as a manual paste-link MVP (invoices.payment_link_url + email button + PDF QR/link) instead of a full Stripe Connect integration: a day of work as a demand probe vs a week for Connect (OAuth onboarding, pay page, webhook auto-booking to 1686). Same column/UI is the upgrade path: Connect would auto-fill payment_link_url later, so nothing is throwaway. Field is PSP-agnostic ("Betalningslänk", any https URL) since the effort is identical and it also covers PayPal/Zettle. Derived documents (credit note, proforma convert, recurring) deliberately do NOT copy the link: a pasted link encodes one amount for one invoice. MCP tools/list token ceiling bumped 45K->45.5K (headroom was <10 tokens; ledger entry in payload-size.bench.test.ts).
 [2026-07-09] Issue #916 (disconnect orphans ledger accounts): release claims by demoting cash_accounts rows to manual (bank_connection_id = null), never deleting: transactions.cash_account_id and ledger history reference the rows, and upsertFromPsd2 promotes a manual holder in place on reconnect so the bank lands back on its original BAS slot. Orphans predating the fix self-heal via a revoked-status filter in the allocator + collision guard (not data repair). When a promote collides with a duplicate row for the same connection+uid (callback mirrored onto an overflow slot pre-fix), the duplicate is deleted only if it has zero linked transactions, otherwise demoted: preserves FK links while freeing the slot. Picker-save rejections now render inline in the picker instead of routing to the sync-progress modal, whose parent-unmount-on-close made every save outcome invisible.
 [2026-07-09] #917 fix scoped to the current-year suggestion: "Sedan räkenskapsårets början" now resolves from the fiscal_periods row containing today, but the "Föregående räkenskapsårets start" custom option still derives from the recurring fiscal_year_start_month: the issue only covers the current-year date and a first-year company has no previous period row to resolve against.