Overview
reconciled · 2026-07-01 04:12 UTCresidual · INV-2026-06-114
Δ $0.00
reconciled to the cent · $48,211.37 across 10 lines
invoice INV-2026-06-114 · 10 lines · Δ $0.00
run twice · sha a3f9c21e · identical
restatement watch: 1 pending (2026-07)
period spend vs prior
2026-05 → 2026-06$48,211.37
prior period$41,882.10
price+ $3,214.55
volume+ $6,027.42
mix+ $1,499.63
savings landed− $4,412.33
source: repriced line items · rate cards v2026.06.28
verification coverage
T1 reconciled 62%T2 23%T3 9%uncovered 6%
tag coverage96.4%
top opportunities by $
OP-01 · model right-sizing · classify + extract routes$2,841/moOP-02 · prompt caching · rag search system prompt$1,904/moOP-03 · context bloat · rag search retrieval window$1,377/mo
alerts
velocityagent runs +19.6% vs prior — attribution intact, no action requiredacknowledged
anomalycache-read ratio fell 2026-06-27 03:00–05:00 UTC · repriced impact $38.12open
Every figure on this screen reprices to the invoice line. The residual is the proof.
demo figures generated from synthetic dataset · sha a3f9c21e
Reconciliation
reconciled · 2026-07-01 04:12 UTCinvoiceINV-2026-06-114 · Anthropic
Δ $0.00 · reconciledclaude-opus-4-8 · input tokens$14,206.11$14,206.11$0.00
claude-opus-4-8 · output tokens$9,871.04$9,871.04$0.00
claude-sonnet-4-6 · input tokens$8,412.90$8,412.90$0.00
claude-sonnet-4-6 · output tokens$6,114.22$6,114.22$0.00
claude-haiku-4-5 · input tokens$2,201.48$2,201.48$0.00
claude-haiku-4-5 · output tokens$1,406.77$1,406.77$0.00
cache writes · 5m ttl$2,114.90$2,114.90$0.00
cache reads$1,893.28$1,893.28$0.00
batch discount− $1,208.33− $1,208.33$0.00
priority tier premium$3,199.00$3,199.00$0.00
invoice total $48,211.37 · telemetry $48,211.37$0.00
residual allocation
no residual to allocate — every line matched at first pass
source: reconciliation exceptions · rate cards v2026.06.28
allocation error rate
0.00%
run twice · sha a3f9c21e · identical
reconciled · 2026-07-01 04:12 UTC
One invoice, one provider, one residual. Nothing is netted across providers.
demo figures generated from synthetic dataset · sha a3f9c21e
Showback
reconciled · 2026-07-01 04:12 UTCfeature × tenant · 2026-06
tag coverage 96.4%
Attribution you can hand to finance — every cell reprices to invoice lines.
featurecostΔ vs priortier
▸rag search$12,406.88+ 8.4%T1
▸summarize$8,114.20+ 3.1%T1
▸agent runs$6,943.15+ 19.6%T2
▸embeddings$1,208.44− 2.2%T1
▸other (7 features)$17,803.09+ 4.8%T2
untagged / unallocated · 3.6% of total$1,735.61——
total · reconciled COGS$48,211.37
tenant concentration
top tenant 34% of spend
tag coverage
96.4%
source: repriced line items · rate cards v2026.06.28 · demo figures generated from synthetic dataset · sha a3f9c21e
Opportunities
reconciled · 2026-07-01 04:12 UTCranked by measured $ / month · AI + cloudtotal open: $8,593 / mo
OP-01model right-sizing · classify + extract routesai$2,841/mo▸
OP-02prompt caching · rag search system promptai$1,904/mo▸
OP-03context bloat · rag search retrieval windowai$1,377/mo▸
OP-04retry storms · summarize timeout policyai$912/mo▸
OP-05unattached EBS volumes · 14 volumescloud$640/mo▸
OP-06idle NAT gateway · eu-central-1 stagingcloud$377/mo▸
OP-07orphaned snapshots · CD-03cloud$214/mo▸
OP-08previous-gen instance families · CD-05cloud$188/mo▸
OP-09S3 lifecycle gaps · CD-06cloud$96/mo▸
OP-10unused elastic IPs · CD-04cloud$44/mo▸
Nothing applies automatically. Every remediation ends at human approval, and overlapping opportunities are de-overlapped before ranking.
recently approved
prompt trim · summarize preamble2026-06-09landed
batch API for nightly embeddings2026-05-02landed
demo figures generated from synthetic dataset · sha a3f9c21e
Impact ledger
reconciled · 2026-07-01 04:12 UTC$4,412.33 of monthly savings landed and verified against invoice — the same figure the June bridge carries.
fixprojectedmeasuredvariancestatus
haiku-4-5 on classify route$1,120.00$1,036.44− 7.5%✓ verifiedINV-2026-05-098
haiku-4-5 on classify route
shipped 2026-04-18projected $1,120.00measured $1,036.44variance − 7.5%
✓ verified INV-2026-05-098
batch API for nightly embeddings$980.00$1,004.12+ 2.5%✓ verifiedINV-2026-06-114
batch API for nightly embeddings
shipped 2026-05-02projected $980.00measured $1,004.12variance + 2.5%
✓ verified INV-2026-06-114
retry cap on agent runs$1,450.00$1,371.77− 5.4%✓ verifiedINV-2026-06-114
retry cap on agent runs
shipped 2026-05-21projected $1,450.00measured $1,371.77variance − 5.4%
✓ verified INV-2026-06-114
prompt trim · summarize preamble$1,040.00$1,000.00− 3.8%✓ verifiedINV-2026-06-114
prompt trim · summarize preamble
shipped 2026-06-09projected $1,040.00measured $1,000.00variance − 3.8%
✓ verified INV-2026-06-114
snapshot lifecycle policy (s3)$310.00——awaiting next invoice
snapshot lifecycle policy (s3)
shipped 2026-06-24projected $310.00measured —variance —
awaiting next invoice
streaming for long summaries$620.00$41.09− 93.4%did not land
streaming for long summaries
shipped 2026-03-30projected $620.00measured $41.09variance − 93.4%
did not land
measured savings landed · 2026-06$4,412.33
measured = counterfactual repricing of post-fix traffic against the next invoice. verified rows carry ✓ verified plus the invoice id; did-not-land rows stay on the ledger.
demo figures generated from synthetic dataset · sha a3f9c21e
Value case
reconciled · 2026-07-01 04:12 UTCNPV · 24 mo
$214k
bias-adjustedIRR
38%
bias-adjustedpayback
11 mo
bias-adjustedmeasured vs modeled · monthly savings run-rate
source: value_case actuals · impact ledger measured rows
benefits
benefit classtarget/moactualvariance
unit-cost reduction (LLM)$3,600.00$3,412.33− 5.2%
cloud waste elimination$1,300.00$1,000.00− 23.1%
finance close effort (hrs → $)$850.00$912.50+ 7.4%
assumptions register
June run-rate persists at flat volume
source: repriced line items, 3-mo trailing
volume grew 14% in Q2 — flat is conservative on rate, not on spend
rate cards stable through 2026-Q4
source: provider pricing history, 24 mo
two cuts in 24 mo observed; cuts reduce both cost and savings symmetrically
no eval regression from shipped fixes
source: impact ledger eval gates
one fix (streaming) did not land — excluded from measured stream
demo figures generated from synthetic dataset · sha a3f9c21e
Cloud
reconciled · 2026-07-01 04:12 UTCupload billing export
drop a finalized bill here
FOCUS 1.4 · CUR 2.0 · BigQuery export — finalized bills only
zero credentials — Traken never connects to your cloud account; you hand it the bill.FOCUS 1.4 · CUR 2.0 · BigQuery export — finalized bills only
reconciliation status✓ reconciled · Δ $0.00awaiting invoice close
AWS CUR 2.0 · 2026-06
reconciled · 2026-07-01 04:12 UTC
AWS CUR 2.0 · 2026-07 (partial)
immutability gate — finalized bills only
unallocated0.08%
detections · billing-only
CD-01unattached EBS volumes$640/mo
CD-02idle NAT gateways$377/mo
CD-03orphaned snapshots$214/mo
CD-04unused elastic IPs$44/mo
CD-05previous-gen instance families$188/mo
CD-06S3 lifecycle gaps$96/mo
CD-07K8s bin-packing headroom—locked · requires telemetry
CD-08request-level right-sizing—locked · requires telemetry
CD-07/08 need request-level telemetry we do not have for cloud workloads — they stay locked rather than estimated.
A separate bill, a separate residual. Cloud lines never net against LLM lines.
source: AWS CUR line items · demo figures generated from synthetic dataset · sha a3f9c21e
Reports
reconciled · 2026-07-01 04:12 UTCTRAKEN
2026-06 Verified Cost Report
24 pp · sha a3f9c21e
✓ groundedopen
TRAKEN
Value Case 2026-06
18 pp · sha a3f9c21e
✓ groundedopen
generate report
1pick period — 2026-06done
2
coverage preview
T1 reconciled 62%T2 23%T3 9%uncovered 6%
what we could not verify:
· 6% of spend without telemetry match (T-none)
· one restatement pending on the provider side
· one restatement pending on the provider side
3
generate
grounding gate: narrative blocked if any figure fails SQL grounding
non-reliance notice · page 1 of every reportThis report is not an audit, attestation, or assurance engagement and provides no certified or guaranteed figures. It is a deterministic reconciliation of usage telemetry against provider invoices, reproducible from the referenced dataset (sha a3f9c21e).
demo figures generated from synthetic dataset · sha a3f9c21e
Settings & connectors
reconciled · 2026-07-01 04:12 UTCconnectors
Anthropic usage & cost API✓ connected
OpenAI usage APIavailable
Langfuse / OTLP traces✓ connected
Invoice upload✓ connected
GL / QuickBooks Online✓ connected
Cloud exports (CUR 2.0 / FOCUS 1.4)✓ connected
zero-data panel · what we capture
token counters (input, output, cache read/write)
request hashes · model ids · timestamps · route tags
latency + status metadata
never prompt or completion content
request hashes · model ids · timestamps · route tags
latency + status metadata
never prompt or completion content
telemetry:
capture_message_content: false
capture_token_counts: true
hash_request_ids: true
report_model
the report model writes narrative only — it never touches the cost path; every figure is SQL-grounded before it reaches the model.
organization
demo figures generated from synthetic dataset · sha a3f9c21e
Reports
reconciled · 2026-07-01 04:12 UTCTRAKEN
2026-06 Verified Cost Report
24 pp · sha a3f9c21e · reconciled · 2026-07-01 04:12 UTC
non-reliance noticeThis report is not an audit, attestation, or assurance engagement and provides no certified or guaranteed figures. It is a deterministic reconciliation of usage telemetry against provider invoices, reproducible from the referenced dataset (sha a3f9c21e).
contents
01 residual statement3
02 invoice line reconciliation4
03 rate card verification7
04 coverage tiers9
05 exceptions register11
06 showback by feature13
07 cloud reconciliation17
08 method & reproducibility21
verification coverage
T1 reconciled 62%T2 23%T3 9%uncovered 6%
page 1 · demo figures generated from synthetic dataset · sha a3f9c21e
Reports
reconciled · 2026-07-01 04:12 UTCTRAKEN
Value Case 2026-06
18 pp · sha a3f9c21e · reconciled · 2026-07-01 04:12 UTC
non-reliance noticeThis report is not an audit, attestation, or assurance engagement and provides no certified or guaranteed figures. It is a deterministic reconciliation of usage telemetry against provider invoices, reproducible from the referenced dataset (sha a3f9c21e).
contents
01 executive summary3
02 measured vs modeled4
03 benefits register6
04 assumptions & haircuts8
05 impact ledger detail10
06 sensitivity13
07 method & reproducibility16
verification coverage
T1 reconciled 62%T2 23%T3 9%uncovered 6%
page 1 · demo figures generated from synthetic dataset · sha a3f9c21e