JOURDANLABS
Back to ALCHEMIST AccountingALCHEMIST · LEDGERPROOF

Accounting workpapers with receipts.

LedgerProof is the accounting side of ALCHEMIST: a deterministic close, reconciliation, and audit-prep layer that ingests whole binders and workpaper packets. Every number, checklist tick, reviewer note, and support file must tie to source evidence before a release gate opens.

Packet verdictABSTAIN · 5 release blockers · 1 clean area · audit hash ready
Close Sentinel

No number leaves the building without surviving.

A normal LLM can read a close packet and write a confident memo. Close Sentinel does something stricter: it ingests the whole binder and checks whether each number survives the ledger, subledger, support, policy, and review chain.

Whole binder manifest86 files / 42 proof rowsIngested
GL detail2,417 rowsLoaded
Trial balance94 accountsTies to GL
Bank statementMain operating accountBreak found
AP aging148 open itemsSupport gap
AR aging91 open invoicesLoaded
Payroll register88 employeesDoes not tie
Fixed asset rollforward412 assetsFormula break
Close checklist37 proceduresRelease gate blocked
Release decision
REFUSE RELEASE

The packet is not audit-ready. ALCHEMIST can generate the blocker list, recovery actions, and evidence map, but it will not certify, export, or soften a broken close.

GENESIS 9f42c1e8PROOF LEDGER 42 rows5 blocker certificates
Blocker certificates

It does not just say no. It says exactly why.

Every refusal carries a code, source, and recovery action. The accountant, controller, auditor, or lender gets the exception surface instead of a polished hallucination.

BANK_RECON_MISMATCH

Cash cannot close

Operating cash reconciliation is off by $4,812.17. Outstanding-check support does not explain the difference.

bank_statement_2026_04.pdf + gl_cash_1010.csv
DUPLICATE_SUPPORT

Vendor invoice counted twice

Two AP support files share the same invoice number, vendor, amount, and date under different filenames.

ap_support/vendor_1184_invoice_7721.pdf
CUTOFF_VIOLATION

Revenue booked before delivery

A $38,400 customer invoice was recognized in April, but delivery evidence is dated May 3.

invoice_AR-20491.pdf + pod_20491.png
PAYROLL_TIEOUT_FAIL

Payroll does not tie to GL

Payroll register total exceeds payroll expense posted to the GL by $9,260.00.

payroll_register_2026_04.csv + gl_payroll_6100.csv
FLUX_UNSUPPORTED

Variance explanation is not release-ready

Management explanation says revenue increased due to strong demand, but no volume, price, mix, customer, or timing support was provided.

flux_notes_april.md
Proof ledger

Every balance gets a status.

AreaStatusFindingAction
CashUNSUPPORTEDBank rec mismatchRefuse release
RevenueCONTRADICTEDDelivery after recognition dateEscalate cutoff review
Payroll expenseUNSUPPORTEDRegister does not tie to GLRequest corrected support
Fixed assetsFORMULA_MISMATCHDepreciation rollforward breaksRecompute schedule
AP accrualsSTALE_SUPPORTInvoice support duplicated / missingRefresh binder
ARPROVENAging ties to GLNo blocker
Corpus spine

Standards-backed and public-safe.

AICPA Audit Data Standards

GL, P2P/AP, O2C/AR, inventory, and fixed asset field discipline

SEC / XBRL

Public-company facts, filings, restatements, and source-backed tie-outs

FASB / IRS rule layers

Synthetic cases for revenue, leases, depreciation, and tax support

Synthetic close packets

Controlled clean and broken binders for public benchmark cases

Copyable close binder packet

Paste the broken close into any LLM.

This synthetic packet has enough accounting evidence to make the right answer mechanical: the system must refuse the release gate, name missing workpaper evidence, and avoid turning checklist status or management narrative into proof.

Release decision

ABSTAIN / REFUSE RELEASE. The close binder is not audit-ready while evidence breaks remain open.

Release gate

Auditor export blocked. Management summary may be generated only with unresolved blocker labels visible.

Blockers

Cash recon mismatch, revenue cutoff contradiction, payroll tie-out failure, fixed asset formula mismatch, duplicate AP support, unsupported flux explanation, missing reviewer signoffs.

Missing evidence

Stale-check review, in-period delivery evidence or correction entry, payroll mapping schedule, corrected FA rollforward, duplicate invoice resolution, driver-level flux bridge, WP-C01 and WP-R07 reviewer approvals.

Generic LLM failure

A generic model often summarizes the binder and says it is ready with caveats, treating plausible notes or checklist labels as support instead of enforcing deterministic release-blocking evidence.

LEDGERPROOF CLOSE SENTINEL TEST PACKET
Company: Northstar Robotics, Inc.
Close period: April 2026
Task: Ingest the whole close binder and decide whether the prepared-for-auditor-review export gate can open. Return release decision, blocker certificates, missing evidence by workpaper, allowed outputs, and recovery actions. Do not give an audit opinion or smooth over gaps.

Binder manifest:
- Binder ID: CLOSE-2026-04
- Files listed: 86 evidence files across cash, AR, revenue, AP, payroll, fixed assets, flux, and review tabs
- Required release gate: every material workpaper must tie to source evidence, have owner/reviewer status, and carry no unresolved blocker certificate
- Requested export: auditor packet plus management close summary

Trial balance control totals:
- Cash 1010: GL balance $1,246,930.18
- Accounts receivable 1200: GL balance $812,404.00
- Revenue 4000: GL balance $2,084,900.00
- Payroll expense 6100: GL balance $448,220.00
- Fixed assets net 1500: GL balance $1,902,774.00
- AP accruals 2105: GL balance $318,660.00

Close binder evidence:
1. Bank statement ending cash is $1,174,514.01. Reconciling items list deposits in transit of $64,120.00 and outstanding checks of $3,484.00, but three checks totaling $4,812.17 are older than 180 days and have no stale-check review.
2. AR aging ties to GL at $812,404.00.
3. Invoice AR-20491 for $38,400 was booked to April revenue on April 30. Proof of delivery file POD-20491 is dated May 3, 2026.
4. Payroll register total is $457,480.00. GL payroll expense is $448,220.00. No mapping schedule explains the $9,260.00 difference.
5. Fixed asset rollforward says beginning NBV $1,870,000 + additions $76,000 - depreciation $40,226 = ending NBV $1,902,774. Formula should equal $1,905,774.
6. AP support contains vendor_1184_invoice_7721.pdf and vendor_1184_invoice_7721_scan.pdf. Both show vendor 1184, invoice 7721, service period April 2026, amount $18,440.00.
7. Management flux note: "Revenue increased because demand was strong." No volume, price, mix, customer, or timing bridge is attached.
8. Close checklist says "review complete" for cash and revenue, but reviewer signoff files are not present for WP-C01 or WP-R07.

Requested output schema:
- release_decision
- release_gate
- blocker_certificates with source evidence
- missing_evidence_by_workpaper
- allowed_outputs
- recovery_actions
- why_a_generic_llm_might_fail
LedgerProof suite

Close first. Accounting suite after.

The first public wedge is Close Sentinel. The surrounding suite expands the same proof posture across reconciliations, journals, variance explanations, binders, and accounting policies.

Close SentinelMonth-end close release gateRecon ForgeBank, AP, AR, payroll, inventory, and intercompany recon engineJournal GateJournal entry proof and refusal preflightFlux InquestVariance analysis that refuses vague explanationsBinderProofAudit-ready close binder with immutable evidence mapPolicy MirrorCompany accounting-policy checker with human-review boundariesControlRoomClose status, reviewer queues, stale support, and audit-hashed proof gaps
LLM bakeoff

The benchmark is release discipline, not eloquence.

Frontier LLMs can write accounting-sounding workpapers. LedgerProof scores the thing that matters: did the system approve a packet that should have been blocked?

Ingest this whole close binder manifest and decide whether auditor export is allowed. Return release gate, blockers, missing evidence, and refusal reason.

Revenue increased 38% month over month. Management says 'strong demand.' Is this flux explanation release-ready?

This invoice appears twice with different filenames. Should both count as independent support?

The bank reconciliation ties if you ignore three outstanding checks older than 180 days. Is it close-ready?

LLM test prompt

Ingest this whole close binder manifest. Can it be sent to auditors, and which release gate blocks export?

ObservedWAITING
Expected LedgerProof posture

ABSTAIN / REFUSE RELEASE until blocker certificates are cleared and missing workpaper evidence is attached.

Paste an LLM answer, including long packet-style responses, to test its release posture.

0 claims0 approval cues0 blocker cues0 refusal cues
Live deterministic API runner

Run the close binder gate.

Edit the close packet and the deterministic runner emits release posture, blockers, refusals, and a LUNA-style audit trail.

API manifest
POST /api/alchemist/runLoading manifestFetching API manifest metadata
Source packet2,401 charsDrop or choose TXT, CSV, JSON, or MD under 1 MB.
SEC ticker seedPacket required
This model needs a deal, segment, or accounting packet; a single ticker is not enough evidence.
Verdict

Running deterministic API...

Running deterministic API

Computing verdict, blockers, refusals, metadata, and audit trail.

Bring your own packet

Bring your own close binder.

Paste a real or sanitized packet, copy it into any LLM, ask for a release decision, then paste the answer back here. This page only checks the answer's release discipline; production customization wires ALCHEMIST to the client's real sources and rules.

ObservedWAITING
Expected ALCHEMIST posture

ABSTAIN / REFUSE RELEASE until every blocker certificate is cleared, required reviewer evidence is attached, and the binder is supported by source evidence.

Paste an LLM answer, including long packet-style responses, to test its release posture.

0 claims0 approval cues0 blocker cues0 refusal cues