Skip to main content
01

Nervos · Decision Intelligence Operating System for finance and procurement

AI can recommend an action.
Nervos governs the decision.

Connect evidence, policy, human authority, execution, and outcomes in one accountable decision system.

For finance and procurement teams on NetSuite. In the sample on this page, a vendor bill waits until the people who own it approve.

Request a private demoOperate the full demo

Pilot: $15,000, 90 days, one approval workflow in your NetSuite. The second half is due on written acceptance against agreed criteria. The full fee is credited to a first-year annual contract.

The NetSuite connector is implemented and not yet exercised against a live account.

A sample decision, replayed

Sample decisionComputed when this page was built

Replayed with Nervos’s own simulation and hashing code. Nothing is sent to an ERP. The record stops at stage 5 for your authorization.

Decision record · SAMPLE

v2 of 10 · evidence

Should we consolidate 14 packaging vendors into 3?

Evidence · observed inputs

  1. Verified

    14 active packaging vendors. $6.4M trailing-twelve-month spend.

    NetSuite vendor ledger, account 5110 · 2 Mar 2025

  2. Verified

    Three vendors hold 71% of that spend.

    Deterministic aggregation of the same ledger rows · 2 Mar 2025

  3. Verified

    Vendor B on-time delivery: 94% over 12 months.

    NetSuite receipts, 412 lines · 2 Mar 2025

  4. Inferred

    Vendor C states it can absorb 40% more volume within one quarter.

    One email from the vendor’s account lead · 3 Mar 2025

4 of 7 evidence items shown · 1 contradiction held

Stage 1 of 10. The record opens with what is known. Each observation keeps its source and its time. A ledger read is VERIFIED. A vendor’s email is INFERRED. Nervos keeps the difference.

Operate all 17 stepsRequest a private demo

The organization and vendors are fictional. This page recomputes the futures and the ledger hashes in your browser. Not every control shown here is enforced in production yet. The claims register says which. Nothing is sent to an ERP.

Read the ten stages as text
  1. Stage 1. The record opens with what is known. Each observation keeps its source and its time. A ledger read is VERIFIED. A vendor’s email is INFERRED. Nervos keeps the difference. VERIFIED: 14 active packaging vendors. $6.4M trailing-twelve-month spend. (NetSuite vendor ledger, account 5110) VERIFIED: Three vendors hold 71% of that spend. (Deterministic aggregation of the same ledger rows) VERIFIED: Vendor B on-time delivery: 94% over 12 months. (NetSuite receipts, 412 lines) INFERRED: Vendor C states it can absorb 40% more volume within one quarter. (One email from the vendor’s account lead) 7 evidence items in all.
  2. Stage 2. Each path is run forward. Two alternatives are run 10,000 times each. Every run draws the beliefs the path rests on. The result is a range, drawn dashed. Nothing here has happened. Keep all 14: not simulated. Consolidate to 6: P10 $209k, P50 $392k, P90 $563k. Consolidate to 3: P10 $119k, P50 $583k, P90 $895k. Computed when this page was built, seeds 717699, 967700.
  3. Stage 3. The recommendation is challenged. The simulation alone says 75%. Vendor C’s capacity rests on one email. In this sample, the weakest source caps confidence at 60%. The risk perspective prefers six vendors, and that stays on the record. Strategic prefers “Consolidate to 3”: Fewer suppliers concentrate negotiating leverage. Market prefers “Consolidate to 3”: Quoted tiers favour the larger volume. Automation prefers “Consolidate to 3”: Fewer vendor records and approval paths to keep. Risk prefers “Consolidate to 6”: Single source in two categories. Keep a second source. Return prefers “Consolidate to 3”: Largest median savings of the paths simulated. These positions are written into the sample record.
  4. Stage 4. A person decides. The recommendation is not the decision. The Director of Procurement holds both sides of the contradiction and declares the assumption. Then makes the choice: Consolidate to 3. Accepted as competing by Director of Procurement. Retain Vendor B as secondary supplier for two quarters while the overcharge is reconciled.
  5. Stage 5. Two approvals, then the write. Nothing is written yet. The policy asks for two approvers on a high-risk bill of $486k. The Controller has approved. You are the VP Finance. A high-risk write that is irreversible or at least $100,000 needs two approvers. The policy evaluator returned: high risk and amount $486,400 ≥ $100,000: a second, distinct approver is required. In the sample, 2 of 2 distinct approvers, neither the proposer.
  6. Stage 6. Each step is chained to the one before. A claim, then preflight, execute, persist. Your browser computes each SHA-256 hash with the production ledger’s rule. Nothing is sent to an ERP. Step 1, authorize, Controller. Step 2, authorize, VP Finance. Step 3, claim, Controller. Step 4, preflight, checked under the grants. Step 5, execute, sent · no answer. Step 6, persist, recorded unknown.
  7. Stage 7. Sent is not confirmed. In the sample, the write was sent. NetSuite has not answered, so the record says UNKNOWN. Nervos does not guess. It waits for the system of record. NetSuite lookup found NS-VB-38207 by idempotency key demo-cons-14-3-vb-1. The system of record answered. In the sample, a NetSuite lookup found the vendor bill by its idempotency key. The record moves from UNKNOWN to VERIFIED. Nervos did not guess.
  8. Stage 8. Prediction. Reality. Difference. Twelve months later, the Controller records the outcome from the ledger. It lands inside the registered band, below the median. The gap is scored, not explained away. Registered band P10 $119k, P50 $583k, P90 $895k. Observed in the sample $468k. CRPS 0.51, PIT 0.35, computed when this page was built.
  9. Stage 9. The outcome changes what is believed. The pricing-tier belief moves from 67% to 60%. The next consolidation starts from a different prior. Calibration needs 20 resolved predictions. This is one. Consolidated pricing tier lands at −8% to −12%: 67% to 60%. No single-source disruption within 12 months: 80% to 82%. The next consolidation decision starts with a tier-discount prior of −6% to −10%, not −8% to −12%.
  10. Stage 10. Recompute the record. Your browser rehashes every ledger step and compares each with the hash this page was built with. Alter a step to see the chain break there. 7 ledger steps. If the approver on step #2 is changed after the fact, its hash no longer matches. The chain breaks at that row.

The problem

Organizations have more intelligence than ever. The reasoning behind a decision is scattered across people, documents, systems, models, spreadsheets, meetings, approvals, emails, and operational software. A week later nobody can say what was assumed, what the alternatives were, or whether the prediction came true.

The problem is not a lack of information. It is a lack of continuity.

Sample decisionBelow, one decision is followed from its first evidence to what it taught the organization. Every figure is the engine’s own output on that record.

02

The idea

A decision should not disappear after it is made.

It should stay connected to what informed it, what challenged it, what was assumed, what was authorized, what happened, and what was learned. Nervos is built around that record.

Decision record

Simulated
Question
Should we consolidate 14 packaging vendors into 3?
Final phase
outcome recorded
Versions
10
Evidence
7
Contradictions held
1
Alternatives
3
Assumptions declared
3
Prediction · outcome
1 · 1
Prediction error (CRPS, lower is better)
0.51
03

The system

Frame. Understand. Explore. Challenge. Govern. Act. Learn.

One continuous lifecycle, not seven features. Each verb is a phase of the same record. A new phase is added as a new version, and Nervos never edits an earlier one.

  1. 01Frameevidence
  2. 02Understandbelief
  3. 03Exploreoptions · simulation
  4. 04Challengecontradiction · critique
  5. 05Governdecided · authorized
  6. 06Actexecuting · verified
  7. 07Learnoutcome recorded
  8. The dashed return from Learn to Frame closes the loop.
Seven verbs, one line. Beneath each, the phases of the decision record it covers; under Challenge, the checks that run inside it. The dashed return is learning: an outcome re-enters as evidence for the next decision.
04

The decision

Should we consolidate 14 packaging vendors into 3?

Fourteen packaging vendors. $6.4M a year. Three of them hold most of the spend. This is the record as it stood before anyone decided anything.

Decision record · v4

phase · optionsSimulated

Evidence

  1. 14 active packaging vendors. $6.4M trailing-twelve-month spend.

    NetSuite vendor ledger, account 5110

    Verified
    2 Mar 2025

  2. Three vendors hold 71% of that spend.

    Deterministic aggregation of the same ledger rows

    Verified
    2 Mar 2025

  3. Vendor B on-time delivery: 94% over 12 months.

    NetSuite receipts, 412 lines

    Verified
    2 Mar 2025

  4. Vendor B unit price feed runs 7% above the contracted rate.

    Price feed connector, last 90 days

    Conflicted
    3 Mar 2025

  5. Vendor B contract rate is fixed through 2027.

    Signed contract, clause 4.2 (document ingest)

    Conflicted
    3 Mar 2025

  6. Vendor C states it can absorb 40% more volume within one quarter.

    One email from the vendor’s account lead

    Inferred
    3 Mar 2025

  7. 11 quality incidents across vendors D–N in the last 12 months.

    Incident log, quality team

    Verified
    3 Mar 2025

Assumptions

  1. A consolidated pricing tier lands between −8% and −12%.

    Two prior negotiations with vendors A and C. No quote yet.

    Inferred

  2. No single-source disruption within 12 months.

    No evidence either way. Recorded as an assumption, not a belief.

    Pending

  3. Switching completes inside one quarter.

    Operations lead estimate. Two comparable transitions took 10 and 14 weeks.

    Inferred

Alternatives

  1. Keep all 14

    No switching cost and no savings. The 11 incidents a year continue.

    Baseline
    —

  2. Consolidate to 6

    Retain a second source in every category. Smaller tier discount.

    Simulated
    P50 $392k

  3. Consolidate to 3

    Largest tier discount. Single source in two categories.

    Simulated
    P50 $583k
05

The futures

Nervos reasons about possible outcomes without pretending a simulation is reality.

Each alternative is run forward ten thousand times, grounded in the beliefs it rests on. Thin evidence widens the band on its own. The result is drawn dashed and labeled POSSIBLE. Nothing here has happened.

Keep all 14

No simulation. No switching cost, no savings. The 11 incidents a year continue.

06

The challenge

Good decision intelligence does not produce an answer. It tests whether the answer deserves to survive.

Before anyone could authorize the preferred path, the system tried to break it.

  1. Conflicted

    The price feed and the contract disagree about Vendor B’s rate.

    The belief is marked CONFLICTED. Every conclusion that rests on Vendor B’s price is capped.

  2. Weak source

    Vendor C’s capacity claim rests on one email.

    The recommendation’s confidence cannot exceed that source. Ceiling: 60%.

  3. Assumption

    “No single-source disruption” has no evidence either way.

    Recorded as an assumption, not a belief. The downside scenario keeps a 25% weight.

  4. Disagreement

    The risk perspective prefers six vendors. The savings perspective prefers three.

    Recorded as a material disagreement on the record. Not averaged away.

The simulation alone would have said 75%. The weakest source it rests on — vendor c can absorb 40% more volume in one quarter, one email — says 60%. The lower number wins. Confidence cannot emerge from abstraction.

07

Trust

Inspect why Nervos believes what it believes.

Every figure opens. Source, status, ceiling, model, time. Not a footnote. The record.

Simulated
source
Monte Carlo, 10,000 iterations, seed derived from the scenario inputs
model
three scenarios (as quoted 55% · volume tier 20% · single-source disruption 25%) over three grounding beliefs
status
simulated
ceiling
none on the quantity; the recommendation built on it is capped at 60%
time
10 Mar 2025 · v5 of the record
08

Consequence

Reason. Govern. Authorize. Act. And when the world does not answer: UNKNOWN.

The recommendation cannot authorize itself. In the sample, two people cross the boundary, and neither is the proposer on record. Then the action enters an external system, and Nervos waits for that system to say what happened.

Authorization

10 Mar 2025
proposed by
Director of Procurement
approved by
Controller · 14:01:12 UTC
approved by
VP Finance · 14:01:40 UTC
quorum
2 of 2 distinct approversHIGH_IMPACT_SECOND_APPROVER_V1In this sample. Not yet enforced in production. AUTHZ-03
separation
Nervos refuses an authorization from the proposer on record, at approval and again at dispatch. Until a pending migration is applied, a member can set who is on record.
action
Post Vendor A’s first bill under the consolidated contractnetsuite · createVendorBill · $486k · risk high · reversible: A compensating write deletes the bill. It does not restore the released vendors.

Execution ledger

SHA-256 hash-chained
  1. genesis · 0000000000…000000
  2. #1authorize · ControllerHashed
    prev 0000000000…000000hash c7dd6eb7d0…ca2049
  3. #2authorize · VP FinanceHashed
    prev c7dd6eb7d0…ca2049hash 2da0380b45…4b7347
  4. #3claimHashed
    prev 2da0380b45…4b7347hash 3f894d75b8…5d4d37
  5. #4preflightHashed
    prev 3f894d75b8…5d4d37hash b0a9e4cfde…f46658
  6. #5execute · unknownHashed
    prev b0a9e4cfde…f46658hash 4d00cb866c…a36215
  7. #6persist · unknownHashed
    prev 4d00cb866c…a36215hash 72dcbd7f26…1e1982
  8. External stateUnknown

    The write was sent. NetSuite did not answer. Nervos does not assume success or failure. The decision stays in executing.

  9. #7verify · reconciled 41 seconds laterVerified

    NetSuite lookup · verifyExecution · NS-VB-38207 · idempotency key demo-cons-14-3-vb-1

    prev 72dcbd7f26…1e1982hash ca99c068a1…6fde48

Entry hash = SHA-256(previous hash · canonical content), computed on this page with the same rule the production ledger uses. Reproducible, tamper-evident. This is a demonstration; no system of record was written.

Here the chain holds and you are reading it. In the demo you alter a row and watch it break at exactly that row — and you authorize this consequence yourself, as the person who did not propose it.

09

Outcome

Prediction. Reality. Difference.

Twelve months later a person recorded the outcome from the system of record. Members cannot edit the registered band to fit it. Once a pending migration is applied, neither can the server.

Possible · registered before the outcome
P10 $119k · P50 $583k · P90 $895k
Observed
$468k
Prediction error · observed − P50
−$115k
The band was registered before the outcome was knowable. Members cannot edit it, and once a pending migration is applied neither can the server. The observed value is drawn solid. The line between them is the prediction error.
observed
$468k
predicted P50
$583k
error · CRPS
0.51
where it landed · PIT
0.35

Inside the 80% band, below median. The tier discount landed at −6.9%, below the assumed −8% to −12%. No disruption occurred. Recorded by the Controller, from NetSuite account 5110 at the twelve-month close, on 2 Apr 2026. CRPS scores the whole predicted distribution against the outcome, in multiples of the switching cost; lower is better. PIT is the share of that distribution at or below the outcome; 0.5 is the median. One prediction is one data point. Nervos gives a calibration verdict only after twenty have resolved; until then it shows interval coverage, with its count.

10

Learning

Outcome changes knowledge. Knowledge changes the next decision.

The outcome entered memory as evidence, at full weight. Two beliefs moved. The next consolidation decision starts from a different prior.

Belief revisions

cause · outcome

  1. Consolidated pricing tier lands at −8% to −12%

    Observed −6.9%: one contradicting observation at full weight.

    67% → 60%

  2. No single-source disruption within 12 months

    No disruption occurred: one supporting observation.

    80% → 82%

  3. Memory claim written

    Verified

    Consolidating 14 packaging vendors to 3 returned 2.6× switching cost in 12 months. The tier discount was −6.9%.

    outcome.observed · half-life 365 days

The next consolidation decision starts with a tier-discount prior of −6% to −10%, not −8% to −12%.

Now operate it yourselfHow it works

The full demo has seventeen steps. You record how the contradiction is handled, authorize the write, and wait for the system of record.

This page followed one sample decision. The organization and vendors are fictional. The figures are the engine’s output on that record — deterministic Monte Carlo, the production hash rule, and proper scoring rules — and no system of record was written. Where Nervos has not yet proven something, it says not yet.

The pilot

Bring one approval workflow you have to get right.

90 days in your NetSuite, for $15,000. The second half is due on written acceptance against agreed criteria. The full fee is credited to a first-year annual contract. The NetSuite connector is implemented and not yet exercised against a live account. Nervos has no customers yet, and says so.

Request a private demoRead the pilot terms