Konstant × Acero + Altus · working conversation · confidential

Two companies. One learning system.

Acero + Altus + Konstant can become an AI-native precision growth company. Acero supplies the shop, quality record, and forty years of machinist judgment. Altus supplies an owned product and market loop. Konstant makes both reusable, then carries the learning into a permissioned network. Precise shows what each person, workflow, and member added.

the working thesis AI native has a practical meaning here: every finished job improves the next operator, quote, setup, product decision, and network model. The first proof is inside Acero. The network earns its way in later.

Konstant · prepared for Michael Fitzgerald · August 2026

What we see from outside

The two companies you built: precision, proven twice over.

Fig. 1 — the two companies, our readacero + altus · public record

Fig. 1. Our read from the public record — the plant, the fleet, the certificates, the circuits. Correct anything; that is what this conversation is for.

why we start here Everything in this deck respects what is already true: both companies run, and they run on judgment that took forty years to earn. The question is where that judgment lives, and what it earns when it is written down.

The AI-native growth case

Every finished job can improve the next person, part, product, and shop.

The combined systemfounder thesis · proposed

The order matters. Workforce and operating gains create the proof. Altus makes the learning loop unusually tight because one founder controls both a precision manufacturer and an owned medical-device product. The network and new revenue come after the local evidence.

founder thesisThe super-growth thesis is compounding productive capacity plus evidence-backed product revenue. It is an operating hypothesis; forecast status remains unbuilt.

Company basis: Acero history; Altus Spine. Konstant and Precise roles are proposed.

You already built this once

You proved this with Sponge. Konstant is Sponge for the decisions.

Fig. 2 — the same instinct, one level upsponge, as profiled 2011 → konstant

Fig. 2. Sponge watches the parts and refuses to let a drift hide. Konstant watches everything above the floor and holds decisions to the same standard the shop holds tolerances.

the one-liner Konstant gives the company a memory that shows its sources, a plan that shows its math, and the compression to ship both. The human decides. The waiting disappears.

Specimen 01 · the decision record

Every part gets a traveler. A serious decision should get one too.

Decision traveler · DT-0712

Thursday Swiss capacity · Acero ↔ Altus

Illustrative specimen · structure only
Signed · outcome measured

Decision ownerM. Fitzgerald

Decision deadlineWednesday · 14:00

Affected windowThursday Swiss capacity

Evidence state6 sources locked · 1 judgment

Question to decideWhich work gets the constrained Swiss window? Two setups are already queued; labor coverage is fixed; no unvalidated process change is permitted.

Option	Value case	Exposure	Value / evidence	What remains judgment
OEM purchase order	Contract revenue now; promised date held.	Consumes the same labor and Swiss window.	0.44 · 85/100	Relationship cost of a three-day move.
Altus Valencia run	Integrated margin; supports committed fall cases.	Moves OEM completion and uses internal capacity.	0.57 · 78/100	Weight placed on future case commitments.

Signed decision · op 030Run Valencia, conditionally. Ask the OEM to acknowledge a three-day move by 16:00. If it does not, revert to the OEM run. Reason: the integrated case commitment outweighs the modeled spread only while the relationship condition is satisfied.

Outcome + calibration · op 040/050Condition satisfied. Valencia sets landed on time; the OEM accepted the move. Forecast value 0.57; measured value 0.61. Keep this as a precedent, but do not generalize beyond the named capacity and demand conditions.

Evidence locked: production schedule · accepted PO and due-date thread · Altus case plan · recent setup actuals · labor coverage · QMS status. Identifiers and numbers are illustrative; the live record links every field to its source and version.

Refused: no cash-margin claim until landed cost is present; no prints, geometry, patient, surgeon, or raw customer data in this decision record.

Specimen 02 · the owner brief

The morning brief should say what changed, what is blocked, and who must decide.

Owner morning control brief · MB-0841

Tuesday · decisions before the floor meeting

As of 06:45 · refreshed 06:39
3 items · 1 refusal · illustrative

Briefing ruleChanged fact, expiring window, or breached threshold

Source health11 current · 1 stale · 0 failed

Decisions due09:00 · 10:00 · 12:00

Release authorityNamed humans only

Changed since yesterdayA third setup overrun joined the same job-family pattern; the RFQ response clock entered its final six hours; the Thursday Swiss conflict is now confirmed on both books.

Priority	What changed	Why it matters	Decision / next action	Evidence · owner
decide	RFQ-2214 enters final response window at noon.	High-fit 5-axis family; current load supports two defensible prices.	Approve premium / 24-hour draft or revise by 09:00.	Quote history + load snapshot locked · M.F.
verify	Job 8841 setup is 2.1 hours over plan for the third recorded run.	A recurring cause may be consuming the same senior intervention.	Floor lead confirms cause code before any schedule or standard-work change.	2/3 travelers complete · floor lead
decide	Swiss Thursday now has an OEM–Altus collision.	One choice moves an external promise; the other moves committed fall cases.	Open DT-0712 and choose by 10:00.	Both books linked · M.F.

What the system preparedTwo quote drafts with assumptions exposed; the three setup records aligned on one timeline; DT-0712 with alternatives, conditions, and source links.

What only people can doSet the price, determine whether the setup pattern is real, choose between the two books, and authorize any process or quality action.

Not briefed: no capacity recommendation beyond Thursday because the overtime roster is stale. The brief names the missing source instead of filling the gap.

Safely — in the language of your quality system

You already run governed processes. We run AI the same way.

Fig. 3 — your QMS, applied to AIiso 13485 → governed agents

On your floor	In Konstant	What it guarantees
Validated special processes	Qualified capabilities	An agent touches only tools that passed qualification and stay under continuous evaluation
Approved supplier list	Capability bindings	Every tool call is bound first: capability, version, policy, terms, all frozen into a signed receipt
Work instructions / SOPs	Declared workflows	A job is declared once — inputs, stages, release gate — and runs the same way every time
Final inspection	Release authority	Outputs are checked against the declaration; a separate gate approves exactly what leaves
Device History Record	Decision travelers	Predicted, actual, delta — per decision, signed, re-runnable. The audit trail writes itself

Fig. 3. The mapping holds because the instinct is the same: processes you can trust are declared, bound, gated, and documented. Your auditors already believe in this shape. So do we.

custody Runs where you choose: your building or a private tenant, one per company, with the owner's bridge between them. Customer prints, Penske drawings, surgeon relationships: all of it stays in your walls, and every governed output carries its receipt.

Specimen 03 · the monthly value report

A value report should expose the comparison, not merely print a score.

Quote desk value report · QD-JUL

Resolved RFQ decisions · decision-time evidence only

Illustrative specimen · not Acero performance
Method frozen before outcomes

PeriodJuly · one-month close

Decision unitOne RFQ disposition

ComparisonNext-best use of constrained hours

Completeness5 resolved · 5 outcome-linked

Executive readIn this specimen, the losses are not explained by win rate alone. One repeat job reused an obsolete price; one high-fit RFQ missed its response window. The report identifies the control failure, the decision owner, and the source record behind each conclusion.

Decision	Comparison fixed at decision time	Observed outcome	Incremental value	Reliability
RFQ-2214 · 5-axis	Premium / 24-hour response vs standard price.	Won; target margin held; hours fit the constrained window.	0.71 · positive	A− · cost + hours + outcome linked
RFQ-2231 · Swiss	Fill otherwise idle hours vs hold the margin floor.	Won thin; unused capacity converted to contribution.	0.58 · positive	B+ · demand alternative estimated
RFQ-2238 · outside envelope	Decline / refer vs consume scarce five-axis time.	Hours protected; referral accepted.	0.52 · positive	B · relationship value bounded
RFQ-2226 · repeat	Reuse prior price vs refresh current cost basis.	Won; material delta gave back most of the value.	0.07 · weak	A− · landed cost complete
RFQ-2219 · new logo	Respond inside target window vs quote on day six.	Lost after the customer's decision window.	0.04 · weak	B · counterfactual win remains uncertain

Control to add · repeat pricingRequire a current cost delta before reusing a prior price. Owner: quote desk. Advance when material, outside processing, setup, and constrained hours are refreshed.

Control to add · response windowStart a visible clock when a high-fit RFQ enters. Owner: commercial lead. Escalate before the evidence window expires, not after the opportunity is lost.

Report floor: no ranking for an unresolved outcome, an unlocked comparison, or missing actual hours. Learned serving estimates may prioritize attention; they do not settle value.

Specimen 04 · the field-asset review

A deployment report should show which capital is earning, which is blocked, and why.

Field asset review · FA-12M

Instrument-set deployment · hold, investigate, redeploy, or add

Illustrative specimen · not Altus data
Identity-protected owner view

Decision unitOne set in one territory

HorizonRolling 12 months + next 90 days

ComparisonCurrent shelf vs best permitted placement

Required contextUse · calendar · rights · move cost

Question to decideWhere should each constrained instrument set sit next? Case count is evidence, not the answer: coverage obligations, surgeon pipeline, training, contract rights, transfer friction, and capital basis remain in the comparison.

Deployment	Observed use	Context that changes the read	Value / reliability	Next decision
Hospital A · 2 Valencia sets	31 cases · stable cadence	Coverage supports two rooms; no near-term conflict.	0.66 · A−	hold Test whether a third set clears a real constraint.
Territory N · distributor	42 cases · 30% share	High use, but economics depend on distributor terms.	0.48 · B+	renew Reprice the agreement before adding inventory.
Hospital C · Suzuka tray	11 cases · launch cohort	Training and surgeon-development period remains open.	0.41 · B	observe Recheck after the next 90-day calendar closes.
Hospital B · Monaco set	3 cases · no booked use next 60 days	Contract permits movement; alternate site has a documented queue.	0.06 · A	review move Human owner confirms continuity and transfer cost.

Earliest decisionReview the Monaco move; do not automate it. The record should arrive with the receiving site's queue, contract rights, logistics, sterilization path, and continuity owner already attached.

Why cases alone failA low-use set may protect launch coverage or training; a high-use set may still underperform its terms. The declared value function carries capital, availability, relationship, and move costs together.

Privacy + operating floor: no patient identity in this report; no redeployment until contract rights, clinical continuity, receiving readiness, and transfer cost are confirmed.

The decision only one person can make

The transfer: one owner, two books, one price.

Fig. 6 — Thursday, the Swiss slotsacero ↔ altus · priced in one head, written nowhere

Fig. 6. The certificates draw the line: Altus designs and distributes, Acero manufactures. If Acero's hours make Altus's sets, every machine-hour on an Altus run is a decision against outside work, and you are the only person who stands on both sides of it. Konstant prices that decision, every week, with the reasoning written down.

why it compounds Every transfer decision scored makes the next one sharper — capacity, mix, and margin learned across both books at once. A forty-year instinct, calibrated on your own record, inside your walls.

The pit wall

Forecasts that rebuild their rooms as reality lands.

R1 The capacity room

A 24/7 floor short of machinists: which mix earns as seats fill — more Swiss turning, more 5-axis, more Altus transfer? Each future priced from your own job history.

Each step names how much history is under it.

R2 The Altus room

A navigation system cleared in 2025 and sixteen systems on the shelf: territory-by-territory futures for the channel — where a set, a rep, or a launch earns most next.

The launch scenario, before the launch.

R3 The transfer room

Integrated margin versus contract revenue, walked forward eight quarters: what both books look like if the Altus mix grows, and what they look like if it does not.

Both futures, on the record.

how it works You stand in a room built from what already happened. One step forward is a projection with its evidence shown; two steps is lossier, and it says so. When reality lands — a quote won, a case covered, a hire made — the walls move forward and the projections rebuild. Course correction is continuous. Every step shows its evidence.

We publish our own forecast accuracy against your outcomes, misses included. On a pit wall, that is just called telemetry.

Where this goes next

Small steps, in the order they compound.

01 The memory step one

Ninety days of email, meetings, and quote threads captured — per company, read-only, in your walls. The corpus starts compounding from day one.

Everything else stands on this.

02 One value report step two

Pick the desk: a month of quotes, the Swiss schedule, or one Altus territory's sets. Priced, ranked, travelers included — like Figs. 4 and 5, on real numbers.

Proof on your own numbers before anything scales.

03 The morning brief step three

Gary live for you and the people you choose — the Tuesday 6:45 brief, with every number clickable to its source.

The day the wait starts disappearing.

04 The working team alongside

An owner on each side and a thirty-minute weekly session, starting whenever you say.

Working conversations become systems when someone owns them.

the terms that matter Read-only until you say otherwise, scope set by you, and if you stop, the corpus exports to you and deletes here. If any board here is wrong about your business, that is the first thing we want to learn.

The next Acero growth constraint

A machine can be bought. Independent judgment has to be built.

Forty machines and a 24/7 schedule create usable capacity only when enough people can set up, adjust, inspect, diagnose, document, and stop the job at the right moment.

Acero · 202624/7 productionMore than 30 machinist positions announced, with second- and third-shift premiums.

U.S. BLSLong-term trainingMachinist proficiency develops on the job; trainees begin as operators and take on harder assignments over time.

Acero qualityJudgment under controlFirst-article and in-process inspection, lot traceability, gauge control, and ISO 13485 sit around every claim of independence.

the constraint in one sentence Acero does not merely need more people at machines. It needs a faster, inspectable path to bounded independent judgment—without lowering the standard or asking the same senior machinist to teach the same lesson forever.

Sources: Acero workforce expansion; Acero Careers; U.S. BLS machinist profile; Acero Quality.

What could be true

Every difficult job can leave a better starting point for the next shift.

When a senior machinist corrects a first article, recognizes chatter, changes an offset, or stops a questionable job, the useful asset is not a recording. It is the smallest signed case that explains what changed, why, and what happened next.

The expert correction becomes a case with context, rationale, outcome, rights, and trainer review.

The case becomes rehearsal for a new hire, a second set of eyes for the next shift, and evidence of readiness.

The next result shows whether the lesson actually improved quality, time, rework, or trainer burden.

What this is not: autonomous machining or a parallel quality system. The model can retrieve, explain, rehearse, assist, and refuse. Qualified Acero people retain setup, process-change, inspection, release, and machine authority.

The whole thesis, from first principles

Turn judgment into capacity. Then turn proven capacity into a network asset.

Inside Acero Capture and qualify.

Attach low-burden evidence capture to real production and quality events. Use signed cases to train people and one local workflow model. Prove one bounded path to independent work.

Across Acero + Altus Close the product loop.

Connect manufacturing outcomes to Altus design, instrument, surgeon, and market feedback. The combined company learns from both the part and the product it becomes.

Across a network Federate and earn.

Admit permitted contributions from matched peers only after local proof. Konstant governs identity, rights, versions, and rollup. Precise shows who improved the declared result. Acero owns the domain product and economics.

why this can become a different company The first return is more capable shifts and less repeated expert interruption. The second is a faster Acero–Altus learning loop. The third—only if the first two are proved—is a governed context product other shops and industrial partners may pay to use.

Proof order: local workforce value first; product-loop value second; network economics last. No external member, transferable model, or new revenue is claimed today.

The evidence behind the constraint

Acero needs independent machining judgment on every shift.

Forty machines and a 24/7 schedule create capacity only when enough people can set up, adjust, inspect, document, troubleshoot, and stop the job at the right moment.

Acero's public hiring record + an illustrative capability laddervalidate the ladder with Acero

The job is wider than running the control. BLS describes a progression from machine operator into more difficult assignments and calls machinist training long-term. Acero adds tight-tolerance medical work, first-article and in-process inspection, traceability, and its own quality system.

the workforce opportunity The craft stays hard. A visible, evidence-backed progression can widen the entry pool, focus senior trainers on the hard calls, support cross-training, and give Acero a defensible standard for independent operation.

Sources: Acero Careers; Acero workforce expansion; U.S. BLS machinist profile.

Extract the exception, not the person

Capture judgment at the moment it changes the job.

A low-burden evidence unitproposed · local by default

01 Shadow mode

The model predicts silently. The expert works normally. The gap becomes the lesson; nothing writes to the machine.

02 Exception replay

Rare, high-value cases return in the next shift huddle with the evidence and eventual outcome attached.

03 Teach-back

The trainee predicts the move and explains the basis. A qualified trainer signs the correction and the next practice case.

04 Skill passport

Readiness is shown by handled job families, recognized exceptions, correct escalations, and trainer-approved independence.

Floor rule: capture must fit an existing checkpoint, and the added operator time must be measured. If the floor rejects the burden, stop. Customer prints, images, part histories, and voice records remain inside Acero unless a specific right permits another use.

Research basis: ICLS 2024, tacit knowledge in machinists; NIST MEP, digital training capture; NIST MEP, Training Within Industry.

The first model stays inside Acero

Build a trainer and second set of eyes for one workflow.

Fig. 8 · one local training and operating loopfact + proposed capability

Start narrow: one machine cohort and one bounded decision, such as first-article dial-in or a recurring setup. The model retrieves matched cases, explains the evidence, and shows where a trainee still needs a qualified person.

public factAcero reports paperless workflow, real-time quality records, traceable inspection data, and an on-site CNC fleet.proposalJoin those signals into one evidence-carrying training and decision model.

Authority boundary: the model never signs a quality record, changes a validated process, releases a part, or writes to a machine. Acero's qualified people and quality system retain those decisions.

Sources: Acero Quality; Acero Precision Machining.

A permissioned federation

Each shop keeps its records. Konstant rolls up the permitted network history.

Fig. 9 · proposed network seamraw records remain local

proposedAcero owns the network product and member rules. Konstant supplies the identity, permissions, rollup history, and governed distribution rail.

Privacy gate: federated training does not make a network private by itself. Secure aggregation, input and output privacy tests, membership controls, revocation, and model-leakage checks are entry conditions.

Research basis: McMahan et al., federated learning; NIST on secure aggregation and privacy attacks.

Contribution before accrual

Precise shows which contribution made the network more valuable.

Fig. 10 · from contribution to a governed value rightexisting method family · proposed Acero binding

The score does not write the check. Precise measures contribution under a declared comparison. Acero sets the member contract that converts an admitted contribution record into access, credit, revenue share, or no accrual.

method existsPrecise already distinguishes exact value from learned estimates and refuses settlement when the method cannot support it.binding proposedThe Acero network contract and data adapters still need to be defined.

Settlement gate: exact or bounded contribution may inform settlement. Learned serving estimates guide operations only. Participation is not contribution, and contribution alone creates no payment right.

Research basis: Federated Shapley; GTG-Shapley. Precise status is based on the current internal capability description; no Acero integration is claimed.

The product above the network

Acero can become the context provider for precision manufacturing.

The commercial asset is an evidence-carrying context product: what tends to work, for which machine and workflow cohort, under which rights, with what reliability.

P1 Peer shops hypothesis

Matched-machine warm starts, cohort benchmarks, and model packs that carry provenance and site-calibration limits.

A faster start without exporting a shop's raw history.

P2 OEM customers hypothesis

Supplier-fit, capacity, quality, and program-risk context built from permitted network summaries.

A qualified view of execution conditions, not a data dump.

P3 Equipment + tooling hypothesis

Cohort performance, failure patterns, and product feedback with member and customer identities protected.

Downstream learning with a visible chain of rights.

the product contract Every output names its cohort, model version, provenance, reliability, allowed use, and expiry. Acero owns the domain product, customer relationship, participation rules, and commercial terms.

Default boundary: prints, geometry, customer identities, part-specific histories, raw shop records, and unprotected model updates stay local. A separate rights decision is required for every broader use.

Market signal only; Acero demand remains untested. MachineMetrics terms reserve rights for aggregate service analyses; Catena-X quality management demonstrates sovereign cross-company quality exchange.

The questions Michael should ask first

Six fair objections. Six conditions the system has to earn.

Fair objection	Credible answer	Pass condition
Every job is different.	Begin with a cohort defined by machine, material, feature, operation, and site. Recalibrate before any transfer.	The signal survives an unseen job and site calibration. Otherwise it stays local.
The important judgment is absent from Sponge.	Capture the exception, action, rationale, and later outcome at the work point. Ambient chat does not become training evidence.	No context or outcome means no admitted learning record.
Medical parts are no place to learn.	Run in shadow and read-only modes. Qualified people retain setup, inspection, release, and quality authority.	Acero quality approves the use, trace, validation path, and rollback.
Machinists will reject another system.	Attach capture to an existing traveler, inspection, correction, or handoff. Credit the trainer contribution visibly.	The floor owner accepts the measured burden and the record improves a later decision.
The customer owns the data.	Keep raw records local. Apply rights at the program and field level. Admit only approved updates, outcomes, and provenance.	Unclear rights mean exclusion. Leakage tests pass before any shared output.
This is another AI pilot.	Freeze the present workflow, value function, and comparison before the model runs. Report misses and refusals.	Independent evidence shows incremental value. A null result stops the expansion.

what earns credibility The system starts as an instrument for trainers and operators. It earns more authority only when the record shows equal or better quality, useful learning, acceptable floor burden, and a clean rights chain.

Controls basis: FDA Quality Management System Regulation; Acero Quality; NIST on federated-learning privacy risk.

A gated path from one workflow to one peer

Start local. Add one matched peer only after the model earns its keep.

proposedAll timing is an estimate. Each phase has an advance gate and a clean stop.

Window	Build	Advance only when
0–90 days	Map rights and provenance. Choose one training-heavy Acero workflow and one measurable outcome. Record the present trainer burden and path to signed independence.	The records are usable, the outcome is attributable, and Acero controls the permitted use.
3–6 months	Run one local model in shadow and teach-back modes on Acero-controlled compute. Compare it with the current workflow and grade each material contribution.	It adds out-of-sample value, a qualified trainer accepts the evidence path, and refusal behavior works.
6–12 months	Admit one non-competing, matched-machine peer. Add secure aggregation, site calibration, membership rules, and a signed accrual agreement.	The peer receives measurable value and privacy tests find no unacceptable leakage.
12–18 months	Offer one bounded context product to one defined buyer group.	Rights, reliability, participant value, and buyer utility survive independent review.

Stop when: rights are unclear, the model adds no incremental value, a leakage test fails, a member cannot see its value, or the attribution method cannot support the proposed accrual.

Operating proposal informed by the cited Acero, NIST, federated-learning, and contribution-evaluation sources. No revenue, launch date, external member, or model-performance result is claimed.

The headline at the end

Forty years of judgment can train the next forty years of Acero.

Acero plus Altus already contains the core of an AI-native precision company: the shop, quality discipline, machinists, an owned product, and a market loop. Konstant makes the learning reusable. Precise shows what each contribution added. Prove it first by helping one operator handle one difficult workflow with less senior interruption and the same quality authority.

One workflow. One trainer. One operator. One signed readiness report. Then one admitted peer.