Future Dispersion Protocol: Artifacts

The six Future Dispersion Protocol (FDP) artifacts are the working file of the protocol. If they are not written down, versioned, and allowed to fail, there is no protocol, only a conversation about better.

They form a loop, not a stack. The near-term portfolio can invalidate an end-state. A forecast check can force a rewrite of the better-statement. That is intended.

How the Six Pieces Fit

---

1. Horizon frame

Job. Freeze the problem so the rest of the work is about one locality, one clock, and constraints that will still exist when current executives are gone.

Required Fields

Field	What good looks like
Locality	A bounded system with a population, rules, and a budget: “North American plants + contract packers,” not “the industry.”
Horizon date	~100 years, plus two intermediate gates (often ~25 and ~50).
Binding constraints	Energy, water, skills, installed capital, regulation, customer switching costs. Things that do not vanish because a slide says “transformation.”
Non-goals	What this exercise will not decide.
Decision rights	Who can change the frame. If anyone can widen it midstream, the work dissolves.

Good fragment

Locality: regional grocery manufacturer, 4 plants, 2,100 employees, private label + one branded line.

Horizon: 2126, gates 2050 and 2075.

Binding: grid reliability in two states, aging retort capacity, retailer data lock-in, USDA/FDA path-dependence, a workforce whose median age is 47.

Non-goal: national food policy.

Fails when the locality is “society,” the horizon is “long term,” or constraints are only the ones the team already likes.

AI use. Retrieve comparable capital-stock lives (plants, brands, standards) and list which current assets will physically outlive the people approving the plan. Do not let a model invent a blank-slate locality.

---

2. Better-statement

Job. Make better expensive to say. This is the moral and commercial core. Every later artifact is a test of this page.

Required Fields

1. 1. Whose welfare counts: named groups, including one group that does not sit in the room.
   2. Dimensions: which conditions must move, in language a plant manager could measure.
   3. Trade-offs: what you will accept getting worse.
   4. Distribution rule: how you will know the floor did not fall while the average rose.
   5. Reversibility rule: which commitments must remain unwindable.
   6. Contest clause: who can challenge the statement and what evidence is enough.

Minimum Dimensions Worth Forcing Onto the Page

• • Preventable harm
  • Material security (wages, price to the end customer, continuity of supply)
  • Useful competence (time to replace a critical skill)
  • Agency (who can inspect, refuse, or leave the system)
  • Time (hours captured by the operating model)
  • Ecological load the locality actually controls
  • Error-correctability (time from failure to a reversible fix)

Good Fragment

Better for this manufacturer means: fewer unplanned line stops that empty a retailer’s shelf; no increase in recordable injuries; a shop-floor skill that can be taught to a new hire in 18 months rather than 7 years; private-label customers able to leave in 12 months without losing their spec history; water use per finished case down, not shifted to a co-packer in a drier basin. We will accept slower SKU proliferation and lower novelty in seasonal lines. We will not count “margin up/headcount down” as better if agency and replaceable skill fall.

Fails when it says “customer-centric excellence,” names no loser, or treats average EBITDA as a welfare measure.

Quality test. Hand the page to someone who will pay the cost (a line lead, a small retailer, a town water board) and ask what is missing. If nothing is missing, the statement is still advertising.

AI use. Ask for excluded stakeholders and historical cases where the same KPI was gamed. Require citations. Discard uplift language the model cannot tie to a group and a metric.

---

3. End-state sketches

Job. Prevent a single official future. You need a small set of internally consistent pictures of the locality at the horizon, including the ugly ones.

Usually three to five sketches

Sketch type	Purpose
Floor met	The better-statement is mostly achieved.
Average-up/floor-down	Headlines look good; the named floor group is worse off.
Lock-in success	The firm wins commercially by becoming un-leavable.
Competence collapse	Automation worked; teaching capacity did not.
Constrained physical world	Energy, water, or permitting bind harder than software hoped.

Each sketch needs the same skeleton so they can be compared:

• • Who operates the locality
  • What the core process looks like
  • Which vendors or agencies are unavoidable
  • Which skills still exist in humans
  • Who lost
  • What would be required to exit

Good Fragment (lock-in success)

2126: the branded line is a rounding error. The company is the default spec engine for three national retailers. Recipes, quality events, and labor scheduling live in a vendor model that retailers also use. Margin is stable. A new manufacturer cannot quote the same retailers without buying into that model. Shop-floor judgment is thin. The better-statement’s agency and reversibility clauses have failed. Commercially this is a win. Under FDP it is a failed sketch, not a target.

Fails when every sketch is a success story, or when sketches differ only by adjectives (“optimistic / pessimistic”) instead of by mechanism.

Quality test. If you cannot say who is worse off in a sketch, it is not internally consistent. Someone always pays.

AI use. Generate sketches only after the better-statement and constraints are pasted in. Then run a second pass: “Break sketch 2 using only the binding constraints.” Keep the breaks.

---

4. Backcast

Job. Turn a 2126 picture into present-tense requirements. This is where vision becomes a list of things that must already be true in 2050, 2035, and next year.

Work one sketch at a time, in bands of 15-25 years, asking four questions at each band:

1. 1. What capability must exist here?
   2. What standard, contract, or norm must already be normal?
   3. What must not have been locked in?
   4. Who had to learn something they do not know in the previous band?

Output. A requirement list, not a prediction timeline.

Good Fragment

For the “floor met” sketch to be reachable in 2126:
2050: spec history and quality events are portable across vendors; two independent maintenance training pipelines exist; water intensity is a board-level constraint, not a CSR paragraph.
2035: a second model vendor can run the scheduling stack; craft skills are documented as teachable units, not tribal knowledge.
2027-28: stop signing retailer data clauses that make portability illegal; pilot a portable spec packet on one co-packer; start a paid teach-back program on the two oldest lines.

Fails when bands contain slogans (“become AI-native by 2035”) or skip the “must not lock in” question. Lock-ins are usually contracts, schemas, and training that stopped.

Quality test. Every 2050 requirement must be caused by something in 2035. If 2050 assumes a portable data right that 2035 never created, the backcast is a wish.

AI use. Consistency checking is the high-value job: “Which 2050 items have no 2035 predecessor?” Humans still choose which requirements are worth wanting.

---

5. Forecast check

Job. Run time forward from today with only trends you can defend, then collide that path with the backcast. Divergence is the product.

Inputs that Belong Here

• • Age and replacement cost of physical assets
  • Workforce age and time-to-proficiency
  • Vendor concentration and switching cost
  • Energy / water / permitting reality in the actual places you operate
  • Regulatory direction you can cite, not hope
  • Customer power (retailer, hospital system, OEM)

Tag Every Divergence

Tag	Meaning	Typical action
Already underway	The backcast requirement is forming without extra heroics	Monitor; do not over-fund
Contingent this decade	A contract, hire, standard, or capex decides it	Put it in the portfolio
Blocked	A named interest or constraint sits on the path	Negotiate, route around, or retire the sketch
Derailment risk	Security, monopoly, skill collapse, ecological overshoot, capture	Redesign or kill the sketch

Good Fragment

Backcast wants portable spec history by 2035. Forecast: two retailers are moving the opposite direction in current Master Service Agreements (MSAs); switching cost is rising. Tag: blocked / contingent this decade. Either win a portability clause in the 2027 renewal or retire the “agency preserved” sketch as decorative.

Fails when the forward view is a market-size chart, or when every divergence is labeled “opportunity.”

Quality test. At least one sketch should die or get rewritten. If all sketches survive first contact with the forecast, the check was polite, not a check.

AI use. Watch a short indicator set: cost curves, overtime on bottleneck skills, vendor share, incident rates, permit times. A model can flag threshold breaks. It cannot decide that a blocked path is “strategic.”

---

6. Near-term portfolio

Job. Spend the next 12-36 months on requirements that survived the collision of backcast and forecast. This is the only artifact the P&L will notice.

Each Item Needs

• • The surviving requirement it serves
  • Owner with actual authority
  • Cost and the thing you will not do instead
  • Reversibility (can you unwind in 12 months?)
  • Kill criterion (what evidence ends it)
  • Leading indicator (what you will look at before the annual review)

A tight portfolio is usually four to eight items, not a transformation office.

Good Items

• • Rewrite the 2027 retailer data clause so spec history remains exportable.
  • Pay two retiring technicians to produce teachable units on the retort lines; measure time-to-proficiency for the next three hires.
  • Run the scheduling model on a second vendor for one plant for six months.
  • Cap new SKUs that add changeover time without moving a floor metric in the better-statement.

Bad Items

• • “Stand up an AI center of excellence.”
  • “Explore future markets.”
  • Anything with no kill criterion.

Fails when the portfolio is just the existing strategy deck with horizon language taped on.

Quality test. If the CEO left and the six artifacts were the only briefing, could a successor see why these eight actions exist and which sketch dies if they are cancelled? If not, the portfolio is still a project list.

AI use. Rank options by switching cost, time-to-kill, and which better-statement clause they protect. Execution stays human.

---

Ownership and Cadence

Artifact	Natural owner	Refresh trigger
Horizon frame	CEO / GM	Acquisition, new geography, regime change in regulation
Better-statement	CEO + one operator + one outsider	A floor metric moves the wrong way
End-state sketches	Strategy / futurist + operations	A sketch’s mechanism is disproved
Backcast	Strategy + finance + engineering	Capex or contract that would lock a band
Forecast check	Finance + ops + a designated skeptic	Indicator breach
Near-term portfolio	Line owners, not a Project Management Office (PMO)	Kill criterion or annual budget, whichever comes first

Do not assign all six to “innovation.” Innovation will protect the attractive sketches.

Version the set. The point of FDP, against the Antikythera problem, is that the reasoning does not sink with the project team.

For more information about how our work in Applied Artificial Intelligence can help you, please contact us HERE.