Generation Future Real-Options Model

Issue: #1666. Stream: gen/future. Status: research memo / scoring model for future-generation issues.

This page gives a future agent the valuation model needed to decide whether a gen/future issue should be carried, exercised, or retired, without rereading the generation epic (#1625).

Core Rule

A gen/future issue is an option, not a backlog entry. The fleet holds the right, but not the obligation, to invest in the direction later. Keeping that right open costs something; exercising it requires evidence. This model makes both explicit so the future lane pays its carry deliberately instead of accumulating permanently-parked bets (the anti-pattern docs/generation.md already names).

The model is a disposition model, not a priority model. Priority still answers how valuable or urgent an item is. Shared trunk still requires one main, explicit-path commits, and witnessed closes. Runtime feature gates still decide whether shipped code is visible. Real-options valuation answers a different question: given what we know now, is it still worth paying carry to keep this future option open, or has the evidence moved enough to exercise, demote, or retire it?

Generation stays orthogonal under this model in three ways:

• Orthogonal to priority. A high-priority gen/future issue (a decision window that is genuinely opening) can receive capacity; a low-value future option can be retired regardless of how “strategic” it sounds. Option value and priority are separate axes that multiply, not substitute.
• Orthogonal to shared trunk. Exercising an option still lands through main by explicit path with the normal witness and DCO rules. The model never authorizes a branch, a worktree escape, or stale trunk hygiene.
• Orthogonal to runtime feature gates. An exercised future option may ship inert behind a default-off gate; the gate decides exposure, the option model decided whether the investment was worth making at all.

Do not solve carry pressure with generation branches, hidden worktrees, or ungated runtime exposure.

Option Variables

Each gen/future issue is described by six variables. They are estimated ordinally (better/worse, cheaper/dearer), not cardinally: these are not tradeable assets and no honest cardinal price exists.

Symbol	Name	What it measures	Source
V	underlying value	the decision, product, or market value if exercised with current evidence	the issue’s ## Why; sharpened at each recheck
X	exercise cost	cost to promote and build: engineering, gate, dogfood, review capacity	leaf tier + docs/generation-agent-capacity-model.md
C	carry cost	cost to keep the option open per recheck cycle: attention, recheck overhead, report tax, context pollution	named explicitly below
σ	uncertainty	how much V could move before the decision window closes	count of open assumptions / invalidating-assumption lines
T	time to decision	horizon until the option must be exercised, promoted, or retired	the recheck cadence / market window
S	strike evidence	the witness condition that makes exercise rational	the issue’s promotion criteria

Carry cost in detail

C is the variable #1666 explicitly asks to be valued. It is the price of not deciding. It compounds each recheck cycle an option stays open:

C_accumulated(issue) = sum of per-cycle carry across rechecks since intake

Carry is denominated in the scarce resources the capacity model already names:

• Attention/context — every open future issue occupies a slot in the milestone report, the dispatch views, and every operator/agent scan of the backlog. More open future issues = more scan tax = thinner focus for the items that matter.
• Recheck overhead — a future option that names a recheck cadence must be re-witnessed on that cadence or it goes stale. Each re-witness is a turn of human or agent work.
• Assumption drift — the longer an option is carried, the more its stated V and S drift from the live repo/market, raising the chance the next recheck is a surprise rather than a confirmation.
• Model tax — future issues that survive in the lane without a named exercise trigger bloat the lane and erode the signal value of the gen/future label itself (priority laundering becomes easier to hide).

A future issue with no named carry cost is not free; it has simply externalized its carry onto the lane. The model refuses that: every carried option states its C.

Disposition Rules

The six variables fold into one of four dispositions. This is the scoring surface — it produces a verb, not a number.

Disposition	Promotion verb	Fires when
exercise	promote	S has fired, V − X − C_accumulated is positive, and σ has fallen enough to commit. Move toward now/next.
hold	park	C is affordable, σ is high, T is open, S not yet fired. Keep open; recheck on the named cadence.
retire	retire	C_accumulated ≥ V, or S is structurally unreachable (an invalidating assumption fired), or a nearer stream superseded it.
hedge	demote	σ fell but V fell with it; keep as context but move to a cheaper/deeper carry tier.

The economically load-bearing rule is retire on accumulated carry. An option that never exercises and never dies is not a free option; it is a leak in the carry budget. The retirement disposition is what stops gen/future from becoming a graveyard of undead bets.

Carry budget

The gen/future lane has a finite carry budget. The sum of per-issue C across the lane must stay under it, or the lowest value-per-carry options retire first. Concretely:

if  Σ C(issue) over open gen/future issues  >  carry_budget:
        retire the issues with the lowest  V / C_accumulated
        until the lane is back under budget

carry_budget is set by operator judgement against the attention/recheck capacity the lane actually has; it is not a fixed number. The discipline is that the budget exists and that retirement is the release valve, not that the number is precise.

Exercise trigger contract

Every gen/future issue should name three things for the model to apply. These are the same three things the issue template already asks for; the model just makes them first-class rather than incidental:

1. Strike witness S — the concrete evidence (test, command, benchmark, market signal, standards shift) that would make exercise rational.
2. Recheck cadence — when carry C is next paid and the variables are re-estimated. A future issue with no recheck date is permanently parked, which is itself a retirement signal.
3. Kill condition — which named invalidating assumption, if it fires, retires the option immediately rather than waiting for the next recheck.

Worked Example

A hypothetical gen/future issue: “Research whether fak should expose a hosted multi-tenant gateway as a product surface.”

• V — high if it changes the go-to-market; the decision it could influence is “do we build a hosted tier.”
• X — large: a hosted surface means auth, tenancy, billing, SLOs, and the L3 share-scope machinery the kernel already gates (L3_CROSS_TENANT_SCOPE_DENIED).
• C — moderate per quarter: one recheck of the market + one witness that the multi-tenant isolation proofs still hold.
• σ — high: the hosting market, the model-serving economics, and the isolation guarantees could all move.
• T — open: no external deadline forcing a decision.
• S — strike: a repeatable customer ask AND a passed ScopeTenant/ScopeFleet isolation proof AND a unit-economics memo showing positive contribution.

Disposition today: hold. Carry is affordable, uncertainty is high, the strike has not fired. Recheck quarterly. Retire if the isolation proof cannot be strengthened within two recheck cycles (kill condition), or if C_accumulated exceeds the expected decision value before the strike fires.

This is the shape a gen/future memo should have: variables named, a disposition, a recheck date, and a kill condition.

Relationship To Existing Surfaces

This model composes with, and does not duplicate, the existing generation surfaces:

• debt_score (docs/generation.md §Debt Metric) is the lane-aggregate signal; this model is the per-issue valuation. The lane’s unpromoted_bets term is the coarse signal that carry is accumulating; the per-item model explains why a given bet is still worth carrying. Falling back to debt_score alone is the documented retirement path if per-item C proves unmeasurable (see Invalidating Assumptions).
• Capacity model (docs/generation-agent-capacity-model.md) sizes the fleet resources each horizon consumes; gen/future “receives small, expiring research/decision capacity.” Carry C is denominated in exactly that capacity, and the carry budget is bounded by it.
• Promotion verbs (docs/generation.md §Promotion Verbs) — promote / demote / retire / park map one-to-one onto exercise / hedge / retire / hold. The option model adds the valuation that justifies the verb; it does not introduce a new verb set.
• Sunset and kill criteria (docs/generation-future-sunset-criteria.md) is the operational companion to this model. It defines the closed sunset-trigger vocabulary that fires the retire disposition defined here (CARRY_EXHAUSTED, STRIKE_UNREACHABLE, ASSUMPTION_FIRED, …), the default recheck cadence, and the retirement-evidence contract. This model says whether to retire; the sunset criteria say when, why, and with what witness.

Promotion And Retirement Evidence

Promotion (exercise) evidence for this model is a re-estimation readout:

• A recheck names the new V, X, C_accumulated, and σ and shows the strike witness S has fired.
• The promoted issue lands through main with a witnessed commit, a feature gate if runtime exposure is involved, and a generation sidecar.
• A held option’s recheck produces a dated note confirming carry was paid and the disposition is unchanged (hold ≠ silence).

Demotion or retirement evidence is equally concrete:

• C_accumulated has reached or exceeded V and no near-term strike is plausible.
• A named invalidating assumption fired against live repo, issue, benchmark, or market evidence.
• A nearer stream shipped a design that supersedes the option.
• The strike witness S is structurally unreachable (e.g. a runtime gate or compatibility policy proved the surface cannot be safely exposed).
• The issue has no recheck date and no owner — permanent parking, which the model treats as a retirement signal, not a valid disposition.

Do not promote, demote, or retire by changing labels alone. Name the variable that moved and the witness that moved it.

Invalidating Assumptions

This model depends on these assumptions:

• Carry C is estimable ordinally. The model assumes a future issue’s carry can be ranked cheaply enough (higher/lower than the next issue) to drive the carry-budget retirement rule. If carry proves genuinely unmeasurable per-item, the model degrades to the lane-aggregate debt_score and should be retired in favor of it rather than defended with fabricated precision. This is the single most important assumption.
• Strike S can be stated up front. If most gen/future issues cannot or will not name a concrete strike witness, the exercise-trigger contract is vacuous and the model collapses into “hold everything forever.”
• A carry budget is enforceable. If operators never actually retire low-value-per-carry options, the budget is decorative and the lane still becomes a graveyard.
• Ordinal valuation resists laundering. The model assumes an operator can tell a high-uncertainty/high-value option from priority laundering. The existing anti-patterns (priority laundering, current-work laundering) apply here unchanged.

If these assumptions fail, replace the model with the stronger measured surface. Do not keep ordinal option math as an operator-facing fact once its inputs have stopped being honest.

Future Implementation Hooks

Reasonable next slices, each a separate issue:

• Add a gen/future issue template section for S (strike), recheck cadence, and kill condition, so the exercise-trigger contract is captured at intake.
• Add a fak future audit (or a gen/future view) that lists open future issues with their last recheck date and flags any with no recheck in the cadence window as retirement candidates.
• Add a carry-budget field to the milestone report’s gen/future lane, reported alongside the existing debt_score.
• Add a fixture that walks three synthetic future issues through exercise / hold / retire and asserts the disposition matches the rule.

Use docs/generation.md for the contract, docs/generation-agent-capacity-model.md for the capacity envelope, and docs/generation-loop-scheduling.md for admission and conflict arbitration.