Foundation / checks (deploy) (push) Successful in 6s
Foundation / checks (ecs) (push) Successful in 1m56s
Foundation / checks (postgres) (push) Canceled after 0s
Foundation / checks (tsan) (push) Canceled after 0s
Foundation / postgres-tools (push) Canceled after 0s
Foundation / server-image (push) Canceled after 0s
Foundation / image (push) Canceled after 0s
Foundation / checks (network) (push) Canceled after 5m28s
Foundation / checks (checks) (push) Canceled after 7m39s
Foundation / checks (client) (push) Canceled after 7m39s
184 lines
10 KiB
Markdown
184 lines
10 KiB
Markdown
# ADR-0098: M11 first cellular loop decisions
|
||
|
||
## Status
|
||
Accepted for M11 (2026-10-09). M11.1 records these decisions; nothing is implemented
|
||
yet. The maintainer chose each one from presented options on 2026-10-09. They are game
|
||
rules for the first playable loop, and changing one needs a new decision.
|
||
|
||
Numeric values marked **tunable** are starting points for the 11.9 playtests. They are
|
||
not locked rules. Each one is a versioned rule constant, so a change goes through
|
||
simulation-version review (ADR-0089) and is recorded with its playtest evidence.
|
||
|
||
## Context
|
||
M11 turns the M2–M6 fixtures and the M10 session layer into repeated primitive-cell
|
||
lives (ROADMAP 11.1–11.9). The locked requirements come from the founding brief
|
||
(PROJECT_VISION): absolute permadeath, independent wildlife descendants, no offline
|
||
progression, no classes and no linear progression tree, biological costs, and shared
|
||
presentation for players and NPCs. Within those requirements, REPRODUCTION_MODEL,
|
||
GENOME_MODEL and OPEN_QUESTIONS left the first-loop rules open. The engineering
|
||
constitution forbids settling them in code. ADR-0089 already ratified time-based
|
||
biological rates, to be migrated in M11.1.
|
||
|
||
## Decision
|
||
|
||
### 1. Reproduction: budding
|
||
One accepted strategy, budding:
|
||
- The parent survives and keeps its life and control.
|
||
- It pays an explicit investment of matter and energy to build one descendant. The
|
||
investment is the ADR-0032 reproduction investment, priced at the parent's own
|
||
construction rate. Construction overhead goes to heat and waste, as in ADR-0028.
|
||
- The descendant is autonomous wildlife from its first tick. No account owns it and
|
||
nothing labels it.
|
||
|
||
Reproduction is an action under ADR-0092: it has a stable id, a durable outcome, and
|
||
duplicates resolve to the first outcome. The server decides prerequisites, cost,
|
||
placement and the child's genome; the client sends only the intention. An organism
|
||
that cannot fund the investment in every ledger channel is refused, and nothing is
|
||
spent. Abundant energy never pays for missing nitrogen. Placement capacity may refuse
|
||
or delay a bud, but it never grants a free or reduced one.
|
||
|
||
Binary fission, eggs, cohorts and spores stay open.
|
||
|
||
### 2. Mutation: earned during life, funded
|
||
A player's living organism can change its genome during its life:
|
||
- Eligibility is earned by surplus that has been fed and retained. The exact trigger
|
||
is the versioned, playtestable M11.4 trial in
|
||
[ADR-0120](0120-cellular-mutation-eligibility-trial.md), not a final balance rule.
|
||
- The player is offered seeded candidates, using the M3.4 generator (ADR-0033): stable
|
||
context, no free rerolls, and stale offers rejected.
|
||
- A chosen candidate is applied through funded development (ADR-0031/0032). It
|
||
reserves and spends matter and energy for the change. It can fail for lack of
|
||
resources, and then nothing changes.
|
||
- Each applied mutation creates an immutable successor genome with ancestry (ADR-0030,
|
||
ADR-0034). The organism keeps its life id; only its genome revision advances.
|
||
|
||
There is no tree, no class and no guaranteed sequence. Offers are contextual, and every
|
||
candidate trades one cost against another.
|
||
|
||
### 3. Heredity: exact copy; wildlife does not mutate in M11
|
||
- A bud inherits its parent's current genome exactly at the moment of birth, and its
|
||
ancestry records the parent organism and genome revision.
|
||
- Wildlife never mutates on its own in M11. Open question 11 (NPC mutation) stays open
|
||
for M12. Lineages diverge in M11 only through mutations players choose before
|
||
budding.
|
||
|
||
### 4. Logout and disconnect: linger, then freeze
|
||
On a clean logout or a detected disconnect, the organism enters **SuspendPending**:
|
||
- Its biology freezes at once. It cannot move, feed, grow, mutate or reproduce.
|
||
Maintenance, gestation, construction and every other biological timer stop, because
|
||
an offline organism may never progress (locked).
|
||
- It stays in the world for a fixed linger time (**tunable**, starting at 10 s). During
|
||
that time it is perceivable and vulnerable: other organisms can engulf it, and lethal
|
||
outcomes apply.
|
||
- If it survives the linger time, it becomes **Suspended** durably and leaves the
|
||
world. Its inventory stays in global accounting and cannot interact.
|
||
- A reconnect during the linger time resumes it in place.
|
||
- A death decided before the freeze boundary is committed first (REPRODUCTION_MODEL
|
||
state diagram).
|
||
|
||
Lingering costs the organism nothing biologically and gains it nothing; it only stops
|
||
logging out from being an escape. Detection bounds for a disconnect (heartbeat and
|
||
timeout) are the session layer's: they reuse the M10 session timeouts and are stated
|
||
in 11.7's contract.
|
||
|
||
### 5. Resume: the same spot, else the nearest valid spot
|
||
A resume first replays any committed death, and a dead life never resumes. Then:
|
||
- The organism resumes at its frozen position if that position is still valid.
|
||
- Otherwise it resumes at the nearest valid position within a small bounded radius
|
||
(**tunable**, starting at a few cell diameters). It is never placed farther away:
|
||
there is no free travel.
|
||
- If no valid position exists within the radius, the life stays suspended and the
|
||
player is told why.
|
||
|
||
Resume performs no elapsed-time biology. It needs a unique control grant: two sessions
|
||
never resume the same life (ADR-0091).
|
||
|
||
### 6. Feeding and predation: absorb and engulf
|
||
- **Absorption.** A cell takes up environmental nutrients through surface contact, at
|
||
most at the genome's intake budget (ADR-0032), channel by channel. It takes only
|
||
what the environment holds (atomic transfers, ADR-0053).
|
||
- **Engulfment.** A cell can engulf a contacted cell that is sufficiently smaller.
|
||
- The size-ratio threshold is **tunable**, starting at 1.5 by mass.
|
||
- The prey's membrane thickness contests it: a thicker membrane raises the ratio
|
||
required, so defence is a built trait with a cost, not a percentage modifier.
|
||
- The engulfed cell dies. Its whole inventory moves to the predator exactly once,
|
||
bounded by the predator's capacity. Any excess becomes detritus at the contact
|
||
point, so nothing is destroyed or created.
|
||
- **Corpses.** A cell that dies another way leaves a corpse or detritus with its
|
||
inventory (ADR-0027, ADR-0056), and any cell can absorb it.
|
||
|
||
The server resolves contact, perception, motion and intake (11.3). Clients send only
|
||
movement and engulf intentions.
|
||
|
||
### 7. Birth cost: reachable within a session
|
||
GENOME_MODEL found that a birth can be funded within a plausible lifespan only near
|
||
the schema's minimum reproduction fraction. M11 re-derives the development-version
|
||
constants (ADR-0032) so a typical primitive cell can fund its first bud in roughly 5
|
||
to 15 minutes of competent feeding (**tunable**). This is a new development version
|
||
with migration and regression evidence, not an edit of the existing one. The target
|
||
is a playtest hypothesis, and 11.9 adjusts it.
|
||
|
||
### 8. Friend recognition: deferred
|
||
M11 has no friend mechanism. Players and wildlife look identical (locked shared
|
||
presentation). Open question 31 is deferred explicitly, to be revisited after the 11.9
|
||
playtests.
|
||
|
||
### Ratified earlier: time-based biological rates
|
||
Ratified in ADR-0089. M11.1 migrates per-tick genome and checkpoint rate fields to
|
||
time-based units, with deterministic per-tick conversion and explicit units,
|
||
precision, remainder carry, rounding and overflow bounds. Regression evidence must
|
||
show unchanged 50 ms behaviour where the version is unchanged.
|
||
|
||
## Scope
|
||
These decisions bound M11 only. Still open:
|
||
- binary fission, eggs, cohorts and spores;
|
||
- NPC mutation (open question 11) and speciation;
|
||
- recombination and multiple parents;
|
||
- how long suspension is retained, and what happens to the matter of a life suspended
|
||
indefinitely;
|
||
- friend recognition;
|
||
- evolvable behaviour.
|
||
|
||
Wildlife behaviour in M11 is a bounded, deterministic controller over the same
|
||
intentions players use (open question 33). It has no special biological rules.
|
||
|
||
## Alternatives considered
|
||
- **Reproduction.** Binary fission needs a rule for which half the player controls.
|
||
Spores and cohorts add cohort materialisation to the first loop.
|
||
- **Mutation.** Mutating only at reproduction would leave a life with no evolutionary
|
||
choices of its own. Mutating both during life and at reproduction would double the
|
||
rule surface before the first playtest.
|
||
- **Heredity.** Random inherited drift would settle NPC mutation now, without M12's
|
||
ecology evidence.
|
||
- **Logout.** An instant freeze makes logging out an escape from predation. Blocking
|
||
logout while engaged punishes clean exits and still freezes on a disconnect.
|
||
- **Feeding.** Absorption alone delays the predation pillar. Damage over contact adds a
|
||
combat model with more invented constants.
|
||
- **Birth cost.** A rare birth makes descendants nearly unobservable in a playtest.
|
||
Deferring the cost leaves budding effectively unreachable.
|
||
|
||
## Benefits
|
||
- **Little new rule surface.** Budding, the candidate generator, funded development and
|
||
the corpse transfers already exist as fixtures.
|
||
- **Conservation stays checkable.** Every change spends ledger channels, and engulfment
|
||
and corpses move inventories exactly once.
|
||
- **Offline fairness.** The locked rule holds even during the linger time, because
|
||
freezing happens before lingering.
|
||
|
||
## Costs and risks
|
||
- **Lingering can still be gamed.** A player could log out where nothing can reach
|
||
them. 11.7's tests cover this, and the linger time is tunable.
|
||
- **Mutation could look like progression.** Funded in-life mutation risks reading as a
|
||
progression system. Offers must stay contextual tradeoffs, and 11.4's tests check
|
||
that no candidate dominates.
|
||
- **Size could dominate predation.** Engulfment by size ratio may favour size; the
|
||
membrane contest and the cost of size (ADR-0032) are the counterweights, and 11.9
|
||
watches for monopoly.
|
||
- **Tunables could harden into rules.** Each tunable value is versioned and must be
|
||
changed with evidence, never silently.
|
||
|
||
## Consequences
|
||
- Updated: REPRODUCTION_MODEL and OPEN_QUESTIONS rows 7, 10, 14, 31 and 33.
|
||
- 11.2–11.7 implement against this ADR, and each states its own contract.
|
||
- The first implementation increment is the ratified time-based rate migration.
|