Commit Graph
18 Commits
Author SHA1 Message Date
dingo 0798410dae Complete M5.2 compact chunk layout comparison 2026-09-26 13:57:39 -04:00
dingo 69bbf364a9 Define M5.1 local voxel contract and expand M5-M9 gates 2026-09-26 10:06:22 -04:00
dingo c237e4da97 Measure aggregate cost and fidelity across scarcity scenarios 2026-09-26 09:39:53 -04:00
dingoandClaude Opus 5 72ed03f4dd Stop calling the end of watching a death
Three defects from review of 89ad95b.

An organism still alive when an episode stopped was recorded dead. A one-tick
run reported no extinctions while writing terminal records for founders that had
not died, and their watched time was added to the same total as completed
lifetimes. Observation ending is a fact about the run, not about the organism,
and ADR-0034 makes a recorded death terminal, so there was nothing to be
terminal about. Survivors stay living now and are counted apart, with their
watched time as the lower bound it is; lived_ticks and observed_ticks answer
different questions and are never summed together.

A genome without the reproduction capability derives a zero investment, which
the driver refused outright, so seeds 10, 12 and 15 failed the whole run.
Filtering those variants out before sampling would have been the worse repair:
the measured set would then be the fertile subset a guard approved of, and being
unable to reproduce is a result rather than grounds for exclusion. The driver
runs such a strategy and proposes no birth for it. ADR-0029 is amended rather
than quietly contradicted.

A half's thermal shortfall was assigned where it was documented as a total, so
two episodes at 15,000 reported 15,000.

Only configurations the driver already rejected behave differently, so the M2.6
pin 2cd4a15f9fbe9949 was checked rather than assumed and still holds. The
ensemble pin moved because the reported facts changed, not to pass a test; the
reason sits beside it. Each fix was confirmed load-bearing by reverting it alone.

Release 33/33, analysis, ASan 33/33 and clang-format all ran and passed here.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-24 05:58:22 -04:00
dingoandClaude Opus 5 89ad95bf12 Let a genome decide, and report where each axis actually acts
M3.6 joins the M3 pieces: strategy_for derives a StrategyConfig from the
metabolic budget, founders are pre-diverged variants admitted to the lineage
archive, and evo-sim --ensemble runs them over a resource and temperature grid
with every cell's seeds split in half and never merged. ADR-0029's driver is
untouched, so the M2.6 fingerprint still holds. Nothing mutates during an
episode, so NPC mutation stays open.

Three things measurement found, kept rather than smoothed over. The report
aggregates were default-initialized, leaving their scalar counters
indeterminate. The resource level never reaches strategy derivation at all, so
the header now says where each axis acts instead of claiming both shape a
strategy. And the resource axis is not monotonic: a sweep peaks near 8000
femtograms and falls away on both sides, because only one consumer may bite a
target per tick, so scarcity concentrates mass into individuals that reach the
threshold while abundance spreads the same income too thin. The test asserts
that an axis moves the outcome and never a direction; an earlier draft asserted
a direction that held at two cells and the sweep disproves.

Earning a birth at all required pinning the investment fraction to the schema
floor, since intake scales with surface and investment with mass. That is a
finding about the ADR-0031/0032 cost model, now the second priority in STATUS.

Release 33/33, analysis, ASan 33/33 and clang-format all ran and passed here;
no Clang or GCC compile happens on this host, so Linux CI is still the first
result for this increment.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-23 23:03:52 -04:00
dingoandClaude Opus 5 8020347f9b Keep every lineage, and never call absence extinction
M3.5 retains the genomes M3.4 produces and the organisms that carried
them. The named risk is deleting extinct genomes, so the answer is
structural: LineageArchive has no removal operation at all, not a guarded
one and not a private one. Retention is a property of the interface
rather than a rule someone has to remember, which is the only form of
"keep extinct genomes indefinitely" that a later change cannot quietly
undo. Four concepts assert that, with a positive control so the check is
known to be live rather than vacuous.

Admission requires the parent to be present, so ancestry never dangles
and the parent graph is acyclic by construction: a cycle would need one
of its members admitted before itself. A duplicate identity is refused
rather than overwritten, because overwriting is deletion renamed.

Organism identity and genome identity stay separate. A death is terminal
for the organism and touches nothing about the genome it carried;
ancestry, descendants and divergence all resolve afterwards.

The quieter risk is calling something extinct that is merely unobserved.
The archive reports living or absent and has no third answer: terminal
extinction is a policy judgement over a reconciled watermark covering
suspended members, eggs, spores and in-flight transfers, none of which
this type can see, so the word does not appear in the API. Absence is
also reversible, and a test shows a lineage regaining a carrier.

Capacity is bounded and admission fails loudly at the limit, since
refusing to record is not deleting but a silent drop would be.

The roadmap lists an NPC-mutation decision as a dependency, and
OPEN_QUESTIONS leaves it open until M12. It stays open: the archive
records an origin event and never an actor, so whether an NPC may
diverge changes nothing here.

Two clang-tidy findings were fixed rather than suppressed. std::map
members made the archive's move potentially throwing, because MSVC's map
move constructor allocates; sorted vectors replaced them. A wrapper
struct holding a Genome by value had the same problem, since ADR-0030
made Genome copy-only, so genomes are stored directly under their own
id(). The redundant carrier tally went with it: carriers are counted from
the organism records, so no second tally can drift.

Native Release, clang-tidy analysis, ASan and format checks pass 21/21.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-23 16:55:36 -04:00
dingoandClaude Opus 5 5497f4d74c Propose mutations without proposing a best build
M3.4 adds the candidate-generation experiment GENOME_MODEL describes: a
static catalog of four effects over the seven capabilities, drawn from a
recorded context with an explicit stream.

The named risk is a hidden best build or offerings biased toward what
would help, so the defence is structural rather than asserted. Nothing is
ranked: candidates are drawn without replacement by rejection sampling,
making every eligible operator exactly as likely as any other, and the
result is returned sorted by operator ID. No score exists in the record,
position carries no meaning, and nothing about an organism's situation
makes a helpful mutation more likely to be offered.

Eligibility is not a second rulebook. An operator is eligible when the
genome it would produce passes the M3.1 validation unchanged, so
prerequisites, required capabilities, ranges and the acyclic development
graph are enforced once, where they were defined. Shedding a capability
another declares as a prerequisite fails that check with no rule written
here to catch it. Range boundaries are refused rather than clamped, since
a clamp would make two operators agree and offer a no-op as a choice.

The draw reads the parent digest and the context and nothing else, so the
same context always returns the same candidates and a retry cannot
reroll; a new event index is what earns a new draw. select_mutation
regenerates the offer and refuses unless it still matches, so a changed
context is rejected rather than answered with a different mutation.

Cost is not invented: a candidate reports the successor's own maintenance
and mass under ADR-0032, and reserves nothing.

The eligible set is pinned by hand from the schema rather than observed,
and the uniformity check is confirmed to fail against a draw that takes
the first eligible operators. The rejection sampling that removes modulo
bias is correct by construction, not test-verified: at this scale the
bias is one part in 10^18. That is recorded rather than claimed.

No organism mutates and no fixture calls any of this. Candidate count,
mutation cost as a game rule, inherited versus acquired expression, NPC
mutation and recombination all remain open, as GENOME_MODEL requires.

Native Release, clang-tidy analysis, ASan and format checks pass 20/20.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-23 15:27:21 -04:00
dingoandClaude Opus 5 be71d4f441 Give capability a cost without giving it a multiplier
M3.3 turns a phenotype into the budgets a ledger can hold: body
composition, maintenance, movement, the next construction step, the
per-tick intake limit and the reproduction investment. metabolism is the
single place the genome's description of capability becomes accounting
quantities, so the record itself stays free of those types and M3.1's
boundary is crossed once, explicitly.

The named risk is reintroducing RPG modifiers, where a capability
parameter becomes a flat percentage over every budget. Each budget is
therefore derived from a different phenotype quantity: maintenance from
mass plus membrane upkeep, movement from mass and speed, intake from
surface, construction from the mass a growth step adds, investment from
the body a parent already has. One parameter moves some budgets and not
others, by different amounts. Doubling membrane thickness costs more
matter to build, more energy to hold and more to haul, while leaving
uptake, speed, assimilation and structural mass identical, and the mass
and maintenance ratios differ, so no single multiplier could produce it.
The test asserts that non-uniformity rather than describing it.

Composition is declared per compartment, and membrane is the carbon-rich
one, so a thicker membrane changes what a body is made of and not only
how much. A split assigns the first three channels by share and gives the
remainder to the last, so parts always sum back to the whole and a
derived body can be spent through the ledger without a rounding leak.
That is checked at four awkward masses where every share floors.

processes::consumer_for is the first path from a genome into the M2
fixtures. Opening inventory stays the caller's: a genome says what an
organism can do, never what it holds, and a derived config that invented
an opening balance would create matter. A derived consumer is shown
driving a real feeding tick with the ledger conserving exactly, and a
bite one femtogram past the derived cap refused by the M2.3 contract.

Every budget is asserted as a hand-calculated literal. The constants are
chosen so budgets are of comparable magnitude, which is legibility for
tests and not tuning. No existing fixture reads genomes, and no organism
spends any budget: there is still no starvation, death or repair rule.

Native Release, clang-tidy analysis, ASan and format checks pass 19/19;
every existing golden and fingerprint is unchanged. Linux CI has not run.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-23 11:31:01 -04:00
dingoandClaude Opus 5 3f86dada0e Derive a phenotype from the genome without touching it
M3.2 adds derive_phenotype and plan_development as pure integer
functions of a genome, a growth state and an environment. Nothing is
stored: a phenotype is recomputed from retained inputs rather than
cached, so a cached copy can always be checked against a recomputation.

The risk this increment was scoped against is confusing the genome with
runtime damage. Growth and damage live in a separate state that no
function may write back, and no entry point takes a mutable genome.
Damage changes the phenotype and leaves the encoded bytes and content
digest identical, so heredity cannot carry acquired condition.

The genome's development version selects a rule set, which is what makes
that field worth carrying; an unknown version fails closed rather than
deriving under the wrong rules. Version 1 declares every constant. The
shape model is explicit rather than implied: the body is a cube of
uniform declared density, so extent is an exact integer cube root and
surface is six extents squared. Membrane mass follows surface and the
genome's thickness, maintenance follows total mass plus a membrane
upkeep term, and uptake scales with surface, so size trades uptake
against maintenance. A capability the genome does not carry yields zero.

A plan advances growth by at most the rule step, reduced to the largest
step the available material funds exactly, never a partial one and never
a grant. It reserves nothing and pays for nothing: the matching transfer
does not exist until M3.3, and an apply that moved no inventory would
look like a committed operation while conserving nothing.

The fixture genome is an exact 20 micrometre cube, so every derived value
is asserted as a hand-calculated literal, as are the half-built, damaged,
maximum and cube-root-boundary cases. Determinism is checked over 45
growth and temperature combinations. The declared flooring is pinned
rather than hidden: a body whose mass floors to zero advances its counter
for no material, granting no structure because none exists either way.

Omitted physics is a documented model limitation, not a modifier. Nothing
here is balance, and no organism yet develops. Native Release, clang-tidy
analysis, ASan and format checks pass 18/18; every existing golden and
fingerprint is unchanged. Linux CI has not run yet.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-22 21:48:24 -04:00
dingoandClaude Opus 5 8584eb1f6f Add the immutable genome v1 record and its canonical encoding
Open M3 with the record that development and mutation will read, in a new
evo::biology library that nothing else links yet. A genome is validated
once at construction and exposes no mutator, so an invalid one cannot
exist at runtime and callers never revalidate. A draft carries
unvalidated input; create and decode apply identical semantic checks, so
a decoded record is never less validated than a constructed one.

Schema 1 fixes the seven v0.1 capability kinds. Each has a unit, an
inclusive range, whether every genome must carry it, and the capabilities
it mandates, all held as schema facts rather than record data, so no
record can contradict its own units. A capability adds one parameter and
a bitmask of declared developmental edges; the union of declared and
mandated edges must be acyclic, which seven nodes make exact and bounded.

Identity is a nonzero world-scoped genome ID distinct from every runtime
identity, a nonzero origin event, and zero or one parent. Zero parents
identifies a documented founder. The format reserves room for a second so
recombination needs a schema version rather than a layout change.

The encoding follows ADR-0008: explicit little endian, bounded before any
allocation, an exact length match so trailing bytes are rejected, records
strictly ascending by kind so the bytes are canonical and duplicates fall
out of the same comparison, a stored digest recomputed on decode, and a
trailing checksum. The digest covers identity and ancestry, so identical
capabilities from different origins never share one. Neither value is an
authenticity claim. Unknown schema versions fail closed.

Tests cover ancestry rules, every declared range boundary from both
sides, self, mutual and three-node cycles, a cycle closed by a mandated
edge, truncation at every length, trailing and oversized buffers, every
single-byte corruption, and three resealed edits that reach the digest,
version and ordering checks rather than stopping at the checksum. One
branch is recorded as uncovered rather than given a test that cannot
fail: in schema 1 every mandated capability is also required.

No genome affects any simulation. Native Release, clang-tidy analysis,
ASan and format checks pass 17/17; golden d61ad5bf0c928c76 and ecosystem
fingerprint 2cd4a15f9fbe9949 are unchanged. Linux CI has not run yet.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-22 20:53:23 -04:00
dingoandClaude Opus 5 eaf58f2b85 Drive the energy cycle with measured ecosystem episodes
Close M2 with a deterministic driver that owns one community and one
seeded stream, holds no inventory and emits only the terminal, feeding
and reproduction intents the earlier increments already accept. No
simulation contract changed to make the cycle run unattended.

Two limits shaped the design. Identities are never reused and slots are
capped, so one community supports a bounded number of consumer lifetimes
and is not a population. No maintenance cost exists, so a consumer that
is never eaten never dies, and inventing a starvation rule here would
settle a question ENERGY_MODEL leaves open inside a measurement harness.
The harness therefore measures many bounded episodes, taking mortality
from predation and from a configured lifespan delivered as an ADR-0027
host terminal fact. Lifespan bounds an episode; it is not senescence.

The driver must never submit a batch the community rejects, so it probes
a discarded copy each tick to size bites against the state feeding will
actually see, claims each target for one feeder, and excludes a bitten
consumer from acting. A rejected tick is reported as a driver defect
rather than absorbed as ecology.

Across 64 seeds the suite steps tick by tick, requiring exact conserved
totals at every tick and an identical report on replay, and covers both
mortality paths, slot exhaustion, tick-limit termination, lineage
attribution, episode independence and configuration rejection.
evo-sim --ecosystem reports seeded ensembles: 3,000 episodes advanced
1,093,007 ticks, all reaching extinction, with deaths reconciling exactly
against founders plus births. That demonstrates arithmetic and turnover,
not ecological plausibility; strategy constants are chosen inputs.

Native Release, clang-tidy analysis, ASan and format checks pass 11/11;
golden d61ad5bf0c928c76 is unchanged. Linux CI has not run yet.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-22 17:38:34 -04:00
dingo 8fd3591b0e Add atomic conserved energy and elemental inventory ledger 2026-09-19 19:42:29 -04:00
dingo 9b19509673 Add exact periodic neighborhood queries and measured scan baseline 2026-09-19 19:21:45 -04:00
dingo b620edfd6f Add deterministic transactional lifecycle command scheduler 2026-09-19 16:35:46 -04:00
dingo bf7d60b7cd Evaluate row storage against pinned EnTT in an isolated deterministic benchmark 2026-09-19 14:16:45 -04:00
dingo 1c6e6e119a Pin verified dev image and document completed foundation checks 2026-09-19 09:29:10 -04:00
dingo 9e412736d9 Complete foundation delivery and record verified project status 2026-09-19 09:16:26 -04:00
dingoandClaude Opus 5 d5b4bd068e Initial commit: architecture docs and headless engineering proof
Aeternalis Evolution, a custom-engine persistent evolutionary MMO.

This first commit establishes the project skeleton:

- Architecture and design docs: project vision, system overview,
  engineering constitution, roadmap, 19 ADRs, and the simulation,
  world, network, persistence and infrastructure architectures.
- Headless engineering proof: deterministic synthetic particles,
  component lifecycle, explicit PRNG streams, canonical snapshots
  and restored execution.
- Build and tooling: CMake 3.28+ / Ninja presets, C++23 feature
  check, clang-format and clang-tidy config, format check script,
  Windows check script, and a Docker build context.
- Tests: unit tests, benchmarks, a snapshot fuzz target and CLI
  integration wiring.

No gameplay, renderer, network service or running deployment yet.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-14 15:44:41 -04:00