Two hooks, Mars::RNG::Seed 0x0049fdf0 and StrategyApp::RunAI 0x008706f0, and a
config that turns everything else off. Two launches from the same save, load
only -- the AI clients are constructed on load, so no End Turn is needed.
Result: every AI client's generator seed differs between processes (net 32,
496 and 512 all move), while the record structure is byte-for-byte the same
shape and one Seed call with seed=0 produces an identical state in both runs.
So the turn1-state -> turn2 nondeterminism is a SEED effect, not the ordering
effect that was predicted, and lane AI1's 'every draw from the static generator
returns 0' is falsified by measurement.
The prediction said the opposite and is left in docs/L1-predictions.md with its
outcome underneath.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ARBgSooAfokKUy6wKUKEyZ
New shim module src/shim/hooks/ai_orders.{h,cpp}: one register-transparent entry
stub on StrategySim::ApplyTurnCommandBatch dumps every submitted TurnCommands block
(six gates, 27 list lengths, element bytes) at the point where all of them are
complete in memory; sixteen entry probes, with RunTaskList's stub recording the
pass so every later hit is pass-attributed.
Two workloads on VM145, one End Turn each. The rule-19 control passed with all
seventeen detours installed: both autosaves byte-identical to the published oracle.
What the AI actually emits, and three things no reading had produced:
- a list-23 element on EVERY turn, the first element ever observed in the free
half of the cost table -- and both turns still cost the measured 12;
- the ids in AI commands are client-allocated and travel in the command (design
18, fleet 34; neither exists in the input save);
- pass 0 emits nothing, measured from element counts rather than inferred.
tests/game_ai/test_live_blocks.cpp rebuilds both captured blocks through the public
OrderClient API and asserts the list profile, element values, gate counts and
ModCount total: 44 checks. Kept separate from test_orders.cpp, which stays the
record of what static reading predicted.
Gates: clean_room_check OK, host ctest 55/55, CT111 shim cross-build exit 0.
`ComputeBudget`'s savings-interest term is now compared against the running game at
a treasury the corpus actually contains. Three runs on VM146 from turn1-state.sav:
A (widened floats, as shipped) 3,895 calls, 0 diverged, 0 undeclared writes
B (exact decimals, the control) 2,718 calls, 1,359 diverged
The game fills savingsInterest with 499 at a treasury of 50,000, and with 380 at
38,100 -- the exact decimals pay 500 and 381. Every divergence in B lands on a
treasury that is a multiple of 100 and no other state diverges at all, which is
exactly the arithmetic. G3's rule-23 reading is now measured, not inferred, and the
one-money error is shown to propagate into `available` and `researchMoney` too.
The control also settles why the earlier 4,437-call green run was green: slot 5 IS
diffed and the harness CAN see it, so that run simply presented no boundary state.
Coverage is therefore reported as distinct states, not calls: 5 distinct treasuries,
2 of them on the boundary.
Two further rule-23 constants found in the same routine by an operand-width sweep,
corrected, and honestly marked UNVERIFIED because no reference turn can see them:
- the research-yield factor is a widened 0.85f while its two neighbours in the
same product are exact doubles. Boundary: research money a multiple of 40,000;
the run presented 9 distinct values and none is.
- the three research modifiers are summed in single precision, not double.
Boundary: two of the three non-zero; the corpus has shrm = TRM = 0.
Both are pinned by boundary cases in test_economy.cpp that fail with the decimals.
Also verified live, in the same run:
- T31's difficulty-column recovery. The live ServerPlayer+0xf9 / NPC flags on all
eight players are exactly what lane PL's save-only inversion claims, including
the awkward system-owning player that is still ambiguous because it is an NPC.
- BANKRUPTCY_PROTECTION_LIMIT_FACTOR reads 3.29999995 = (float)3.3. Its file image
is zero because the loader fills it at run time, so lane PL-3 had to assume the
value; it is now measured and the assumption was right.
Falsified, and recorded as such: the difficulty-mods record does NOT sit inline at
ServerPlayer+0x36c -- that field is a heap pointer on all eight players. The row IS
reachable from a ServerPlayer (which corrects the hook's standing coverage note),
but the fitted {3.0,1.5}/{1.0,1.0} pair remains unverified. The hook logs the
pointer and does not follow it.
The `verified` column stays 0, deliberately. Every phase this compare touches is
Partial for reasons upstream of it, and promoting one because part of it was checked
is the drift app_test_catalog exists to catch. What moved is models; see
docs/L5-live-verification.md for each one with its coverage.
Gates run separately: clean-room OK, host ctest 54/54, CT111 shim cross-build exit 0.
Lane SV recovered the script-object subsystem statically and predicted that
SVSOSwarmQueen::RegisterHives takes one RNG_NextInt per new hive inside
StrategyServer::BeginProcessTurn -- which runs inside lane Z's autosave bracket
and outside every one of its subtotals, so a draw there had never been
attributed by anything.
Five nested trace hooks, each declaring the strategic generator as a region and
each carrying a model evaluated at entry so the record can disagree with it:
StrategyServer::BeginProcessTurn the unhooked interval, plus a region
over Frame so the increment is a fact
SVSOSwarmQueen::OnTurnBegin evt 0x13, vtable slot +0x60
SVSOSwarmQueen::RegisterHives predict_new_hives from the original's
own two predicates; reads the LO/HI
config pointers live
SVSOSwarmQueen::TickHives the NextQ slip and its three gates
SVSOSlaversRefuel::UpdateDifficultyTier tail phase 20; a 4-byte region over
CDiff so a store and its ABSENCE are
distinguishable (method rule 20)
Also watch.mode=snlv: one arming line moves slot 1 from the NVO map's _Mysize to
the target system's SnLv, and the arming sweep now prints SnLv with its decoded
per-player 2-bit levels for every system, which costs no debug register.
Measured on VM140: hive creation costs 2 words in BeginProcessTurn and the
residual outside the two turn drivers is 2, not 0; the next turn it is 0 again.
LO=20 HI=30. Spica's SnLv reads 0x200 with AFlags 0. The oracle reproduced
byte for byte with every one of these detours live.
Predictions and outcomes: docs/L1-predictions.md.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ARBgSooAfokKUy6wKUKEyZ
Second mode in lane W2's watchpoint module. No second hook: only different arithmetic on the S the
ApplyAllTurnCommands detour already holds.
- picks the target system by predicate at arm time (AFlags == 0, NVO non-empty) and logs all 28
systems, so the choice is auditable rather than a hard-coded pointer;
- probes both candidate ServerSystem bases and logs how many systems validate under each, which
settled a documentation dispute (+0x274: 9, +0x26c: 0) by measurement;
- prints the trade-route and spy-program vector triples, which is the workload-confirmation
instrument two earlier lanes lacked.
Rule 19 control passed: the armed run reproduced the determinism oracle byte for byte.
Reads the whole colony output chain off the instruction stream (every range
disassembled to the next function start) and compares two of its functions
against the running game.
The population -> output law is linear and is carried by the executable:
output points per head are typeOutputModifier x 1.8 / 500000, and the
three-row population-type table is built in code rather than loaded, so the
imperial (1.0) and civilian (0.33f) modifiers are facts about the binary.
A system's total output is a SUM of three terms, not one multiplicative
chain. The station bonus scales only the imperial term and morale only the
civilian one, so OutputModifiers no longer carries either; they belong to
GroupOutputInputs. The function previously described as the base-output term
is the over-harvest RESOURCE demand, and it is corrected in place.
Live on VM140, both hooks in compare mode over two species and two workloads:
GroupOutput 13,105 calls / 0 divergences; ComputeTotalOutput 11,252 calls /
1 divergence of one ulp, in a value its caller rounds to an integer. Both
functions declare a whole-object Guard: 0 undeclared writes in 24,357 calls,
which is what makes the side-effect-free claim a measurement.
sim::Narrow forces the double rounding a 32-bit x87 build otherwise skips;
without it every civilian row came out one ulp low.
Also fixes ComputeBankruptcyLimits' elimination divisor, which was the
decimal -0.15 rather than the image's widened float -0.15000000596046448.
The two disagree for every maximum income divisible by 3 and for essentially
every empire above ~3,000,000.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ARBgSooAfokKUy6wKUKEyZ
The boundary ledger says which phase spends a turn's words; this says which
call site. Each entry point is detoured with its verified prototype and records
__builtin_return_address(0) plus the word cost from left before/after -- two
4-byte reads, no record per draw, since NextFloat alone has 109 call sites.
Each draw is tagged with WHICH generator it came from. That is not a detail:
the first build counted every Mars::RNG instance in the process and reported 44
words against a bracket of 18. The StrategyClient's generator at +0x134 draws 8
times a turn and must not be in the strategic total.
Also hooks EncounterDetect::AssignContacts, lane I's one inlined-draw site in
ProcessTurn's closure -- invisible to both a call-graph sweep and to the entry-
point detours, so only a boundary hook can see it. It did not fire on this
workload, which is consistent with the site sums reconciling exactly.
Eight End Turns across two saves. A turn costs 18-22 generator words, all of
it inside StrategyServer::ProcessTurn; the tail costs 0; the residual outside
the two drivers is exactly 0 on every complete bracket. The generator does not
move between turns at all, so the interval a standalone has to reproduce is
closed at both ends.
Checked against the save files, not just against itself: the turn-6 autosave
pair gives 18 words read from the two Sim.RNG blobs, with twists == 0 -- so
that number never passes through a twist implementation and the agreement is
about the game rather than about two copies of one algorithm.
P6 was wrong. S+0x8 advances 12-14 times per turn, not twice; both drivers are
hooked so the other increments come from somewhere unidentified. Lane K's 'at
least twice' was right and its conclusion is strengthened.
Node-line decay still has not fired, and the hook now says how far away it is
rather than that it did not happen: 51 of 53 lines are permanent, the mortal
ones are dug ~1/turn by the Zuul, and each is ~40 turns from expiry.
The turn's RNG cost has never been measured end to end. combat-done-tail.md
found two draw sites in OnAllCombatDone_Tail that nothing models and that both
run before the autosave, so a reimplementation that reproduces both ProcessTurn
functions exactly still diverges the first turn a node line expires.
RngLedger recovers an ABSOLUTE WORD POSITION from (mt[624], left) alone, by
indexing the forward-only chain of blocks the twist generates. Word deltas
between any two observations are then exact -- across twists, across NextInt
rejection loops, and across draws nobody hooked. That last point is not
theoretical: the image has four draw entry points, one of which (NextUInt
0x004f7670) appears in no previous lane's primitive set, plus inlined draws in
twelve functions. A primitive-counting hook would have undercounted silently.
Six nested trace hooks bracket one End Turn between the two autosaves and
attribute the words: the two turn drivers, the two tail phases that can draw,
and ProcessNodeSpaceTravel because it runs twice a turn. NodeLineDecay carries
a real model -- one word per expired node line under NodePath::RemainingLife --
so compare mode checks the count rather than reporting it.
Corrections from the instruction stream, both load-bearing:
* StrategyHost::Autosave is ret 8, not ret 4, and returns the std::string* in
EAX. A void-returning hook would have dropped it at both call sites.
* node-line decay's 0x20000-fleet skip runs AFTER the Chance(0.5f) call, not
before, so it cannot change the draw count -- combat-done-tail.md reads as
if it gated the roll.
fpu.sample_turn releases StrategyServer::ProcessTurn, which the fpu sampler and
this ledger both want and MinHook grants to one of them. Default on: no
existing run changes behaviour.
Host ctest 37/37; shim cross-built on CT111; clean-room check OK.
Every Coverage note in this repo said RollResearchEvent draws "exactly one
NextFloat". That is the cost of REACHING its branch. When the roll beats the
odds, the plague path draws a SECOND word (NextInt) to pick an owned system and
posts EVENT_PLAGUE_OUTBREAK, and the rebellion path allocates an AIRebellion at
ServerPlayer+0x3b8 and cancels the current research. Nothing has caught this
because the branch has not fired in three sessions.
Corrected in research.h (two Coverage entries plus the scope comment, and the
branch entry raised to Risk::High), research.cpp, tech_effects.h and
tech_effects.cpp. Ours still models the first word only; the branch stays
declared unmodelled, now accurately.
Header regenerated from sots-re bb0f990 (750 entries) - lane K's map of the
combat-done tail: the autosave and save-file writer, the bankruptcy limits, the
turn-results accumulator and outbox, and the encounter-block callees.
Descriptor + pure adapter + host tests for the per-player turn driver. Not
deployed; the WIN32 half is unbuilt here (no cross-compiler on this host).
The declared boundary is narrower than the function on purpose. Phases 2, 3
and 6 -- the savings apply, the aid records and the research refund -- are pure
functions of ComputeBudget's 22 slots and ProcessResearch's overBudget, and
both live in the original's own stack frame. Reaching them would mean calling
ComputeBudget ourselves (it repairs ships in orbit, audit #6), reading the
nested B1/B3 hooks (audit #5, the self-fulfilling compare), or inferring them
from the Sav delta. So they are guarded, not checked, and the three formulas
are written and unit-tested but not wired into the verdict.
Declared: the phase-7 clear, the RebAI decay, the descending timed-bonus
sweep, plus roll_flags and rng as observations ours never writes. Guards over
the whole ServerPlayer and the TechTree header.
docs/T-turn-driver.md states, before any run: which regions must not diverge,
which checks are weak by construction on the reference save, what falsifies
the ResearchRollPending reading, and the save that would finally fire the
branch nobody has seen.
host ctest 36/36 (was 35/35); clean_room_check OK.
Pure module game/sim/techgraph: PrereqsMet (AND of ORs, empty group fails),
SetResearched (stamps, child-cost sweep, sticky turnAvailable, zero-cost
recursion) and the newly-available collector, all read out of 0x00581e10,
0x0057d8e0 and 0x00587cc3.
Wired into the B3 hook in compare mode only, over the scratch node copies:
four more node write-backs, the EVENT_TECHS_UNLOCKED list (still an input,
still nullptr when it could not be computed), the de-duplicating observed-tech
append and the one RNG word RollResearchEvent draws.
docs/U-unlock.md section 4 is the prediction, written before the build was staged.
New fpu_force module (4 register-transparent asm stubs, same pattern as the M0
Initialize hook, so the [unverified] prototypes of the turn-gate functions are
never relied on) plus two shim.cfg keys:
fpu.force=<cw>|off fldcw at StrategyClient::EndTurn and
StrategyServer::BeginProcessTurn, and nowhere else
fpu.sample_ticks=on|off per-frame sampler, logs only when the word CHANGES
Forcing is deliberately one write per turn: re-forcing inside the pipeline would
guarantee the value is present without proving it ever held, which is the exact
false negative this experiment has to avoid. Verification is kept separate --
StrategyServer::ProcessTurn is hooked sample-only, and its reading plus the
existing per-hook fpu_cw fields (38 samples per turn across phases 4, 6 and 8)
are what establish that the setting lasted the whole turn.
Six shim.cfg variants, identical apart from the fpu.force line, and docs.
Used to settle STATE_CHECKSUM.md 3.5: 53-bit and 64-bit x87 produce byte-
identical turn results, so an x64/SSE port has no double-rounding budget to
preserve; 24-bit and round-up each move exactly one thing. Findings and evidence
live in the notes repo (findings/subsystems/fpu-precision-sensitivity.md).
clean_room_check.sh OK; host ctest 32/32.
The header's ObservedTech notes carried a std::string size of 0x18. That was a
mis-attribution: the element's +0x24 word is the name string's trailing (empty)
allocator, not a data field. sizeof(std::string) is 0x1c binary-wide, confirmed
against every serializer in the original.
ObservedTech is now fully mapped, so the two research-hook comments that said
sizeof and the append site were unpinned are corrected. The region stays
declared-not-modelled -- appending is a behavioural change this hook does not
make -- but its byte delta now has an exact expected value (44 per completion),
and the note records that the original's append de-duplicates by tech name, so a
naive push_back would diverge on re-observation.
clean_room_check: OK. host ctest: 33/33.
TechTree::ProcessResearch's events region now compares a modelled value instead of
reporting a known defect. ours posts the pass's events into its own
sots::events::EventStorage, seeded from a scan of the owner's list taken BEFORE the
original runs, and writes only the counts into the region's scratch copy. The game's
PostEvent is never called and no live byte moves; replace mode still posts nothing,
because a bumped EvNxID with no record behind it would corrupt the oracle's save.
- game/events: PostResearchPassEvents (the decision half, pure) + KeylessEventText
- shim/hooks/event_inputs (new lib shim_events, host-tested): the live<->model adapter,
carrying game pointers as explicit uint32 so a 64-bit host build cannot alias them
- research hook: the wiring, a new observed_techs region for ServerPlayer+0x274, and
turn / events_next_id_in / events_dedup_risk in the args so the count model's own
assumption is measured rather than assumed
- EVENT_TECHS_UNLOCKED is NOT posted: its trigger is pinned but needs SetResearched's
unlock cascade, which ours does not run. The driver takes the unlock list as an input
and is handed 'no list', so a missing input cannot look like a modelled negative.
Predicted residual: next_id short by exactly 1 on a completion call.
ctest 33/33 (shim_events_unit is new), clean_room_check OK. The shim TU is
syntax-checked only: no MinGW cross toolchain on this box.
See docs/P-events-wiring.md for the exact prediction for the next VM run.
The behavioural compare found 8 of 45 StrategyServer::MoveFleet calls diverging by
one ULP on a position component. Read off the instruction stream, the cause is that
the engine's vector normalise narrows to float32 four separate times and we kept
everything in double:
delta.c = f32(dest.c - pos.c) stored back to a float32 slot before normalising
sumsq = f32(x*x + y*y + z*z) products/adds in 53-bit regs, only the SUM stored
len = f32(sqrt(sumsq))
inv = f32(1.0 / len) a reciprocal, MULTIPLIED through, not three divides
dir.c = f32(delta.c * inv)
and the same call returns the leg distance, so it is never recomputed in a wider
precision either. The position tail was already right, which is why the error was a
constant absolute ~1.2e-7 (half an ULP of the inputs) rather than a formula error.
Adds NormalizeVec3 / StraightLeg / StraightLegDistance / AdvanceAlongUnitDirection
and rebuilds AdvanceAlongDirection on them; the movement hook now takes both the
direction and the distance from one StraightLeg call, as the original does. The
arrival test is an exact float compare, so the distance has to be that same float32.
Tests pin float32 BIT PATTERNS, not tolerances: one case per narrowing plus four
independent legs component by component. A CHECK_NEAR would pass against the old
arithmetic.
sim::Distance is left in double on purpose and flagged at its declaration: it now
serves only the node-line/stutter geometry, which very likely needs the same
treatment but has zero behavioural coverage to correct it against.
Live, same VM/save/workload, run twice by this lane:
control recap-7584bad-20260908T0615Z 45 calls, 45 compared, 8 diverged, exit 1
fixed mf-45bdf7d-dirty-20260908T0721Z 45 calls, 45 compared, 0 diverged, exit 0
with identical arguments, identical pos.before and identical ORIGINAL pos.after on
all 45 calls. The control reproduced the eight divergent call_ids exactly.
Coverage unchanged and still thin: all 15 moving calls are the same straight-run
waypoint type; types 2-5 were attempted and could not be reached (the only player
that would travel a node line has no ships on this save). See docs/M-movefleet.md.
ctest 32/32; tools/clean_room_check.sh OK.
Five hooks recaptured; no source change needed. ProcessResearch's over-budget
event now diverges as one line instead of a 609 KB save diff (15 calls, 3
diverged, RNG 15/15). ComputeBudget holds at 0 over 4284 compares but only 20
distinct states and 13 always-zero slots. MoveFleet diverges on 8 of 45 calls by
1 ULP of position - B4's clean verdict was a one-sample verdict. First guarded
captures for OnTechResearched (RollResearchEvent draw finally fired and matched),
the colony turn and MoveFleet. Guards mapped SetResearched live and found an
undeclared observed-tech vector append.
Recovers the game's event-posting API so the engine can post events and the
compare harness can see them. Until now the owner's event list was invisible to
every layer: B3's replace-mode oracle failed by exactly one item across 40,300
(an unposted EVENT_RESEARCH_OVERBUDGET) while its compare read clean, and B2's
clean compare bounds the economy fields only.
New pure module src/game/events:
* EventStorage / TurnEvents / PlayerEvent -- the list is bucketed by TURN, not
flat, which the save-editor struct notes had wrong.
* EventStorage::Post reproducing the original's rules, including the four that
change save bytes: the FLT_MAX (not infinity) default position; action 0 with
no subject and no position storing as 2; per-bucket dedup that compares
message/image/location/position/action but NOT summary; and EvNxID starting
at 0 and being promoted to 1 on the first post.
* PruneOldTurns reproduced with its off-by-one: of a leading run of buckets
older than turn-50 it erases n-1, so one stale bucket always survives. The
survivor is serialized, so correcting it would diverge.
* research_events: the five events the research path raises, their EvImg
identifiers and string-table keys, and the 0.8 completion split evaluated
against (double)0.8f rather than the decimal 0.8.
Localized text is deliberately absent: only the EVENTSUM_/EVENTMSG_ keys are
here and the text resolves through a caller-supplied lookup, as the game does.
The four event offsets the B3 hook carried as local literals now come from the
generated header; they are read off instructions rather than inferred from the
save schema.
tests/game_events: 112 checks including a replay of the event list
turn3-state.sav actually holds. ctest 31/31 -> 32/32.
docs/E-events.md carries the proposed region and Coverage wording for the next
B3 recapture.
