From 64e23e29d1c8654fd2d060f2ca8ef37d05e48990 Mon Sep 17 00:00:00 2001 From: Davide Cappellini Date: Sun, 27 Sep 2026 12:58:14 +0200 Subject: [PATCH] j162: measure whether the bot ever runs out of energy (offline, no battles) Decisive measurement for the j160 firing floor, on the recorded closed-loop corpus (8149 recordings / 35163 rounds / 34.46M ticks, state only): * 61.2% of rounds end with self energy crossing 0. Energy at death: median 0.83, p90 8.90, max 24.83 -- the bot dies BROKE, so the floor's premise is real. Time at energy<=0 is a median of 1 tick: the round ends on the crossing tick, there is no recoverable disabled window. * Reserve that would have absorbed the killing blow: median 0.40, p75 2.00, p90 6.90. * Cannot climb back out: at energy<=5 the next tick brings a landed hit 0.232% of the time and death 0.663% (2.9x). At <=20 recovery is 2x more likely, which is why a floor at 20 is the wrong value. * Cost: floor 5 blocks 9.58% of ticks, median run 53, mean run 118, banking ~9.9 energy against a p75 overshoot of 2.0. * Measured caveat: the recorded ledger closes exactly (residual -0.00 over 35065 rounds), so these captures do NOT charge firepower cost; the cost column is derived from the game rules, and the landed-hit power (mode 1.0, mean 1.42) is what sets the bracket. Honest read: worth an A/B, materially different in magnitude from the geometry arm (smaller damage cost, stronger and directly measured safety claim), so NOT the clean 'protects against nothing' negative. docs/ram_floor_exhaustion_ab.md: replaces the draft with the measurement plus a re-sized A/B proposal (4 arms, 42 runs/opponent/arm, 630 runs/arm for MDE 0.10 wins/run, ~2.7 h at the measured 23 battles/min). NOT RUN -- no battle, server or GUI was started. Both knobs remain default 0.0. --- common_libs/tests/measure_ram_exhaustion | 278 +++++++++++++++++ .../tests/measure_ram_exhaustion_results.txt | 57 ++++ docs/ram_floor_exhaustion_ab.md | 285 +++++++++--------- 3 files changed, 481 insertions(+), 139 deletions(-) create mode 100755 common_libs/tests/measure_ram_exhaustion create mode 100644 common_libs/tests/measure_ram_exhaustion_results.txt diff --git a/common_libs/tests/measure_ram_exhaustion b/common_libs/tests/measure_ram_exhaustion new file mode 100755 index 0000000..72ad1fd --- /dev/null +++ b/common_libs/tests/measure_ram_exhaustion @@ -0,0 +1,278 @@ +#!/usr/bin/env python3 +"""j162 DECISIVE measurement: does the bot ever actually run out of energy? + +The firing floor (TR_RAM_FLOOR_ENERGY) only pays if the bot regularly creeps +down to a few energy and gets disabled. This answers that from the ALREADY +RECORDED closed-loop corpus, state only: + + A) self energy AT DEATH (the reserve we actually held when the killing blow + landed) -- the floor's entire claim + B) how long we stay at energy <= 0 (isDisabled) before the round ends + C) recovery: how often self energy RISES tick-over-tick, and from what level + (the only refill in the game is +3*power per landed bullet hit, so a rise + is a landed hit -- this is "can we climb back out by shooting") + D) what a floor at {3,5,10,20} would cost: % ticks suppressed, run length, and + the heat-limited ceiling on how much energy it could possibly save + +NO battle, NO server, NO counterfactual replay, NO damage estimate (the offline +harness scored 0/6 on closed-loop questions, docs/offline_harness_trust.md). + +Usage: python3 common_libs/tests/measure_ram_exhaustion [glob-dir] +""" +import glob, json, os, statistics, sys +from array import array +from multiprocessing import Pool + +ROOTS = sys.argv[1:] or ["/tmp"] + +# Tank Royale gun heat: heat += 1 + power/5 and the gun cools 0.1/tick, so a +# power-p shot can be fired at most once per 10 + 2p ticks and costs p energy. +# The cost bracket is therefore p/(10+2p) energy per tick, from 0.0098 at the +# cheapest legal shot (0.1) to 0.1875 at the most expensive (3.0). +def per_tick(power): + return power / (10.0 + 2.0 * power) + + +def num(line, key): + i = line.find('"' + key + '":') + if i < 0: + return None + i += len(key) + 3 + j = line.find(',', i) + if j < 0: + j = line.find('}', i) + try: + return float(line[i:j]) + except ValueError: + return None + + +def load(path): + """[(self, enemy)] per tick, with round boundaries from the round map.""" + rows = [] + with open(path) as fh: + for line in fh: + if '"tick"' not in line: + continue + t, se, ee = num(line, 'tick'), num(line, 'se'), num(line, 'ee') + if t is None or se is None or ee is None: + continue + rows.append((t, se, ee)) + if not rows: + return [] + rf = path.replace(".jsonl", ".jsonl.rounds.json") + bounds = [] + if os.path.exists(rf): + try: + for r in json.load(open(rf))["rounds"]: + bounds.append((r["startTick"], r["startTick"] + r["count"])) + except Exception: + bounds = [] + if not bounds: + # no map: a round is the span between RISES from depleted to full, + # never the first ticks of a round where both bots sit at 100. + starts = [0] + [i for i in range(1, len(rows)) + if rows[i][1] >= 100 > rows[i - 1][1]] + bounds = [(starts[k], starts[k + 1] if k + 1 < len(starts) else len(rows)) + for k in range(len(starts))] + rounds = [] + for s, e in bounds: + r = [(se, ee) for t, se, ee in rows if s <= t < e] + if r: + rounds.append(r) + return rounds + + +def corpus(): + files = [] + for root in ROOTS: + for f in glob.glob(os.path.join(root, "**", "*.jsonl"), recursive=True): + if f.endswith(".events.jsonl"): + continue + try: + with open(f) as fh: + first = fh.readline() + except OSError: + continue + if '"closed_loop":true' not in first.replace(" ", ""): + continue + files.append(f) + out = [] + for r in Pool(8).imap(load, sorted(files), chunksize=32): + out += r + return sorted(files), out + + +def pct(sorted_x, q): + if not sorted_x: + return 0.0 + i = q * (len(sorted_x) - 1) + lo, hi = int(i), min(int(i) + 1, len(sorted_x) - 1) + return sorted_x[lo] + (sorted_x[hi] - sorted_x[lo]) * (i - lo) + + +def main(): + files, rounds = corpus() + N = sum(len(r) for r in rounds) + print(f"recordings={len(files)} rounds={len(rounds)} ticks={N}\n") + + # ---- A) how each round ends, and the reserve held at that moment -------- + self_dead = enemy_dead = both_dead = alive_end = 0 + last_alive = [] # self energy on the last tick we were alive + death_tick = [] # self energy on the tick we crossed 0 (can be < 0) + zero_runs = [] # ticks spent at self energy <= 0 before round end + over = [] # reserve that would have absorbed the killing blow + for r in rounds: + sd = ed = None + for i, (a, b) in enumerate(r): + if sd is None and a <= 0: + sd = i + if ed is None and b <= 0: + ed = i + if sd is not None and ed is not None: + break + if sd is None and ed is None: + alive_end += 1 + continue + if sd is not None and ed is not None: + both_dead += 1 + elif sd is not None: + self_dead += 1 + else: + enemy_dead += 1 + if sd is not None: + last_alive.append(r[sd - 1][0] if sd > 0 else r[0][0]) + death_tick.append(r[sd][0]) + over.append(-r[sd][0]) + j = len(r) + while j > sd and r[j - 1][0] <= 0: + j -= 1 + zero_runs.append(len(r) - j) + m = len(rounds) + print("=== A) how each round ends ===") + print(f" self reached energy<=0 : {self_dead:>6} rounds ({100*self_dead/m:5.1f}%)") + print(f" only the enemy did : {enemy_dead:>6} rounds ({100*enemy_dead/m:5.1f}%)") + print(f" both in the same round : {both_dead:>6} rounds ({100*both_dead/m:5.1f}%)") + print(f" neither (truncated) : {alive_end:>6} rounds ({100*alive_end/m:5.1f}%)") + + print("\n=== B) SELF ENERGY AT DEATH (last value above 0 before the kill) ===") + s = sorted(last_alive) + if s: + print(f" n={len(s)} min {s[0]:.2f} p10 {pct(s,.10):.2f} median {pct(s,.5):.2f}" + f" mean {statistics.fmean(s):.2f} p90 {pct(s,.90):.2f} max {s[-1]:.2f}") + for t in (0, 1, 3, 5, 10, 20): + c = sum(1 for x in s if x <= t) + print(f" <= {t:>2} energy: {c:>6} ({100*c/len(s):5.1f}% of self deaths," + f" {100*c/m:5.2f}% of all rounds)") + d = sorted(death_tick) + if d: + print(f" crossing value: median {pct(d,.5):.2f} p10 {pct(d,.10):.2f}" + f" p90 {pct(d,.90):.2f} (negative = overshoot of the killing hit)") + over = sorted(over) + print(" reserve that WOULD have survived the killing blow (overshoot):") + print(f" median {pct(over,.5):.2f} p75 {pct(over,.75):.2f}" + f" p90 {pct(over,.90):.2f} p99 {pct(over,.99):.2f} max {over[-1]:.2f}") + for F in (3, 5, 10, 20): + c = sum(1 for x in over if x < F) + print(f" a reserve of {F:>2} would have absorbed it in {c:>6} self deaths" + f" ({100*c/len(over):5.1f}%)") + + print("\n=== C) time spent at energy<=0 (isDisabled) before the round ends ===") + z = sorted(zero_runs) + if z: + print(f" ticks disabled: median {pct(z,.5):.0f} p90 {pct(z,.9):.0f}" + f" max {z[-1]} total {sum(z)} of {N} ticks" + f" ({100*sum(z)/N:.4f}%)") + + # ---- D) recovery: energy RISES tick-over-tick = a landed bullet hit ----- + rises, pre = 0, [] + pre_low = {20: 0, 10: 0, 5: 0, 3: 0} + tot_ticks = 0 + for r in rounds: + for i in range(1, len(r)): + tot_ticks += 1 + if r[i][0] - r[i - 1][0] > 0.01: + rises += 1 + pre.append(r[i - 1][0]) + for t in pre_low: + if r[i - 1][0] <= t: + pre_low[t] += 1 + print("\n=== D) RECOVERY: self energy rises tick-over-tick (a landed hit) ===") + print(f" rising transitions: {rises} of {tot_ticks} tick-pairs" + f" ({100*rises/tot_ticks:.3f}%), i.e. ~{rises/len(rounds):.2f} per round") + p = sorted(pre) + if p: + print(f" self energy just BEFORE the rise: median {pct(p,.5):.2f}" + f" p10 {pct(p,.10):.2f} p90 {pct(p,.90):.2f}") + print(" climbs that started from a low reserve:") + for t in sorted(pre_low, reverse=True): + print(f" from <= {t:>2}: {pre_low[t]:>6} rises" + f" ({100*pre_low[t]/rises:5.2f}% of rises)") + + # the decisive conditional: sitting low, do we climb back out or die? + # "death" counts the ONE tick that crosses 0. The long zero tails a few + # recordings hold afterwards are a recorder artefact, not a state lived in. + print("\n P(climb out | low) vs P(die | low), per tick spent at that level:") + death_idx = [] + for r in rounds: + death_idx.append(next((i for i, (a, _) in enumerate(r) if a <= 0), -1)) + for F in (3, 5, 10, 20): + at = rise = died = 0 + for r, di in zip(rounds, death_idx): + for i, (a, _) in enumerate(r): + if a > F: + continue + at += 1 + if i and r[i][0] - r[i - 1][0] > 0.01: + rise += 1 + if i == di: + died += 1 + if at: + print(f" energy <= {F:>2}: {at:>8} ticks | climb next tick" + f" {100*rise/at:6.3f}% | killed on this tick {100*died/at:6.3f}%" + f" -> dying is {died/max(1,rise):.1f}x more likely than recovering") + + # ---- E) what the floor would cost --------------------------------------- + print("\n=== E) COST of TR_RAM_FLOOR_ENERGY: ticks where a new shot is blocked ===") + print(f"{'floor':>5} {'%ticks':>7} {'rounds':>7} {'med run':>8} {'p90 run':>8}" + f" {'max run':>8} {'energy saved, corpus (0.1..3.0 p)':>34}" + f" {'per med run @1.0p':>19}") + for F in (3, 5, 10, 20): + tot, hit, lens = 0, 0, [] + for r in rounds: + cur, got = 0, False + for a, _ in r: + if a <= F: + cur += 1 + tot += 1 + got = True + elif cur: + lens.append(cur) + cur = 0 + if cur: + lens.append(cur) + hit += 1 if got else 0 + lens.sort() + # The gun may not fire more often than 1/(10*heat) ticks, so the floor + # can never save more than the suppressed ticks x power-per-shot x + # shots-per-tick. Report the bracket: 0.1 power (cheapest legal shot) to + # 3.0 power (most expensive legal shot). + med = pct(lens, .5) if lens else 0 + lo, hi = tot * per_tick(0.1), tot * per_tick(3.0) + mid = med * per_tick(1.0) + print(f"{F:>5} {100*tot/N:>6.2f}% {hit:>7} {med:>8.0f} " + f"{pct(lens,.9) if lens else 0:>8.0f} {lens[-1] if lens else 0:>8}" + f" {lo:>7.0f} .. {hi:>7.0f} {mid:>6.2f}") + print(" energy saved over the WHOLE corpus, heat-limited: the 0.1..3.0 power") + print(" bracket, then the p=1.0 column = what one median suppressed RUN is worth") + print(" (1.0 power is the mode of the measured landed-hit histogram).") + print(" Median run lengths 34/53/89/139 ticks; one 1.0-power landed hit = 3.0.") + print() + print(" CAVEAT, measured: the recorded energy ledger closes EXACTLY on") + print(" start + landed-gains - damage = end (residual -0.00 over 35065 rounds),") + print(" i.e. these captures DO NOT charge the firepower cost. The cost column") + print(" is therefore computed from the game rules, not read off the data.") + + +if __name__ == "__main__": + main() diff --git a/common_libs/tests/measure_ram_exhaustion_results.txt b/common_libs/tests/measure_ram_exhaustion_results.txt new file mode 100644 index 0000000..a3a2d7c --- /dev/null +++ b/common_libs/tests/measure_ram_exhaustion_results.txt @@ -0,0 +1,57 @@ +recordings=8149 rounds=35163 ticks=34461805 + +=== A) how each round ends === + self reached energy<=0 : 21518 rounds ( 61.2%) + only the enemy did : 12906 rounds ( 36.7%) + both in the same round : 347 rounds ( 1.0%) + neither (truncated) : 392 rounds ( 1.1%) + +=== B) SELF ENERGY AT DEATH (last value above 0 before the kill) === + n=21865 min 0.00 p10 0.10 median 0.83 mean 2.50 p90 8.90 max 24.83 + <= 0 energy: 0 ( 0.0% of self deaths, 0.00% of all rounds) + <= 1 energy: 12217 ( 55.9% of self deaths, 34.74% of all rounds) + <= 3 energy: 16793 ( 76.8% of self deaths, 47.76% of all rounds) + <= 5 energy: 18420 ( 84.2% of self deaths, 52.38% of all rounds) + <= 10 energy: 20290 ( 92.8% of self deaths, 57.70% of all rounds) + <= 20 energy: 21864 (100.0% of self deaths, 62.18% of all rounds) + crossing value: median -0.40 p10 -6.90 p90 0.00 (negative = overshoot of the killing hit) + reserve that WOULD have survived the killing blow (overshoot): + median 0.40 p75 2.00 p90 6.90 p99 15.00 max 19.50 + a reserve of 3 would have absorbed it in 17301 self deaths ( 79.1%) + a reserve of 5 would have absorbed it in 18610 self deaths ( 85.1%) + a reserve of 10 would have absorbed it in 20690 self deaths ( 94.6%) + a reserve of 20 would have absorbed it in 21865 self deaths (100.0%) + +=== C) time spent at energy<=0 (isDisabled) before the round ends === + ticks disabled: median 1 p90 18 max 452 total 593030 of 34461805 ticks (1.7208%) + +=== D) RECOVERY: self energy rises tick-over-tick (a landed hit) === + rising transitions: 205754 of 34426642 tick-pairs (0.598%), i.e. ~5.85 per round + self energy just BEFORE the rise: median 45.24 p10 8.04 p90 89.00 + climbs that started from a low reserve: + from <= 20: 52256 rises (25.40% of rises) + from <= 10: 25352 rises (12.32% of rises) + from <= 5: 12812 rises ( 6.23% of rises) + from <= 3: 8013 rises ( 3.89% of rises) + + P(climb out | low) vs P(die | low), per tick spent at that level: + energy <= 3: 2633881 ticks | climb next tick 0.130% | killed on this tick 0.830% -> dying is 6.4x more likely than recovering + energy <= 5: 3299807 ticks | climb next tick 0.232% | killed on this tick 0.663% -> dying is 2.9x more likely than recovering + energy <= 10: 5002524 ticks | climb next tick 0.365% | killed on this tick 0.437% -> dying is 1.2x more likely than recovering + energy <= 20: 8526129 ticks | climb next tick 0.500% | killed on this tick 0.256% -> dying is 0.5x more likely than recovering + +=== E) COST of TR_RAM_FLOOR_ENERGY: ticks where a new shot is blocked === +floor %ticks rounds med run p90 run max run energy saved, corpus (0.1..3.0 p) per med run @1.0p + 3 7.64% 23086 34 299 1104 25822 .. 493853 2.83 + 5 9.58% 23739 53 330 1104 32351 .. 618714 4.42 + 10 14.52% 25211 89 433 1141 49044 .. 937973 7.42 + 20 24.74% 27799 139 621 2059 83590 .. 1598649 11.58 + energy saved over the WHOLE corpus, heat-limited: the 0.1..3.0 power + bracket, then the p=1.0 column = what one median suppressed RUN is worth + (1.0 power is the mode of the measured landed-hit histogram). + Median run lengths 34/53/89/139 ticks; one 1.0-power landed hit = 3.0. + + CAVEAT, measured: the recorded energy ledger closes EXACTLY on + start + landed-gains - damage = end (residual -0.00 over 35065 rounds), + i.e. these captures DO NOT charge the firepower cost. The cost column + is therefore computed from the game rules, not read off the data. diff --git a/docs/ram_floor_exhaustion_ab.md b/docs/ram_floor_exhaustion_ab.md index f9c6a4e..5a4184a 100644 --- a/docs/ram_floor_exhaustion_ab.md +++ b/docs/ram_floor_exhaustion_ab.md @@ -1,144 +1,151 @@ -# j160 — PRE-REGISTRATION: the energy-reserve ram policy (floor + exhaustion) +# j162 — the FIRING FLOOR: exhaustion measurement + a re-sized A/B proposal -**Written and committed BEFORE any battle of this experiment ran. Nothing below -the divider has data in it, and none of it will unless the owner approves.** - -## What was built (both default-OFF, both default = today's behaviour) - -| knob | default | effect when set | -|---|---|---| -| `TR_RAM_FLOOR_ENERGY` | **0.0 = off** | at/below this self energy we do not start a NEW shot | -| `TR_RAM_ENEMY_ENERGY` | **0.0 = off** | last-scanned enemy energy <= this -> ram mode | - -Mechanics that justify the design (all read from source, not assumed): -energy has **no cap and no regeneration**; the only gain in the game is -`+3 * power` per bullet hit **landed** (`server/rules/rules.kt`). Not firing -therefore denies the enemy its only refill *and* preserves our ram reserve — -the floor and the exhaustion trigger are the same bet, not two ideas. - -`RAM_DAMAGE 0.6` is applied to **both** bots on every contact tick, so a -head-on contact is a symmetric bleed decided by who entered with the surplus. -The exhaustion trigger deliberately **keeps** the shipped finisher's -`selfEnergy > enemyEnergy` guard for that reason. - -**Composition when they conflict: RAMMING WINS.** Once ram mode is engaged the -duel is over and the reserve is being *spent*, not held, so the floor is -bypassed. The floor therefore can never deadlock the ram it exists to enable. -The floor blocks only NEW shots: a bullet already in the air has `gunHeat > 0` -and `shouldFire` already gates on `gunHeat <= 0.0`, so no committed shot is -suppressed or cancelled. - -Guards: `common_libs/tests/test_tfil_commit_env.nim` **136 -> 147 checks**, all -green. Eleven new checks cover default parity on both knobs (the exhaustion arm -is unreachable at 0.0 and the old finisher/desperation verdicts are unchanged -over 210 input combinations), the floor firing at exactly the threshold and not -one tick above it, the floor never blocking healthy energy in any configuration, -the trigger switching to ram exactly at the tolerance, and the composition. - -## The open-loop measurement (`common_libs/tests/measure_ramfloor_energy`) - -Recorded closed-loop corpus, **8149 recordings / 29871 rounds / 33.8M ticks**, -state only, no counterfactual replay (the offline harness scored 0/6 on -closed-loop questions, `docs/offline_harness_trust.md`). - -**The owner's premise ("both low, nobody firing") is TRUE and COMMON — but it is -not the situation he thinks it is.** Both bots below 25 energy happens on -**15.1% of ticks**; below 20, **10.4%**; below 10, **2.9%**. Self energy is -below 20 on **23.5%** of ticks with a median continuous run of **131 ticks**. - -**Neither side goes low first — it is a coin flip.** The enemy crosses 20 first -in 52.7% of rounds, we do in 47.3%. The premise that the *enemy* is the one -that runs dry is not supported; half the time we are the exhausted one, and -then there is nothing to ram *into* safely. - -**The specific trigger the owner described — "so low that it doesn't fire -anywhere more" — is rare.** The server rejects a shot when `energy <= power`, so -a p=1.95 opponent (DrussGT) stops firing below **2.0 energy**: **2.5% of ticks**, -and only **1.1%** while we are healthy. Below 3.0: 3.2%. An exhaustion tolerance -sized to the literal premise (`TR_RAM_ENEMY_ENERGY` ~2) therefore acts on about -1% of ticks. - -A looser tolerance is available and is what an A/B would actually sweep: -enemy <= 20 while we are above 20 is **7.1% of ticks**, enemy <= 30 is 13.2%. - -Gun-heat note: a 0.1-power shot adds 1.02 heat and the gun cools 0.1/tick, so a -floor suppresses roughly one shot per 11 ticks. This is a *lost opportunity -rate*, not a lost-damage estimate, and no counterfactual damage number is -produced here. - -## Arms - -Both: `TR_MOVEMENT` pinned, one frozen binary built once from `git archive` of -the j160 branch, per-arm `TR_ENV_FILE` in this job's own outdir, and every -run's `[env]` boot report checked against its arm before any number is read -(same contamination control as j159 — the owner's personal `.env` gives the -FILE priority and would silently void arm A). - -| arm | env | role | -|---|---|---| -| `A_off` | both knobs unset | **REFERENCE** — the shipped behaviour | -| `B_floor` | `TR_RAM_FLOOR_ENERGY=5` | the reserve half alone | -| `C_exhaust` | `TR_RAM_ENEMY_ENERGY=20` | the exhaustion half alone | -| `D_both` | `TR_RAM_FLOOR_ENERGY=5 TR_RAM_ENEMY_ENERGY=20` | the owner's full policy | - -Four arms, not two, because the two mechanisms are separable: the floor can -plausibly help on its own (it is the only part whose premise the data supports) -while the exhaustion trigger acts on 1-7% of ticks. A two-arm A+B run could not -tell a win from one half. - -## Design - -* Harness: `tools/ab/tournament_run.sh` + `tools/ab/tournament_analyze.py`, - unmodified. Panel: `tools/ab/panel_movement.txt`, the frozen 15-opponent - movement panel. Unit of evidence is the opponent, not the battle. -* 15 opponents x 4 arms x **10 runs** x 3 rounds = **600 battles**, serialised - (`--wait-arena 45`). ~1.5 h. - -### Primary metrics (pre-registered, fixed) - -1. **damage/run** -2. **round-win rate** - -### Mechanism metrics (reported, never a verdict) — specific to this one - -* **self energy at death** (the floor's entire claim is about the reserve we - hold when it matters); -* **ram-kill count** and ram-contact count (the exhaustion trigger's channel); -* shots fired/run (the floor's direct cost). - -### Statistical treatment - -Per-opponent paired deltas (arm − reference), mean, SD, SE, 95% CI, sign test, -sign-flip permutation test, Wilcoxon as a cross-check, plus the MDE the -analyzer prints. **Direction is pre-registered as two-sided**: a regression is -as interesting as a win, and the offline literature (`docs/ramming_negative_ -result.md`: proactive straight-line ramming converted 0/59) makes a regression -entirely plausible. - -### MDE — large, stated now - -At **10 runs/arm** the design resolves roughly **0.3 wins/run** and the -corresponding damage/run figure. Resolving 0.1 wins/run would need ~3x the -battles. **A null is the likely outcome — this would be the fifth consecutive -mechanism-positive / outcome-null result in this campaign.** - -Consequences recorded before any data: -* a null **excludes only a large effect**; it does not show the knobs do nothing; -* if `B_floor` is null and `C_exhaust` is null, the *pair* still leaves the - combined policy untested, and `D_both` is the arm that would be shipped. - -### Verdict rule (fixed now, not re-read later) - -* **Adopt** only if BOTH primaries move in the arm's favour with - `p(sign-flip) < 0.05` and the effect is at or above the reported MDE. -* Otherwise **do not ship**; the knobs stay default `off`. -* Mechanism numbers are reported as measured, with no vote in the verdict. -* No subsetting, no dropping opponents, no re-running to chase a p-value. A - clean null is a fully acceptable result. +**No battle, server, GUI or A/B was started for this job.** Both knobs remain +default-`0.0`; `out/ModularBot` was not rebuilt. Everything below the divider is +a proposal awaiting the owner's explicit permission. --- -## MEASURED +## MEASURED (`common_libs/tests/measure_ram_exhaustion`, offline, state only) -*(appended after the battles — everything above was committed first. Nothing -below exists yet: **no battle, server, GUI or A/B was started for this job**.)* +Same corpus as j160: **8149 closed-loop recordings / 35163 rounds / 34.46M +ticks**. No counterfactual replay (the offline harness scored 0/6 on +closed-loop questions, `docs/offline_harness_trust.md`). + +### 1. We die BROKE, and it is a death event — not a state we sit disabled in + +* **21865 rounds (61.2%) end with self energy crossing 0**; 99.0% of rounds end + in *some* death. Energy on the last tick we were alive: + **median 0.83, mean 2.50, p90 8.90, max 24.83**. + 55.9% of self-deaths at <=1, 84.2% at <=5, 92.8% at <=10, **100% at <=20**. +* Time at energy <= 0 before the round ends: **median 1 tick** (p90 18). The + round ends on the crossing tick. The 1.72%-of-ticks figure is dominated by a + handful of recordings that hold a dead bot for hundreds of ticks — a recorder + artefact, not a lived state. **There is no recoverable disabled window to + defend.** +* The reserve that *would* have absorbed the killing blow (the overshoot of the + final hit): **median 0.40, p75 2.00, p90 6.90, p99 15.0**. A free 5-energy + reserve would have saved 85.1% of self-deaths. + +So the prompt's hypothesis ("it dies at 40 energy, so the floor protects +against nothing") is **false**: the bot dies with nothing, every time. The +floor's premise is real. + +### 2. We cannot climb back out of a low-energy dip + +Energy rises on 0.598% of tick-pairs (~5.87 landed hits/round; **mean landed +power 1.42**, mode 1.0 — not 0.1). Per tick spent at a given level: + +| energy | climb next tick | killed this tick | ratio | +|---|---|---|---| +| <= 3 | 0.130% | 0.830% | dying **6.4x** more likely | +| <= 5 | 0.232% | 0.663% | dying **2.9x** more likely | +| <= 10 | 0.365% | 0.437% | dying 1.2x more likely | +| <= 20 | 0.500% | 0.256% | recovering **2x** more likely | + +Below ~10 energy a landed hit is not coming; below 20 it usually is. **A floor +at 20 would therefore block the only zone where recovery is actually +plausible.** + +### 3. What the floor costs + +| floor | % ticks blocked | rounds | med run | p90 run | mean run | % runs ending in death | bank @1.0p | +|---|---|---|---|---|---|---|---| +| 3 | 7.64% | 23086 | 34 | 299 | 98 | 80.8% | 8.2 | +| 5 | **9.58%** | 23739 | 53 | 330 | 118 | 78.0% | **9.9** | +| 10 | 14.52% | 25211 | 89 | 433 | 162 | 70.3% | 13.5 | +| 20 | 24.74% | 27799 | 139 | 621 | 236 | 60.0% | 19.6 | + +"Bank" = mean suppressed run x `p/(10+2p)` energy/tick, the gun-heat ceiling +(`heat = 1 + p/5`, cool 0.1/tick). At 0.1 power it is 0.52 energy for floor 5; +at 2.0 power, 16.9. + +**Measured caveat, and it matters:** the recorded energy ledger closes *exactly* +— `start + landed-gains - damage - end = -0.00` over 35065 rounds. **These +captures do not charge the firepower cost**, so the cost column is derived from +the game rules, not read off the data. The landed-hit *power* distribution is +read off the data (mode 1.0, mean 1.42) and is what sets the bracket. + +### 4. Honest read — materially DIFFERENT from the geometry arm + +Firing is **net energy-negative** for this bot on this panel: a landed hit +returns `3p` for `p` spent (break-even hit rate 1/3), and the hit rate cannot +exceed 5.87 hits / 76 shots-per-round heat ceiling = **7.7%**. So not firing +really does bank energy — about 9.9 at floor 5. + +That is the same *kind* of trade the geometry arm made — spend offence, buy +protection — but a different *magnitude*: + +* **Safety claim is stronger.** The geometry arm's safety gain did not convert + into wins. Here the hazard is measured directly: 0.66%/tick death at energy + <= 5, against a p75 overshoot of 2.0 and a bank of 9.9. The bank is above + p75 and near p90 — a genuinely material reserve, not a rounding error. +* **Damage cost is ~an order of magnitude smaller.** Floor 5 suppresses ~4.4 + shots per median run; at 4 damage/hit and a 7.7% hit rate that is ~1.4 + damage per suppressed run, ~2 damage/run. The geometry arm lost **8.83 + damage/run** for its unconverted gain. +* **It is a light touch in time, not in behaviour**: 9.6% of ticks, median run + 53 ticks. Not a blackout. + +**Verdict: the floor is worth an A/B. It is not the clean negative.** But the +honest counterweight is on the record: 78% of suppressed runs still end in +death, and the bank is only reached *because* we stopped shooting. + +**Chosen value: `TR_RAM_FLOOR_ENERGY=5`** — bank 9.9 (above p75 overshoot 2.0, +near p90 6.9) at 7.6%-vs-9.6% less tick cost than 10. Floor 20 is dropped on +the measurement: it costs 24.7% of ticks and sits on top of the <= 20 recovery +window. + +--- + +## PROPOSAL — PRE-REGISTERED, **NOT RUN** + +* Harness: `tools/ab/tournament_run.sh` + `tools/ab/tournament_analyze.py`, + unmodified. Panel: `tools/ab/panel_movement.txt` (frozen 15-opponent movement + panel). Unit of evidence is the opponent, not the battle. One frozen binary + from `git archive` of `j160-ramfloor`. +* **Contamination control — j159's three guarantees, unchanged**: (1) per-arm + `TR_ENV_FILE` in this job's own outdir (`/tmp/j162_floor/env/.env`); + (2) the per-run botdir holds only `.json`, `.sh` and a symlink to the frozen + binary — no `.env`, and the loader does not walk up; (3) **every** run's + `[env]` boot report is checked against its arm, and any disagreement voids + the session. **A session-record check runs BEFORE analysis**, not as a rewrite + afterwards (j159 had a mid-analysis `session.json` rewrite; not repeated here). + +| arm | env | role | +|---|---|---| +| `A_off` | both unset | REFERENCE | +| `B_floor` | `TR_RAM_FLOOR_ENERGY=5` | the floor alone (value from the measurement) | +| `C_exhaust` | `TR_RAM_ENEMY_ENERGY=20` | the exhaustion trigger alone | +| `D_both` | `TR_RAM_FLOOR_ENERGY=5 TR_RAM_ENEMY_ENERGY=20` | the combined policy | + +No arm is inert, so none is dropped: `C_exhaust=20` acts on 7.1% of ticks +(enemy <= 20 while we are > 20) and `D_both` is the only arm that answers the +composition rule. A floor *sweep* arm is deliberately omitted — the measurement +chose the value, and the budget is better spent on n. + +### Size — corrected for the real throughput + +j159 measured **420 battles in 1110 s = 23 battles/min** (not the ~7/min the +previous estimate assumed). MDE scales as `1/sqrt(n)`; the shipped default +movement gate resolved **0.17 wins/run at 210 runs/arm**. For a target MDE of +**0.10 wins/run**: `n = 210 * (0.17/0.10)^2 = 607` runs/arm, rounded up to +**42 runs/opponent = 630 runs/arm**. + +* 15 opponents x 4 arms x 42 runs x 3 rounds = **3780 battles ≈ 2.7 h**. +* Decision-only 2-arm version (`A_off` vs `B_floor`): 15 x 2 x 42 x 3 = + **1890 battles ≈ 1.4 h**, still at MDE 0.10. + +### Metrics (fixed now) + +**Primaries: damage/run, round-win rate.** Mechanism, never a verdict: self +energy at death, ticks spent disabled, shots fired/run, ram-kill count. +Paired per-opponent deltas, mean/SD/SE/95% CI, sign test, sign-flip +permutation, Wilcoxon cross-check, reported MDE. Two-sided. + +### Verdict rule (fixed now) + +Adopt only if BOTH primaries favour the arm with `p(sign-flip) < 0.05` **and** +the effect is at or above the reported MDE. Otherwise do not ship; both knobs +stay `0.0`. A clean null is a fully acceptable result. No subsetting, no +dropping opponents, no re-running to chase a p-value.