pog-mcp 0.9.6 → 0.9.8

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pog-mcp",
-  "version": "0.9.6",
+  "version": "0.9.8",
   "type": "module",
   "description": "MCP server that lets an AI agent play Proof of Goal — wallet, sign-in, squad building, and matches as typed tools.",
   "license": "MIT",

package/skill/reference/measurements.md

@@ -2544,17 +2544,52 @@ that was an artefact of an unattainable +4 buff and is withdrawn.
   the short side drops by the same Δ. A real fatigued squad is uneven.
 
 
-## What growth-neutral load would a player accrue? (`engagement-load.mts` + `c2-team-load.ts`)
-
-**This section does NOT close C-2.** An earlier version said it did; that was wrong. `τ` was
-never measured — the probe's 24h default was simply used — and the neutral point was left
-unchosen between its two branches. An F-1 implementer following that decision would have made
-a probe default the production recovery rate. And `τ` cannot be measured from this data: F-1
-does not exist, so production has no fatigue and there is no observed recovery curve to fit.
-What measurement CAN do is price the choices, which is what the τ sweep does. A third product
-choice sits beside τ and the neutral point: F-1 has not selected predicate C versus D. The
-published values are conditional on C; switching to D moves τ=24h median/p95 `L` by
-−5.3%/−10.4%. That measures the price of the choice, not which workload meaning the product wants.
+## What growth-neutral load would a player accrue? — measurement and product decision
+
+The first #783/#784 measurement did **not** close C-2. `τ` was not measurable — the probe's 24h
+default had merely been used — and the neutral point was left between two branches. Production
+had no fatigue curve to fit before F-1; measurement could price the choices but could not make the
+product choice. Predicate C versus D was also open at publication and F-1 subsequently persisted
+**D** on 2026-09-04. The C-conditioned tables and correction trail below remain historical evidence;
+they are not relabelled as D.
+
+**Product decision, 2026-09-06 — F-2 reference curve.** Re-integrating the same owner-only export
+and stored seed column under D, then independently validating only the selected 4× curve, chooses
+**`τ=24h`**, active-appearance neutral load **`L_neutral=8.076776`**,
+**`L₀=90.717176`** (4× D p95 `22.679294`), **`b=0.5399`**, and therefore
+**`a=b·L_neutral/(L_neutral+L₀)=0.044139`**. The versioned equation is
+`cond=5+a−bL/(L+L₀)`. This does not activate fatigue: F-2 first lands behind a default-OFF flag,
+and G-3 can flip only after the maximum legal role×condition interaction, 8–11-match cup depletion,
+the launch-amplitude prize/cost ordering, H ceiling, and I-c CAP clear against this same curve.
+The D product rerun is recorded below; the older C run remains the audit trail.
+
+### F-1 implementation status — predicate D is now durable
+
+F-1 persists version `f1-d-growth-neutral-v1`: actor plus responder selections minus only the
+actor/responder selections produced by a zero-weight `weightedPick` fallback. Legitimate direct
+defender selections, fixed set-piece/post-play actors, and penalty keepers remain counted. This
+keeps the measured defensive participation while preventing the known all-zero ZP table from
+turning the first non-GK slot into a fatigue lever.
+
+The engine instrumentation is opt-in and consumes no RNG. The existing growth-buffed simulation
+still decides the real result; one separate engine instance projects `total` back to the public
+attribute sum and emits only unsigned slot counters. Scheduled processing claims the complete due
+batch before either real or shadow simulation, freezes one external batch epoch into
+`MatchInputSnapshot v3.capturedAt`, and reuses it on retries. A first locked transaction commits the
+snapshot and durable per-player reservation; a second commits the real result, replay events,
+`match_workloads` root, 11/22 `player_workload_events`, and reservation release. Interactive ladder
+play holds the same player locks from snapshot through both simulations and the durable write.
+
+Cup classification is structural (`tournament_id IS NOT NULL`) and credits both sides. Ladder
+credits home, plus away only when the shared `isAI || isSystemOwnedWallet` rule identifies a bot;
+human away produces no workload rows. Durable players retain `playerId`; daily cup fill-bots use a
+separate salt over cup date, team and slot. Listing, fixed purchase, auction settlement, custody
+activation/delivery finalization, and shared ownership transfer all participate in the same lock
+and reservation fence. Recovery discovers exactly
+`(tournament_id IS NOT NULL OR match_type='playoff') AND status='complete' AND completed_at > cursor`,
+never filters on kickoff, and records explicit legacy exemptions. Friendly, paid trial and sandbox
+write no workload and pin both route-local engine copies to `cond=5`. EWMA decay, neutral point and
+actual condition derivation remain F-2; F-1 alone changes no competitive input.
 
 ### Growth-leak gate: use a zero-growth shadow, not the real action trace
 
@@ -2578,7 +2613,7 @@ stored seed and `gameFlg`. F-1 consumes unsigned engagement counters only from t
 The real result still uses the growth-buffed input; the shadow cannot change its score, events,
 ratings or growth. It is therefore a **zero-growth counterfactual workload**, never the action
 record of the match that happened. Runtime cost is one extra pure simulation per competitive match;
-F-1 still has to persist its counters atomically with completion. The workload timestamp must be
+F-1 persists its counters atomically with the computed result. The workload timestamp must be
 growth-invariant too: fix every due match's first-condition-snapshot (claim-start) time before ANY
 real or shadow engine in that scheduler batch runs, then use that match's one stored value for both.
 A sequential loop that timestamps match B only after simulating match A would carry A's
@@ -2604,8 +2639,8 @@ projection to the synthetic variations above and got byte-identical candidate-F-
 **1,201/1,201 matches and all 2,402 sides**. Unit tests pin erasure, preservation, non-mutation and
 idempotence; `match-growth.test.ts` pins that the real `applyMatchGrowth` producer projects exactly
 back to its innate team, and the existing squad-hash test independently pins that `total` is outside
-the hash. This closes only the §3-5 prerequisite, not F-1/F-3/G-1/F-4's own implementation and
-exposure contracts.
+the hash. That closed the §3-5 prerequisite; F-1 now uses it and G-1 is complete, while F-3 and F-4
+retain their own implementation and exposure contracts.
 
 **The gate is not a per-mode ceiling — it is defined on the exposure-weighted mix.**
 `band-worstcase.mts` holds the canonical statement (`01-engine-expansion.md:319-329` puts
@@ -2765,7 +2800,8 @@ counts; on expected rates it is 14%. Within-match noise had been read as between
 structure. The argument survives in a different form.
 
 The keeper still accrues almost nothing — it enters a contest only at the A0 one-on-one and
-at penalties — so no rule here makes one tired, and F-1 has to fill that blank deliberately.
+at penalties. F-1 deliberately accepted that limitation when it selected D rather than adding a
+separate goalkeeper source; whether it yields usable rotation belongs to F-2.
 
 Predicate D now subtracts fallbacks on BOTH sides of the contest: zone press picks its ACTOR
 through the fallback too, and an earlier version recorded only the responder's, so "C without
@@ -3007,8 +3043,9 @@ conversion.
 
 ### τ cannot be measured — only priced
 
-`τ` is the recovery rate §6 C-2 asks for. F-1 does not exist, so production has no fatigue
-and there is no observed curve to fit: `τ` is a DESIGN choice, not a measurement. What
+`τ` is the recovery rate §6 C-2 asks for. This measurement window predates F-1 and fatigue,
+so there is no observed curve to fit. The later F-1 ledger still does not derive condition;
+`τ` remains a DESIGN choice, not a measurement. What
 measurement can do is re-run everything at each candidate and report what follows. Each row
 below re-integrates the timeline, re-runs that τ's full cell set, and re-searches for the safe
 amplitude (one run, one snapshot).
@@ -3126,7 +3163,7 @@ it leaves open — does an opponent outside the basket exceed? — is asked sepa
 only **62 distinct XIs**. The old pool held 95 objects, of which only **47** were distinct
 lineups — 48 were duplicates. And **35 of the 95 teams fielded more than one XI** inside the
 window, every later one discarded. The new pool CONTAINS the old one and adds 15 lineups.
-(Snapshot `cond` is 5 for everyone in this window — F-1 does not exist yet — so "fielded XI"
+(Snapshot `cond` is 5 for everyone in this historical window — it predates F-1 — so "fielded XI"
 is squad identity here, not condition.)
 
 **Taking only the top three was arbitrary — all 13 teams that played are measured.** Two
@@ -3387,14 +3424,64 @@ rule-of-three bound was bolted on for it. Under empirical Bernstein the range te
 `3R ln(3/δ)/n` survives at SD = 0, so the interval never collapses in the first place: no
 guard, no special case, and one fewer argument resting on its own iid assumption.
 
+### Post-F-1 product rerun — certify only the selected D · 4× curve
+
+Results 6 through the safe-amplitude section above preserve the historical predicate-C run and
+its joint 1×/2×/4× candidate family. The product values were not obtained by scaling that table.
+All four τ values were re-integrated and tested from scratch with
+**`LOAD_PRED=D L0_SWEEP=4`**, using the same stored seed column and the same owner-only export,
+SHA-256 `1c7335ba9fa11388363de57628207ecc4820dc9bb02285066bda849adf571b6f`.
+Letting discarded 1×/2× curves choose the worst cell or enlarge the family would charge the
+shipped curve for hypotheses the product does not ship. The 4× choice was fixed from input-load
+geometry, not outcome seeds, before the result bounds; its family is
+`(26+26+26+24)×6 opponents = 612`.
+
+| τ(h) | D `L` median | D `L` p95 | 0.2-derived `b` (4×) | one match (pp)\* | cells over | **product-curve safe `b`** | worst upper at low end |
+| --- | --- | --- | --- | --- | --- | --- | --- |
+| 6 | 2.51 | 10.52 | 3.548 | 0.2948 | **14/16** | 0.4296–0.4331 | 3.74pp |
+| 12 | 4.07 | 14.29 | 3.009 | 0.1876 | **7/16** | 0.4672–0.4701 | 3.44pp |
+| **24** | **8.08** | **22.68** | **2.446** | **0.0980** | **0/18** | **0.5399–0.5423** | **3.69pp** |
+| 48 | 16.69 | 40.03 | 2.119 | 0.0488 | **0/17** | 0.5546–0.5567 | 3.70pp |
+
+\* “One match” is the secant conversion at each row's `b` derived from E's 0.2 median spread,
+kept only for comparison. It is not the one-match cost at the selected safe `b`. The safe range
+is not a confidence interval either: its left endpoint is the largest grid point independently
+established below for every actual cell, while the right endpoint is the smallest grid point not
+established below. No row established a lower bound above 5pp, so none gets an upper threshold.
+
+The product choices are:
+
+- **`τ=24h`.** The 6h and 12h 0.2-derived curves put 14 and 7 cells over; 24h is the first with
+  zero. 48h is also zero but halves the one-match secant again, 0.0980→0.0488pp. This is a product
+  compromise between daily rotation and fairness, not a recovery estimate fitted to observations.
+- **`L₀=90.717176`.** This is 4× the D/τ24 pre-appearance p95 `22.679294`, and 2.12× the observed
+  maximum `42.87`. At that maximum the marginal slope retained relative to zero load is
+  12.0%/26.4%/46.1% for 1×/2×/4×. Only 4× meaningfully follows §3-1's prescription to keep the
+  curvature scale above the realistic range so marginal cost does not die early.
+- **Neutral `L=8.076776`, `b=0.5399`, `a=0.044139`.** The median active appearance maps to
+  condition 5. Zero load maps to 5.044, D p95 to 4.936, this window's observed maximum to 4.871,
+  and the asymptote to 4.504. At `b=0.5399`, the independent worst upper bound is 3.69pp across
+  all 18 actual cells.
+- **Prize versus cost remains unresolved.** At the safe amplitude's maximum recoverable deficit
+  0.173, the prize is 0.19pp [−0.07, 0.45], the selected cheapest one-point replacement costs
+  0.16pp, and paired cost−prize is −0.03pp [−0.50, 0.43]. Together with the unmeasured legal
+  role×condition and cup extremes, that keeps F-2 default OFF until the follow-up gates and G-3.
+
+The canonical constants are therefore `τ=24h` and
+`cond = 5 + 0.044139 − 0.5399·L/(L + 90.717176)`. Implementation derives `a` from the other
+three constants so documentation rounding cannot move the exact neutral point.
+
 ### Stage 3 — a multi-wallet feeder is exempt, not merely advantaged
 
 F-1's side rule writes nothing for a ladder away side that is human; `playoff-finalize.ts`
 calls `applyResult` for both; `selectOpponentByStrength` draws from the five strength-nearest
 in-division; the cooldown is initiator-only. Observed at-appearance `L` reaches 55 for an
 honest player while a team taking the same match count as the away side accumulates zero. That
-is exemption, not a ratio — bounded only by what bounds the feature, since the prize is 0.59pp at g=0.4 (0.16pp at the τ=24h safe amplitude)
-and the team-level gap is 3.43pp at the safe amplitude, under the gate.
+is exemption, not a ratio — bounded only by what bounds the feature. The historical `g=0.4`
+single-player prize is 0.59pp [0.25, 0.93]; at the selected D product amplitude the maximum
+recoverable deficit gives 0.19pp [−0.07, 0.45], and the independently certified team-level worst
+upper bound is 3.69pp across the 18 observed cells. The unmeasured legal role×condition extreme
+is why that bound is not yet an activation verdict.
 
 ### What this does not settle
 
@@ -3428,10 +3515,10 @@ states while `allCells` did not. A deficit of exactly zero is not evidence of cl
 gate either, so such teams are excluded outright and the output says how many. None occur in
 this window.
 
-The prize-versus-cost ordering AT THE AMPLITUDE THAT WOULD SHIP — it splits at `g=0.4`
-(−0.43pp [−0.76, −0.10]) but not at the safe amplitude's own deficit (0.133; prize 0.16pp
-[−0.09, 0.40], cheapest cost 0.16pp, paired difference 0.00pp [−0.47, 0.47]), and that second
-one is the decision. The neutral-point fork. Maximum LEGAL role concentration (the
+The prize-versus-cost ordering at the D product amplitude remains unresolved: at its maximum
+recoverable deficit 0.173, prize is 0.19pp [−0.07, 0.45], cheapest selected cost is 0.16pp, and
+paired cost−prize is −0.03pp [−0.50, 0.43]. The neutral-point fork is now closed at the D active-appearance median;
+what remains is maximum LEGAL role concentration (the
 replay is what happened, not what the rules permit). Player identity is the snapshot's `actorIds` playerId, keyed GLOBALLY rather than per team —
 a minted player sold mid-window would otherwise get two chains starting from zero, understating
 the buyer's vector. A missing actor map or even one player that fails to join is rejected by the
@@ -3455,9 +3542,15 @@ itself. The cost is that nothing with a period longer than 11 days is visible.
 C2_ENV_FILE=<main checkout>/.env.deploy DAYS=30 WINDOW_END=2026-09-01 \
   EXPORT_SNAPSHOTS=/tmp/prod-tl.json \
   pnpm --filter @sws26/api exec tsx scripts/c2-team-load.ts
+# Post-F-1 product curve: predicate D, testing only the selected L₀=4×p95
+MATCHES=3000 SNAPSHOTS=/tmp/prod-tl.json REPLAYS=5 TAU_SWEEP=6,12,24,48 \
+  LOAD_PRED=D L0_SWEEP=4 LAMBDA_MED=6 LAMBDA_MAX=21 \
+  pnpm --filter pog-mcp exec tsx skill/reference/probes/engagement-load.mts
+
+# Historical C-conditioned 1×/2×/4× table only
 MATCHES=3000 SNAPSHOTS=/tmp/prod-tl.json REPLAYS=5 TAU_SWEEP=6,12,24,48 \
-  LAMBDA_MED=6 LAMBDA_MAX=21 \
-  npx tsx packages/mcp/skill/reference/probes/engagement-load.mts
+  LOAD_PRED=C L0_SWEEP=1,2,4 LAMBDA_MED=6 LAMBDA_MAX=21 \
+  pnpm --filter pog-mcp exec tsx skill/reference/probes/engagement-load.mts
 ```
 
 **The window's end is now stated explicitly, so these figures can be rebuilt.** The default is
@@ -3536,9 +3629,12 @@ from an artificially low `L` and `Lmed`, the gate vectors and the safe amplitude
 understated. Engine changes are input-dependent, so interspersed failures are a normal
 possibility rather than a clean release boundary.
 
-`LOAD_PRED` selects the integration predicate (default `C`). C is published, and the probe also
-integrates at D at every τ and prints the difference — F-1 has not chosen, so the calibration
-has to carry which predicate it is conditional on.
+`LOAD_PRED` selects the integration predicate (default `D`). D is the F-1 product predicate, and
+the probe also integrates the opposite predicate C at every τ and prints the difference. Set
+`LOAD_PRED=C` explicitly to reproduce the historical C-conditioned figures: calibration must
+always carry the predicate it used. `L0_SWEEP` selects the p95 multiples under test. Its default
+`1,2,4` preserves the historical comparison; the product amplitude is certified separately with
+`L0_SWEEP=4`, so an unselected curve neither enlarges the family nor chooses the worst cell.
 
 Without `TAU_SWEEP` only `TAU_H` runs. **Calling a single-τ run a "decision" makes a probe
 default the production recovery rate** — that is why the sweep exists.