pog-mcp 0.8.8 → 0.8.10

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pog-mcp",
-  "version": "0.8.8",
+  "version": "0.8.10",
   "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",
@@ -39,8 +39,8 @@
   "scripts": {
     "build": "tsc",
     "dev": "tsc --watch",
-    "type-check": "tsc --noEmit",
-    "lint": "tsc --noEmit",
+    "type-check": "tsc --noEmit -p tsconfig.typecheck.json",
+    "lint": "pnpm run type-check",
     "start": "node dist/index.js",
     "test": "vitest run",
     "test:watch": "vitest",

package/skill/SKILL.md

@@ -361,9 +361,22 @@ the tables are in `reference/measurements.md`. These describe the engine as it
 currently stands and could change if it is rebalanced.
 
 **Give every player a shape.** A squad with all four attributes equal on every
-player is the worst thing you can build — it lost to every differentiated squad
-tested, most by 3:1. Defenders want defense, forwards want shoot. This is the
-single largest effect measured, and it is what a naive even split gets wrong.
+player is the worst thing you can build — last of eleven in the round-robin, and
+beaten by nine of the ten differentiated squads tested. In the worst of those
+pairings it wins 9% of matches and loses 53%, over 20,000 matches. Defenders
+want defense, forwards want shoot. This is the single largest effect measured,
+and it is what a naive even split gets wrong.
+
+**The tenth pairing is worth knowing, and it is not a reprieve.** The flat squad
+edges `stars-433` — the build that pours its budget into a few 26–29 players —
+winning 39.5% of 20,000 regulation matches to 34.6% lost, the rest drawn (the
+two are level in knockouts). Piling a budget onto a handful of players is the
+one mistake that costs more than not shaping at all, because that squad concedes
+1.30 goals a match, roughly four times what the leaders concede. Spread the
+points across eleven, *then* shape them. (Corrected 2026-08-22: this passage used
+to say the flat squad lost to every differentiated squad. It never did — the
+probe seeder was too weak to show the exception. See
+`reference/measurements.md`.)
 
 **The goalkeeper slot reads exactly one attribute when defending.** Every save
 a keeper ever makes scores `defense + cond/2 + total/3`, at three sites: the 1v1
@@ -461,8 +474,14 @@ total / 3`, `isPkKicker` by `(pass + shoot) / 2 + total / 3`. One player may
 hold both roles if he wins both criteria. (Neither nomination applies to a shootout — that order is computed
 from `shoot + pass` across the whole squad.)
 
-**There is no home advantage.** Identical squads win equally often from either
-side. Never explain a result by which side you were on.
+**Home and away are not measurably different.** Identical squads win 25.1%
+against 24.9% over 200,000 matches, on each of three squad shapes — which bounds
+any edge to under 0.6 of a percentage point. Do not explain a result by which
+side you were on: an effect that small cannot show up in the number of matches
+you will ever play. (The engine is not literally symmetric — one attacking
+branch reads a different player's total at home — so a squad deliberately built
+around it is worth measuring rather than assuming. Nothing measured so far
+moves.)
 
 ## Measuring a change — `simulate_batch`, not thirty friendlies
 

package/skill/reference/measurements.md

@@ -50,9 +50,27 @@ measured) — the cycle probe additionally applies a Bonferroni-corrected
 threshold, and its re-run is recorded in "Strategy round-robin in BOTH modes"
 below. `sample-size.mts` was re-validated the same day (SHA-256; both tables below
 updated — shape unchanged, rows a touch weaker; its identical-squads section is
-its own true-zero control). Still queued: `strategy-probe.mts`,
-`skill-value.mts`, `gk-sweep.mts`, and any published verdict resting on
-|z| < 10 from the old seeder.
+its own true-zero control).
+
+**2026-08-22: the last three were re-seeded and re-run** — `strategy-probe.mts`,
+`skill-value.mts` and `gk-sweep.mts`, each now on SHA-256 with a standing mirror
+cell, at the sample sizes their published tables used. Every table they feed is
+updated below, changed values marked. **No probe in this directory uses the old
+seeder any more.**
+
+That re-run also produced the sharpest demonstration yet of why it was needed.
+Take the round-robin's `flat-433` vs `stars-433` fixture, put the SAME squad on
+both sides — a cell whose true value is zero by construction — and seed it from
+the tag pair those two labels generate. Over 20,000 matches the SHA-256 mirror
+reads **z −0.41**; the FNV mirror reads **z −4.17**. On the real fixture the old
+seeder then scored the pairing 50.3% (z 1.03 — a coin flip) where SHA-256 scores
+it 52.5% (z 8.11). So the old seeder does not only invent effects on empty
+cells; it also buries real ones. A sentence in `SKILL.md` had been resting on
+the buried half — see the round-robin correction below.
+
+`strategy-probe.mts` section ⑤ prints that four-cell comparison on every run. It
+is the one place the retired seeder still exists in this directory, kept as the
+control that justifies its retirement.
 
 **A probe that scores by comparing `homeScore` to `awayScore` is wrong under
 `gameFlg: 1`** and silently reports decided matches as draws. `MatchResult` has
@@ -95,27 +113,92 @@ matches than a per-call cap allows and they carry their mirror controls.
 
 ## Round-robin: 11 strategies × home/away × 300 seeds = 33,000 matches
 
-Ordered by points per match (3 for a win, 1 for a draw).
+Re-measured 2026-08-22 with SHA-256 seeding (mirror control z −0.43). Ordered by
+points per match (3 for a win, 1 for a draw).
 
 | Strategy | Win % | Draw % | Goals for /match | Goals against /match |
 | --- | --- | --- | --- | --- |
-| `gkmin-433` — minimum keeper, spread outfield | 37.2 | 47.2 | 0.68 | 0.33 |
-| `def-433` — defense-leaning roles | 32.4 | 52.6 | 0.52 | 0.27 |
-| `role-433` — attack-leaning roles | 32.9 | 48.1 | 0.58 | 0.36 |
-| `def-532` — five at the back | 27.4 | 62.8 | 0.39 | 0.16 |
-| `bal-442` | 25.3 | 52.4 | 0.46 | 0.40 |
-| `shootonly-433` — shoot on everyone | 26.6 | 41.9 | 0.57 | 0.63 |
-| `gkheavy-433` — 29-point keeper | 22.9 | 50.4 | 0.43 | 0.47 |
-| `atk-352` | 22.9 | 48.4 | 0.46 | 0.52 |
-| `passonly-433` — pass on everyone | 22.1 | 46.9 | 0.47 | 0.57 |
-| `stars-433` — a few 26–29 players | 25.8 | 26.0 | 0.88 | **1.30** |
-| `flat-433` — all four attributes equal | **15.3** | 41.9 | 0.35 | 0.76 |
+| `gkmin-433` — minimum keeper, spread outfield | 37.1 | 46.7 | 0.67 | 0.33 |
+| `def-433` — defense-leaning roles | 31.7 | 52.6 | 0.51 | 0.28 |
+| `def-532` — five at the back | 28.0 | 62.5 | 0.40 | 0.15 |
+| `role-433` — attack-leaning roles | 31.5 | 48.8 | 0.57 | 0.39 |
+| `bal-442` | 25.9 | 51.3 | 0.46 | 0.40 |
+| `gkheavy-433` — 29-point keeper | 24.7 | 49.7 | 0.45 | 0.45 |
+| `shootonly-433` — shoot on everyone | 25.4 | 42.3 | 0.57 | 0.66 |
+| `passonly-433` — pass on everyone | 23.2 | 47.3 | 0.48 | 0.53 |
+| `atk-352` | 21.9 | 48.6 | 0.44 | 0.53 |
+| `stars-433` — a few 26–29 players | 25.4 | 26.9 | 0.85 | **1.30** |
+| `flat-433` — all four attributes equal | **15.3** | 42.9 | 0.36 | 0.74 |
+
+**What the re-seed changed here: three adjacent swaps, no movement at either
+end.** `def-532` now finishes above `role-433`, `gkheavy-433` above
+`shootonly-433`, and `passonly-433` above `atk-352`.
+
+The first has independent corroboration, **but only for the direct matchup**.
+`cycle-probe.mts` plays that pair at 4,000 matches with a Bonferroni-corrected
+threshold and has `def-532` beating `role-433` 60.2% of decided matches,
+significant. That establishes the two are genuinely separated *head to head*.
+It does not settle which belongs higher in this table, because these rows rank
+by points aggregated over ten opponents, and a head-to-head result is blind to
+the other nine — a squad can win a pairing and still finish below its opponent
+on aggregate. So: the direct matchup is decided, the aggregate ordering is not.
+
+The other two swaps have no corroboration at all. They separate rows 0.05 and
+0.03 points-per-match apart; a 6,000-match row here carries a standard error
+near 0.014, so those gaps sit at roughly 2.6 and 1.3 standard errors
+uncorrected — and this table makes 55 pairwise comparisons, for which
+`cycle-probe`'s bar would be |z| ≈ 3.3. Neither clears it.
+
+So read the eleven rows as bands, not as a ranking: `gkmin-433` clear at the
+top, then a top group, a middle group, then `stars-433` and `flat-433`. The top
+and bottom of the table did not move at all. **No aggregate ordering inside a
+band is decidable at 300 seeds, before or after the re-seed** — that is what
+`cycle-probe.mts` exists to do, and it does it pairing by pairing.
 
 Two things stand out. `flat-433` — the squad a naive even split produces — is
 last by a wide margin, which is why "give every player a shape" is the first
 piece of advice. And `stars-433` scores the most goals of any squad and still
 finishes near the bottom, because it concedes 1.30 per match.
 
+### Correction: `flat-433` does NOT lose to everything
+
+`SKILL.md` said the flat squad "lost to every differentiated squad tested". It
+does not, and it did not under the old seeder either — the claim was never true
+of the data, and the weak seeder is why nobody could see it. Head to head over
+20,000 regulation matches (`strategy-probe.mts` section ④, which runs by default
+so this correction reproduces from a checkout rather than citing a scratch run):
+
+| `flat-433` vs | Points share % | W % | D % | L % | z |
+| --- | --- | --- | --- | --- | --- |
+| `gkmin-433` | 28.2 | 9.2 | 38.0 | 52.8 | −78.2 |
+| `def-433` | 30.7 | 6.1 | 49.2 | 44.7 | −76.5 |
+| `def-532` | 32.6 | 3.6 | 58.1 | 38.3 | −75.9 |
+| `role-433` | 31.5 | 10.5 | 42.1 | 47.4 | −68.7 |
+| `bal-442` | 34.8 | 11.8 | 46.0 | 42.2 | −58.4 |
+| `gkheavy-433` | 37.5 | 13.7 | 47.7 | 38.6 | −48.7 |
+| `shootonly-433` | 38.5 | 19.8 | 37.3 | 42.9 | −41.2 |
+| `passonly-433` | 39.5 | 17.8 | 43.4 | 38.8 | −39.5 |
+| `atk-352` | 38.4 | 16.6 | 43.7 | 39.7 | −43.7 |
+| `stars-433` | **52.5** | **39.5** | 25.9 | 34.6 | **+8.1** |
+
+**The points column counts a draw as half a win; it is not a win rate.** Half of
+all regulation matches end level, so the two columns sit far apart — `flat-433`
+takes 32.6% of the points against `def-532` while winning 3.6% of the matches.
+Quote the wrong one and the table says something it does not say (Codex P2,
+#736). The W/D/L columns are printed by the probe for exactly that reason.
+
+Nine crushing losses and one real win. `cycle-probe.mts` — already on SHA-256,
+already published — had the same cell at 54.3% of decided matches, significant
+under its Bonferroni bar; the fact was in this repository and unread. In
+knockouts the pairing is level (52.5% of decided matches, not significant), so
+the win is a regulation-only result.
+
+This is not a defence of the flat split, and the reason matters more than the
+row: **pouring a budget into a few 26–29 players is the one mistake that costs
+more than not shaping at all.** `stars-433` concedes 1.30 goals a match, roughly
+four times what the top squads concede. Being beaten by the second-worst build
+in the table is not evidence for the worst one.
+
 Note the draw column: with regulation-only scoring, roughly half of all matches
 end level. That is the single most important fact for anyone trying to read a
 short run of friendlies.
@@ -124,24 +207,32 @@ short run of friendlies.
 
 Only the goalkeeper's total changes; the remaining points are spread evenly over
 the ten outfield players. Every squad plays the same fixed opponent (a keeper of
-20, spread outfield), home and away, 3,000 matches per row.
+20, spread outfield), home and away, 3,000 matches per row. Re-measured
+2026-08-22 with SHA-256 seeding (mirror control z 0.28).
 
 | GK total | Points per match | Goals against /match |
 | --- | --- | --- |
-| 10 | 1.406 | 0.38 |
-| 13 | 1.363 | 0.40 |
-| 17 | 1.333 | 0.39 |
-| 19 | 1.314 | 0.39 |
-| 23 | 1.144 | 0.47 |
-| 29 | 0.991 | 0.51 |
-
-The trend falls across the whole range, but **adjacent totals do not separate**.
-Re-running the sweep, the order reverses at five of the twenty values — 13, 15,
-17, 19 and 28 each score above the total directly below them. Three thousand
+| 10 | 1.365 | 0.38 |
+| 13 | 1.284 | 0.40 |
+| 17 | 1.324 | 0.38 |
+| 19 | 1.235 | 0.41 |
+| 23 | 1.142 | 0.49 |
+| 29 | 1.037 | 0.53 |
+
+The trend falls across the whole range, but **adjacent totals do not separate**,
+and the re-seed made that more obvious rather than less. Over the full
+twenty-value sweep the order now reverses at **eight** values — 14, 15, 17, 18,
+20, 22, 26 and 28 each score above the total directly below them, where the old
+seeder showed five (13, 15, 17, 19, 28). Neither list is a finding; the fact
+that the list changes wholesale between seeders is the finding. Three thousand
 matches cannot split a one-point difference in keeper budget. What it does split
-is the range: the ~0.4 pt/match gap between the low teens and the high twenties
-is far larger than any reversal. Read this table as "the low end beats the high
-end", never as "19 beats 20".
+is the range: the ~0.33 pt/match gap between total 10 and total 29 is far larger
+than any reversal. Read this table as "the low end beats the high end", never as
+"19 beats 20".
+
+Note also that the effect is smaller than published: 1.406 → 1.365 at the cheap
+end and 0.991 → 1.037 at the expensive end, so the spread narrowed from ~0.42 to
+~0.33. The direction and the mechanism are unchanged.
 
 Note the second column: a *better* keeper concedes *more*, because the points
 came out of the ten players in front of it.
@@ -307,7 +398,9 @@ dominance or cycle verdict.
 | 3-cycles (A>B>C>A) | **0** | **0** |
 | Dominant strategy | none | **`role-433`** beats all seven |
 
-Mode reversals that survived this re-measurement:
+Mode reversals that survived this re-measurement. **Every percentage in these
+three bullets is a share of DECIDED matches** — draws dropped, not counted as
+half — which is the metric that probe's matrix prints:
 
 - **Attack vs defense lean** (same formation, same budgets): `def-433` beats
   `role-433` 56.5% in regulation; the SAME pairing flips to 53.6% for the
@@ -323,27 +416,93 @@ Mode reversals that survived this re-measurement:
 Monotone in both modes, still: role differentiation (`flat-433` last
 everywhere) and spreading over star-concentration (`stars-433` next-to-last).
 
+**Those are statements about the RANKING, not about every pairing** — and the
+difference is exactly where a `SKILL.md` sentence went wrong. This probe's own
+regulation matrix has `flat-433` beating `stars-433` 54.3% of decided matches,
+significant under the Bonferroni bar above. Last in the table does not mean
+loses to everyone. See "Correction: `flat-433` does NOT lose to everything".
+
 Scope: all squads tenure=0, as everywhere in this file. Production applies
 `applyMatchGrowth()` before the engine, raising effective totals — these
 dominance/reversal claims are about zero-buff squads until the buff-0/max
 control runs (docs/tactics/01, track 0.1).
 
-## Home advantage: none
+## Home advantage: none measurable
 
-Identical squads, 3,000 matches: home 24.1%, draw 51.9%, away 24.0%. Symmetric
-to within a tenth of a point.
+Identical squads, home side fixed (`gk-sweep.mts` section ②):
+
+| Squad · sample | Home | Draw | Away | z |
+| --- | --- | --- | --- | --- |
+| baseline · 3,000 (the published sample) | 26.63% | 49.53% | 23.83% | 2.16 |
+| baseline · 200,000 (the deciding sample) | 25.06% | 50.10% | 24.85% | 1.34 |
+| `stars-433` · 200,000 | 38.87% | 22.25% | 38.87% | −0.01 |
+| `atk-352` · 200,000 | 25.40% | 49.43% | 25.16% | 1.49 |
+
+**The verdict holds; the old evidence for it did not.** The previous entry read
+"home 24.1%, draw 51.9%, away 24.0% — symmetric to within a tenth of a point",
+which sounds like a tight measurement and was luck: at 3,000 matches the
+standard error on the home−away gap is about 1.3 points, so a tenth of a point
+is three decimal places past what that sample can see. Re-seeded, the same
+sample size lands 2.8 points apart instead — also luck, in the other direction.
+The section now carries a sample that can actually decide it, and at 200,000
+matches home and away are within 0.2 points.
+
+**Worth knowing that this is an empirical result and not a structural one.** The
+engine is *not* symmetric in code: `getPoint` has an `isHome` branch for DMF
+attackers that scores the ball carrier's total (`this.kiten`) instead of the
+acting player's — a faithful port of the original Perl bug, marked as such in
+`engine.ts`. It can only bite when the acting DMF is *not* the carrier, which
+rules out the ordinary pass/dribble/shoot path and leaves free-kick receivers,
+counter-attacks and post-play. That is why the last two rows are there: they are
+the shapes most able to trigger it — `stars-433` spreads its slot totals from 12
+to 29, so "the carrier's total" and "this player's total" are far apart, and
+`atk-352` fields three DMFs. Both come back level.
+
+**Be precise about what that establishes.** Inside an affected play the branch
+does change the number: the home side scores it off the carrier's total, the
+away side off the acting DMF's, so this is not an inert code path and the two
+sides are not computing the same thing. What the rows above show is that no
+effect survives to the *match result* for the three squad layouts sampled. The
+legal roster space is far larger than three, and none of the rest is tested
+here — a squad shaped to exploit the branch, or a rebalance that widens DMF
+involvement, has to be measured rather than assumed. The claim is "no aggregate
+effect detected in these squads", not "the asymmetry is harmless".
+
+**The practical advice survives that narrowing, for a reason worth separating
+out.** Do not explain a result by which side you were on — not because the
+effect is proven to be exactly zero everywhere, but because of its size. Across
+the three 200,000-match runs the 95% interval on home-minus-away tops out at
++0.52, +0.39 and +0.55 percentage points of all matches — so even the loosest
+reading leaves under 0.6 of a point on the table, and a 3,000-match sample
+cannot see something that small (it read 2.8 points the wrong way on pure
+noise).
+Anyone reading a handful of friendlies is orders of magnitude further away. So
+side is never the explanation for a result you can actually observe — while a
+squad built to lean on DMF free-kick receivers or counters is untested here, and
+should be measured rather than assumed either way.
 
 ## Set-piece takers
 
 Same squad, only the kicker slots move. Opponent fixed, 3,000 matches each.
+Re-measured 2026-08-22 with SHA-256 seeding (`gk-sweep.mts` section ③, mirror
+control z 0.28).
 
 | FK / PK taker | Win % | Points per match |
 | --- | --- | --- |
-| FW / FW (same player) | 25.5 | 1.271 |
-| FW / FW (two players) | 24.1 | 1.221 |
-| OMF / FW | 20.4 | 1.145 |
-| DF / DF | 20.4 | 1.149 |
-| GK / GK | 17.4 | 1.066 |
+| FW / FW (same player, slot 9) | 25.4 | 1.249 |
+| FW / FW (same player, slot 8) | 24.9 | 1.235 |
+| FW / FW (two players) | 24.2 | 1.241 |
+| OMF / FW | 19.5 | 1.115 |
+| DF / DF | 18.9 | 1.112 |
+| GK / GK | 17.6 | 1.075 |
+
+The ordering survived: forwards first, the keeper last by a wide margin. Two
+things moved, both inside the noise this sample can resolve. `OMF / FW` and
+`DF / DF` swapped places — they are 0.003 pt/match apart, which is nothing. And
+the slot-8 row is new: it is the same *kind* of arrangement as the slot-9 row
+(one forward taking both kicks), and it scores 0.014 lower. That difference has
+no cause, and it is the honest floor for reading any other gap in this table.
+The old five-row version silently dropped it.
 
 Conversion tracks the taker's shooting — for FREE kicks. This table moved both
 roles together and never crossed the two criteria, so it says nothing about the
@@ -447,7 +606,8 @@ decisive" and "is decisive".
 ## Does the Skill actually help? — and which part of it
 
 Three hand-built squads, each exactly 212 and validator-clean, 8,000 matches per
-pairing, home and away balanced (`probes/skill-value.mts`):
+pairing, home and away balanced (`probes/skill-value.mts`). Re-measured
+2026-08-22 with SHA-256 seeding (mirror control z −0.55):
 
 - **naive** — 5/5/5/4 on everyone, the squad the tool descriptions alone produce.
 - **shaped** — roles differentiated, defense-leaning, both set pieces on the best
@@ -457,15 +617,16 @@ pairing, home and away balanced (`probes/skill-value.mts`):
 
 | Matchup | Win % | Draw % | Loss % |
 | --- | --- | --- | --- |
-| shaped + cheap keeper **vs** naive | 52.1 | 40.4 | 7.5 |
-| shaped (keeper 20) **vs** naive | 43.6 | 45.9 | 10.6 |
-| shaped + cheap keeper **vs** shaped | 25.7 | 55.0 | 19.3 |
+| shaped + cheap keeper **vs** naive | 51.7 | 40.4 | 7.9 |
+| shaped (keeper 20) **vs** naive | 44.0 | 46.3 | 9.7 |
+| shaped + cheap keeper **vs** shaped | 27.3 | 53.4 | 19.3 |
 
-Read the second and third rows together: **role shaping alone accounts for most
-of the advantage**, and the keeper budget adds a real but secondary edge on top.
-An agent that only differentiates its roles and never touches the keeper beats a
-naive squad 43.6% of the time and loses 10.6% — a four-to-one record, with the
-rest drawn.
+Every row moved by less than a point and a half; this is the table the re-seed
+touched least. Read the second and third rows together: **role shaping alone
+accounts for most of the advantage**, and the keeper budget adds a real but
+secondary edge on top. An agent that only differentiates its roles and never
+touches the keeper beats a naive squad 44.0% of the time and loses 9.7% — better
+than four wins to one loss, with the rest drawn.
 
 ---
 
@@ -495,8 +656,8 @@ So even the two facts do not compose into one recommended shape. What survives
 is the negative advice: never leave points in the keeper's pass/dribble/shoot,
 and never scale a balanced keeper down.
 
-Keep it in proportion. Against a naive squad, role shaping alone wins 43.6% and
-adding a cheap keeper takes that to 52.1%. Role differentiation still decides
+Keep it in proportion. Against a naive squad, role shaping alone wins 44.0% and
+adding a cheap keeper takes that to 51.7%. Role differentiation still decides
 more matches than the keeper does.
 
 `SKILL.md` states the two facts separately and warns against scaling a balanced