pi-mega-compact 0.20.84 → 0.20.85

package/dist/extensions/dashboard-server/routes-rag-settings-helpers.js

@@ -126,6 +126,12 @@ export const SETTINGS = [
             num("MEGACOMPACT_EMBEDDING_CHARS_PER_TOKEN", "Embedding Chars per Token", "Estimated characters per token used for embedder chunking size", 4, 1, 32),
         ],
     },
+    {
+        name: "Compaction",
+        settings: [
+            num("MEGACOMPACT_THRESHOLD_PCT", "Compaction Threshold", "Fraction of the actual model context window at which compaction fires — 0.80 fires at 80% used (leaves 20% free). Applies to any model size; a per-model Model Thresholds row overrides it", 0.8, 0.1, 0.95),
+        ],
+    },
     VECTOR_CORTEX_SETTINGS,
     {
         name: "Cost API",

package/dist/extensions/mega-config.js

@@ -66,11 +66,30 @@ function resolveThreshold() {
     }
     const raw = (process.env.MEGACOMPACT_TIER ?? "low").toLowerCase();
     const tier = (raw in COMPACT_TIERS ? raw : "low");
-    const tierPct = TIER_PCT[tier];
+    let tierPct = TIER_PCT[tier];
+    // 3WF-2 threshold invariant: under the umbrella, when no named tier is set
+    // the fire point is the configurable % of the ACTUAL model window (default
+    // 0.80 — "20% free remaining"). Tiered (named preset) keeps its preset pct;
+    // both paths still compute the legacy 200k boot fallback below as a display
+    // placeholder + the custom-tier absolute companion. Umbrella OFF stays
+    // byte-identical to v0.20.83 (default tier=low 0.5).
+    const umbrella = envBool("MEGACOMPACT_THREE_WAY_FAILBACK", true);
+    if (umbrella && !(process.env.MEGACOMPACT_TIER && process.env.MEGACOMPACT_TIER !== "")) {
+        tierPct = clamp(envFlag("MEGACOMPACT_THRESHOLD_PCT", 0.8), 0.1, 0.95);
+    }
     // Boot fallback: sane gate before the first context event provides a window.
+    // (NO hardcoded window in the firing path — effectiveThresholdImpl defers
+    // when window unknown; this remains only a display placeholder + custom
+    // companion under the umbrella.)
     const thresholdTokens = Math.round(tierPct * 200_000);
     return { tier, tierPct, thresholdTokens };
 }
+/** Clamp `n` into [lo, hi]; non-finite → fallback. */
+function clamp(n, lo, hi) {
+    if (!Number.isFinite(n))
+        return lo;
+    return Math.min(hi, Math.max(lo, n));
+}
 /**
  * Pure helper: the real compaction fire point, given the model context window.
  *

package/dist/extensions/mega-events/context-handler/gateCheck.js

@@ -40,11 +40,23 @@ export function evaluateGate(runtime, config, opts) {
     }
     else {
         // custom tier OR tiered-but-pct-unavailable → token gate (S27 fallback).
-        if (currentTokens < runtime.effectiveThreshold) {
+        // 3WF-2: under the umbrella, when the window is known AND the config is
+        // tiered, honor the per-model Dashboard override exactly like the percent
+        // branch (firePointPct % of the actual window) instead of the bare boot
+        // fallback. custom / window-unknown / umbrella-OFF keep the bare
+        // runtime.effectiveThreshold (window-unknown defers via +Infinity from
+        // effectiveThresholdImpl; custom is the explicit absolute).
+        let gateThreshold = runtime.effectiveThreshold;
+        if (config.threeWayFailback &&
+            config.tierPct != null &&
+            runtime.lastCtxWindow > 0) {
+            gateThreshold = Math.round((perModelThreshold.firePointPct / 100) * runtime.lastCtxWindow);
+        }
+        if (currentTokens < gateThreshold) {
             runtime.diagCtxFastGate++;
             return { kind: "return", view: tailResult() ?? undefined };
         }
-        const check = autoCompactCheck(currentTokens, runtime.effectiveThreshold); // SERVER-STYLE CONFIRM (local)
+        const check = autoCompactCheck(currentTokens, gateThreshold); // SERVER-STYLE CONFIRM (local)
         if (!check.shouldCompact) {
             runtime.diagCtxNoCompact++;
             return { kind: "return", view: tailResult() ?? undefined };

package/dist/extensions/mega-runtime/pressure-getters.js

@@ -51,6 +51,18 @@ export function pressureImpl(self) {
  * always below pi's native auto-compaction (~80% of window).
  */
 export function effectiveThresholdImpl(self) {
+    // 3WF-2 threshold invariant: under the umbrella, a tiered config with an
+    // UNKNOWN window (lastCtxWindow <= 0) DEFERS — auto-compaction must never
+    // substitute a guessed window. Returning +Infinity keeps every downstream
+    // `tokens >= threshold` comparison false (gateCheck token path,
+    // agent_end durable trigger, live-trim re-compact), so no compaction fires
+    // until the provider reports a real window. custom (tierPct null) and
+    // umbrella-OFF fall through to the legacy helper (byte-identical).
+    if (self.config.threeWayFailback &&
+        self.config.tierPct != null &&
+        self.lastCtxWindow <= 0) {
+        return Number.POSITIVE_INFINITY;
+    }
     return effectiveThresholdTokens({
         tierPct: self.config.tierPct,
         fallbackThreshold: self.config.thresholdTokens,

package/extensions/dashboard-server/routes-rag-settings-helpers.ts

@@ -274,6 +274,19 @@ export const SETTINGS: ReadonlyArray<SettingGroup> = [
 			num("MEGACOMPACT_EMBEDDING_CHARS_PER_TOKEN", "Embedding Chars per Token", "Estimated characters per token used for embedder chunking size", 4, 1, 32),
 		],
 	},
+	{
+		name: "Compaction",
+		settings: [
+			num(
+				"MEGACOMPACT_THRESHOLD_PCT",
+				"Compaction Threshold",
+				"Fraction of the actual model context window at which compaction fires — 0.80 fires at 80% used (leaves 20% free). Applies to any model size; a per-model Model Thresholds row overrides it",
+				0.8,
+				0.1,
+				0.95,
+			),
+		],
+	},
 	VECTOR_CORTEX_SETTINGS,
 	{
 		name: "Cost API",

package/extensions/mega-config.ts

@@ -78,12 +78,31 @@ function resolveThreshold(): {
 	}
 	const raw = (process.env.MEGACOMPACT_TIER ?? "low").toLowerCase();
 	const tier = (raw in COMPACT_TIERS ? raw : "low") as CompactTier;
-	const tierPct = TIER_PCT[tier];
+	let tierPct = TIER_PCT[tier];
+	// 3WF-2 threshold invariant: under the umbrella, when no named tier is set
+	// the fire point is the configurable % of the ACTUAL model window (default
+	// 0.80 — "20% free remaining"). Tiered (named preset) keeps its preset pct;
+	// both paths still compute the legacy 200k boot fallback below as a display
+	// placeholder + the custom-tier absolute companion. Umbrella OFF stays
+	// byte-identical to v0.20.83 (default tier=low 0.5).
+	const umbrella = envBool("MEGACOMPACT_THREE_WAY_FAILBACK", true);
+	if (umbrella && !(process.env.MEGACOMPACT_TIER && process.env.MEGACOMPACT_TIER !== "")) {
+		tierPct = clamp(envFlag("MEGACOMPACT_THRESHOLD_PCT", 0.8), 0.1, 0.95);
+	}
 	// Boot fallback: sane gate before the first context event provides a window.
+	// (NO hardcoded window in the firing path — effectiveThresholdImpl defers
+	// when window unknown; this remains only a display placeholder + custom
+	// companion under the umbrella.)
 	const thresholdTokens = Math.round(tierPct * 200_000);
 	return { tier, tierPct, thresholdTokens };
 }
 
+/** Clamp `n` into [lo, hi]; non-finite → fallback. */
+function clamp(n: number, lo: number, hi: number): number {
+	if (!Number.isFinite(n)) return lo;
+	return Math.min(hi, Math.max(lo, n));
+}
+
 /**
  * Pure helper: the real compaction fire point, given the model context window.
  *

package/extensions/mega-events/context-handler/gateCheck.ts

@@ -82,11 +82,27 @@ export function evaluateGate(
 		gatePassed = pct / 100 >= firePct;
 	} else {
 		// custom tier OR tiered-but-pct-unavailable → token gate (S27 fallback).
-		if (currentTokens < runtime.effectiveThreshold) {
+		// 3WF-2: under the umbrella, when the window is known AND the config is
+		// tiered, honor the per-model Dashboard override exactly like the percent
+		// branch (firePointPct % of the actual window) instead of the bare boot
+		// fallback. custom / window-unknown / umbrella-OFF keep the bare
+		// runtime.effectiveThreshold (window-unknown defers via +Infinity from
+		// effectiveThresholdImpl; custom is the explicit absolute).
+		let gateThreshold = runtime.effectiveThreshold;
+		if (
+			config.threeWayFailback &&
+			config.tierPct != null &&
+			runtime.lastCtxWindow > 0
+		) {
+			gateThreshold = Math.round(
+				(perModelThreshold.firePointPct / 100) * runtime.lastCtxWindow,
+			);
+		}
+		if (currentTokens < gateThreshold) {
 			runtime.diagCtxFastGate++;
 			return { kind: "return", view: tailResult() ?? undefined };
 		}
-		const check = autoCompactCheck(currentTokens, runtime.effectiveThreshold); // SERVER-STYLE CONFIRM (local)
+		const check = autoCompactCheck(currentTokens, gateThreshold); // SERVER-STYLE CONFIRM (local)
 		if (!check.shouldCompact) {
 			runtime.diagCtxNoCompact++;
 			return { kind: "return", view: tailResult() ?? undefined };

package/extensions/mega-runtime/pressure-getters.ts

@@ -81,6 +81,20 @@ export function pressureImpl(self: PressureContext): number {
  * always below pi's native auto-compaction (~80% of window).
  */
 export function effectiveThresholdImpl(self: PressureContext): number {
+	// 3WF-2 threshold invariant: under the umbrella, a tiered config with an
+	// UNKNOWN window (lastCtxWindow <= 0) DEFERS — auto-compaction must never
+	// substitute a guessed window. Returning +Infinity keeps every downstream
+	// `tokens >= threshold` comparison false (gateCheck token path,
+	// agent_end durable trigger, live-trim re-compact), so no compaction fires
+	// until the provider reports a real window. custom (tierPct null) and
+	// umbrella-OFF fall through to the legacy helper (byte-identical).
+	if (
+		self.config.threeWayFailback &&
+		self.config.tierPct != null &&
+		self.lastCtxWindow <= 0
+	) {
+		return Number.POSITIVE_INFINITY;
+	}
 	return effectiveThresholdTokens({
 		tierPct: self.config.tierPct,
 		fallbackThreshold: self.config.thresholdTokens,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-mega-compact",
-  "version": "0.20.84",
+  "version": "0.20.85",
   "description": "Layered, local, vector-backed context compressor for pi — supersede/collapse/cluster compaction with deduped inline recall.",
   "type": "module",
   "license": "BSD-3-Clause",