pi-zip 0.2.1 → 0.2.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-zip",
-  "version": "0.2.1",
+  "version": "0.2.2",
   "description": "Keeps long Pi sessions cheap without losing anything: folds old tool output only when the prompt cache has already gone cold, and every fold can be recalled byte for byte. Zero config.",
   "type": "module",
   "license": "MIT",

package/src/plan.ts

@@ -282,6 +282,7 @@ export interface PlanOpts {
 	relax?: boolean; // default RELAX_PREV_TURN
 	minGain?: number;
 	inturnAge?: number; // default settings().inturnAge (PI_ZIP_INTURN_AGE, 60); 0 = outputs of the protected turns never fold on age
+	noSummary?: boolean; // a summary was already made while this cache stayed warm: no second one unless the context is at/above the compaction room
 	rereadOnly?: boolean; // zip_recall is not available to the model: fold only outputs that can be re-read (classify "rereadable")
 	pastTtl?: boolean; // a user return after the declared TTL: a warm plan may relax into the previous user turn too (RELAX_PREV_TURN)
 }
@@ -417,7 +418,13 @@ export function planContext(entries: Any[], o: PlanOpts): PlanResult | null {
 		}
 	}
 	let ctxAfterFolds = ctxEst - savings();
-	const sumTrigger: "cold" | "valve" | null = ctxAfterFolds > coldCap ? trig : null;
+	// what no summary can remove: the system prompt, the protected turns (after their folds) and the previous summary, which the next
+	// one carries forward; the cap is never chased below it (a context that is mostly this floor would be re-summarised for nothing)
+	const prevSum = blocks[0]?.kind === "summary" ? blocks[0].tokens : 0;
+	const floor = o.sys + prevSum + blocks.reduce((a, b) => a + (protectedTurn(b) ? (tokOverride.get(b.idx) ?? b.tokens) : 0), 0);
+	const target = Math.max(coldCap, floor + SUMMARY_FLOOR);
+	const atRoom = room !== null && k * ctxAfterFolds >= room;
+	const sumTrigger: "cold" | "valve" | null = ctxAfterFolds > target && (!o.noSummary || atRoom) ? trig : null;
 	let cutIdx: number | null = null;
 	let prefixTokens = 0;
 	let summaryTokensPlanned = 0;
@@ -436,7 +443,7 @@ export function planContext(entries: Any[], o: PlanOpts): PlanResult | null {
 			acc += tokOverride.get(i) ?? blocks[i].tokens;
 		}
 		const total = o.sys + acc;
-		const S = (x: number) => clamp(SUMMARY_RATIO * x, SUMMARY_FLOOR, SUMMARY_CAP);
+		const S = (x: number) => Math.max(clamp(SUMMARY_RATIO * x, SUMMARY_FLOOR, SUMMARY_CAP), prevSum); // the next summary carries the previous one forward
 		if (cuts.length) {
 			let pick = cuts[cuts.length - 1];
 			for (const c of cuts) if (total - pre[c] + S(pre[c]) <= coldCap) { pick = c; break; }

package/src/run.ts

@@ -94,6 +94,8 @@ export class Zip implements ZipControl {
 
 	/** zip_recall declared to the model? Checked at session start and before every run (index.ts); a `--tools` allowlist hides it. */
 	recallOk = true;
+	/** A summary was persisted since the last cold run start: later warm plans make none unless at the compaction room. */
+	summarizedWarm = false;
 	setRecallOk(ok: boolean) {
 		if (ok !== this.recallOk) this.ledger({ type: "recall_available", ok });
 		this.recallOk = ok;
@@ -146,7 +148,7 @@ export class Zip implements ZipControl {
 	private opts(ctx: Any, pending: boolean, mode: "cold" | "warm", entries: Any[], pWarm = mode === "cold" ? 0 : 1): { o: PlanOpts; cal: Calibration } {
 		const sys = this.sysTokens(ctx);
 		const cal = calibrate(entries, sys);
-		return { cal, o: { mode, sys, k: cal.k, cwd: (ctx?.cwd as string) ?? process.cwd(), recalled: this.recalled, model: ctx?.model, promptPending: pending, rereadOnly: !this.recallOk, reserve: this.reserve(ctx), steerIds: this.steerIds, law: this.law(ctx, pWarm), trace: [] } };
+		return { cal, o: { mode, sys, k: cal.k, cwd: (ctx?.cwd as string) ?? process.cwd(), recalled: this.recalled, model: ctx?.model, promptPending: pending, rereadOnly: !this.recallOk, noSummary: mode === "warm" && this.summarizedWarm, reserve: this.reserve(ctx), steerIds: this.steerIds, law: this.law(ctx, pWarm), trace: [] } };
 	}
 
 	private law(ctx: Any, pWarm: number): Law {
@@ -234,6 +236,7 @@ export class Zip implements ZipControl {
 		if (!this.prevKey) ({ key: this.prevKey, total: this.prevTotal } = lastPrompt(branch)); // fresh process: the session's newest response
 		const { cold, reason, ttl, pWarm: p, src, gapS, pastTtl } = detectCold(ctx.model, this.lastReqMs, branch, Date.now(), this.lastModelKey, (g, prior) => pWarm(this.ent, g, prior));
 		this.cold = cold;
+		if (cold) this.summarizedWarm = false; // a cold return may summarise again
 		this.pastTtl = pastTtl;
 		this.pWarm = p;
 		this.survSrc = src;
@@ -512,6 +515,7 @@ export class Zip implements ZipControl {
 			return undefined;
 		}
 		const bad = checkEdits(o.blocks, o.userTurns, live, cut);
+		if (!bad && cut) this.summarizedWarm = true; // one summary per warm stretch (two in a row cost two uncached calls; the second gains little)
 		if (bad) {
 			this.ledger({ type: "guard_drop", turnIndex: e.turnIndex, reason: bad, folds: live.length, source: plan.source });
 			plan.persisted = false; // keep the request-local view: never switch the fold set mid-run