@letterstory/cli 0.66.0 → 0.69.0

package/README.md

@@ -369,14 +369,17 @@ letterstory admin tuning trial-run <trial> --repeats 2  # runs advance on the se
 letterstory admin tuning trial-save <trial> --note "…"  # a trial as a candidate version (never live); then `admin tuning compare`
 letterstory admin tuning promote <experiment-id> --versions <service-kernel-id>:21   # from a concluded "ship" decision
 letterstory admin experiments list --q colons          # what we've tried
+letterstory admin experiments list --primitive write --stage rarity   # what we learned about one stage
+letterstory admin experiments new "A different writer, voice rewritten, rarity rewrite off" --primitive write   # an investigation in Write's notebook
 letterstory admin experiments new "We use too many colons" --as "claude-code"
 letterstory admin experiments update <id> --plan-file plan.json --budget 120
-# ...a person approves the plan in Kernel Tuning -> Experiments...
+# ...a person approves the plan on the investigation's page (approval gates spending a budget; a person's
+# investigation needs none and is concluded straight away)...
 letterstory admin experiments spend <id> 54
 letterstory admin tuning-import lab/results/colons/publish/<brief>.json --experiment <id>
 letterstory admin experiments conclude <id> --decision drop \
   --conclusion "Colons fell 60%; P(machine) did not move; quality held." \
-  --evidence lab:"kernels lab/results/colons/report.md"
+  --evidence lab:"kernels lab/results/colons/report.md" --stages rules
 ```
 
 The ledger behind Kernel Tuning's Experiments tab
@@ -728,8 +731,10 @@ AI visibility and page access — the Measure tabs' reads, for scripts and agent
 ```bash
 # The Measure Overview: how often AI answers name you and cite your pages (rates
 # with 95% intervals and gated trends), share of voice, search for your site and
-# the phantom network kept apart, and each ranking target's standing.
-letterstory insights summary [--period 14d|30d|90d]
+# the phantom network kept apart, and each ranking target's standing. --part
+# (repeatable) returns only those blocks: audience, ai, shareOfVoice, pageAccess,
+# search, actions, targets — `ai` is the slow one on a large org.
+letterstory insights summary [--period 14d|30d|90d] [--part <block>]...
 
 # The AI Visibility Bank: which questions named you, cited your pages, or both (a
 # breakthrough). All time unless --period; --source picks the phantom network

package/lib/cli.mjs

@@ -56,7 +56,7 @@ import {
 } from "./commands.mjs";
 
 // Keep in sync with cli/package.json.
-export const VERSION = "0.66.0";
+export const VERSION = "0.69.0";
 
 // Flags that never take a value. Listing them explicitly means `deploy get --json <id>`
 // can't accidentally swallow the id as --json's value.
@@ -554,19 +554,24 @@ Admin (STAFF key — --staff-key / LETTERSTORY_STAFF_API_KEY; acts before an org
                                                        Live answer replays: breakthroughs re-asked in ChatGPT /
                                                        Gemini / Google / Perplexity / Claude, with screenshots.
                                                        Omit --org for the whole fleet. Needs fleet:read.
-  admin replay --org <uuid> (<response-id> | --target <uuid>) [--surface openai|anthropic|google-aio|google-gemini|perplexity]
+  admin replay --org <uuid> [<response-id> | --target <uuid>] [--surface openai|anthropic|google-aio|google-gemini|perplexity]
                                                        Replay a breakthrough live now (background; poll
-                                                       admin replays). Needs fleet:replay.
-  admin experiments list [--q colons] [--status proposed] | show <id>
-          | new "<question>" [--hypothesis …] [--plan-file p.json] [--tags a,b] [--as "<agent>"]
-          | update <id> [--plan-file …] [--result-file …] [--evidence kind:ref,…] [--budget N]
+                                                       admin replays). With neither, the client's newest
+                                                       breakthrough. Needs fleet:replay.
+  admin experiments list [--q colons] [--status proposed] [--primitive write] [--stage rarity] | show <id>
+          | new "<what you are investigating>" [--primitive write|research|shred|enhance] [--stages a,b]
+                [--hypothesis …] [--plan-file p.json] [--tags a,b] [--as "<agent>"]
+          | update <id> [--plan-file …] [--result-file …] [--evidence kind:ref,…] [--stages a,b] [--budget N]
           | note|started <id> "<text>" | spend <id> <runs>
-          | conclude <id> --decision ship|drop|follow-up|inconclusive --conclusion "…" [--evidence …]
+          | conclude <id> --decision ship|drop|follow-up|inconclusive --conclusion "…" [--stages a,b] [--evidence …]
           | abandon <id> --reason "…"
-                                                       The kernel experiment ledger: what we tried,
-                                                       how, what it found and why. Approving a plan
-                                                       stays with a signed-in person. Needs the
-                                                       tuning:experiment capability.
+                                                       Each primitive's notebook of investigations:
+                                                       what we tried, how, what it found and why,
+                                                       and the pipeline stages it is about. A plan
+                                                       is approved by a signed-in person before a
+                                                       budget is spent; concluding needs evidence,
+                                                       not approval. "ship" hands off: a promoter
+                                                       makes it live. Needs tuning:experiment.
   admin tuning-import <brief-results.json> [--experiment <id>]
                                                        One brief's kernels lab results as a console
                                                        session (a trial per arm), linked to its
@@ -630,7 +635,11 @@ Admin (STAFF key — --staff-key / LETTERSTORY_STAFF_API_KEY; acts before an org
                                                        candidate, compare, comparisons, picks, pick,
                                                        reading-list, pair, side-by-side, promote,
                                                        sessions, session-new, propose, trial-add,
-                                                       trial-run, trial-save, and the arena, shredder
+                                                       trial-run, trial-save, the Write input bank
+                                                       (inputs, input, inputs-harvest, input-save),
+                                                       screens (screen-new, screens, screen,
+                                                       screen-advance),
+                                                       and the arena, shredder
                                                        and enhancer twins (every flag is listed in
                                                        cli/lib/commands/admin.mjs).
   admin create-org --domain <bare-domain> [--name "Acme"] [--test] [--no-key]
@@ -882,7 +891,7 @@ Mail:
   mail postal-address "<address>"                      Set the org's CAN-SPAM mailing address (email footers)
 
 Insights:
-  insights summary [--period 14d|30d|90d]              Measure Overview: AI named/cited rates, share of voice, search, targets
+  insights summary [--period …] [--part <block>]...    Measure Overview: AI named/cited rates, share of voice, search, targets
   insights site [--period 14d|30d|90d] [--collection <uuid>]
   insights post <article-id> [--period 14d|30d|90d]
   insights top [--period 14d|30d|90d] [--collection <uuid>] [--limit <n>] [--sort clicks|impressions]

package/lib/commands/admin.mjs

@@ -11,7 +11,7 @@
 //   admin evals scoreboard|cases|runs|run|triage|guard|retire …                      (fleet:read / tuning:experiment)
 //   admin crm import|capture|signals|record|signal|accounts|account|score|slack-channels|link-slack|white-glove|set-white-glove|churn|record-churn|undo-churn|onboard|onboarding|retry-onboarding|onboard-plan|payment-link …  (crm:read / crm:write)
 //   admin replays [--org <id>] [--target <id>] [--status …] [--limit n]                (fleet:read)
-//   admin replay --org <id> (<response-id> | --target <id>) [--surface …]           (fleet:replay)
+//   admin replay --org <id> [<response-id> | --target <id>] [--surface …]           (fleet:replay)
 //   admin create-org --domain acme.com [--name "Acme"] [--test] [--no-key] [--key-name N]
 //                    [--capabilities a,b] [--onboard] [--about "…"] [--sitemap <url>]
 //                    [--cms <provider>] [--posts N] [--seers suggest|skip] [--no-visuals]
@@ -268,7 +268,8 @@ function evidenceFrom(flags) {
 
 function experimentLine(e) {
 	const decision = e.decision ? ` → ${e.decision}` : "";
-	return `${e.id}  [${e.status}${decision}]  ${e.question}${e.gist ? `\n    ${e.gist}` : ""}`;
+	const where = e.primitive ? `${e.primitive}${e.stages?.length ? `: ${e.stages.join(", ")}` : ""}  ` : "";
+	return `${e.id}  [${e.status}${decision}]  ${where}${e.question}${e.gist ? `\n    ${e.gist}` : ""}`;
 }
 
 /**
@@ -293,7 +294,13 @@ async function cmdAdminExperiments(ctx) {
 	switch (sub) {
 		case "list": {
 			const params = new URLSearchParams(
-				compact({ q: flagStr(flags.q), status: flagStr(flags.status), limit: flagStr(flags.limit) })
+				compact({
+					q: flagStr(flags.q),
+					status: flagStr(flags.status),
+					primitive: flagStr(flags.primitive),
+					stage: flagStr(flags.stage),
+					limit: flagStr(flags.limit),
+				})
 			);
 			const result = await staffRequest(ctx, "GET", `${EXPERIMENTS}?${params}`);
 			return out(result, (r) =>
@@ -306,9 +313,13 @@ async function cmdAdminExperiments(ctx) {
 		case "new": {
 			const question = [id, ...rest].filter(Boolean).join(" ").trim();
 			if (question.length < 3)
-				throw new CliError('Usage: admin experiments new "<question>" [--plan-file plan.json] [--tags a,b]');
+				throw new CliError(
+					'Usage: admin experiments new "<what you are investigating>" [--primitive write|research|shred|enhance] [--stages write,rarity] [--plan-file plan.json] [--tags a,b]'
+				);
 			const body = compact({
 				question,
+				primitive: flagStr(flags.primitive),
+				stages: flagList(flags.stages),
 				restated: flagStr(flags.hypothesis),
 				plan: jsonFile(flags, "plan"),
 				method: flagStr(flags.method),
@@ -331,6 +342,7 @@ async function cmdAdminExperiments(ctx) {
 				result: jsonFile(flags, "result"),
 				method: flagStr(flags.method),
 				tags: flagList(flags.tags),
+				stages: flagList(flags.stages),
 				budget_runs: flagNum(flags.budget),
 				engine_build: flagStr(flags.build),
 				add_evidence: evidenceFrom(flags),
@@ -357,12 +369,13 @@ async function cmdAdminExperiments(ctx) {
 			const conclusion = flagStr(flags.conclusion);
 			if (!id || !decision || !conclusion)
 				throw new CliError(
-					'Usage: admin experiments conclude <id> --decision ship|drop|follow-up|inconclusive --conclusion "<what we learned and why>" [--evidence kind:ref,…] [--result-file r.json]'
+					'Usage: admin experiments conclude <id> --decision ship|drop|follow-up|inconclusive --conclusion "<what we learned and why>" [--stages write,rarity] [--evidence kind:ref,…] [--result-file r.json]'
 				);
 			const body = compact({
 				action: "conclude",
 				decision,
 				conclusion,
+				stages: flagList(flags.stages),
 				result: jsonFile(flags, "result"),
 				add_evidence: evidenceFrom(flags),
 				actor: actorFrom(flags),
@@ -443,13 +456,28 @@ async function cmdAdminTuningContext(ctx) {
  *   admin tuning base-kernels                                  the kernels a session can run on (ids for --kernel)
  *   admin tuning base-versions <kernel-id>                     the versions a session on it can run against
  *   admin tuning sessions | session <id> [--texts]             list / one session: trials, runs and their numbers (--texts adds each trial's latest text)
- *   admin tuning session-new --title "…" --kernel <id|name> (--input-file f.md | --input "…") [--kind article|outline|brief]
+ *   admin tuning session-new --title "…" --kernel <id|name> (--input-file f.md | --input "…" | --bank <input-id>) [--kind article|outline|brief]
  *                            [--version N] [--writers echo/echo,…] [--trials-from <session-id>]
+ *                            (--bank: a bank input's text, and its request's constraints under every trial)
  *   admin tuning propose <session-id> --text "Echo with rarity on; live with the shred after" [--create]
  *   admin tuning trial-add <session-id> [--name "…"] [--patch '{…}'] [--options '{…}'] [--from <trial-id>]
  *   admin tuning trial-run <trial-id> [--repeats 2]            queue runs (guardrails apply)
  *   admin tuning trial-save <trial-id> [--note "…"]            save a trial as a candidate version (never live)
  *
+ * The Write input bank: real kernel requests (above all the arena outlines Auto-Pen sends), as sent.
+ *   admin tuning inputs [--filter "origin=arena_outline;material=thin,typical"] [--q "…"]   list by tag, with coverage per bucket
+ *   admin tuning input <input-id>                              one input in full: text, constraints, tags, run defaults, origin
+ *   admin tuning inputs-harvest [--n 40] [--pool 400] [--callers auto-pen,endpoint] [--since-days 21] [--min-per-bucket 3]
+ *                            freeze recent production runs, stratified; idempotent by key
+ *   admin tuning input-save <session-id> [--note "…"]          save a session's pasted input or arena outline to the bank
+ *
+ * Screens: "is this candidate worth a full comparison?", on a slice of the bank (it does not decide).
+ *   admin tuning screen-new --title "…" (--from <session-id> | --kernel <kernel-id> --versions 23,29) [--version N]
+ *                            [--n 8] [--filter "origin=arena_outline;material=thin"] [--inputs <id>,…] [--repeats 1] [--lift-budget]
+ *                            one session per input with the same trials; runs queue as the console's caps allow
+ *   admin tuning screens | screen <id>                         list / one screen: per candidate, each measure against the Base, with its noise
+ *   admin tuning screen-advance <id>                           queue what the caps allow now (the cron does it every minute)
+ *
  * Research (Phase 3): our arena's versions as files, research comparisons, the deploy split.
  *   admin tuning arenas | arena <version-id>                   our arena's versions / one version's file and diff
  *   admin tuning arena-candidate --patch '{…}' | --file f.json [--of <version-id>] [--note "…"]   save a candidate
@@ -492,6 +520,19 @@ function jsonFlag(raw, name) {
 	}
 }
 
+/**
+ * `--filter "origin=arena_outline;material=thin,typical"` → the bank's query params (one comma list
+ * per tag dimension; lib/tuning/input-tags.ts `parseTagFilter` reads the same form).
+ */
+function tagFilterParams(raw) {
+	const out = {};
+	for (const part of (raw ?? "").split(";")) {
+		const [k, v] = part.split("=");
+		if (k?.trim() && v?.trim()) out[k.trim()] = v.trim();
+	}
+	return out;
+}
+
 /**
  * A session as an agent reads it: each trial's changes and runs with their headline numbers, and
  * (with `texts`) the latest finished text per trial. The full session (traces, records) is `--json`.
@@ -770,9 +811,11 @@ async function cmdAdminTuning(ctx) {
 					throw new CliError(`--input-file ${path}: ${err.message}`);
 				}
 			}
-			if (!title || !kernel || !input)
+			// A bank input instead of a paste: its text, and its request's constraints under every trial.
+			const bank = flagStr(flags.bank);
+			if (!title || !kernel || (!input && !bank))
 				throw new CliError(
-					'Usage: admin tuning session-new --title "…" --kernel <id|name> (--input-file f.md | --input "…") [--kind article|outline|brief] [--version N] [--writers a,b] [--versions 23,29] [--trials-from <session-id>]'
+					'Usage: admin tuning session-new --title "…" --kernel <id|name> (--input-file f.md | --input "…" | --bank <input-id>) [--kind article|outline|brief] [--version N] [--writers a,b] [--versions 23,29] [--trials-from <session-id>]'
 				);
 			let kernelId = kernel;
 			if (!/^[0-9a-f-]{36}$/i.test(kernel)) {
@@ -788,8 +831,9 @@ async function cmdAdminTuning(ctx) {
 				T,
 				compact({
 					title,
-					input_text: input,
+					input_text: bank ? undefined : input,
 					input_kind: flagStr(flags.kind) ?? "unknown",
+					material: bank ? { source: "bank", input_id: bank } : undefined,
 					base_kernel_id: kernelId,
 					base_version: flagNum(flags.version),
 					trials_from_session_id: flagStr(flags["trials-from"]),
@@ -871,6 +915,167 @@ async function cmdAdminTuning(ctx) {
 				`${r.kernel}: saved as candidate v${r.version}${r.baseVersion ? ` (from v${r.baseVersion})` : ""}${r.leftBehind.length ? `; per-run, not saved: ${r.leftBehind.join(", ")}` : ""}`
 			);
 		}
+		// --- The Write input bank: real kernel requests a session can start from ---
+		case "inputs": {
+			const q = new URLSearchParams(
+				compact({ ...tagFilterParams(flagStr(flags.filter)), q: flagStr(flags.q), limit: flagStr(flags.limit) })
+			);
+			const r = await staffRequest(ctx, "GET", `${T}/inputs${q.size ? `?${q}` : ""}`);
+			if (json) return printResult(io, flags, r);
+			const coverage = Object.entries(r.coverage)
+				.map(
+					([dim, buckets]) =>
+						`  ${dim.padEnd(11)} ${Object.entries(buckets)
+							.map(([b, n]) => `${b} ${n}`)
+							.join(" · ")}`
+				)
+				.join("\n");
+			return io.log(
+				[
+					`bank ${r.set}: ${r.inputs.length} of ${r.total} input(s)`,
+					...r.inputs.map(
+						(i) =>
+							`${i.id}  ${i.source.padEnd(10)} ${String(i.tags.origin ?? "").padEnd(13)} ${String(i.tags.material?.bucket ?? "").padEnd(7)} ${String(i.tags.shape?.bucket ?? "").padEnd(5)} ${i.title}`
+					),
+					"coverage (the whole bank):",
+					coverage,
+				].join("\n")
+			);
+		}
+		case "input": {
+			if (!id) throw new CliError("Usage: admin tuning input <input-id>");
+			return printResult(io, { ...flags, json: true }, await staffRequest(ctx, "GET", `${T}/inputs/${id}`));
+		}
+		case "inputs-harvest": {
+			const r = await staffRequest(
+				ctx,
+				"POST",
+				`${T}/inputs/harvest`,
+				compact({
+					n: flagNum(flags.n),
+					pool: flagNum(flags.pool),
+					callers: flagStr(flags.callers)
+						?.split(",")
+						.map((c) => c.trim())
+						.filter(Boolean),
+					since_days: flagNum(flags["since-days"]),
+					min_per_bucket: flagNum(flags["min-per-bucket"]),
+				})
+			);
+			if (json) return printResult(io, flags, r);
+			return io.log(
+				[
+					`added ${r.added.length} input(s) to ${r.set} from a pool of ${r.pool} run(s); ${r.alreadyHeld} already held`,
+					...r.added.map((a) => `  ${a.key}  (${a.reason})`),
+					...Object.entries(r.skipped).map(([why, n]) => `  skipped ${n}: ${why}`),
+				].join("\n")
+			);
+		}
+		case "input-save": {
+			if (!id) throw new CliError('Usage: admin tuning input-save <session-id> [--note "…"]');
+			const r = await staffRequest(
+				ctx,
+				"POST",
+				`${T}/sessions/${id}/bank`,
+				compact({ note: flagStr(flags.note) })
+			);
+			return json ? printResult(io, flags, r) : io.log(`bank input ${r.input.id} (${r.input.key})`);
+		}
+		// --- Screens: candidates across a slice of the bank, read against the Base, beside the noise ---
+		case "screen-new": {
+			const title = flagStr(flags.title);
+			const from = flagStr(flags.from);
+			const versions = (flagStr(flags.versions) ?? "")
+				.split(",")
+				.map((v) => Number(v.trim()))
+				.filter((v) => Number.isInteger(v) && v > 0);
+			if (!title || (!from && !(flagStr(flags.kernel) && versions.length)))
+				throw new CliError(
+					'Usage: admin tuning screen-new --title "…" (--from <session-id> | --kernel <kernel-id> --versions 23,29) [--version N] [--n 8] [--filter "origin=arena_outline;material=thin"] [--inputs <id>,…] [--repeats 1] [--lift-budget]'
+				);
+			const filter = Object.fromEntries(
+				Object.entries(tagFilterParams(flagStr(flags.filter))).map(([k, v]) => [
+					k,
+					v
+						.split(",")
+						.map((x) => x.trim())
+						.filter(Boolean),
+				])
+			);
+			const inputIds = (flagStr(flags.inputs) ?? "")
+				.split(",")
+				.map((x) => x.trim())
+				.filter(Boolean);
+			const r = await staffRequest(
+				ctx,
+				"POST",
+				`${T}/screens`,
+				compact({
+					title,
+					from_session_id: from,
+					base_kernel_id: flagStr(flags.kernel),
+					base_version: flagNum(flags.version),
+					versions: versions.length ? versions : undefined,
+					n: flagNum(flags.n),
+					filter: Object.keys(filter).length ? filter : undefined,
+					input_ids: inputIds.length ? inputIds : undefined,
+					repeats: flagNum(flags.repeats),
+					lift_budget: flagBool(flags["lift-budget"]) || undefined,
+				})
+			);
+			if (json) return printResult(io, flags, r);
+			return io.log(
+				`screen ${r.screen.id}: ${r.screen.inputs} input(s), ${r.aggregate.columns.length} column(s); ${r.screen.pending} run(s) waiting${r.screen.waiting ? ` (${r.screen.waiting})` : ""} → ${ctx.client.url}/superadmin/tuning/screens/${r.screen.id}`
+			);
+		}
+		case "screens": {
+			const r = await staffRequest(ctx, "GET", `${T}/screens?limit=${flagNum(flags.limit) ?? 30}`);
+			if (json) return printResult(io, flags, r);
+			return io.log(
+				r.screens
+					.map(
+						(x) =>
+							`${x.id}  ${x.status.padEnd(7)} ${x.title} · ${x.inputs} inputs${x.pending ? ` · ${x.pending} waiting` : ""}`
+					)
+					.join("\n") || "(no screens)"
+			);
+		}
+		case "screen": {
+			if (!id) throw new CliError("Usage: admin tuning screen <screen-id>");
+			const r = await staffRequest(ctx, "GET", `${T}/screens/${id}`);
+			if (json) return printResult(io, flags, r);
+			// The reading as an agent reads it: per candidate, each measure's difference from the Base and its noise.
+			return printResult(
+				io,
+				{ ...flags, json: true },
+				{
+					screen: r.screen,
+					note: r.aggregate.note,
+					columns: r.aggregate.columns,
+					measures: r.aggregate.measures.map((m) => ({
+						measure: m.label,
+						judge: m.judgeId,
+						standing: m.standing,
+						sigma: m.sigma,
+						per_column: m.perTrial.map((p) =>
+							compact({
+								column: p.column,
+								n: p.n,
+								mean: p.mean,
+								delta: p.delta ?? undefined,
+							})
+						),
+					})),
+				}
+			);
+		}
+		case "screen-advance": {
+			if (!id) throw new CliError("Usage: admin tuning screen-advance <screen-id>");
+			const r = await staffRequest(ctx, "POST", `${T}/screens/${id}/advance`, {});
+			return json
+				? printResult(io, flags, r)
+				: io.log(`queued ${r.queued} run(s)${r.waiting ? `; waiting: ${r.waiting}` : ""}`);
+		}
 		// --- Research: our arena's versions, research comparisons, the deploy split ---
 		case "arenas": {
 			const r = await staffRequest(ctx, "GET", `${T}/arena-files`);
@@ -1977,7 +2182,6 @@ async function cmdAdminReplays(ctx, sub) {
 	const responseId = positionals[0];
 	const target = flagStr(flags.target);
 	if (!org) throw new CliError("admin replay needs --org <org-id>");
-	if (!responseId && !target) throw new CliError("Pass a breakthrough <response-id>, or --target <target-id>");
 	const result = await staffPost(
 		ctx,
 		REPLAYS,

package/lib/commands/insights.mjs

@@ -21,6 +21,7 @@ export const ARGS = {
 	},
 	get_measure_summary: {
 		period: { flag: "period", enum: PERIODS },
+		parts: { flag: "part", type: "list" },
 	},
 };
 

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@letterstory/cli",
-	"version": "0.66.0",
+	"version": "0.69.0",
 	"description": "Spin up and manage Letterstory phantom blogs from your terminal.",
 	"type": "module",
 	"license": "UNLICENSED",