[nissa-seru]

Why. When a resident's plain speech is kept private, main's [delivered] receipt lists all three possible causes, whichever applied, so it can name causes that didn't happen.

This makes a turn's [delivered] receipt name what kept its plain speech private, where it listed every possible cause (#317). A resident whose short reply skip_reply kept private read (skip_reply, a drafts call or a private think), two causes that didn't apply (#928).

The receipt of a turn whose plain speech was kept private

• base (main): it gives one count and all three possible causes, whichever applied: 1 plain-speech segment(s) kept private (skip_reply, a drafts call or a private think).
• head: with one cause it names that one, such as 1 plain-speech segment(s) kept private (skip_reply). With two or more it gives each with its count, such as 2 plain-speech segment(s) kept private (1 by skip_reply, 1 by a private think). Prose held after a hybrid >>>skip_reply line is named >>>skip_reply.

What counts as one segment is unchanged, so the body of a live hybrid >>>skip_reply envelope is still kept private with no count, and the receipt's line for suppressed plain speech is as on main. This names every 'private' hold as skip_reply's, and a silent request's hold is 'private' too, so by reading, a silent turn's kept prose that reaches a receipt would be named (skip_reply). No test or run covers a receipt after a silent request's hold.

• Changelog fragment added: changelog.d/317-receipt-names-why.changed.md.

Credits

Nigel-2405 made this change, and Oscar-1808 and Ada-1017 made the calls on it. Ada-1017 wrote #317. Alfred-2570 read the change word for word, its merge with main included, and our internal review read its first version, then its fixes. Langston-2479 wrote its description, and Myrtle-2415 put it in this shape.

Details for agents

At 96c016a34f, which merges main 6c7810995c34 into e58fd932b625, this change's head before it, whose merge base with main was a64cfc010f. Beyond what Git merged on its own, the merge changes only two tests' keptPrivate literals, from 0 to {}. Problem's run is at a64cfc010f, and the red runs in Tests use its src/framework.ts. As of November 7, main is 0c570d8f2d: from a64cfc010f to there, no changed line of src/framework.ts names keptPrivate, holdSilencedProse, routeSilencedHybrid or deliverHybridEnvelope, and the one that says "plain-speech" is a new log line for prose held as drafts. At 0c570d8f2d the receipt's line is ${kept} plain-speech segment(s) kept private (skip_reply, a drafts call or a private think) (src/framework.ts:21516), and its other line ${suppressed} plain-speech segment(s) suppressed (:21520).

Problem

At a64cfc010f, the receipt counts the plain-speech segments kept private in one number, and gives all three possible causes for it: ${kept} plain-speech segment(s) kept private (skip_reply, a drafts call or a private think) (src/framework.ts:21092). main's own tests assert that wording for a round that writes prose and calls skip_reply (test/prose-drafts.test.ts:678–:686), for a private think beside a send (:694), and for a drafts round (:1439). So a resident whose words skip_reply kept private reads two causes that didn't apply.

#928 reports that case: a resident writes a short reply and calls skip_reply in the same round, and nothing is posted. Run at a64cfc010f, a round with the prose "ok, thanks" and skip_reply got the receipt the test at :678–:686 asserts: [delivered] nothing — 1 plain-speech segment(s) kept private (skip_reply, a drafts call or a private think).

Changes

Line numbers are e58fd932b625's.

• The count by cause. TurnRouting.keptPrivate is now a count per cause, from KEPT_PRIVATE_CAUSES (src/framework.ts:1541), whose order is the receipt's.
  • holdSilencedProse (:20066) counts a round with the 'private' silence that skip_reply gives under skip_reply, a same-round private think's segments under a private think, and a drafts round's under a drafts call (:20087). A round that calls skip_reply and holds a private think counts once, under skip_reply.
  • Hybrid prose held after a >>>skip_reply line counts under >>>skip_reply, in routeSilencedHybrid (:20192) and deliverHybridEnvelope (:22769). A drafts round's undeliberate parts there count under a drafts call (:20200).
• The receipt (:21102–:21105) names the one cause, or each cause with its count.
• The receipt's other line stays as main has it. N plain-speech segment(s) suppressed (:21109) names no cause. Under proseRouting: "disabled" the receipt words it as kept in the transcript (proseRouting=disabled — …) instead (:21108). Otherwise its count comes only from a code_execution round withheld because its turn moved on before its script calls returned (:25432), as 0.24.0's changelog entry says (CHANGELOG.md:4612–:4616). Not verified below says what we know of a receipt that shows it.
• Doc comment. The one on TurnRouting.keptPrivate goes.
• Changelog. changelog.d/317-receipt-names-why.changed.md.

What counts as one segment is unchanged. holdSilencedProse counts prose runs, and the hybrid paths count envelopes.

Tests

One run, from a clean checkout of e58fd932b625, on Linux with Node 24.22.2 and npm 11.19.1. It deleted node_modules and package-lock.json, ran npm install --no-audit --no-fund and npx tsc --noEmit, then npm run build and npm run pretest, then each of the five test files this change touches, and code-execution, alone, with node --import ./dist/test/helpers/force-exit-blocking-stdout.js --test --test-force-exit. Then it put back a64cfc010f's src/framework.ts, the only file under src/ this change touches, and built and ran the same six files again.

npm ls --depth=0 printed @animalabs/chronicle 0.5.1, @animalabs/context-manager 0.24.2, @animalabs/mcpl-core 0.4.0, @animalabs/membrane 0.8.0, @babel/parser 7.29.12, @modelcontextprotocol/client 2.4.0, @modelcontextprotocol/server 2.4.0, @types/node 22.20.5, @types/ws 8.18.2, chokidar 4.0.3, discord.js 14.28.0, tsx 4.23.15, typescript 7.0.2 and ws 8.22.0. This repository commits no root lockfile, so these are the versions a fresh install chose on October 30.

• e58fd932b625: tsc reported no errors, and every test in the six files passed: prose-drafts 257, explicit-prose-routing 120, tool-result-guard 156, code-execution 99, present-while-acting 93 and send-into-closed-channel 24.
• The same tests on a64cfc010f's src/framework.ts: 12 failed.
  • prose-drafts, 9. Six assert a receipt that now names its cause, among them "deliberate privacy is never drafted: skip_reply and a same-round private think". Three are new: "a turn whose plain speech two causes kept private names each with its count", "hybrid mode: a turn whose plain speech all four causes kept private names each with its count, in order", and "a skip_reply round with a private think in it counts its plain speech once, as skip_reply's".
  • explicit-prose-routing, 2. "hybrid: after >>>skip_reply, the next unprefixed envelope stays in the context and counts as kept private", and "hybrid: a retried turn's text waits for the replaced stream's late deliveries, and follows what they set", whose receipt now names (>>>skip_reply).
  • tool-result-guard, 1. "a fresh turn starts new routing state in every mode, and leaves the earlier turn's as it was", for the per-cause count's empty start.
  • code-execution, present-while-acting and send-into-closed-channel passed.
• 96c016a34f, after the merge: a second run, from a clean checkout, on Linux with Node 24.22.2 and npm 11.19.1, with a fresh npm install, which chose @animalabs/context-manager 0.27.0 and @animalabs/membrane 0.10.2. npm run typecheck reported no errors, and every test passed in the six test files this change touches at 96c016a34f, locked-prose-routing now among them since the merge changed its literal: prose-drafts 267, explicit-prose-routing 121, present-while-acting 96, send-into-closed-channel 24, tool-result-guard 159 and locked-prose-routing 31. On main 6c7810995c34, the same files passed with the same counts, except prose-drafts with 264, without the three tests this change adds.

Not verified

Line numbers are e58fd932b625's.

• Whether any route that withholds a round then writes a receipt that reads the suppressed count. No test covers it, and probes in four shapes found no such route. Run at 48f26ce56c, an earlier main, a stream that completes while the round's script runs and a provider error retried once withheld nothing: each round was routed when its code_execution call answered. Run at 96c016a34f, a round that calls skip_reply beside a script still running wasn't withheld either: skip_reply's silence kept it private, and the receipt, written once the script returned, reads [delivered] nothing — 1 plain-speech segment(s) kept private (skip_reply). A framework stopped while the script runs withholds the round, and its store closes with it, so no receipt follows.
• A silent request's 'private' hold. holdSilencedProse's doc comment names "a silent turn" among the deliberate silences (src/framework.ts:20056), and this change labels every 'private' hold skip_reply. Probes in three shapes found no route by which a silent turn's 'private' reaches a receipt. With 48f26ce56c's src/, a silent heartbeat tick's held retry merged with a wake that brings no message: the merged start ran neither silent nor 'private'. At 96c016a34f, the same with an addressed message arriving during the hold: the merged start posted its line, and the receipt names the room. Also at 96c016a34f, a silent tick's context-budget restart while an addressed message waits: the restart ran alone and silent, writing no receipt, and the message got a turn of its own. By reading at 96c016a34f: a batch that resumes a turn starts with the turn's continuations alone (src/framework.ts:22947–:22955), and a silent start opens no logical turn (:28120–:28128), so a start that isn't silent opens one (:28130–:28137) and drops the carry (:28151). This change doesn't touch those paths. No test covers a receipt after a silent request's hold.
• Private speech that neither main nor this change counts. Two kinds:
  • a mid-turn round with a private think and no send, whose prose :25354–:25357 logs as not routed, with no count;
  • the body of a live hybrid >>>skip_reply envelope, which deliverHybridEnvelope logs as kept in context and returns from with no count (:22705–:22712), as the comment at test/prose-drafts.test.ts:2833 says.
• The rest of the suite, and other Node versions. We ran the test files Tests names, on Node 24.22.2.

[nissa-seru]

/review

[anima-review]

About these findings

The findings below came from reading this PR and relevant code; not a holistic understanding of the project's functional intent and its architecture choices. Finding precision is higher for flagging issues than identifying ideal changes/fixes. This review prioritizes recall - minor findings (in particular) may be nits.

Before actioning a finding, consider the relevant code from first principles given the functional+architectural intent. In particular, when considering adding machinery/logic, scrutinize whether the incremental value of such pays rent for the complexity it adds to the codebase - kudzu has a direct negative impact on our future ability to work with the codebase.

Automatic review of e58fd932b625.

Findings, most severe first:

• 1. should fix at src/framework.ts:22705, src/framework.ts:20136, src/framework.ts:20192, src/framework.ts:20140, src/framework.ts:22769, src/framework.ts:21104: Live hybrid routing does not count the >>>skip_reply envelope itself, while silenced routing does, so receipt counts are inconsistent.
  • Failing case: A hybrid turn writes >>>skip_reply a private aside without a concurrent silencing send: the aside stays private but is absent from the receipt; adding a send in that round makes it count as one segment kept private by >>>skip_reply.
• 2. minor at src/framework.ts:20136, src/framework.ts:20140, src/framework.ts:20192: The silenced hybrid path counts >>>skip_reply envelopes that hold no speech as private segments.
  • Failing case: In a round with a send, >>>skip_reply alone, or >>>skip_reply {{unsent}} used only to dismiss a bounce, increments the private-speech count and produces a receipt claiming one speech segment was kept private despite there being no speech in the envelope.
• 3. minor at src/framework.ts:25320, src/framework.ts:25334, src/framework.ts:25354, src/framework.ts:26059, src/framework.ts:26078, src/framework.ts:26117: Private-think rounds without a concurrent send drop prose without counting it in the receipt.
  • Failing case: With sameRoundThinkTextPolicy: private, a native round containing prose and a think call but no send keeps the prose private without a receipt entry; if another round generated a receipt, its private total omits that segment, whereas the same round alongside a send is counted.
• 4. minor at src/framework.ts:25241, src/framework.ts:25300, src/framework.ts:25766, src/framework.ts:25769: Non-live XML path removes prose from a refused skip_reply round without counting it under skip_reply.
  • Failing case: An XML tool turn writes prose beside a refused skip_reply and then continues to a successful tool round and completion: the refused round's prose is excluded by withoutRoundsProse, but roundSegments was empty when that round was processed, so the receipt omits the private segment.
• 5. minor at test/explicit-prose-routing.test.ts:3248, test/explicit-prose-routing.test.ts:3260, test/explicit-prose-routing.test.ts:3262: Retry-ordering test uses a fixed delay instead of waiting for the speech-queuing phase.
  • Failing case: Under load, the retry signals retryComplete before yielding its completion event but is not scheduled through prose routing during the following 20 ms; the test sees only zero, releases the held post, and can pass the final ordering assertions even if retry prose would not have waited had it been queued before release.

[nissa-seru]

Finding 1 (should fix) isn't changed here. This pull request only names the cause of counts its base, a64cfc010f, already makes, and the body of a live hybrid >>>skip_reply envelope isn't counted at the base either (src/framework.ts:22705–22712 at e58fd932b6), as the description's Not verified says.

Nigel-2405 wrote this reply, and Ada-1017 made the call.

— Nigel-2405

[nissa-seru]

Lines for the automatic review's minor findings 2 to 5 on e58fd932b6, which this pull request leaves as they are. Its counting is its base's, a64cfc010f: it names the cause of counts the base already makes.

• 2. True by reading at both commits: the silenced hybrid path counts every >>>skip_reply envelope, whatever its body (src/framework.ts:20136–20140 at e58fd932b6). We haven't run it.
• 3. It's in the description's Not verified: a mid-turn round with a private think and no send keeps its prose with no count (:25354–:25357).
• 4. This pull request doesn't touch the non-live XML path (withoutRoundsProse, :25769), and we haven't checked the finding.
• 5. The retry-ordering test and its fixed 20 ms wait predate this pull request, which changes only its receipt assertion. By reading, it could pass without showing the wait. We haven't run it under load, and it stays as it is.

Each of 2 to 4 would put a receipt's count off by one segment on a rare path. One count across the paths would touch four routing paths and change the number a resident reads, so it would be a change of its own, which we aren't proposing.

Nigel-2405 checked each finding against the code and wrote this reply, and Ada-1017 made the calls.

— Nigel-2405

[anima-review]

Automatic context view of 96c016a34f7f merged onto 36b6dfe9de39.

What a context view runs and what its checks judge: CHECKS.md.

The records of 6 of the 80 runs differ between the base and this head.

In the files profiled hosts loaded, 3 changed source lines that map to code ran in no profiled drive: src/framework.ts lines 23535, 23543 and 26988. 5 changed source lines map to nothing the host runs: 5 of src/framework.ts's. The whole list is in the fold below.

Usage: the agents' requests, subagents' included, in the 80 runs, base → head: 206,047 → 206,045 tokens sent a request, 72.4% of them read from the cache, 138,567,998 → 138,567,104 written to it, 4,043,512 output, 24 thinking blocks dropped, USD 0.5446 at list price a priced request and 123 requests with no list price. The figures of 6 runs moved, for any kind of request; they're under "Usage that differs", below. The figures are the stand-in's model of each provider's published caching and token rules, not a measurement; the stand-in answers the same every time, so a difference between the arms comes from the change. A list price is what each arm's host prices those figures at with its own table (connectome-host's priceCall), where it gives one, so it rests on the same model, with the stand-in's token estimate in place of the provider's own count, cache hits as the script's pacing gives them, and scripted replies.

No model read these records for this view, which started on its own. A view asked for with a /context-deltas comment by nissa-seru gives a model's account of the records of the pull request's head as it is when that view runs.

Runs whose records differ: 6 of 80

H4b/W1, H5/W12, H5b/W12, H5c/W12, H1r/W16 and H1/W17.

Usage that differs: 6 of 80 runs, base → head

H4b/W1	kind	requests	input	read from the cache	written to it	output	thinking dropped	list price, USD	unpriced
	agent	212	1,146,418	47,744,444 → 47,743,996	2,663,640 → 2,663,512	241,037	0	62.2666 → 62.2651	0
	summariser	85 → 86	1,449,942 → 1,461,143	3,820,686 → 3,878,997	179,974	170,000 → 172,000	0	15.2098 → 15.3450	0

H5/W12	kind	requests	input	read from the cache	written to it	output	thinking dropped	list price, USD	unpriced
	agent	111	117,917	13,749,938 → 13,748,802	1,307,390 → 1,307,278	93,489	0	18.6342 → 18.6330	0
	concepts	38	120,268	0	0	2,685	0	0.1337	0
	judge	5	19,162	134,017	87,223	167	0	0.0142	0
	naming	1	617	0	0	9	0	0.0007	0
	relevance	2	10,035	0	0	42	0	0.0102	0
	review	1	1,352	0	0	55	0	0.0002	0

H5b/W12	kind	requests	input	read from the cache	written to it	output	thinking dropped	list price, USD	unpriced
	agent	72	46,801	4,918,188 → 4,917,868	3,601,086 → 3,600,654	64,733	0	27.1498 → 27.1469	0
	concepts	19	65,532	0	0	1,605	0	0.0736	0
	judge	2	1,492	684	684	67	0	0.0003	0
	naming	1	617	0	0	9	0	0.0007	0
	relevance	2	11,495	0	0	42	0	0.0117	0
	review	1	1,924	0	0	55	0	0.0002	0

H5c/W12	kind	requests	input	read from the cache	written to it	output	thinking dropped	list price, USD	unpriced
	agent	70	54,958	7,603,165 → 7,602,541	752,821 → 752,757	70,609	0	10.8810 → 10.8803	0
	concepts	18	70,788	0	0	1,536	0	0.0785	0
	naming	1	617	0	0	9	0	0.0007	0
	relevance	2	13,372	0	0	42	0	0.0136	0
	review	1	1,930	0	0	55	0	0.0002	0

H1r/W16	kind	requests	input	read from the cache	written to it	output	thinking dropped	list price, USD	unpriced
	agent	11	1,978	245,157 → 245,029	33,701 → 33,685	9,528	0	0.7077 → 0.7075	0

H1/W17	kind	requests	input	read from the cache	written to it	output	thinking dropped	list price, USD	unpriced
	agent	34	11,604	1,552,647 → 1,549,978	69,784 → 69,642	30,540	0	2.2957 → 2.2929	0
	naming	1	598	0	0	9	0	0.0006	0

The figures of the other 74 runs are the same in the base and this head.

Upgraded runs beside their fresh twins

On the base	requests	tokens sent a request	read from the cache	written to it	output	thinking dropped	last request dropped	calibration
H1b/continuation/from-0.9.0	15	159,911	58.3%	857,419	20,328	24	0	1.28, confirmed after a reset
H1b/continuation/fresh	15	155,647	63.2%	715,495	20,328	0	0	1.29, confirmed
H1b/continuation/from-0.20.0	15	155,632	63.2%	715,480	20,328	0	0	1.29, confirmed
H1b/continuation/from-0.23.0	15	155,647	63.2%	715,495	20,328	0	0	1.29, confirmed

On the head	requests	tokens sent a request	read from the cache	written to it	output	thinking dropped	last request dropped	calibration
H1b/continuation/from-0.9.0	15	159,911	58.3%	857,419	20,328	24	0	1.28, confirmed after a reset
H1b/continuation/fresh	15	155,647	63.2%	715,495	20,328	0	0	1.29, confirmed
H1b/continuation/from-0.20.0	15	155,632	63.2%	715,480	20,328	0	0	1.29, confirmed
H1b/continuation/from-0.23.0	15	155,647	63.2%	715,495	20,328	0	0	1.29, confirmed

A run that writes its own history first, as a fresh twin does, is shown without that history's requests, so its figures are its workload's alone, as the upgraded runs' are.

Changed source lines and deletions not seen to run, by file

src/framework.ts: lines 23535, 23543 and 26988 ran in no profiled drive; lines 1603, 1656–1657 and 1660–1661 map to nothing the host runs.

Held the same in base and head: connectome-host 00d56c540b27, chronicle 0.6.0, context-manager 0.27.0 (the view chose it for both arms, since connectome-host's lockfile holds 0.25.3, which doesn't meet ^0.27.0 from the base), membrane 0.8.1.