8cda.6.6 left two telegram.Gateway seams nil on purpose, both blocked on the same missing fact: no contract lets one process learn another component's real running version.
RunningVersions: turns an installer's claim into a checked one. Only a component's own compiled-in build can vouch for itself. The Messaging Service knows neither archied's build nor the observed archie-agent version (daemon.AgentStatus); the Gateway's gatewayVersion is a different binary's stamp, so reporting it as the daemon's manufactures the false success the check exists to catch. Update reports currently relay as unverified claims.
ReleaseAnnouncements: releaseannounce.Announcer announces nothing unless each Component carries a parseable version (hasReleasedComponent), so wiring it without trustworthy versions is a no-op that reads as configured.
Consequence: internal/releaseannounce now has no production consumer at all (the daemon's wiring went in 8cda.6.4). Keep or delete is part of this decision, not a separate cleanup.
Likely shape: a component-version report on the contract each service already exposes, carrying (component id, version, install type) that the reporting process compiled in or directly observed. ChatSnapshot.Version (added in 8cda.6.6) is the rendered display block and is deliberately not that contract.
Decision record: docs/architecture/migration-decisions.md, 'Telegram operator surface after extraction'.
Notes
ASSESSMENT (session 01a0c4d7, 2026-09-22) -- scoped from the code, not implemented.
The orphan is real and confirmed: internal/releaseannounce holds one file and has zero production consumers, so keep-or-delete is genuinely undecided rather than implied.
The shape this bead proposes is consistent with the code: ChatSnapshot.Version exists and is a rendered display block; daemonRunningVersions in internal/app/archied/chat_setup.go exists precisely so a process vouches only for what it compiled in, and its comment already says the agent version is the agent's own report, not archied's stamp.
THE MISSING DECISIONS, named:
Which contract carries the report. A new field on the Snapshot each service already serves, or a new RPC -- and, more importantly, which value each component reports for ITSELF rather than for its neighbours, since that is the property that makes the report trustworthy.
What makes a relayed update report VERIFIED: compared against the reporting component's own build, the installer's claim, or the observed agent status. daemonRunningVersions is the existing precedent for the first.
Keep or delete internal/releaseannounce.
RECOMMENDED ORDER: decide 3 first. It is a delete-or-keep with no design in it, and it removes the only consumer question in the bead before any contract is designed.
SCOPING ONLY -- not started, per the release manager.
DELETE QUESTION SETTLED FROM THE EVIDENCE: KEEP. It is not dead code resting on a superseded design; it is unwired code with a named blocker and a stated intent to restore it, and this bead's own title is that statement.
EVIDENCE
The intent is recorded twice and is explicit. docs/architecture/migration-decisions.md's telegram operator-surface table says ReleaseAnnouncements is "Deliberately left nil ... Wiring the announcer without them yields a no-op that reads as configured, so it stays unwired UNTIL COMPONENT SELF-REPORTING EXISTS". The blocker is named as the subject of this bead, not as a lost reason.
The implementation is complete and tested, not a stub: internal/releaseannounce (287 lines) parses versioned CHANGELOG sections, tracks per-recipient announced state, detects changed components and sends one-time upgrade notes; 230 lines of tests cover it, including the version comparison and the changelog section parse.
The consumer seam is intact and tested: telegram.Gateway.ReleaseAnnouncements (internal/channels/telegram/telegram.go) with its own announce call path and TestReleaseAnnouncementTargetsAuthorizedUsers. Deleting the announcer would leave that seam with no implementation at all.
The one argument I expected to make for keeping it does NOT hold, and that is worth recording: the CHANGELOG heading format is no longer dependent on this package. tools/newsgen parses the same version headings and IS wired (it publishes docs/news and the per-version release notes), so the format keeps its live consumer either way. Deleting releaseannounce would cost the feature, not the contract.
WHAT DELETION WOULD ACTUALLY DECIDE
Dropping the feature -- that authorized Telegram users stop receiving one-time upgrade notes -- which is a product decision no evidence here supports. The register's wording, this bead's title and the intact seam all point the other way. So this is not a cleanup; it is a removal of wanted behaviour.
TWO DECISIONS THE IMPLEMENTATION FACES, neither of which is deletion
The version contract, which is the other half of this bead: the announcer announces nothing unless each Component carries a parseable version (hasReleasedComponent), and only a component's own build can vouch for itself.
There is NO operator switch today: grep finds no announce key in internal/config or config.example.toml, and once wired the telegram gateway announces to every authorized user on a version change with no way to opt out. If an opt-out is wanted it is a new [chat.telegram] key, which means it must also survive the control-plane round trip -- the tool-settings projection trap, where a field added to one site is silently dropped when a control plane is configured.
RECOMMENDATION: keep the package, and treat "keep or delete" as closed. The remaining work is the version contract plus, if wanted, an opt-out with its control-plane projection.
DECISION (2026-09-22): KEEP - do NOT delete internal/releaseannounce. Recorded from the assessment: it is not dead code resting on a superseded design, it is UNWIRED code with a named blocker and a stated intent to restore it, and this bead's title IS that statement (register: ReleaseAnnouncements 'Deliberately left nil ... until component self-reporting exists'). 287 lines of complete, tested implementation plus a tested telegram seam; deleting the announcer would leave that seam with no implementation. NOTE the justification that was checked and does NOT hold: the CHANGELOG heading format is no longer dependent on this package, because tools/newsgen parses the same headings and IS wired (it publishes docs/news), so the format keeps a live consumer either way - deleting would cost the FEATURE, not the contract. Deletion would decide that authorised Telegram users stop receiving one-time upgrade notes, which no evidence supports.
REMAINING DECISIONS (neither is deletion):
The version contract: the announcer announces nothing unless each Component carries a parseable version, and only a component's own build can vouch for itself.
THERE IS NO OPERATOR SWITCH. No announce key exists in internal/config or config.example.toml, so once wired the gateway announces to every authorised user on every version change with no opt-out. If an opt-out is wanted it is a new [chat.telegram] key, and it must survive the control-plane round trip - i.e. the tool-settings projection trap where a field added at one site is silently dropped whenever a control plane is configured. Verify the consumer exists before adding the key (the repo rejects a config field nothing reads).
Close reason
RunningVersions reads the daemon version from archied's own presence record (agent stays unverified). ReleaseAnnouncements wired in archie-messaging from its own build and a compiled-in CHANGELOG.md; announcer accepts v-prefixed and bare versions. No opt-out key added.
Imported from beads archie-core-8cda.6.9 · GitHub #923 · type task · priority P2 · created 2026-09-19 · closed 2026-10-03
8cda.6.6 left two telegram.Gateway seams nil on purpose, both blocked on the same missing fact: no contract lets one process learn another component's real running version.
- RunningVersions: turns an installer's claim into a checked one. Only a component's own compiled-in build can vouch for itself. The Messaging Service knows neither archied's build nor the observed archie-agent version (daemon.AgentStatus); the Gateway's gatewayVersion is a different binary's stamp, so reporting it as the daemon's manufactures the false success the check exists to catch. Update reports currently relay as unverified claims.
- ReleaseAnnouncements: releaseannounce.Announcer announces nothing unless each Component carries a parseable version (hasReleasedComponent), so wiring it without trustworthy versions is a no-op that reads as configured.
Consequence: internal/releaseannounce now has no production consumer at all (the daemon's wiring went in 8cda.6.4). Keep or delete is part of this decision, not a separate cleanup.
Likely shape: a component-version report on the contract each service already exposes, carrying (component id, version, install type) that the reporting process compiled in or directly observed. ChatSnapshot.Version (added in 8cda.6.6) is the rendered display block and is deliberately not that contract.
Decision record: docs/architecture/migration-decisions.md, 'Telegram operator surface after extraction'.
### Notes
ASSESSMENT (session 01a0c4d7, 2026-09-22) -- scoped from the code, not implemented.
The orphan is real and confirmed: internal/releaseannounce holds one file and has zero production consumers, so keep-or-delete is genuinely undecided rather than implied.
The shape this bead proposes is consistent with the code: ChatSnapshot.Version exists and is a rendered display block; daemonRunningVersions in internal/app/archied/chat_setup.go exists precisely so a process vouches only for what it compiled in, and its comment already says the agent version is the agent's own report, not archied's stamp.
THE MISSING DECISIONS, named:
1. Which contract carries the report. A new field on the Snapshot each service already serves, or a new RPC -- and, more importantly, which value each component reports for ITSELF rather than for its neighbours, since that is the property that makes the report trustworthy.
2. What makes a relayed update report VERIFIED: compared against the reporting component's own build, the installer's claim, or the observed agent status. daemonRunningVersions is the existing precedent for the first.
3. Keep or delete internal/releaseannounce.
RECOMMENDED ORDER: decide 3 first. It is a delete-or-keep with no design in it, and it removes the only consumer question in the bead before any contract is designed.
SCOPING ONLY -- not started, per the release manager.
DELETE QUESTION SETTLED FROM THE EVIDENCE: KEEP. It is not dead code resting on a superseded design; it is unwired code with a named blocker and a stated intent to restore it, and this bead's own title is that statement.
EVIDENCE
1. The intent is recorded twice and is explicit. docs/architecture/migration-decisions.md's telegram operator-surface table says ReleaseAnnouncements is "Deliberately left nil ... Wiring the announcer without them yields a no-op that reads as configured, so it stays unwired UNTIL COMPONENT SELF-REPORTING EXISTS". The blocker is named as the subject of this bead, not as a lost reason.
2. The implementation is complete and tested, not a stub: internal/releaseannounce (287 lines) parses versioned CHANGELOG sections, tracks per-recipient announced state, detects changed components and sends one-time upgrade notes; 230 lines of tests cover it, including the version comparison and the changelog section parse.
3. The consumer seam is intact and tested: telegram.Gateway.ReleaseAnnouncements (internal/channels/telegram/telegram.go) with its own announce call path and TestReleaseAnnouncementTargetsAuthorizedUsers. Deleting the announcer would leave that seam with no implementation at all.
4. The one argument I expected to make for keeping it does NOT hold, and that is worth recording: the CHANGELOG heading format is no longer dependent on this package. tools/newsgen parses the same version headings and IS wired (it publishes docs/news and the per-version release notes), so the format keeps its live consumer either way. Deleting releaseannounce would cost the feature, not the contract.
WHAT DELETION WOULD ACTUALLY DECIDE
Dropping the feature -- that authorized Telegram users stop receiving one-time upgrade notes -- which is a product decision no evidence here supports. The register's wording, this bead's title and the intact seam all point the other way. So this is not a cleanup; it is a removal of wanted behaviour.
TWO DECISIONS THE IMPLEMENTATION FACES, neither of which is deletion
1. The version contract, which is the other half of this bead: the announcer announces nothing unless each Component carries a parseable version (hasReleasedComponent), and only a component's own build can vouch for itself.
2. There is NO operator switch today: grep finds no announce key in internal/config or config.example.toml, and once wired the telegram gateway announces to every authorized user on a version change with no way to opt out. If an opt-out is wanted it is a new [chat.telegram] key, which means it must also survive the control-plane round trip -- the tool-settings projection trap, where a field added to one site is silently dropped when a control plane is configured.
RECOMMENDATION: keep the package, and treat "keep or delete" as closed. The remaining work is the version contract plus, if wanted, an opt-out with its control-plane projection.
DECISION (2026-09-22): KEEP - do NOT delete internal/releaseannounce. Recorded from the assessment: it is not dead code resting on a superseded design, it is UNWIRED code with a named blocker and a stated intent to restore it, and this bead's title IS that statement (register: ReleaseAnnouncements 'Deliberately left nil ... until component self-reporting exists'). 287 lines of complete, tested implementation plus a tested telegram seam; deleting the announcer would leave that seam with no implementation. NOTE the justification that was checked and does NOT hold: the CHANGELOG heading format is no longer dependent on this package, because tools/newsgen parses the same headings and IS wired (it publishes docs/news), so the format keeps a live consumer either way - deleting would cost the FEATURE, not the contract. Deletion would decide that authorised Telegram users stop receiving one-time upgrade notes, which no evidence supports.
REMAINING DECISIONS (neither is deletion):
1. The version contract: the announcer announces nothing unless each Component carries a parseable version, and only a component's own build can vouch for itself.
2. THERE IS NO OPERATOR SWITCH. No announce key exists in internal/config or config.example.toml, so once wired the gateway announces to every authorised user on every version change with no opt-out. If an opt-out is wanted it is a new [chat.telegram] key, and it must survive the control-plane round trip - i.e. the tool-settings projection trap where a field added at one site is silently dropped whenever a control plane is configured. Verify the consumer exists before adding the key (the repo rejects a config field nothing reads).
### Close reason
RunningVersions reads the daemon version from archied's own presence record (agent stays unverified). ReleaseAnnouncements wired in archie-messaging from its own build and a compiled-in CHANGELOG.md; announcer accepts v-prefixed and bare versions. No opt-out key added.
---
_Imported from beads `archie-core-8cda.6.9` · GitHub #923 · type task · priority P2 · created 2026-09-19 · closed 2026-10-03_
sam
added the mediumtask labels 2026-10-27 13:35:01 +00:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
8cda.6.6 left two telegram.Gateway seams nil on purpose, both blocked on the same missing fact: no contract lets one process learn another component's real running version.
Consequence: internal/releaseannounce now has no production consumer at all (the daemon's wiring went in 8cda.6.4). Keep or delete is part of this decision, not a separate cleanup.
Likely shape: a component-version report on the contract each service already exposes, carrying (component id, version, install type) that the reporting process compiled in or directly observed. ChatSnapshot.Version (added in 8cda.6.6) is the rendered display block and is deliberately not that contract.
Decision record: docs/architecture/migration-decisions.md, 'Telegram operator surface after extraction'.
Notes
ASSESSMENT (session 01a0c4d7, 2026-09-22) -- scoped from the code, not implemented.
The orphan is real and confirmed: internal/releaseannounce holds one file and has zero production consumers, so keep-or-delete is genuinely undecided rather than implied.
The shape this bead proposes is consistent with the code: ChatSnapshot.Version exists and is a rendered display block; daemonRunningVersions in internal/app/archied/chat_setup.go exists precisely so a process vouches only for what it compiled in, and its comment already says the agent version is the agent's own report, not archied's stamp.
THE MISSING DECISIONS, named:
RECOMMENDED ORDER: decide 3 first. It is a delete-or-keep with no design in it, and it removes the only consumer question in the bead before any contract is designed.
SCOPING ONLY -- not started, per the release manager.
DELETE QUESTION SETTLED FROM THE EVIDENCE: KEEP. It is not dead code resting on a superseded design; it is unwired code with a named blocker and a stated intent to restore it, and this bead's own title is that statement.
EVIDENCE
WHAT DELETION WOULD ACTUALLY DECIDE
Dropping the feature -- that authorized Telegram users stop receiving one-time upgrade notes -- which is a product decision no evidence here supports. The register's wording, this bead's title and the intact seam all point the other way. So this is not a cleanup; it is a removal of wanted behaviour.
TWO DECISIONS THE IMPLEMENTATION FACES, neither of which is deletion
RECOMMENDATION: keep the package, and treat "keep or delete" as closed. The remaining work is the version contract plus, if wanted, an opt-out with its control-plane projection.
DECISION (2026-09-22): KEEP - do NOT delete internal/releaseannounce. Recorded from the assessment: it is not dead code resting on a superseded design, it is UNWIRED code with a named blocker and a stated intent to restore it, and this bead's title IS that statement (register: ReleaseAnnouncements 'Deliberately left nil ... until component self-reporting exists'). 287 lines of complete, tested implementation plus a tested telegram seam; deleting the announcer would leave that seam with no implementation. NOTE the justification that was checked and does NOT hold: the CHANGELOG heading format is no longer dependent on this package, because tools/newsgen parses the same headings and IS wired (it publishes docs/news), so the format keeps a live consumer either way - deleting would cost the FEATURE, not the contract. Deletion would decide that authorised Telegram users stop receiving one-time upgrade notes, which no evidence supports.
REMAINING DECISIONS (neither is deletion):
Close reason
RunningVersions reads the daemon version from archied's own presence record (agent stays unverified). ReleaseAnnouncements wired in archie-messaging from its own build and a compiled-in CHANGELOG.md; announcer accepts v-prefixed and bare versions. No opt-out key added.
Imported from beads
archie-core-8cda.6.9· GitHub #923 · type task · priority P2 · created 2026-09-19 · closed 2026-10-03