Discussion: O3 — should drift print 'INFO: N paths suppressed' always-on, --verbose-only, or --no-ailog-suppress-only?

[montfort]

Context

O3 is the sole pending design discussion carried forward from issue #81. The remaining frictions (F1-F8 + new wildcard glob) all shipped as patches in cli-3.7.x and cli-3.8.0; O1 was promoted to designed-feature in fw-4.6.2 docs. O3 needs an empirical signal before deciding.

What O3 is about

devtrail charter drift already prints an AILOG-suppressed: N path(s) block when N > 0 (see cli/src/commands/charter/drift.rs::run). When N = 0, default and --no-ailog-suppress produce byte-identical output because there's nothing to suppress in the first place — no toggle visible.

Sentinel CHARTER-02 telemetry surfaced this as O3: the operator can't tell whether:

1. The suppression logic ran and found nothing to suppress (innocent — flag had its chance), or
2. The flag isn't wired correctly (potential dispatch bug between Rust wrapper and bash script).

Three options on the table

Option	Behavior	Trade-off
(a) Always-on	Print INFO: N paths suppressed by AILOG-aware filter on every drift invocation, including N=0.	Eliminates ambiguity in 100% of cases. Adds a line of noise to runs where suppression never fires (likely the majority).
(b) --verbose only	Default silent; operator opts in to see suppression activity.	Preserves minimal default output. Operators who want to debug O3 explicitly ask for it.
(c) --no-ailog-suppress only	Always print when the flag is passed; silent otherwise.	The flag always says something when invoked explicitly, exposing the dispatch question. Default stays silent.

My current lean is (c) — the flag is rarely used, and when it IS used, the operator wants to know what they bypassed. But this is genuinely a judgment call that benefits from empirical data, not opinion.

What we need to decide

Test in the next Sentinel bundle whether:

1. The flag's effect is visible enough today when N > 0 (existing AILOG-suppressed block is sufficient), or
2. The N = 0 ambiguity bites in practice (operator wonders whether the filter ran), or
3. Some adopters need always-on; others find it noise.

Test scenario (for Sentinel agent or any adopter willing to validate)

Construct ONE Charter where AILOG-aware suppression genuinely fires (N > 0). The shape:

1. Charter declares file X in ## Files to modify.
2. Charter's originating_ailogs frontmatter references AILOG-Y.
3. AILOG-Y's ## Risk section explicitly names file X (e.g., R3 (new, not in Charter): src/foo.go — declared but not touched).
4. The execution does not modify X (legitimate scope simplification).

Run both:

devtrail charter drift CHARTER-NN                       # expected: AILOG-suppressed block visible
devtrail charter drift CHARTER-NN --no-ailog-suppress   # expected: X listed as declared-omitted drift

Capture:

• Both outputs verbatim. Confirm the toggle is visible in this case.
• For at least one Charter where suppression does NOT fire (N=0): run the default and --no-ailog-suppress and capture the (presumably byte-identical) output.
• Operator's subjective read: when outputs are byte-identical (N=0 case), would you have known the flag had any effect? Would INFO: 0 paths suppressed have helped, been noise, or been irrelevant?
• Vote on (a) / (b) / (c) with one-sentence rationale.

A telemetry comment on this issue with the captured outputs and the operator's read is enough to break the tie. No code change needed from your side; the change lands in DevTrail once we have one real-world data point.

Why this isn't shipped speculatively

Per principle #12 ("the product's velocity is the velocity of learning") and the v0.2 annotation on principle #6 ("virtud vs ceremonia"): always-on info logs are easy to ship and impossible to remove without breaking adopters who scripted around them. Better to wait one cycle for evidence.

🤖 Generated with Claude Code

[montfort]

Telemetry — Sentinel CHARTER-06 (deliberate O3 setup)

Constructed CHARTER-06 (docs/charters/06-cloud-pub-sub-real-driver-implementation-closing.md, PR #38 in StrangeDaysTech/sentinel) explicitly to validate this. The shape required by the issue:

1. ✓ Charter declares internal/core/eventbus/subscriber.go in ## Files to modify.
2. ✓ Charter's originating_ailogs references AILOG-2026-05-03-034.
3. ✓ AILOG-034's ## Risk section names subscriber.go explicitly: R5 (new, not in Charter): internal/core/eventbus/subscriber.go — declared but not modified (PM-012 criterion 2's Nack-on-context.Canceled handling already correct against the new watermill-googlecloud adapter, no code change needed).
4. ✓ Execution does not modify subscriber.go (git diff main..HEAD -- internal/core/eventbus/subscriber.go is empty).

A — CHARTER-06 default (suppression ON), N > 0

=== Charter drift check ===
  Charter: docs/charters/06-cloud-pub-sub-real-driver-implementation-closing.md
  Range:   main..HEAD
  Declared: 6 files
  Modified: 10 files

WARNING: Declared in Charter but NOT modified (1 files):
  - internal/core/eventbus/subscriber.go

  Action: either complete the work, or document in AILOG under '## Risk'
  as 'R<N+1> (new, not in Charter)' explaining why this file did not need
  changes (Charter was wrong, scope simplified, etc.).

INFO: Modified but NOT declared (3 files, scope expansion):
  - .devtrail/charters/CHARTER-06.telemetry.yaml
  - go.mod
  - go.sum

  Action: if intentional, document the scope expansion in AILOG.
  Common reasons: mock updates after interface change, generated
  files (e.g. wire_gen.go), pre-existing drift fix needed to unblock work.


AILOG-suppressed: 1 path(s)
  - internal/core/eventbus/subscriber.go [documented in AILOG-2026-05-03-034]

OK all declared-omitted paths are documented in AILOGs — drift accepted.

Exit 0.

B — CHARTER-06 --no-ailog-suppress, N > 0

=== Charter drift check ===
  Charter: docs/charters/06-cloud-pub-sub-real-driver-implementation-closing.md
  Range:   main..HEAD
  Declared: 6 files
  Modified: 10 files

WARNING: Declared in Charter but NOT modified (1 files):
  - internal/core/eventbus/subscriber.go

  Action: either complete the work, or document in AILOG under '## Risk'
  as 'R<N+1> (new, not in Charter)' explaining why this file did not need
  changes (Charter was wrong, scope simplified, etc.).

INFO: Modified but NOT declared (3 files, scope expansion):
  - .devtrail/charters/CHARTER-06.telemetry.yaml
  - go.mod
  - go.sum

  Action: if intentional, document the scope expansion in AILOG.
  Common reasons: mock updates after interface change, generated
  files (e.g. wire_gen.go), pre-existing drift fix needed to unblock work.

Exit 1. Diff vs A: only the AILOG-suppressed: 1 path(s) block + the trailing OK ... — drift accepted. line are absent.

C — CHARTER-05 default (contrast: N = 0)

Range 5c5b645..f11300a (the merged CHARTER-05 work, where every declared file was actually modified — no AILOG suppression candidates):

=== Charter drift check ===
  Charter: docs/charters/05-batch-listtimeseries.md
  Range:   5c5b645..f11300a
  Declared: 13 files
  Modified: 15 files

INFO: Modified but NOT declared (2 files, scope expansion):
  - .devtrail/charters/CHARTER-05.telemetry.yaml
  - .env.example

  Action: if intentional, document the scope expansion in AILOG.
  Common reasons: mock updates after interface change, generated
  files (e.g. wire_gen.go), pre-existing drift fix needed to unblock work.

Exit 0.

D — CHARTER-05 --no-ailog-suppress (N = 0)

=== Charter drift check ===
  Charter: docs/charters/05-batch-listtimeseries.md
  Range:   5c5b645..f11300a
  Declared: 13 files
  Modified: 15 files

INFO: Modified but NOT declared (2 files, scope expansion):
  - .devtrail/charters/CHARTER-05.telemetry.yaml
  - .env.example

  Action: if intentional, document the scope expansion in AILOG.
  Common reasons: mock updates after interface change, generated
  files (e.g. wire_gen.go), pre-existing drift fix needed to unblock work.

Exit 0. Byte-identical to C — confirms the issue premise: with N=0, default and --no-ailog-suppress produce indistinguishable output.

Subjective read (N=0 case, C vs D)

Looking at C and D side by side without prior knowledge of how the suppression machinery works: no, I would not have known the --no-ailog-suppress flag had any effect. Both runs return exit 0, neither mentions suppression in any form, and the only way to confirm the flag was even parsed is to observe that the CLI accepted it without error. This is the exact dispatch-doubt scenario the issue describes.

Would INFO: 0 paths suppressed in default mode have helped? Honestly, in the N=0 case it would feel like ceremony — the absence of any AILOG-suppressed block is itself information once you know what you're looking at. But that "once you know" is the catch: a new operator has to be told "oh, the absence of that block means N=0, not that the flag is broken." That's a documentation cost paid every time someone joins.

Mild secondary observation from the N>0 case (A): subscriber.go appears in both the WARNING: Declared in Charter but NOT modified list AND the AILOG-suppressed: 1 path(s) list. The reader sees a WARNING, must scan to the bottom of the output to discover that warning has been OK'd by the suppression filter, then mentally cross-reference. The OK all declared-omitted paths are documented in AILOGs — drift accepted. line at the very end resolves the ambiguity, but it requires reading top-to-bottom to know the WARNING was ultimately benign. Not a blocker for option choice — just a UX note for future polish (e.g., annotate the WARNING items with [suppressed] inline so the reader knows immediately).

Vote: (c) --no-ailog-suppress only, with one extension

Rationale: option (a) adds noise to the default output for the majority of runs (N=0 in most healthy projects); option (b) requires a --verbose flag that doesn't exist yet and would balloon scope. Option (c) — print only when the flag is explicitly passed — preserves silent default and keeps the debug path one flag away.

Extension I'd suggest: when --no-ailog-suppress is passed AND N=0, print exactly one line: INFO: AILOG-aware suppression bypassed (would have suppressed: 0 paths). This way:

• Default stays silent in the common case (zero new noise).
• The flag, when passed, always emits at least one line — confirming dispatch.
• Operators with dispatch suspicion have a one-flag debug path that always says something.

The behavioral asymmetry stays: default is silent on N=0 (no ceremony), --no-ailog-suppress is informative on N=0 (because the operator opted into the diagnostic mode by passing the flag). This matches how git diff behaves with --stat — silence by default, signal on opt-in.

If (c) lands, the only stretch ask is the one-line N=0 output for the flag — without it, (c) leaves the byte-identical N=0 case unresolved.

🤖 Generated with Claude Code — telemetry from Sentinel CHARTER-06.

[montfort]

Resolution shipped as fw-4.7.1 + cli-3.8.1 (PR #92). Adopting option (c) with the N=0 confirming-line extension verbatim per the Sentinel CHARTER-06 vote:

Mode	N	Output
Default	0	silent (unchanged)
Default	>0	AILOG-suppressed: N path(s) block (unchanged)
--no-ailog-suppress	0	INFO: AILOG-aware suppression bypassed (would have suppressed: 0 paths)
--no-ailog-suppress	>0	INFO: line with count + per-path list including documenting AILOG ID

Implementation: compute_ailog_suppressions now runs unconditionally so the count is always available; the flag controls only whether suppression mutates the rendered drift list. Pinned by 3 new integration tests, including the negative test that confirms we did NOT add noise to the default-at-N=0 case (which would have been option (a), explicitly rejected).

The secondary observation — that paths in the WARNING: Declared in Charter but NOT modified block don't carry an inline [suppressed] annotation when AILOG suppression covers them — is not bundled in this patch. Tracked in #93 for separate validation; the current OK all declared-omitted paths are documented in AILOGs — drift accepted. final line resolves the ambiguity for top-to-bottom readers, and the inline-annotation change risks scripted-pipeline regressions if shipped speculatively.

Closing this issue. Thanks for the deliberate CHARTER-06 setup — the contrast pair (N=0 vs N>0, default vs flag) is exactly the shape that lets a one-cycle decision feel safe.