feat(enrich): confidence score in workflow result + configurable thresholds (bookshelf-rppy) #908

Merged

zombor merged 4 commits from bd-bookshelf-rppy into main 2026-07-03 21:06:40 +00:00

Conversation 10 Commits 4 Files changed 11 +919 -71

zombor commented 2026-07-03 18:24:46 +00:00

Owner

Copy link

Summary

• LowConfidenceError struct added to internal/books carrying book_id, confidence, threshold, book_type, winner_provider so the wfengine activity adapter can surface them in the EnrichResult payload (rather than an opaque error string).
• GetConfidenceThresholds curried function reads ENRICH_EBOOK_CONFIDENCE_THRESHOLD / ENRICH_COMIC_CONFIDENCE_THRESHOLD from app_settings, defaulting to DefaultEbookConfidenceThreshold=0.5 / DefaultComicConfidenceThreshold=0.5 when unset or invalid (validated 0.0–1.0). Wired through RefetchMetadata → applyConfidenceGate.
• EnrichWorkflow return type changed from error to (EnrichResult, error); low-confidence skips become successful results with SkippedLowConfidence *ConfidenceSkipDetails populated instead of permanent errors.
• boundedFanOutWithResult[T, R any] generic added to fanout.go for (R, error) sub-workflows; BoundedFanOutEnrichInput routes through it.
• 100% coverage: LowConfidenceError.Error(), GetConfidenceThresholds all branches, resolveConfidenceThresholds success + error path via RefetchMetadata, boundedFanOutWithResult empty/clamp/alreadyExists paths, boundedFanOut onItemError callback via BoundedFanOutLLMSweepInput.

Test plan

• make test passes (all packages)
• make coverage passes — check-coverage: OK — zero uncovered statement blocks
• golangci-lint run ./internal/books/... ./internal/wfengine/... — 0 issues
• New tests cover LowConfidenceError.Error(), GetConfidenceThresholds all branches, resolveConfidenceThresholds error path, boundedFanOutWithResult empty/concurrency-clamp/alreadyExists, BoundedFanOutLLMSweepInput onItemError

Closes bead bookshelf-rppy on merge.

## Summary - `LowConfidenceError` struct added to `internal/books` carrying `book_id`, `confidence`, `threshold`, `book_type`, `winner_provider` so the wfengine activity adapter can surface them in the `EnrichResult` payload (rather than an opaque error string). - `GetConfidenceThresholds` curried function reads `ENRICH_EBOOK_CONFIDENCE_THRESHOLD` / `ENRICH_COMIC_CONFIDENCE_THRESHOLD` from `app_settings`, defaulting to `DefaultEbookConfidenceThreshold=0.5` / `DefaultComicConfidenceThreshold=0.5` when unset or invalid (validated 0.0–1.0). Wired through `RefetchMetadata` → `applyConfidenceGate`. - `EnrichWorkflow` return type changed from `error` to `(EnrichResult, error)`; low-confidence skips become successful results with `SkippedLowConfidence *ConfidenceSkipDetails` populated instead of permanent errors. - `boundedFanOutWithResult[T, R any]` generic added to `fanout.go` for `(R, error)` sub-workflows; `BoundedFanOutEnrichInput` routes through it. - 100% coverage: `LowConfidenceError.Error()`, `GetConfidenceThresholds` all branches, `resolveConfidenceThresholds` success + error path via `RefetchMetadata`, `boundedFanOutWithResult` empty/clamp/alreadyExists paths, `boundedFanOut` onItemError callback via `BoundedFanOutLLMSweepInput`. ## Test plan - [x] `make test` passes (all packages) - [x] `make coverage` passes — `check-coverage: OK — zero uncovered statement blocks` - [x] `golangci-lint run ./internal/books/... ./internal/wfengine/...` — 0 issues - [x] New tests cover `LowConfidenceError.Error()`, `GetConfidenceThresholds` all branches, `resolveConfidenceThresholds` error path, `boundedFanOutWithResult` empty/concurrency-clamp/alreadyExists, `BoundedFanOutLLMSweepInput` onItemError Closes bead bookshelf-rppy on merge.

recompute-match-score-kebab-open.png

109 KiB

wf-detail-older-execution.png

55 KiB

recompute-match-score-kebab-open.png

109 KiB

wf-detail-older-execution.png

56 KiB

wf-detail-older-execution.png

55 KiB

recompute-match-score-kebab-open.png

109 KiB

wf-detail-older-execution.png

56 KiB

recompute-match-score-kebab-open.png

109 KiB

zombor added 1 commit 2026-07-03 18:24:46 +00:00

feat(enrich): surface confidence score in workflow result + configurable thresholds (bookshelf-rppy) ... 1b316c5dc2

- Add LowConfidenceError struct carrying book_id, confidence, threshold,
  book_type, winner_provider so the wfengine activity can map them into
  the structured EnrichResult payload instead of an opaque error string
- Add GetConfidenceThresholds curried function reading ebook/comic thresholds
  from app_settings (ENRICH_EBOOK_CONFIDENCE_THRESHOLD /
  ENRICH_COMIC_CONFIDENCE_THRESHOLD), falling back to the DefaultXxx consts
- Thread getConfidenceThresholds through RefetchMetadata → applyConfidenceGate
  replacing the hardcoded DefaultEbookConfidenceThreshold const
- Wire GetConfidenceThresholds in wire.go and build_extended_deps.go
- Add EnrichResult / ConfidenceSkipDetails structs in wfengine; EnrichWorkflow
  now returns (EnrichResult, error); low-confidence skips become successful
  results with SkippedLowConfidence populated instead of permanent errors
- Add boundedFanOutWithResult generic variant for (R, error) sub-workflows;
  BoundedFanOutEnrichInput routes through it
- 100% coverage: tests for LowConfidenceError.Error(), GetConfidenceThresholds
  all branches, resolveConfidenceThresholds error path via RefetchMetadata,
  and boundedFanOutWithResult empty/clamp/alreadyExists/onItemError paths

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

forgejo-actions commented 2026-07-03 18:27:25 +00:00

Copy link

Recompute Match Score — kebab open screenshot (recompute-match-score-kebab-open)

**Recompute Match Score — kebab open screenshot** (recompute-match-score-kebab-open) 

forgejo-actions commented 2026-07-03 18:27:51 +00:00

Copy link

Workflow Detail page screenshot (wf-detail-older-execution)

Older completed ContinueAsNew epoch detail — execution ID and state visible, Cancel absent.

**Workflow Detail page screenshot** (wf-detail-older-execution) Older completed ContinueAsNew epoch detail — execution ID and state visible, Cancel absent. 

zombor commented 2026-07-03 18:35:00 +00:00

Author

Owner

Copy link

Security Review — bookshelf-rppy (PR #908)

Findings

[MAJOR] internal/books/enrich_confidence.go:107 — NaN bypasses the [0.0, 1.0] range clamp in readThreshold

strconv.ParseFloat("NaN", 64) returns (math.NaN(), nil) — no error, no panic. IEEE 754 comparisons with NaN are always false, so the guard v < 0.0 || v > 1.0 evaluates to false || false = false, and NaN is returned as the threshold. Downstream in applyConfidenceGate, confidence < math.NaN() is always false for any finite confidence value, silently disabling the confidence gate for that book type.

Any admin (or DB operator) who stores the literal string "NaN" for ENRICH_EBOOK_CONFIDENCE_THRESHOLD or ENRICH_COMIC_CONFIDENCE_THRESHOLD causes the confidence gate to stop firing — metadata below the intended quality bar is persisted without rejection or log warning.

The project already guards against this exact risk in analogous float-parsing paths:

• internal/books/metadata_handler.go:399: if f < 0 || f > 10 || math.IsNaN(f) || math.IsInf(f, 0)
• internal/books/service.go:341: if err != nil || math.IsNaN(f) || math.IsInf(f, 0)

Fix: add || math.IsNaN(v) (and optionally || math.IsInf(v, 0), though +Inf > 1.0 and -Inf < 0.0 already catch infinity via the range check):

if err != nil || v < 0.0 || v > 1.0 || math.IsNaN(v) {
    return defaultVal
}

---

Other checked areas (no findings)

• Parse safety: strconv.ParseFloat cannot panic; the error path is handled. ±Infinity is correctly rejected by the existing range check (both fall outside [0.0, 1.0]). Only NaN escapes.
• Threshold = 0.0 design: Allowing 0.0 as a valid threshold is intentional ("accept everything") and is explicitly covered by tests. Not a gate bypass — it requires admin/DB-level write access to set.
• Auth-Z on threshold write surface: No HTTP write surface for ENRICH_EBOOK_CONFIDENCE_THRESHOLD or ENRICH_COMIC_CONFIDENCE_THRESHOLD is introduced in this PR. These keys are read-only from the application's perspective; they can only be set via direct DB manipulation, which already implies elevated access.
• Info leak in EnrichResult: The result payload (ConfidenceSkipDetails) carries book_id, confidence, threshold, book_type, winner_provider. No API keys, tokens, or cross-user data. The workflow detail surface that exposes this (GET /admin/workflows/{instanceID}) is correctly gated behind AdminRequired.
• Multi-user scoping: The enrich workflow result is exposed only via the admin diagnostics UI (all workflow routes under /admin/workflows/ carry adminRequired). No new per-user-scoped surface is introduced.
• Architecture boundary: internal/books/enrich_confidence.go imports only stdlib (database/sql, strconv, etc.) and internal domain packages (internal/metadata, internal/middleware). No workflow engine import. Clean boundary preserved.
• Secrets in logs: The new warn log in applyConfidenceGate emits book_id, book_type, confidence, threshold, winner_provider, trace_id. No credentials or PII.

---

REVIEW VERDICT: 0 blocker, 1 major, 0 minor

## Security Review — bookshelf-rppy (PR #908) ### Findings [MAJOR] internal/books/enrich_confidence.go:107 — NaN bypasses the [0.0, 1.0] range clamp in readThreshold `strconv.ParseFloat("NaN", 64)` returns `(math.NaN(), nil)` — no error, no panic. IEEE 754 comparisons with NaN are always false, so the guard `v < 0.0 || v > 1.0` evaluates to `false || false = false`, and NaN is returned as the threshold. Downstream in `applyConfidenceGate`, `confidence < math.NaN()` is always false for any finite confidence value, silently disabling the confidence gate for that book type. Any admin (or DB operator) who stores the literal string `"NaN"` for `ENRICH_EBOOK_CONFIDENCE_THRESHOLD` or `ENRICH_COMIC_CONFIDENCE_THRESHOLD` causes the confidence gate to stop firing — metadata below the intended quality bar is persisted without rejection or log warning. The project already guards against this exact risk in analogous float-parsing paths: - `internal/books/metadata_handler.go:399`: `if f < 0 || f > 10 || math.IsNaN(f) || math.IsInf(f, 0)` - `internal/books/service.go:341`: `if err != nil || math.IsNaN(f) || math.IsInf(f, 0)` Fix: add `|| math.IsNaN(v)` (and optionally `|| math.IsInf(v, 0)`, though `+Inf > 1.0` and `-Inf < 0.0` already catch infinity via the range check): ```go if err != nil || v < 0.0 || v > 1.0 || math.IsNaN(v) { return defaultVal } ``` --- ### Other checked areas (no findings) - **Parse safety**: `strconv.ParseFloat` cannot panic; the error path is handled. ±Infinity is correctly rejected by the existing range check (both fall outside [0.0, 1.0]). Only NaN escapes. - **Threshold = 0.0 design**: Allowing 0.0 as a valid threshold is intentional ("accept everything") and is explicitly covered by tests. Not a gate bypass — it requires admin/DB-level write access to set. - **Auth-Z on threshold write surface**: No HTTP write surface for `ENRICH_EBOOK_CONFIDENCE_THRESHOLD` or `ENRICH_COMIC_CONFIDENCE_THRESHOLD` is introduced in this PR. These keys are read-only from the application's perspective; they can only be set via direct DB manipulation, which already implies elevated access. - **Info leak in EnrichResult**: The result payload (`ConfidenceSkipDetails`) carries `book_id`, `confidence`, `threshold`, `book_type`, `winner_provider`. No API keys, tokens, or cross-user data. The workflow detail surface that exposes this (`GET /admin/workflows/{instanceID}`) is correctly gated behind `AdminRequired`. - **Multi-user scoping**: The enrich workflow result is exposed only via the admin diagnostics UI (all workflow routes under `/admin/workflows/` carry `adminRequired`). No new per-user-scoped surface is introduced. - **Architecture boundary**: `internal/books/enrich_confidence.go` imports only stdlib (`database/sql`, `strconv`, etc.) and internal domain packages (`internal/metadata`, `internal/middleware`). No workflow engine import. Clean boundary preserved. - **Secrets in logs**: The new warn log in `applyConfidenceGate` emits `book_id`, `book_type`, `confidence`, `threshold`, `winner_provider`, `trace_id`. No credentials or PII. --- REVIEW VERDICT: 0 blocker, 1 major, 0 minor

zombor commented 2026-07-03 18:36:34 +00:00

Author

Owner

Copy link

[MAJOR] internal/wfengine/simple_workflows_test.go:290 — Missing EnrichWorkflow-level test for low-confidence result propagation
The EnrichActivities.Refetch unit tests (line 207-287) prove the LowConfidenceError → EnrichResult conversion logic, but there is no EnrichWorkflow workflow-tester-level test that sets stubErr to a *books.LowConfidenceError and asserts workflowResult.SkippedLowConfidence is populated with the correct confidence/threshold. go-workflows serializes the activity result to JSON and deserializes it back into the workflow result; a type registration bug or JSON tag mismatch would make the workflow tester return a zero-valued EnrichResult{} while all activity tests stay green. Per the workflow-integration-test-must-assert-success rule: "go-workflows marks errored-finished as completed — assert Result IS the success value." Add a Context inside the EnrichWorkflow Describe block: set stubErr = &books.LowConfidenceError{BookID:5, Confidence:0.42, Threshold:0.50, BookType:"ebook", WinnerProvider:"google"}, execute, assert workflowResult.SkippedLowConfidence != nil and that the Confidence/Threshold fields are propagated.

[MINOR] internal/wfengine/export_test.go:714 — Redundant type aliases for already-exported types
EnrichResultExport = EnrichResult and ConfidenceSkipDetailsExport = ConfidenceSkipDetails re-alias types that are already exported (uppercase). Tests in wfengine_test package can reference wfengine.EnrichResult directly without the Export alias. The aliases add no capability. Consider removing them and having the tests use the canonical names directly; keeping them just trains readers to expect an Export alias when there is none of the unexported-symbol-bridging that export_test.go aliases normally provide.

[MINOR] internal/books/metadata_service_test.go:358 — Misaligned nil arguments (3 spaces vs tabs)
Several + nil, lines added throughout the test file use 3-space indentation instead of the tab-based indentation that gofmt expects (matching the surrounding lines). Affects lines 358, 492, 500, 508, 533, 541, 549, 557. Run gofmt -w internal/books/metadata_service_test.go to fix.

[MINOR] internal/wfengine/simple_workflows_test.go:264 — Standalone It("returns no error") should be folded into value assertion
The low-confidence context has a separate It("returns no error (skip is reported as result, not failure)") { Expect(err).NotTo(HaveOccurred()) } followed by It("returns a result with SkippedLowConfidence populated") { Expect(result.SkippedLowConfidence).NotTo(BeNil()) }. Per conventions, the nil-error check should be folded into the first value assertion: Expect(result.SkippedLowConfidence, err).NotTo(BeNil()) — asserts both err==nil AND the field is non-nil in a single Expect, eliminating the standalone no-error It.

---

What this PR gets right:

• Arch boundary is clean: internal/books has no go-workflows import; LowConfidenceError is a plain struct; wfengine does the mapping in Refetch. ✓
• Permanent-vs-retryable semantics are intact: LCE is intercepted before isPermanentEnrichErr, transient errors still return as errors and get retried. ✓
• boundedFanOutWithResult uses the same sliding-window bounded-fan-out semantics as boundedFanOut — concurrency cap is enforced. ✓
• readThreshold correctly validates 0.0–1.0, falls back to defaults on error/missing/invalid. ✓
• UI surface: workflow_detail.html:93-97 already renders .Detail.Result as a <pre class="workflow-payload-block"> generic JSON block — the EnrichResult.SkippedLowConfidence JSON will be visible there without template changes. The bead spec "reuse canonical result rendering" is satisfied. ✓

REVIEW VERDICT: 0 blocker, 1 major, 3 minor

[MAJOR] internal/wfengine/simple_workflows_test.go:290 — Missing EnrichWorkflow-level test for low-confidence result propagation The `EnrichActivities.Refetch` unit tests (line 207-287) prove the `LowConfidenceError → EnrichResult` conversion logic, but there is no `EnrichWorkflow` workflow-tester-level test that sets `stubErr` to a `*books.LowConfidenceError` and asserts `workflowResult.SkippedLowConfidence` is populated with the correct confidence/threshold. go-workflows serializes the activity result to JSON and deserializes it back into the workflow result; a type registration bug or JSON tag mismatch would make the workflow tester return a zero-valued `EnrichResult{}` while all activity tests stay green. Per the `workflow-integration-test-must-assert-success` rule: "go-workflows marks errored-finished as completed — assert Result IS the success value." Add a Context inside the `EnrichWorkflow` Describe block: set `stubErr = &books.LowConfidenceError{BookID:5, Confidence:0.42, Threshold:0.50, BookType:"ebook", WinnerProvider:"google"}`, execute, assert `workflowResult.SkippedLowConfidence != nil` and that the Confidence/Threshold fields are propagated. [MINOR] internal/wfengine/export_test.go:714 — Redundant type aliases for already-exported types `EnrichResultExport = EnrichResult` and `ConfidenceSkipDetailsExport = ConfidenceSkipDetails` re-alias types that are already exported (uppercase). Tests in `wfengine_test` package can reference `wfengine.EnrichResult` directly without the Export alias. The aliases add no capability. Consider removing them and having the tests use the canonical names directly; keeping them just trains readers to expect an Export alias when there is none of the unexported-symbol-bridging that export_test.go aliases normally provide. [MINOR] internal/books/metadata_service_test.go:358 — Misaligned nil arguments (3 spaces vs tabs) Several `+ nil,` lines added throughout the test file use 3-space indentation instead of the tab-based indentation that gofmt expects (matching the surrounding lines). Affects lines 358, 492, 500, 508, 533, 541, 549, 557. Run `gofmt -w internal/books/metadata_service_test.go` to fix. [MINOR] internal/wfengine/simple_workflows_test.go:264 — Standalone `It("returns no error")` should be folded into value assertion The low-confidence context has a separate `It("returns no error (skip is reported as result, not failure)") { Expect(err).NotTo(HaveOccurred()) }` followed by `It("returns a result with SkippedLowConfidence populated") { Expect(result.SkippedLowConfidence).NotTo(BeNil()) }`. Per conventions, the nil-error check should be folded into the first value assertion: `Expect(result.SkippedLowConfidence, err).NotTo(BeNil())` — asserts both err==nil AND the field is non-nil in a single Expect, eliminating the standalone no-error It. --- **What this PR gets right:** - Arch boundary is clean: `internal/books` has no go-workflows import; `LowConfidenceError` is a plain struct; `wfengine` does the mapping in `Refetch`. ✓ - Permanent-vs-retryable semantics are intact: LCE is intercepted before `isPermanentEnrichErr`, transient errors still return as errors and get retried. ✓ - `boundedFanOutWithResult` uses the same sliding-window bounded-fan-out semantics as `boundedFanOut` — concurrency cap is enforced. ✓ - `readThreshold` correctly validates 0.0–1.0, falls back to defaults on error/missing/invalid. ✓ - UI surface: `workflow_detail.html:93-97` already renders `.Detail.Result` as a `<pre class="workflow-payload-block">` generic JSON block — the `EnrichResult.SkippedLowConfidence` JSON will be visible there without template changes. The bead spec "reuse canonical result rendering" is satisfied. ✓ REVIEW VERDICT: 0 blocker, 1 major, 3 minor

zombor referenced this pull request from a commit 2026-07-03 18:45:30 +00:00

fix(review): address MAJOR/MINOR findings from PR #908 review

zombor added 1 commit 2026-07-03 18:45:31 +00:00

fix(review): address MAJOR/MINOR findings from PR #908 review ... 62db17d252

- [MAJOR security] reject NaN/Inf in readThreshold (enrich_confidence.go):
  strconv.ParseFloat("NaN",64) returns (NaN,nil); NaN<0||NaN>1 is always
  false so a stored "NaN" silently disabled the confidence gate. Add
  math.IsNaN(v)||math.IsInf(v,0) to the rejection condition, matching
  the IsNaN/IsInf guards in metadata_handler.go and service.go. Add
  tests for NaN and Inf falling back to the 0.5 default.

- [MAJOR code] add EnrichWorkflow tester-level test for LowConfidenceError:
  proves SkippedLowConfidence survives go-workflows JSON serialize/
  deserialize round-trip (type-registration or JSON-tag bug would yield
  zero-valued EnrichResult{} while activity tests stayed green).

- [MINOR] remove EnrichResultExport/ConfidenceSkipDetailsExport aliases
  from export_test.go — both EnrichResult and ConfidenceSkipDetails are
  already exported types; the aliases add no bridging value. Update
  simple_workflows_test.go and fanout_test.go to reference the types
  directly.

- [MINOR] fold standalone It("returns no error") into value assertion
  Expect(result.SkippedLowConfidence, err).NotTo(BeNil()) per
  project-conventions one-Expect-per-It rule.

- [MINOR] run gofmt -w on metadata_service_test.go to fix 3-space
  indent on nil args (mixed spaces/tabs).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

forgejo-actions commented 2026-07-03 18:48:05 +00:00

Copy link

Recompute Match Score — kebab open screenshot (recompute-match-score-kebab-open)

**Recompute Match Score — kebab open screenshot** (recompute-match-score-kebab-open) 

forgejo-actions commented 2026-07-03 18:48:34 +00:00

Copy link

Workflow Detail page screenshot (wf-detail-older-execution)

Older completed ContinueAsNew epoch detail — execution ID and state visible, Cancel absent.

**Workflow Detail page screenshot** (wf-detail-older-execution) Older completed ContinueAsNew epoch detail — execution ID and state visible, Cancel absent. 

zombor added 1 commit 2026-07-03 19:04:12 +00:00

fix(coverage): add cancel-before-seeding test for BoundedFanOutEnrichInput ... 3bde3ba9b7

The boundedFanOutWithResult pre-seeding ctx.Err() guard (fanout.go:248)
was not covered by unit tests — only BoundedFanOutInt64 had a cancel-before-
seeding test (which covers boundedFanOut's equivalent check). Add a parallel
test for BoundedFanOutEnrichInput that cancels the workflow context immediately
before calling BoundedFanOutEnrichInput, exercising the return-on-cancel path
in the result-typed generic variant.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

zombor referenced this pull request 2026-07-03 19:11:03 +00:00

fix(e2e/browser): gate 2 ungated screenshot helpers; add screenshot-policy-check guard (bookshelf-6krh) #910

forgejo-actions commented 2026-07-03 19:11:10 +00:00

Copy link

Workflow Detail page screenshot (wf-detail-older-execution)

Older completed ContinueAsNew epoch detail — execution ID and state visible, Cancel absent.

**Workflow Detail page screenshot** (wf-detail-older-execution) Older completed ContinueAsNew epoch detail — execution ID and state visible, Cancel absent. 

forgejo-actions commented 2026-07-03 19:11:48 +00:00

Copy link

Recompute Match Score — kebab open screenshot (recompute-match-score-kebab-open)

**Recompute Match Score — kebab open screenshot** (recompute-match-score-kebab-open) 

zombor referenced this pull request from a commit 2026-07-03 19:43:56 +00:00

fix(review): address MAJOR/MINOR findings from PR #908 review

zombor force-pushed bd-bookshelf-rppy from 3bde3ba9b7 to 019bd55a3c 2026-07-03 19:43:57 +00:00 Compare

forgejo-actions commented 2026-07-03 19:45:56 +00:00

Copy link

Workflow Detail page screenshot (wf-detail-older-execution)

Older completed ContinueAsNew epoch detail — execution ID and state visible, Cancel absent.

**Workflow Detail page screenshot** (wf-detail-older-execution) Older completed ContinueAsNew epoch detail — execution ID and state visible, Cancel absent. 

forgejo-actions commented 2026-07-03 19:46:00 +00:00

Copy link

Recompute Match Score — kebab open screenshot (recompute-match-score-kebab-open)

**Recompute Match Score — kebab open screenshot** (recompute-match-score-kebab-open) 

zombor referenced this pull request from a commit 2026-07-03 21:00:49 +00:00

fix(review): address MAJOR/MINOR findings from PR #908 review

zombor force-pushed bd-bookshelf-rppy from 019bd55a3c to 3c95be3322 2026-07-03 21:00:50 +00:00 Compare

zombor merged commit a7075ae338 into main 2026-07-03 21:06:40 +00:00

zombor referenced this pull request from a commit 2026-07-03 21:06:42 +00:00

Merge pull request 'feat(enrich): confidence score in workflow result + configurable thresholds (bookshelf-rppy)' (#908) from bd-bookshelf-rppy into main

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

2 participants

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

zombor/pergamum!908

No description provided.