fix(riva): avoid stale interim carryover in segment assembly

[rbright]

Summary

• stop pre-committing divergent interim ASR hypotheses in the Riva receive loop
• keep only the latest interim hypothesis until a final result is emitted
• add regression tests to prevent stale/duplicated leading text in assembled transcripts

Why

Riva interim hypotheses can reset boundaries across updates. Pre-committing previous interim text could prepend stale text or duplicate leading phrases in final output.

Testing

• go test ./apps/sotto/internal/riva
• just ci-check
• nix build 'path:.#sotto'

Summary by CodeRabbit

• Bug Fixes

  • Improved handling of interim speech-recognition results: stale interim text is no longer prepended to finals; prior interim is cleared or preserved based on stability, suffix/continuation checks, and sentence-ending punctuation. Introduces a high-confidence stability threshold (~0.85) and resets interim stability on final results.

• Tests

  • Added and updated tests covering divergent/stale interim scenarios, stability-threshold behavior, suffix-correction handling, and final-result consolidation.

[coderabbitai]

📝 Walkthrough

Walkthrough

Stream interim handling now tracks interim stability and uses a stability-plus-punctuation and suffix/prefix similarity rule to decide when to commit a prior interim on divergence; stream state and processing were updated and tests added/renamed to validate replacement and commit behaviors.

Changes

Cohort / File(s)	Summary
Core stream state apps/sotto/internal/riva/client.go	Added lastInterimStability field to Stream and adjusted related field comments/ordering to track interim stability.
Stream processing logic apps/sotto/internal/riva/stream_receive.go	Reset lastInterimStability on final results; replaced prior interim-commit check with shouldCommitPriorInterimOnDivergence(...); set lastInterim and lastInterimStability after interim results and commit prior interim conditionally.
Interim decision logic apps/sotto/internal/riva/transcript_segments.go	Added stableInterimBoundaryThreshold = 0.85; extended isInterimContinuation with suffix checks and stronger prefix/suffix heuristics; added shouldCommitPriorInterimOnDivergence(previous, previousStability, current) bool and endsWithSentencePunctuation(transcript) bool to decide commits based on stability or sentence-ending punctuation.
Tests apps/sotto/internal/riva/client_test.go	Renamed TestRecordResponseCommitsInterimAcrossPauseLikeReset → TestRecordResponseReplacesDivergentInterimWithoutPrecommit; added tests TestRecordResponseCommitsStableDivergentInterimForPartialRecovery, TestRecordResponseDoesNotPrependStaleInterimBeforeFinal, TestRecordResponseTreatsSuffixCorrectionAsContinuation; added assertions about clearing segments, interim replacement, and stability-driven commit behavior.

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~45 minutes

Poem

🐇 I hopped through interims, sniffed each line,
Kept the steady ones, let the shaky resign,
If punctuation stood proud or stability high,
I planted it firm — else the old would fly,
Fresh hops, no stale echoes — that's my sign.

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 46.67% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely summarizes the main change: fixing stale interim carryover in segment assembly, which aligns with the PR's core objective of preventing pre-committed divergent interim hypotheses from creating stale text in assembled transcripts.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

• 📝 Generate docstrings (stacked PR)
• 📝 Generate docstrings (commit on current branch)

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch fix/riva-interim-hypothesis-reset

Warning

There were issues while running some tools. Please review the errors and either fix the tool's configuration or disable the tool if it's a critical failure.

🔧 golangci-lint (2.5.0)

Error: can't load config: unsupported version of the configuration: "" See https://golangci-lint.run/docs/product/migration-guide for migration instructions
The command is terminated due to an error: can't load config: unsupported version of the configuration: "" See https://golangci-lint.run/docs/product/migration-guide for migration instructions

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 38d2798505

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[coderabbitai]

🧹 Nitpick comments (2)

apps/sotto/internal/riva/transcript_segments.go (2)

72-75: Consider clarifying the minimum word count rationale.

The shorter >= 3 condition prevents false positives on very short transcripts, but this intent isn't immediately obvious. A brief inline comment would help.

📝 Suggested comment

+	// Require at least 3 words to use suffix similarity; shorter transcripts
+	// risk coincidental trailing-word matches.
 	commonSuffix := commonSuffixWords(prevWords, currWords)
 	if shorter >= 3 && commonSuffix*2 >= shorter {
 		return true
 	}

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@apps/sotto/internal/riva/transcript_segments.go` around lines 72 - 75, Add a
brief inline comment above the conditional that uses commonSuffixWords to
explain the rationale for the minimum-word cutoff: clarify that the check
(shorter >= 3 && commonSuffix*2 >= shorter) is intended to avoid false positives
on very short transcripts by requiring at least 3 words before applying the
"majority suffix" heuristic; reference the variables commonSuffix, shorter and
the helper commonSuffixWords so reviewers can quickly understand why the
threshold exists.

---

5-6: Consider documenting the threshold rationale.

The 0.85 threshold is a magic number that determines when interim results are stable enough to commit on divergence. A brief comment explaining the empirical basis or design rationale would help future maintainers understand when this value might need adjustment.

📝 Suggested documentation

-const stableInterimBoundaryThreshold = 0.85
+// stableInterimBoundaryThreshold is the minimum stability score at which a prior
+// interim hypothesis is considered reliable enough to commit when the next
+// hypothesis diverges (i.e., is not a continuation).
+const stableInterimBoundaryThreshold = 0.85

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@apps/sotto/internal/riva/transcript_segments.go` around lines 5 - 6, The
constant stableInterimBoundaryThreshold = 0.85 is a magic number; add a short
inline comment above the stableInterimBoundaryThreshold declaration that
documents its purpose and rationale (e.g., how 0.85 was chosen—empirically tuned
for X dataset/latency vs stability tradeoff—or a pointer to tests/experiments
that justify it), note the units/meaning (probability/confidence), and describe
when to adjust it or where to find related tests/benchmarks that should be
updated if it changes.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Nitpick comments:
In `@apps/sotto/internal/riva/transcript_segments.go`:
- Around line 72-75: Add a brief inline comment above the conditional that uses
commonSuffixWords to explain the rationale for the minimum-word cutoff: clarify
that the check (shorter >= 3 && commonSuffix*2 >= shorter) is intended to avoid
false positives on very short transcripts by requiring at least 3 words before
applying the "majority suffix" heuristic; reference the variables commonSuffix,
shorter and the helper commonSuffixWords so reviewers can quickly understand why
the threshold exists.
- Around line 5-6: The constant stableInterimBoundaryThreshold = 0.85 is a magic
number; add a short inline comment above the stableInterimBoundaryThreshold
declaration that documents its purpose and rationale (e.g., how 0.85 was
chosen—empirically tuned for X dataset/latency vs stability tradeoff—or a
pointer to tests/experiments that justify it), note the units/meaning
(probability/confidence), and describe when to adjust it or where to find
related tests/benchmarks that should be updated if it changes.