You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
test(search): compare rankers and measure storage-backed discovery - #5
Keep the search experiments in one synthetic, offline research area and measure the complete lightweight path rather than only ranking manually anchored candidates. Preserve the original transferred fixtures/reports, model explanations, tool research, 23-variant matrix and frozen-feature robustness validation.
Add a model-free question path, fresh authored wording, simple ranking ablations and a build-driven storage-backed discovery benchmark. The frozen planner/BM25/context prototype drops from 92.9% on shared topic templates to 56.3% (9/16) on fresh actual-answer questions. Partial candidate retrieval recovers the missing evidence. The integrated shared service reaches 16/16 in the fresh top ten, 14/16 in the top three and 2/16 at rank one, with MRR@10 0.521 and nDCG@10 0.748. These examples became development cases; they are not blinded or independently human-labeled quality proof.
The older template remains recorded: 48/56 actual answers in the top ten, eight retention-paraphrase ranking failures and 160 partial hits on missing-fact/unknown-qualifier queries. Discovery is evidence retrieval, not an answer-availability detector; the strict default-switch gates remain unmet.
Separate archive ingestion from query RAM by launching fresh Node query processes against marked synthetic databases. The 100,341-message report records about 125 MB query-process peak RSS and about 108 ms question p95 on the measured Ryzen laptop, with no neural weights. Preserve exact question inputs, result ids, build/fixture/runner hashes and launch metadata. The launcher stages only the selected harness and transparently adapts staged core scope/native package aliases after the @wirecat rename, without changing original scripts.
The shared implementation is cli-messaging PR #803; public CLI/MCP strict defaults and SDK compatibility are preserved. This repo publishes no live account data and changes no existing conversation-search benchmark or published suite API.
Validation after rebase: repository lint, types, all 59 tests with coverage and document checks; targeted research Markdown; historical/matrix/validation/model-free harness typechecks; 12 planner tests; built SDK integration/control tests in cli-messaging. Reports retain failures. See performance/search/model-free/FRESH-VALIDATION.md for complete comparison, resources, limitations and reproduction.
leemour
changed the title
test(search): centralize offline ranking experiments and explain models
test(search): compare lightweight rankers and centralize search research
Oct 9, 2026
Next steps: Take a moment to review the security alert above. Review
the linked package source code to understand the potential risk. Ensure the
package is not malicious before proceeding. If you're unsure how to proceed,
reach out to your security team or ask the Socket team for help at
support@socket.dev.
Suggestion: Remove or replace dependencies that include known critical CVEs. Consumers can use dependency overrides or npm audit fix --force to remove vulnerable dependencies.
Mark the package as acceptable risk. To ignore this alert only
in this pull request, reply with the comment
@SocketSecurity ignore pypi/nltk@3.9.2. You can
also ignore all packages with @SocketSecurity ignore-all.
To ignore an alert for all future pull requests, use Socket's Dashboard to
change the triage state of this alert.
Warn
Critical CVE: NLTK: Allowlisted pickle loaders still permit code execution in current source
Next steps: Take a moment to review the security alert above. Review
the linked package source code to understand the potential risk. Ensure the
package is not malicious before proceeding. If you're unsure how to proceed,
reach out to your security team or ask the Socket team for help at
support@socket.dev.
Suggestion: Remove or replace dependencies that include known critical CVEs. Consumers can use dependency overrides or npm audit fix --force to remove vulnerable dependencies.
Mark the package as acceptable risk. To ignore this alert only
in this pull request, reply with the comment
@SocketSecurity ignore pypi/nltk@3.9.2. You can
also ignore all packages with @SocketSecurity ignore-all.
To ignore an alert for all future pull requests, use Socket's Dashboard to
change the triage state of this alert.
Next steps: Take a moment to review the security alert above. Review
the linked package source code to understand the potential risk. Ensure the
package is not malicious before proceeding. If you're unsure how to proceed,
reach out to your security team or ask the Socket team for help at
support@socket.dev.
Suggestion: Remove or replace dependencies that include known critical CVEs. Consumers can use dependency overrides or npm audit fix --force to remove vulnerable dependencies.
Mark the package as acceptable risk. To ignore this alert only
in this pull request, reply with the comment
@SocketSecurity ignore pypi/nltk@3.9.2. You can
also ignore all packages with @SocketSecurity ignore-all.
To ignore an alert for all future pull requests, use Socket's Dashboard to
change the triage state of this alert.
Warn
Critical CVE: NLTK: JVM argument injection bypass via per-call options in the NLTK Stanford wrappers (incomplete fix of CVE-2026-12841)
Next steps: Take a moment to review the security alert above. Review
the linked package source code to understand the potential risk. Ensure the
package is not malicious before proceeding. If you're unsure how to proceed,
reach out to your security team or ask the Socket team for help at
support@socket.dev.
Suggestion: Remove or replace dependencies that include known critical CVEs. Consumers can use dependency overrides or npm audit fix --force to remove vulnerable dependencies.
Mark the package as acceptable risk. To ignore this alert only
in this pull request, reply with the comment
@SocketSecurity ignore pypi/nltk@3.9.2. You can
also ignore all packages with @SocketSecurity ignore-all.
To ignore an alert for all future pull requests, use Socket's Dashboard to
change the triage state of this alert.
leemour
changed the title
test(search): compare lightweight rankers and centralize search research
test(search): compare lightweight ranking and validate reply context
Oct 9, 2026
leemour
changed the title
test(search): compare lightweight ranking and validate reply context
test(search): compare lightweight ranking and evaluate model-free question search
Oct 10, 2026
leemour
changed the title
test(search): compare lightweight ranking and evaluate model-free question search
test(search): compare rankers and measure storage-backed discovery
Oct 10, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Keep the search experiments in one synthetic, offline research area and measure the complete lightweight path rather than only ranking manually anchored candidates. Preserve the original transferred fixtures/reports, model explanations, tool research, 23-variant matrix and frozen-feature robustness validation.
Add a model-free question path, fresh authored wording, simple ranking ablations and a build-driven storage-backed discovery benchmark. The frozen planner/BM25/context prototype drops from 92.9% on shared topic templates to 56.3% (9/16) on fresh actual-answer questions. Partial candidate retrieval recovers the missing evidence. The integrated shared service reaches 16/16 in the fresh top ten, 14/16 in the top three and 2/16 at rank one, with MRR@10 0.521 and nDCG@10 0.748. These examples became development cases; they are not blinded or independently human-labeled quality proof.
The older template remains recorded: 48/56 actual answers in the top ten, eight retention-paraphrase ranking failures and 160 partial hits on missing-fact/unknown-qualifier queries. Discovery is evidence retrieval, not an answer-availability detector; the strict default-switch gates remain unmet.
Separate archive ingestion from query RAM by launching fresh Node query processes against marked synthetic databases. The 100,341-message report records about 125 MB query-process peak RSS and about 108 ms question p95 on the measured Ryzen laptop, with no neural weights. Preserve exact question inputs, result ids, build/fixture/runner hashes and launch metadata. The launcher stages only the selected harness and transparently adapts staged core scope/native package aliases after the @wirecat rename, without changing original scripts.
The shared implementation is cli-messaging PR #803; public CLI/MCP strict defaults and SDK compatibility are preserved. This repo publishes no live account data and changes no existing conversation-search benchmark or published suite API.
Validation after rebase: repository lint, types, all 59 tests with coverage and document checks; targeted research Markdown; historical/matrix/validation/model-free harness typechecks; 12 planner tests; built SDK integration/control tests in cli-messaging. Reports retain failures. See performance/search/model-free/FRESH-VALIDATION.md for complete comparison, resources, limitations and reproduction.