recipes: add adaptive capture classification with confidence gating#139
recipes: add adaptive capture classification with confidence gating#139mjh756 wants to merge 4 commits intoNateBJones-Projects:mainfrom
Conversation
|
Hey @mjh756 — welcome to Open Brain Source! 👋 Thanks for submitting your first PR. The automated review will run shortly and check things like metadata, folder structure, and README completeness. If anything needs fixing, the review comment will tell you exactly what. Once the automated checks pass, a human admin will review for quality and clarity. Expect a response within a few days. If you have questions, check out CONTRIBUTING.md or open an issue. |
matthallett1
left a comment
There was a problem hiding this comment.
Review — Adaptive Capture Classification
Thoughtful recipe. The per-type threshold learning loop is a genuinely useful pattern, and the A/B model comparison is a nice bonus. Schema has proper RLS. Well-documented.
Issues to fix
1. metadata.json author mismatch
Author is listed as "Matt Hardy" / "matthardy" but the PR is from mjh756. Fix to match your GitHub account.
2. Classifier prompt uses non-OB1 types
The default types in classifier_prompt.md are:
["thought", "task", "note", "decision", "question", "event"]OB1's canonical types are: idea, task, person_note, reference, decision, lesson, meeting, journal. The schema.sql and README should reference the standard types, even if the recipe allows customization.
3. No implementation code
This is pseudocode + schema only. That's fine for a recipe, but worth noting in the README that this requires the user to implement the gating logic in their capture interface. The pseudocode is clear enough to translate, but a TypeScript reference implementation (even a minimal one) would make this much more actionable.
What's good
- RLS on all 4 tables
ab_comparisonsCHECK constraint onwinnerfield- Threshold floor (0.50) and ceiling (0.95) prevent the system from becoming fully autonomous or fully manual
- Consistency check (run classifier twice when confidence is low) is a smart pattern
- Schema is fully additive, clean rollback story
Fix the author and types, then this is ready.
|
Thanks, my first git PR. The system is working well for me, Thanks to Nate, yourself and the community. |
|
Thank you mjh! It is a fun ride. Glad to have you part of it! /|##!##| |
|
That is meant to be a rocket... |
|
All good! For better or worse, I am old enough to remember the joys of ASCII drawing. My first coding project was a graphical choose your own adventure game written in BASIC on a Commodore 64, with graphics drawn using ASCII characters and print statements!
From: Matt Hallett ***@***.***>
Sent: Tuesday, 31 March 2026 4:34 PM
To: NateBJones-Projects/OB1 ***@***.***>
Cc: mjh756 ***@***.***>; Mention ***@***.***>
Subject: Re: [NateBJones-Projects/OB1] recipes: add adaptive capture classification with confidence gating (PR #139)
[https://avatars.githubusercontent.com/u/50158096?s=20&v=4]matthallett1 left a comment (NateBJones-Projects/OB1#139)<#139?email_source=notifications&email_token=B6QDT25OUXXL6N3TJN2AFND4TNKCZA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTIMJWGAYDGMZUG44KM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJNLQOJPWG33NNVSW45C7N5YGK3S7MNWGSY3L#issuecomment-4160033478>
That is meant to be a rocket...
-
Reply to this email directly, view it on GitHub<#139?email_source=notifications&email_token=B6QDT25OUXXL6N3TJN2AFND4TNKCZA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTIMJWGAYDGMZUG44KM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJNLQOJPWG33NNVSW45C7N5YGK3S7MNWGSY3L#issuecomment-4160033478>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/B6QDT27B5KDWAW3YSUBVFJ34TNKCZAVCNFSM6AAAAACXG67D5SVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DCNRQGAZTGNBXHA>.
You are receiving this because you were mentioned.Message ID: ***@***.******@***.***>>
|
|
Ha! BASIC on a C64? That’s legend. You were basically shipping before indie games existed. I bet you had PETSCII art that put my rocket ship to shame 😂 were you a Zork fan? I was a "ladder" freak on Kapro. :-) |
|
I didn't remember Kapro, but a quick google was a nice walk down memory lane.
I certainly enjoyed Zork, but my strongest memories are of Impossible Mission, Defender of the Crown, Carmen Sandiego, and destroying joysticks playing summer games!
…________________________________
From: Matt Hallett ***@***.***>
Sent: Tuesday, 31 March 2026 5:32 PM
To: NateBJones-Projects/OB1 ***@***.***>
Cc: mjh756 ***@***.***>; Mention ***@***.***>
Subject: Re: [NateBJones-Projects/OB1] recipes: add adaptive capture classification with confidence gating (PR #139)
[https://avatars.githubusercontent.com/u/50158096?s=20&v=4]matthallett1 left a comment (NateBJones-Projects/OB1#139)<#139?email_source=notifications&email_token=B6QDT27L5HGZOIX2B2BJ6ST4TNQ73A5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTIMJWGAZDMMJWGYYKM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJNLQOJPWG33NNVSW45C7N5YGK3S7MNWGSY3L#issuecomment-4160261660>
Ha! BASIC on a C64? That’s legend. You were basically shipping before indie games existed. I bet you had PETSCII art that put my rocket ship to shame 😂 were you a Zork fan? I was a "ladder" freak on Kapro. :-)
—
Reply to this email directly, view it on GitHub<#139?email_source=notifications&email_token=B6QDT27L5HGZOIX2B2BJ6ST4TNQ73A5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTIMJWGAZDMMJWGYYKM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJNLQOJPWG33NNVSW45C7N5YGK3S7MNWGSY3L#issuecomment-4160261660>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/B6QDT25NO36PEGB76ZJ7BP34TNQ73AVCNFSM6AAAAACXG67D5SVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DCNRQGI3DCNRWGA>.
You are receiving this because you were mentioned.Message ID: ***@***.***>
|
Contribution Type
/recipes)What does this do?
Adds a confidence gating and learning loop on top of OB1's existing capture flow. The classifier reports a confidence score (0–10); a per-type threshold gates whether to auto-classify or ask the user for confirmation; every user response nudges that threshold up or down. Over time the system learns which capture types it can reliably auto-classify and becomes progressively less intrusive. Includes optional A/B model comparison tables for empirically choosing between classifier models.
Requirements
Checklist
README.mdwith prerequisites, step-by-step instructions, and expected outcomemetadata.jsonhas all required fields