Feedback makes OpenTacit reliable. The rank of each technique, each evidence line, and each decay flag comes from the real outcomes of members. The system collects feedback from your usual actions and words. It also records the confidence of each signal accurately.
What OpenTacit records, and when
OpenTacit monitors four stages for each technique:
| Stage | How OpenTacit records it |
|---|---|
| Shown | Automatically, when you see a technique: a pushed ◆ suggestion, or the one technique that an @tacit answer cites (a playbook search does not count) |
| Adopted | Automatically, when your action on a suggestion is verifiable, for example when you run a command that matches the recipe |
| Helped | From your words ("that worked") or an @tacit that helped; OpenTacit can also infer it carefully when a move that you adopted gets no complaint |
| Dismissed | From your words ("not relevant," "didn't work," "already knew"); OpenTacit maps each phrase to a reason |
No action is necessary from you for this. A reaction in your usual words in your next message is sufficient: OpenTacit records "that helped" and "didn't work" correctly.
Explicit and inferred signals
Each event has a confidence level:
- explicit — you said it in clear words.
- inferred — OpenTacit deduced it from behavior.
- verification — this third level applies to autonomous "applied" sessions, where a build or test result replaces the reaction of a member.
The registry keeps explicit and inferred signals separate. The signal-trust view of the dashboard shows the agreement between inferred signals and explicit signals. Because of this, automation cannot increase the evidence of a technique without a visible record. OpenTacit records inferred "helped" events carefully: one time for each technique, and only after an adoption continues through more turns without a complaint.
Dismissals give as much data as adoptions
When you dismiss a suggestion, the reason is important:
- Not relevant here — the retrieval did not match your work. This signal adjusts the matching.
- Didn't work — evidence against the technique. A sufficient number of these signals flags the technique as decaying.
- Already knew — the technique is good, but you do not need it. OpenTacit does not show it to you again for 30 days.
Record feedback deliberately
This is usually not necessary. OpenTacit already records a reaction in your next message, or the application of the move. A /tacit:feedback picker is available for the deliberate case. Do not use the picker and also give a usual reaction for the same event, because that causes a double count.
In some cases you want to make sure that OpenTacit records an event. An example is a suggestion from an earlier point in the session that you applied much later. Then speak to OpenTacit directly:
@tacit that helpedOpenTacit acknowledges this and records the reaction against the technique that it last suggested. From the command line, you can attach a reaction to a specified technique by id:
tacit feedback <technique-id> --stage helpedA helped also records the adoption. Send one stage, not two.
What leaves your machine, and what does not
Feedback events contain the id of the technique, the stage, the confidence, and your cohort (team, role, harness). They do not contain your name, and the registry has no member identities to connect them to. (Your history of adoptions and dismissals stays on your machine, never in the registry.)
There is one more data flow, and it applies only if your organization configured it. The first adoption of a suggestion can send an anonymized sketch: the situation and the move, with identification data removed. Sketches help the organization find techniques that are not in the playbook. The setting is a visible line in your agent.env (TACIT_SKETCH_URL). OpenTacit tells you when the value is first set. If you clear the value, you opt out permanently. Your raw transcript does not leave your machine in either case.