manage-scoring

name	manage-scoring
description	View, edit, and operate scoring — list profiles and rules, tune point values, add or remove assignments, trigger recompute, and inspect scores with contribution breakdowns. The day-two counterpart to configure-scoring.

Manage Scoring

Use this skill to inspect and adjust scoring after the initial configure-scoring setup. For concept definitions (dimensions, profiles, rules, assignments, point-value shapes, auto-trigger behaviour), see _shared/scoring.md.

For signal subscriptions (the upstream that feeds scores), use manage-signals.

Saber CLI check

Run saber --help. Scoring requires the CLI — if it isn't installed, point the user to saber.app and stop.

Step 1 — Inventory

List profiles first:

saber scoring profile list

Show in a table grouped by type:

ID	Type	Name	Description
...	company	EMEA Enterprise	New-business outbound
...	contact	Decision makers	Senior engineering buyers

Ask which profile the user wants to work with. Then drill in:

saber scoring rule list <profileId>

Show each rule with: signal template, dimension, point-values shape preview.

Step 2 — Choose an action

Present the menu:

1. Inspect scores for one or more objects
2. Edit a rule (change point values)
3. Delete a rule (remove a signal from the profile)
4. Add a rule (cover a new signal template)
5. Manage assignments (add / remove / list)
6. Recompute manually for specific objects
7. Delete the profile (cascades to all its rules, assignments, scores)

Step 3 — Execute

1. Inspect scores

saber scoring scores --type company --object <domain> --detailed
saber scoring scores --type company --object <domain1> --object <domain2>

--detailed returns per-rule contributions plus the previous score (Δ). Use it whenever you want to debug why an object scores the way it does.

2. Edit a rule (change point values)

Upsert is keyed on (profileId, signalTemplateId, dimension) — calling it again replaces the rule. Every upsert triggers a recompute for every assigned object.

saber scoring rule upsert <profileId> \
  --signal-template <signalTemplateId> \
  --dimension fit \
  --answer-type boolean \
  --true 25 --false -10

Confirm with the user before running — the recompute can be expensive on large profiles.

3. Delete a rule

saber scoring rule delete <profileId> <ruleId>

Also triggers a recompute.

4. Add a rule

Same as Step 4 in configure-scoring — pick a signal template, dimension, answer type, point values. See _shared/scoring.md for the answer-type → shape contract.

5. Manage assignments

# What profiles is one object assigned to?
saber scoring assignment list --type company --object acme.com

# Add a single object
saber scoring assignment create --profile <id> --type company --object acme.com

# Add many at once
saber scoring assignment bulk --profile <id> --type company \
  --object acme.com --object stripe.com --object ramp.com

# Remove (clears that assignment's scores too)
saber scoring assignment delete <assignmentId>

Use bulk for full-list assignments — duplicates are skipped silently.

6. Recompute manually

Auto-trigger usually keeps scores fresh. Use compute only when:

• You just edited rules and want immediate confirmation on a specific object
• Underlying signal data was repaired out-of-band
• You want to force a refresh ahead of a meeting

saber scoring compute --type company --object acme.com --object stripe.com

Returns {queued, failed}. failed > 0 means some object dispatches errored — retry to pick those up. Compute is idempotent: duplicate triggers attach to the running workflow.

7. Delete the profile

Cascades to all rules, assignments, and scores under it. Confirm explicitly first:

"Deleting profile [name] will remove [N] rules, [M] assignments, and all historical scores under it. This is not reversible. Continue?"

saber scoring profile delete <profileId>

Useful checks

• Coverage gap — if signalCoverage is much lower than totalRules in a score, the underlying signals haven't run yet for that object. Check manage-signals to make sure subscriptions cover this list.
• Stale scores — computedAt more than a few hours after the latest signal completion suggests auto-trigger may have failed. Run compute manually to recover.
• Identical scores across many objects — usually a rule with too-coarse point values. Use --detailed on a few objects and look at which rules actually differentiate.

Related

• configure-scoring — first-time setup
• score-accounts — rank a list by current scores
• manage-signals — manage the signal subscriptions that feed scores
• _shared/scoring.md — reference for all concepts and CLI shapes