rbrooks/WeatherBot

Watch

Fork 0

Guardrailed timing/impact AI follow-ups (post-validation) #21

New issue

Closed

opened 2026-06-23 23:15:34 +00:00 by rbrooks · 2 comments

rbrooks commented

2026-06-23 23:15:34 +00:00

Owner

Target release: v1.3.0

After accuracy validation, enable cautious, source-grounded timing/impact follow-ups per the AI summarization Phase 3 design — synthesizing only from official text + structured WeatherBot context, preserving uncertainty.

Source: docs/ROADMAP.md → Local LLM Alert Summarization → "Later phases" / AI design notes Phase 3.

**Target release:** v1.3.0 After accuracy validation, enable cautious, source-grounded timing/impact follow-ups per the AI summarization Phase 3 design — synthesizing only from official text + structured WeatherBot context, preserving uncertainty. Source: `docs/ROADMAP.md` → Local LLM Alert Summarization → "Later phases" / AI design notes Phase 3.

rbrooks added the

area:ai

type:feature

v1.3.0

labels

2026-06-23 23:16:51 +00:00

rbrooks added this to the v1.3.0 - AI summarization expansion milestone

2026-06-23 23:23:26 +00:00

claude-bot referenced this issue

2026-07-19 00:20:08 +00:00

Decide quiet-hours AI summaries: supplement vs replace #19

claude-bot commented

2026-07-19 00:43:30 +00:00

Contributor

Approach (guardrailed timing/impact follow-ups)

Implementing in the v1.3.0 summarizer phase, off by default so it changes nothing until an operator opts in:

New setting AI_TIMING_IMPACT_FOLLOWUPS_ENABLED (default false). When off, prompt/validators are byte-for-byte unchanged.
When on, the summary prompt gains a bounded, source-grounded timing/impact clause: synthesize approximate onset/duration/impact only from the official source text and the structured WeatherBot context (SPC/alert timing fields, forecast digest), always preserving uncertainty wording and never inventing towns/roads/distances/exact arrival times not in the inputs.
Guardrailed by the existing validators plus the new output-URL/delimiter hygiene from #16; the "no urgent language not in source" and "no unsupported risk label" checks already constrain over-reach. No new network calls.

Rationale for default-off: this is the most speculative part of the AI pipeline; it ships dark and an operator enables it after they're satisfied with real output quality (#16).

## Approach (guardrailed timing/impact follow-ups) Implementing in the v1.3.0 summarizer phase, **off by default** so it changes nothing until an operator opts in: - New setting `AI_TIMING_IMPACT_FOLLOWUPS_ENABLED` (default `false`). When off, prompt/validators are byte-for-byte unchanged. - When on, the summary prompt gains a bounded, source-grounded timing/impact clause: synthesize approximate onset/duration/impact **only** from the official source text and the structured WeatherBot context (SPC/alert timing fields, forecast digest), always preserving uncertainty wording and never inventing towns/roads/distances/exact arrival times not in the inputs. - Guardrailed by the existing validators plus the new output-URL/delimiter hygiene from #16; the "no urgent language not in source" and "no unsupported risk label" checks already constrain over-reach. No new network calls. Rationale for default-off: this is the most speculative part of the AI pipeline; it ships dark and an operator enables it after they're satisfied with real output quality (#16).

rbrooks referenced this issue from a commit

2026-07-19 00:59:57 +00:00

Summarizer: prompt hygiene, output-URL validator, quiet-hours REPLACE, guarded follow-ups

claude-bot referenced this issue

2026-07-19 01:00:11 +00:00

v1.3.0 Phase B: summarizer hygiene, quiet-hours mode, guarded follow-ups #105

claude-bot commented

2026-07-19 01:07:36 +00:00

Contributor

Implemented in #105 (merged), shipping dark per the approach above. AI_TIMING_IMPACT_FOLLOWUPS_ENABLED (default false) gates a bounded prompt clause: when enabled, summaries may add approximate onset/duration/impact synthesized strictly from the official source text + structured WeatherBot context, always preserving uncertainty and never inventing towns/roads/distances/exact arrival times. With the flag off the prompt is byte-for-byte unchanged. The existing validators (no urgent language absent from source, no unsupported risk labels) plus the new #16 hygiene continue to guard over-reach. Enable after you're satisfied with real output quality (#16).

Implemented in #105 (merged), shipping **dark** per the approach above. `AI_TIMING_IMPACT_FOLLOWUPS_ENABLED` (default `false`) gates a bounded prompt clause: when enabled, summaries may add approximate onset/duration/impact synthesized strictly from the official source text + structured WeatherBot context, always preserving uncertainty and never inventing towns/roads/distances/exact arrival times. With the flag off the prompt is byte-for-byte unchanged. The existing validators (no urgent language absent from source, no unsupported risk labels) plus the new #16 hygiene continue to guard over-reach. Enable after you're satisfied with real output quality (#16).

claude-bot closed this issue

2026-07-19 01:07:39 +00:00

No milestone

No project

No assignees

2 participants

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

rbrooks/WeatherBot#21

No description provided.

Rows
Columns