neesh Inc.

PRODUCT_03 WATCHBASE IN_BUILD

The change you would have found out about on Thursday.

WatchBase watches the pages you care about, diffs them structurally and textually, and scores every change on our own machine before a model ever sees it. The default threshold is set so that about one change in ten earns a sentence. The rest are counted and left alone.

Book Free Assessment See it deployed in a scenario
WatchBase narrating one change and filtering another A radar sweeping eight watched pages. The first pass finds no change. The second finds a pricing change scoring 0.82 against a 0.40 threshold, which blooms and is narrated. The third finds a navigation-copy change scoring 0.31, which stays dim and is filtered. The footer reads eight targets, forty-seven filtered, three narrated, four cents of model cost. SWEEP · 8 TARGETS DETECTION_LOG PASS_01 · 8 TARGETS NO CHANGE PASS_02 · competitor/pricing NARRATE ≥ 0.40 0.82 NARRATED PASS_03 · competitor/nav-copy 0.31 BELOW THRESHOLD · FILTERED Their compare page dropped the Alpine Shell 18% overnight. 8 TARGETS · 47 FILTERED · 3 NARRATED · $0.04 AI COST
WatchBase narrating one change and filtering another A radar sweeping eight watched pages. The first pass finds no change. The second finds a pricing change scoring 0.82 against a 0.40 threshold, which blooms and is narrated. The third finds a navigation-copy change scoring 0.31, which stays dim and is filtered. The footer reads eight targets, forty-seven filtered, three narrated, four cents of model cost. SWEEP · 8 TARGETS DETECTION_LOG PASS_01 · 8 TARGETS NO CHANGE PASS_02 · competitor/pricing NARRATE ≥ 0.40 0.82 NARRATED PASS_03 · competitor/nav-copy 0.31 BELOW THRESHOLD · FILTERED Their compare page dropped the Alpine Shell 18% overnight. 8 TARGETS · 47 FILTERED 3 NARRATED · $0.04 AI COST
READS The 0.40 significance threshold is WatchBase’s default and is configurable per organization. The pages watched and the sentence narrated are illustrative.
CAPABILITY_SET

What it does, and where each part stands

  1. Scheduled checks with a lease shipped

    Each target has its own cadence and is claimed under a lease, so two workers running at once never check the same page twice or miss one.

  2. Two-layer diff shipped

    A structural comparison of the page tree and a text comparison of its content, both over normalised markup with the parts that change on every load stripped first.

  3. Significance scoring, locally shipped

    A model runs in-process on our own hardware, not an API call, and scores every change on four factors. Below 0.40 it is counted and left alone. That gate is the product.

  4. Semantic narration cache shipped

    A change that means the same thing as one already narrated reuses that sentence instead of paying for a new one. This is the suite’s reference implementation of the pattern.

  5. Delivery by significance shipped

    High significance goes out at once to Slack, email, signed webhook and in-app, quiet hours or not. Medium lands in-app, and quiet hours keep everything below high on that silent channel. Each organization can re-route the bands.

  6. Narration in build

    The sentence at the end of the pipeline. Everything up to it is verified end to end; narration itself has not yet run against a live model on our machines, because we have no model key on them.

RECOGNISE_IT_IF

This is probably you

  • Someone has a bookmark folder they open every single morning.
  • You found out about a competitor’s change from one of your own customers.
  • You tried alerting once and turned it off, because everything fired.
  • The pages you care about are public, and checking them is nobody’s actual job.
EXAMPLE_CONFIGURATIONS

Shapes it has been configured into

Wrappers of the same engine, not separate products. Where a line carries a figure, the figure is a design target from the scenario named beside it, not a customer result.

  • REGULATORY WATCH New regulation published Tuesday. Your compliance team knew by Wednesday.
  • COMPETITOR BRIEF Weekly competitive intelligence brief in Slack. AI cost: four cents. DESIGN TARGET · competitor-monitoring scenario
  • METRICS BRIEF Monday morning: a briefing on your business metrics, before anyone asks for it.
  • HIRING SIGNALS A competitor posted 3 senior engineering roles. You see the pattern first.
PROOF

Two numbers we measured. Two we are aiming at.

Measured figures name the run that produced them and come from a review packet in the product’s own repository. Design targets name the scenario they belong to. There is no third kind of number on this page.

MEASURED 0 type errors across 109 source files under mypy strict the WatchBase build, 9 Sep 2026
MEASURED 0.61 significance scored in-process on a change a live run detected the WatchBase build, 9 Sep 2026
DESIGN TARGET 4 days pre-warning on a competitor’s Black Friday pricing competitor-monitoring scenario
DESIGN TARGET $0.04 model cost for the week that caught it competitor-monitoring scenario
PRICING

What it costs

Set on the assessment call.

WatchBase is in build, so the conversation is about what you need watched and when it is ready, not a number we would have to revise once we know.

THE_HONEST_LINE

What we will not claim

In build means everything up to the sentence is verified end to end, and the sentence is not. Scheduling, diffing, scoring, caching and delivery all run and are tested; narration has not been produced against a live model on our machines yet, because there is no model key on them. That is the last mile, and it is the only one left.

Thirty minutes tells you whether WatchBase fits.

A fixed agenda, a one-page map back within two business days, and no pitch. If WatchBase is the wrong tool for your problem, that is what the map will say.

Book Free Assessment

All four instruments The competitor-monitoring scenario