Review Response SLA Breach Audit

Measure how many negative reviews in your multi-location portfolio were answered late or are still unanswered, against the response time you agreed to deliver. The Actor works only on an export you supply. It does not scrape Google Maps or any other site, does not post replies, and never contacts reviewers.

What it answers

• Which locations breach the agreed response SLA, and how often?
• How many negative reviews are still unresolved past the SLA right now?
• How long do answered reviews take (median and P90 hours) per location?
• Which breaches are critical (elapsed time above a multiple of the SLA)?

Input

Copy the prefilled sample in the Actor input, or supply your own reviews array:

{
  "reviews": [
    {"locationId": "store-north", "reviewId": "r001", "rating": 1,
     "reviewTimestamp": "2026-03-01T09:00:00Z", "responseTimestamp": "2026-03-02T09:00:00Z"},
    {"locationId": "store-north", "reviewId": "r003", "rating": 1,
     "reviewTimestamp": "2026-03-05T12:00:00Z"}
  ],
  "asOf": "2026-03-10T00:00:00Z",
  "responseSlaHours": 48,
  "locationSlaHours": {"store-south": 24},
  "minSampleSize": 3
}

Each review needs locationId, reviewId, rating (1-5) and reviewTimestamp; responseTimestamp is empty for unanswered reviews. IDs must be de-identified codes (letters, digits, _ . : -, up to 64 characters). Records with names, emails or free text in the ID fields are rejected, and all other fields are ignored, so review text and reviewer details are never read.

Options: default SLA hours, per-location SLA overrides, scope (negative only or all ratings), highest rating counted as negative, critical multiplier, minimum sample size, attention breach percentage, breach row cap and review cap. Set asOf for reproducible results; otherwise the run time is used.

Output

The dataset contains three record types:

• portfolio - one row of totals, rejected/duplicate counts, truncation flag and caveats.
• location - per-location counts, breach rate, median and P90 response hours, oldest unresolved age and a flag: attention, within_target, low_sample or no_in_scope_reviews.
• review - breached reviews (late or unresolved) with hours over SLA and severity, largest overrun first, capped by maxBreachRows.

A summary and samples of rejected records are also stored in the key-value store record OUTPUT.

Pricing and billing behavior

The code emits one location-audited event per audited location, only after results are stored. Runs that fail input validation or produce no valid location are not charged. No payable plan is configured in this project; any pricing must be set and verified separately.

Limitations and caveats

• Sample size: a location with a handful of negative reviews can show a 50% or 100% breach rate from one case. Locations below minSampleSize (default 10) are flagged low_sample; treat them as anecdotes.
• Data quality: results are only as good as the export. If your source omits responses or records them late, they appear as unresolved or late.
• Calendar hours: business hours, time zones of opening times and holidays are not modelled; timestamps without a zone are treated as UTC.
• Scope: only the first response timestamp is used; edited or repeated replies, review text and sentiment are not analyzed.
• Capacity: up to 50,000 reviews per run, and results are not verified against any platform.
• It does not establish causes, customer outcomes or revenue impact.

Local tests

Offline unit tests are in tests/ and use only the standard library for the analysis module (python -m unittest discover -s tests).