Weekly Delivery Summary

The following weekly signals were delivered to GCS buckets on March 3:

Note: Patent filings now deliver weekly alongside the other weekly signals. 20-F had no new filings.

Upcoming

Monthly signals (LinkedIn Company, Glassdoor, GitHub, Product Reviews, SEO, Twitter/X, YouTube) are due on or around March 13.

For a full list of delivery dates and bucket URIs, see the Delivery page.

Bi-weekly Deliveries

Reddit Mentions (Company) — February 23

LinkedIn Posts (Contact) — March 2

All files are available in both JSONL and Parquet formats.

For a full list of delivery dates and bucket URIs, see the Delivery page.

Weekly Delivery Summary

The following weekly signals were delivered to GCS buckets on February 24:

Note: News (autobound-news-v2) had an empty delivery for this week. 20-F had no new filings.

All files are available in both JSONL and Parquet formats in their respective buckets.

For a full list of delivery dates and bucket URIs, see the Delivery page.

Delivery Summary

• Buckets delivered: 8
• Total signals: 201,669
• Unique companies: 88,784

Bucket-Level Breakdown

What Changed

Hiring Velocity — 8 New Fields

Added net flow rate and velocity change metrics at 7-day, 30-day, and 90-day windows:

No columns were removed. Row count increased modestly from 45,018 to 47,536 (+5.6%).

Work Milestones — Fanout Fix Applied + Recency Filter

The Work Milestones signal was reprocessed on February 18 to fix two data quality issues:

1. Previous-position fanout eliminated. The pipeline was emitting one row per (contact, previous_position) pair instead of one row per career event. This inflated jobChange and promotion counts significantly. The fix now emits one signal per actual event, with previous_company_name set to the role directly beneath the new one (the job they actually left).

2. 6-month recency cutoff applied. Signals with an event_date older than 6 months are now filtered out, removing stale career history surfaced during first-time contact ingestion.

Before vs after:

The remaining ~40 contacts with multiple rows per subtype are legitimate multi-hop career events (e.g., a person who left two different companies for the same new role).

Cross-delivery dedupe remains active: 0% contact overlap between Feb 10 and Feb 17 deliveries.

Coming Soon

• Higher-frequency LinkedIn profile ingestion — building updated pipelines for more frequent realtime profile data pulls, which will increase the volume and freshness of work milestone signals on a weekly basis.

Other Known Issues

• Hiring velocity outliers: Some extreme net flow rate values (100K%+) from small companies with tiny baselines.

The following signals were delivered to GCS buckets on February 13:

Social & Web Signals

• LinkedIn Posts (Company) — next delivery Mar 16
• LinkedIn Posts (Contact) — next delivery Mar 2 (bi-weekly)
• LinkedIn Comments (Contact) — next delivery Mar 16
• Twitter/X Posts (Company)
• Twitter/X Posts (Contact)
• YouTube Activity (Company)
• YouTube Activity (Contact)
• Glassdoor Reviews

Company Intelligence

• GitHub Activity
• Product Reviews (G2)
• SEO & Website Traffic

All files are available in both JSONL and Parquet formats in their respective buckets.

February 17, 2026 — Weekly Signal Batch

The following signals are scheduled for delivery tomorrow, February 17:

• SEC Filings (10-K, 10-Q, 8-K, 6-K, 20-F)
• Earnings Transcripts
• News
• Hiring Trends
• Hiring Velocity
• Work Milestones

These signals now deliver weekly as part of the cadence change announced in early February.

Where to Find Delivery Dates & Bucket URIs

For a full list of last/next delivery dates, frequencies, and the GCS bucket URI for each signal type, see the Delivery page.

If you receive files through a custom delivery method (e.g., S3 sync), your delivery timing may differ — reach out to your account contact with any questions.

Weekly Signals Delivered (9 of 10)

Notes

• All 9 delivered signals landed with 2026-02-10 folder timestamps.
• 20-F did not deliver in this upload as there we re no recent filings.

Coming Up: February 13 Delivery (13 signals)

Thirteen monthly and bi-weekly signals are due on or around Feb 13.

Next Weekly Delivery: February 17

The same 10 weekly signals (10-K, 10-Q, 8-K, 6-K, Earnings, Hiring Trends, Hiring Velocity, News, Work Milestones, and 20-F) are due again on Feb 17.

API Reference: autobound-api.readme.io/docs | Signal Catalog: autobound.ai/signal-data

Work Milestones Data Fix

We identified and resolved a data quality issue in the Work Milestones signal. Approximately 47,505 records (out of ~52 million) had incorrect event_date values due to missing start month data in the source LinkedIn profiles. When LinkedIn only provides a year without a specific month, the system was outputting 00 for the month (e.g., 2011-00-01) instead of handling it properly.

Resolution: Records without a valid start month are now filtered out, as we cannot calculate signals precisely without this data. The corrected dataset has been re-uploaded to gs://autobound-work-milestones/ with validated JSONL and Parquet files.

SEC Filing Excerpts Fix

The issue reported on February 2 regarding the excerpts field returning raw XBRL tag identifiers instead of human-readable text has been resolved. We removed the previous bucket contents and delivered new files on February 3, 2026 containing the corrected data across all affected SEC signals (10-K, 10-Q, 8-K, 20-F, 6-K).

No action is required — simply use the latest delivery folder in each bucket.

We've completed the first weekly delivery for the following signal categories, which were shifted from monthly.

Delivery Summary

Because of an issue with our Jan 29th batch, we decided to remove that delivery and regenerate those signals in this delivery.

Schema Updates

New field: data.filing_date

All SEC filing signals (10-K, 10-Q, 8-K, 6-K) now include a data.filing_date field representing the actual date the filing was submitted to the SEC. This is distinct from detected_at, which indicates when we processed the signal. Use filing_date to understand how fresh the underlying event is.

Example:

{
  "detected_at": "2026-02-03T00:00:00Z",
  "data": {
    "filing_date": "2026-01-30",
    "filing_year": 2025
  }
}

Subtype cleanup: digitalTransformation

The digitalTransformation subtype was producing false positives previously. This issue has been fixed in the current batch.

Date Ranges

Known Limitations

20-F: No new filings this period. There were no new 20-F filings between January 17 and February 3. The 20-F bucket was not updated.

Earnings Transcripts: Limited data. Only 42 signals from 3 companies in this delivery. Earnings transcript availability depends on the earnings calendar; Q4 2025 earnings season is ramping up and volume will increase in coming weeks.

We recently completed a delivery of our Website Intelligence signal that includes some information worth noting.

The latest release is available in gs://autobound-website-intelligence-v1/.

Refresher on Website Intelligence

Website Intelligence detects meaningful changes across company websites to surface buying signals and business priorities. We scan 6 million company websites on a recurring basis, analyzing:

• Text modifications — Changes to messaging, positioning, and product descriptions
• Page additions — New product pages, pricing tiers, compliance certifications, partnerships
• Page removals — Deprecated products and discontinued services

This Delivery

• 2,062,117 signals across 523,145 companies
• 38 standardized subtypes (consolidated from 1,700+ fragmented event types in previous iterations)
• High-value subtypes include: pricingChange, productLaunch, partnershipAnnouncement, executiveChange, geographicExpansion

See the full schema documentation for details.

Known Limitations

Relevance scoring needs tuning. The data.relevance field is generated by an AI model that isn't calibrated as tightly as we'd like. Currently, 71% of signals score between 0.5-0.7, with only 9% scoring above 0.9. We recommend filtering on relevance >= 0.7 for higher-precision use cases. Improved model calibration is planned for the next delivery.

Limited date range. This delivery covers website changes detected between January 2-8, 2026 only. While we've dramatically expanded company and signal coverage, the event date window is narrow. Future deliveries will analyze multiple historical timestamps to provide broader temporal coverage.

❗️

Update (February 2, 2026 — 2:00 PM PT): When this entry was initially published, there was an SEC filing delivery issue. The excerpts field was returning raw XBRL tag identifiers instead of human-readable text from the filings. This was been confirmed in 10-K signals and likely extends to 10-Q, 20-F, 8-K, and 6-K as well. This fix was reprocessed in a batch delivered on February 3, 2026.

Signal delivery has been upgraded from a monthly to a weekly cadence across nine existing signals, providing more timely data to downstream consumers. We've also launched a new Work Milestones signal covering job changes, promotions, and work anniversaries.

For full delivery schedules and next delivery dates, see the Delivery docs.

Delivery Cadence: Monthly → Weekly

The following signals have shifted from monthly to weekly delivery. The January 31 delivery was a one-time push to get data in by end of month for partners who required it. Starting February 3, 2026, these signals will begin their regular weekly cadence.

The output format, schema, and bucket structure remain unchanged - only the frequency has increased.

New Signal: Work Milestones

We launched a brand new signal: Work Milestones (gs://autobound-work-milestones/). This signal consolidates three career event subtypes under a single workMilestone signal type:

• jobChange — Contact moved to a different company
• promotion — Contact changed roles within the same company
• workAnniversary — Contact hit a tenure milestone at their current company

This signal is also on a weekly delivery cadence, starting February 3, 2026.

Access details:

• Bucket: gs://autobound-work-milestones/
• First run: gs://autobound-work-milestones/2026-01-30-16-07-00/
• Files:
  • output.jsonl (54.4 GB)
  • output.parquet (7.8 GB)
• Auth: Authenticate via gcloud auth login with access to the autobound-signal-delivery project

Why the first delivery is unusually large (~54 GB)

The first run produced a significantly larger file than future deliveries will. This is expected behavior.

The delivery window is based on detected_at (when our system identified the event), not event_date (when the career event actually occurred). Since this is the first run under a new process, every historically detected signal was included in a single backlog delivery — even if the underlying career events happened months or years ago.

Going forward, each weekly delivery will only include signals where detected_at falls within the past week, resulting in much smaller files.