feat(spac): AI SPAC classifier, sponsor promote, de-SPAC linkage (#149, #150, #151) by sroussey · Pull Request #196 · workglow-dev/sec

sroussey · 2026-07-15T21:58:30Z

Implements three SPAC roadmap items, based on harness.

#149 — AI content classifier for miscoded SPACs

extractSpacClassification runs behind the existing S1Classification.classifier_source = "ai" seam to catch SPACs filed under a miscoded/absent SIC. It runs in processFormS1 only when the deterministic SIC path did not flag the filing and a cheap keyword heuristic (looksLikeBlankCheck, ≥2 blank-check signals) trips, so the AI call stays rare. A confident spac verdict (it distinguishes a true SPAC from shell/operating) flips is_spac, overwrites the classification row as "ai", and mints the known-SPAC row so lifecycle extractors attach. "Not a SPAC" auto-resolves its MODEL_EMPTY dead-letter (like the LOI detector); a model-unavailable blank-check filing dead-letters spac-classification for retry.

Config: SEC_S1_CLASSIFIER_MODEL / SEC_S1_CLASSIFIER_CONFIDENCE_FLOOR; registered in secModelIds(); added to EVAL_EXTRACTORS.

#150 — Sponsor promote economics

extractSponsorPromote + a new spac_promote_terms table capture founder (Class B) shares & percentage, private-placement warrant count/price/public coverage, and trust per-share/total. Keyed (extractor_id, accession_number) like spac_unit_terms, so S-1-registered and 424-final promote compare across extractor ids. Wired into the shared runOfferingSections (SPAC-only) so both S-1 and priced-424 pipelines populate it; added to the eval harness and DI.

#151 — De-SPAC linkage

On a completed combination, buildSpacRow derives surviving_name/current_name from the deal target, and SpacReportWriter.recordDeSpacLinkage (invoked on the item-2.01 8-K) fills surviving_name/post_merger_sic/post_merger_tickers from the SPAC CIK's post-close entity/entity_tickers metadata — each only when it diverges from the SPAC-era value, so replays stay order-safe. current_cik stays null in the common CIK-survives case. New sec spac backfill-despac refreshes completed SPACs after submissions catch up.

Testing

npx tsc --noEmit clean.
New/updated tests: classifier heuristic + extractor + processFormS1 integration, sponsor-promote extractor + storage + offering-wiring, de-SPAC rollup + writer + 8-K integration.
Affected suites green: src/storage, src/config, src/task/forms, src/sec/forms/registration-statements, src/sec/forms/miscellaneous-filings, src/eval.

No new extractor id was added — the classifier reuses S-1, promote reuses S-1/424, de-SPAC linkage rides the 8-K path. Genuinely separate newco/S-4 CIK linkage remains deferred.

🤖 Generated with Claude Code

https://claude.ai/code/session_01KqMBp35YzRXtkJMu4Xa4zm

Generated by Claude Code

#150, #151) Implements three SPAC roadmap items: #149 AI content classifier for miscoded SPACs: an AI classifier behind the existing S1Classification `classifier_source = "ai"` seam catches SPACs filed under a miscoded/absent SIC. Gated on a cheap blank-check keyword heuristic so the AI call stays rare; a confident SPAC verdict flips is_spac, overwrites the classification row, and mints the known-SPAC row. Env: SEC_S1_CLASSIFIER_MODEL / SEC_S1_CLASSIFIER_CONFIDENCE_FLOOR. Added to the eval harness. #150 Sponsor promote economics: extracts founder (Class B) shares + percentage, private-placement warrant count/price/public coverage, and trust per-share/total from the prospectus into a new spac_promote_terms table, via the shared offering-sections runner (S-1 and priced-424). Added to the eval harness. #151 De-SPAC linkage: on a completed combination the rollup derives surviving_name/current_name from the deal target, and recordDeSpacLinkage populates surviving_name/post_merger_sic/post_merger_tickers from the SPAC CIK's post-close entity metadata on the item-2.01 8-K. New `sec spac backfill-despac` refreshes completed SPACs after submissions catch up. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KqMBp35YzRXtkJMu4Xa4zm

Copilot

Pull request overview

Implements three SPAC roadmap items across the S-1 and 8-K pipelines: (1) an AI content classifier to catch SIC-miscoded SPACs, (2) extraction + storage of sponsor-promote economics from SPAC prospectuses, and (3) de-SPAC linkage enrichment from post-close entity metadata (plus a CLI backfill).

Changes:

Add an AI-gated SPAC classifier (keyword heuristic → model classification → upgrade is_spac / mint SPAC row) and evaluation fixtures.
Add sponsor promote extraction + persistence (spac_promote_terms) and wire it into SPAC offering section processing.
Add de-SPAC linkage derivation/enrichment and wiring on 8-K Item 2.01 close, plus a CLI refresh command.

Reviewed changes

Copilot reviewed 28 out of 28 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/storage/spac/spacRollup.ts	Derives surviving/current identity fields from completed deals and post-merger patches.
src/storage/spac/spacRollup.test.ts	Tests surviving/current name behavior for completed vs pending deals and explicit overrides.
src/storage/spac/SpacReportWriter.ts	Adds `recordDeSpacLinkage` to enrich post-merger identity from entity metadata.
src/storage/spac/SpacReportWriter.deSpac.test.ts	Tests de-SPAC linkage enrichment and no-op conditions.
src/storage/offering/SpacPromoteTermsSchema.ts	New storage schema/token for sponsor promote terms table.
src/storage/offering/SpacPromoteTermsRepo.ts	Repo wrapper for saving/querying sponsor promote terms.
src/storage/offering/SpacPromoteTermsRepo.test.ts	Repo round-trip and query ordering tests.
src/sec/forms/registration-statements/s1/sponsorPromoteSchema.ts	DataPort schema for sponsor promote model output.
src/sec/forms/registration-statements/s1/spacContentHeuristic.ts	Cheap keyword heuristic to gate classifier calls.
src/sec/forms/registration-statements/s1/spacClassifierSchema.ts	DataPort schema/types for SPAC content classification.
src/sec/forms/registration-statements/s1/spacClassifierModel.ts	Model resolution + confidence floor config for classifier.
src/sec/forms/registration-statements/s1/spacClassifier.test.ts	Tests heuristic + classifier extractor behavior.
src/sec/forms/registration-statements/s1/sectionExtractors.ts	Adds `extractSponsorPromote` and `extractSpacClassification`.
src/sec/forms/registration-statements/s1/offeringSections.ts	Wires sponsor-promote extraction + persistence into offering runner (SPAC-only).
src/sec/forms/registration-statements/s1/extractSponsorPromote.test.ts	Tests sponsor promote extractor parsing/null behavior.
src/sec/forms/registration-statements/Form_S_1.storage.ts	Integrates AI classifier upgrade path into `processFormS1`.
src/sec/forms/registration-statements/Form_S_1.storage.offering.test.ts	Integration tests for sponsor-promote persistence and SPAC/non-SPAC gating.
src/sec/forms/registration-statements/Form_S_1.storage.classifier.test.ts	Integration tests for classifier upgrade/non-upgrade/no-call scenarios.
src/sec/forms/miscellaneous-filings/spac8kMilestones.test.ts	Tests de-SPAC linkage wiring on 8-K close (Item 2.01).
src/sec/forms/miscellaneous-filings/Form_8_K.storage.ts	Calls `recordDeSpacLinkage` after completion milestone write.
src/eval/fixtures.ts	Adds eval harness extractors + fixtures for sponsor-promote and spac-classification.
src/config/TestingDI.ts	Registers in-memory storage for `spac_promote_terms` in tests.
src/config/setupAllDatabases.ts	Ensures `spac_promote_terms` storage is set up.
src/config/resetAllDatabases.ts	Ensures `spac_promote_terms` storage is reset.
src/config/registerModels.ts	Includes `SEC_S1_CLASSIFIER_MODEL` in registered model IDs.
src/config/DefaultDI.ts	Registers persistent storage for `spac_promote_terms`.
src/commands/spac.ts	Adds `sec spac backfill-despac` CLI to refresh post-merger identity on completed SPACs.
CLAUDE.md	Documents classifier gating/config, sponsor promote table, and de-SPAC linkage/backfill command.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

sroussey · 2026-07-15T22:13:00Z

+      // Renamed surviving entity: the current entity name once it diverges from
+      // the blank-check shell name. Overrides the deal-target-derived fallback.
+      if (entity?.name != null && entity.name !== existing.spac_name) {
+        patch.surviving_name = entity.name;
+      }


Good catch — all three comments (this, the post_merger_sic, and the post_merger_tickers ones) were the same real gap: the code overwrote on every run whenever entity data diverged, contradicting the "cannot regress a populated linkage" contract. Fixed in 3c6b481 to make these genuine write-once close-time snapshots:

surviving_name: upgrades the deal-target fallback (or fills an empty slot) exactly once; never overwrites an already entity-sourced value on a later replay/rebrand.

post_merger_sic: only set when the slot is still null.

post_merger_tickers: only set when still null, deduped + sorted for determinism, and only when the symbol set diverges from spac_tickers (no more SPAC-era-symbol leak).

Added tests for the later-rebrand no-mutation case and the SPAC-era-ticker non-leak case.

Generated by Claude Code

+      // Operating-company SIC once it diverges from the SPAC-era ~6770.
+      if (entity?.sic != null && entity.sic !== existing.spac_sic) {
+        patch.post_merger_sic = entity.sic;
+      }


+      // New listing symbol(s) — JSON string[] mirroring the spac_tickers shape.
+      const symbols = tickers.map((t) => t.ticker).filter((s): s is string => !!s);
+      if (symbols.length > 0) patch.post_merger_tickers = JSON.stringify(symbols);


Address review: recordDeSpacLinkage overwrote surviving_name / post_merger_sic / post_merger_tickers on every run whenever entity data diverged from the SPAC-era value, contradicting the "cannot regress a populated linkage" contract, and wrote post_merger_tickers without checking divergence from spac_tickers or sorting (non-deterministic order). - surviving_name: upgrade the deal-target fallback (or fill an empty slot) once; never overwrite an already entity-sourced snapshot on a later replay/rebrand. - post_merger_sic: write-once (only when the slot is still null). - post_merger_tickers: write-once, deduped + sorted for determinism, and only when the symbol set diverges from the SPAC-era tickers. Adds tests for the later-rebrand no-mutation case and the SPAC-era-ticker non-leak case. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KqMBp35YzRXtkJMu4Xa4zm

sroussey self-assigned this Jul 15, 2026

sroussey requested a review from Copilot July 15, 2026 22:04

Copilot started reviewing on behalf of sroussey July 15, 2026 22:04 View session

This was linked to issues Jul 15, 2026

[SPAC] AI content classifier for miscoded SPACs #149

Open

[SPAC] Sponsor promote economics extraction #150

Open

[SPAC] De-SPAC linkage via 8-K / merger extraction #151

Open

Copilot AI reviewed Jul 15, 2026

View reviewed changes

sroussey merged commit 45b2025 into harness Jul 15, 2026
1 check passed

sroussey deleted the claude/hopeful-albattani-veyi2k branch July 15, 2026 22:25

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat(spac): AI SPAC classifier, sponsor promote, de-SPAC linkage (#149, #150, #151)#196

feat(spac): AI SPAC classifier, sponsor promote, de-SPAC linkage (#149, #150, #151)#196
sroussey merged 2 commits into
harnessfrom
claude/hopeful-albattani-veyi2k

sroussey commented Jul 15, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

sroussey Jul 15, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

sroussey commented Jul 15, 2026

#149 — AI content classifier for miscoded SPACs

#150 — Sponsor promote economics

#151 — De-SPAC linkage

Testing

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

sroussey Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants