aggregator(charges-match): service and SaaS-marker matching is substring-only — METAMUCIL and METAL ROOFING classify as Meta charges

Summary

matchService and isSaasLooking in pm-stack/aggregator/src/charges-match.ts match on squashed substrings with no token boundary, so ordinary non-SaaS merchants whose names merely contain a marker or service name are classified as SaaS/AI charges — and, when a hub service carries that name, are relabelled as that service in the charges feed.

isSaasLooking (charges-match.ts:158-160): SAAS_MARKERS.some((marker) => flat.includes(marker)) over the space-stripped merchant, with markers as short as 3–4 chars (XAI, META, GROK, CLINE).
matchService (charges-match.ts:164-172): a service name/alias whose squashed form is ≥4 chars matches anywhere inside the squashed merchant (flat.includes(flatNorm)), so the service Meta matches any merchant containing the letters META.

Evidence (verified this run, main `0f84fe08`; laptop probe `~/git/fleet/probe-2026-10-31-laneB-charges/probe.ts` importing the shipped file — `charges-match.ts` md5 `bec11791e5b91a32acf78529d4960087` identical in the sandbox and laptop clones)

normalizeMerchantName("METAMUCIL") → "METAMUCIL"; isSaasLooking → true; matchService(..., [{name:"Meta"}, ...]) → Meta. A full matchCharges run on a single Plaid METAMUCIL $12.50 charge produces one row labelled "Meta" — a pharmacy product reported as a Meta AI charge.
normalizeMerchantName("METAL ROOFING LLC") → "METAL ROOFING"; isSaasLooking → true; matchService → Meta. A roofing contractor is likewise classified as a Meta charge.
Control (existing behaviour that must keep working): OPENAI *CHATGPT SUB matches the OpenAI service via its ChatGPT alias, and CURSOR.COM normalizes to CURSOR (covered by test/charges-match.test.ts).

Why it matters

The charges strip is financial evidence (see #1057/#2130): a mislabelled row attributes one vendor's charge to a hub service, and a false SaaS-looking flag pushes household merchants (pharmacy, trades) into the unmatched-SaaS attention strip. The failure is silent — the row looks authoritative, carries the service's name, and rerun determinism makes the wrong answer stable.

Distinct from #2130 (same module): that issue is the dedupe key collapsing distinct charges; this one is the classification match itself. #2130's acceptance criteria do not touch matching boundaries.

Acceptance criteria

Service names/aliases and SaaS markers match only at token boundaries (a whole normalized token, or a whole multi-token sequence), never as a substring inside a longer token: METAMUCIL and METAL ROOFING match neither the Meta service nor the META SaaS marker.
Existing matches keep working and stay tested: CHATGPT→OpenAI alias, CURSOR.COM→Cursor, multi-word service names, and the token-set path in matchService.
New tests in pm-stack/aggregator/test/charges-match.test.ts cover the boundary cases above (prefix-containing merchants for at least META and one more short marker), and all existing charges-match tests pass.

Related: #1057 (feature), #2130 (dedupe defect in the same module), #1047 (epic).

🤖 Agent note · filed

Agent: new-issue-creator job (lane B, new-issue-creator-code-b)
Session / ID: cron new-issue-creator-code-b, run 2026-10-31 07:20 CDT
Found via: filed this run
Task: Filed after verifying substring-only service/SaaS-marker matching in pm-stack/aggregator/src/charges-match.ts against the shipped module (laptop probe, file md5 identical across clones): METAMUCIL and METAL ROOFING both classify as Meta charges.
Expectation: effort/xs triage — confirm on current main and route a token-boundary fix with the regression tests the issue names.
Links: probe ~/git/fleet/probe-2026-10-31-laneB-charges/ (laptop); sibling defect aggregator(charges-match): dedupe collapses distinct charges — same-source repeats within 3 days merge into one row, and empty-normalized merchants merge across merchants #2130 (dedupe, distinct)

🤖 Design-architect pass 27 (2026-10-31) — agent note: this issue is adopted as part of Phase 2 (Admission & classification) of design #3161 ("a charge is evidence"). Triage: do NOT re-file, re-scope or split it — record the phase structure instead. The design records the rules this phase lands under: admission requires payment evidence (declined/failed and merely-due amounts produce no candidate; the amount is the amount paid — explicit paid/charged labels outrank a bare total, and a bare total must be a standalone label, never a substring of "Subtotal"); classification matches names/aliases/markers at token boundaries only, against the hub's whole catalog (SaaS services plus quota pools and their aliases, from their existing extractors). Phase 2 siblings: #3120, #3121, #3048, #3119. Phase 2 is independent of Phase 1. Details on #3161 and in files/designs/code-charges-pipeline-contract.md (goal workspace).

agent-note:v1 — holder: repo-worker-code (OpenChamber scheduled lane), at 2026-11-02T14:53:00Z. event: picked-up; task: 5620; queue: repo-health-code-pilot. Fresh metadata confirms #3048 OPEN and canonical PR #4306 OPEN, non-draft, authored/assigned to Drew. No PR contents/checks/reviews/branch/source inspected.

agent-note:v1 — holder: repo-worker-code (OpenChamber scheduled lane), at 2026-11-02T14:54:00Z. event: parked; task: 5620; queue: repo-health-code-pilot. Canonical PR #4306 remains OPEN, non-draft, Drew-authored/assigned. Hard stop: this lane did not inspect or touch PR contents, checks, reviews, branch or source and did not create a competing PR. Resume after ownership release or PR disposition, then revalidate fresh.

🤖 Agent note · done

Agent: queue worker (repo-worker-code), lane K
Session / ID: task 5620
Found via: queue claim
Task: revalidate task 5620 for issue aggregator(charges-match): service and SaaS-marker matching is substring-only — METAMUCIL and METAL ROOFING classify as Meta charges #3048 — fresh GitHub state shows issue aggregator(charges-match): service and SaaS-marker matching is substring-only — METAMUCIL and METAL ROOFING classify as Meta charges #3048 is CLOSED and PR fix(aggregator): require token boundaries in charge matching #4306 is MERGED (merge commit c54fd5a), so the token-boundary fix is already done; no competing work created. Removing stale queue/status/work labels.
Expectation: done = task 5620 completed as already_done with this evidence
Links: PR fix(aggregator): require token boundaries in charge matching #4306 fix(aggregator): require token boundaries in charge matching #4306

Navigation Menu

Repository navigation

aggregator(charges-match): service and SaaS-marker matching is substring-only — METAMUCIL and METAL ROOFING classify as Meta charges #3048

Summary

Evidence (verified this run, main `0f84fe08`; laptop probe `~/git/fleet/probe-2026-10-31-laneB-charges/probe.ts` importing the shipped file — `charges-match.ts` md5 `bec11791e5b91a32acf78529d4960087` identical in the sandbox and laptop clones)

Why it matters

Acceptance criteria

51 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

aggregator(charges-match): service and SaaS-marker matching is substring-only — METAMUCIL and METAL ROOFING classify as Meta charges #3048

Description

Summary

Evidence (verified this run, main 0f84fe08; laptop probe ~/git/fleet/probe-2026-10-31-laneB-charges/probe.ts importing the shipped file — charges-match.ts md5 bec11791e5b91a32acf78529d4960087 identical in the sandbox and laptop clones)

Why it matters

Acceptance criteria

Activity

dezren39 commented on Oct 31, 2026

dezren39 commented on Oct 31, 2026

51 remaining items

dezren39 commented on Nov 2, 2026

dezren39 commented on Nov 2, 2026

dezren39 commented on Nov 2, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Evidence (verified this run, main `0f84fe08`; laptop probe `~/git/fleet/probe-2026-10-31-laneB-charges/probe.ts` importing the shipped file — `charges-match.ts` md5 `bec11791e5b91a32acf78529d4960087` identical in the sandbox and laptop clones)