Give each analysis type its own reading, and stop guessing which they are

[adamjohnwright]

Spec 011 Phase 6.

The type enum, read rather than guessed

SPECIES_COMPARISON, OVERREPRESENTATION, EXPRESSION, GSA_REGULATION, GSA_STATISTICS, GSVA. Three are summarised; the three GSA ones are declined.

Reading it closed a gap in Phase 2. is_gsa recognised ReactomeGSA by its gsaMethod field alone — so a result carrying GSA_REGULATION, GSA_STATISTICS or GSVA without that field would have been summarised confidently. That is exactly the outcome D8 exists to prevent, and it was sitting in code I wrote and reviewed two phases ago.

Both signals are checked now. Neither is known to be sufficient alone, and checking both is free.

Each type says what it supports

A summary that ignores the type is likelier to say something wrong than nothing:

• expression carries per-pathway values across columns and must describe behaviour across them — and must not guess what the columns are, since their labels are user-supplied text and are never sent
• species comparison must say findings are inferred by orthology, not observed
• over-representation must not use direction language at all, because that result cannot support up, down, increased or activated

exp joins the allow-list for the expression reading — aggregated per pathway rather than the reader's raw measurements, with the column labels that would make it identifying still excluded. Numbers without labels, which is the point.

Testing

On the served path, because the branch lives in the endpoint: each type gets its own reading and not another's. Verified by sabotage, with the sabotage asserted to have applied first.

All checks gated with set -e.

🤖 Generated with Claude Code